diff --git "a/checkpoint-2400/trainer_state.json" "b/checkpoint-2400/trainer_state.json" new file mode 100644--- /dev/null +++ "b/checkpoint-2400/trainer_state.json" @@ -0,0 +1,14908 @@ +{ + "best_metric": 0.6019096970558167, + "best_model_checkpoint": "./output_v2/34bCodellama_CodeLlama-34b-Python-hf_codellama_blob_1/checkpoint-2400", + "epoch": 1.684801684801685, + "global_step": 2400, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 1.5667, + "step": 1 + }, + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 0.9201, + "step": 2 + }, + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 0.6019, + "step": 3 + }, + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 0.861, + "step": 4 + }, + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 1.064, + "step": 5 + }, + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 1.0007, + "step": 6 + }, + { + "epoch": 0.0, + "learning_rate": 0.0004, + "loss": 1.0533, + "step": 7 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.8498, + "step": 8 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.9189, + "step": 9 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.9369, + "step": 10 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 1.1375, + "step": 11 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.8369, + "step": 12 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.689, + "step": 13 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.8885, + "step": 14 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.9666, + "step": 15 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.744, + "step": 16 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.8998, + "step": 17 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.8371, + "step": 18 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.7615, + "step": 19 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.6963, + "step": 20 + }, + { + "epoch": 0.01, + "learning_rate": 0.0004, + "loss": 0.5974, + "step": 21 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 1.4922, + "step": 22 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 1.1272, + "step": 23 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.4373, + "step": 24 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 1.0598, + "step": 25 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 1.191, + "step": 26 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.6499, + "step": 27 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.7526, + "step": 28 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.7252, + "step": 29 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.8555, + "step": 30 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.479, + "step": 31 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.5273, + "step": 32 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.3907, + "step": 33 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.5284, + "step": 34 + }, + { + "epoch": 0.02, + "learning_rate": 0.0004, + "loss": 0.8696, + "step": 35 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.9264, + "step": 36 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.6812, + "step": 37 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.4482, + "step": 38 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 1.9031, + "step": 39 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 1.3295, + "step": 40 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.7498, + "step": 41 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 1.4452, + "step": 42 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 1.224, + "step": 43 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.3272, + "step": 44 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.7366, + "step": 45 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 1.0331, + "step": 46 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.8471, + "step": 47 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.8171, + "step": 48 + }, + { + "epoch": 0.03, + "learning_rate": 0.0004, + "loss": 0.418, + "step": 49 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.6942, + "step": 50 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.1712, + "step": 51 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.5041, + "step": 52 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.4619, + "step": 53 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.5011, + "step": 54 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.5, + "step": 55 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.917, + "step": 56 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.8142, + "step": 57 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.8708, + "step": 58 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.7987, + "step": 59 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.9699, + "step": 60 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 1.0015, + "step": 61 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.8052, + "step": 62 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 0.8645, + "step": 63 + }, + { + "epoch": 0.04, + "learning_rate": 0.0004, + "loss": 1.1444, + "step": 64 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 1.0087, + "step": 65 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.5679, + "step": 66 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.5507, + "step": 67 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 1.1764, + "step": 68 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.6332, + "step": 69 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.8759, + "step": 70 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.8385, + "step": 71 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.7869, + "step": 72 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 1.4457, + "step": 73 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.9331, + "step": 74 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.8943, + "step": 75 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.5783, + "step": 76 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.6433, + "step": 77 + }, + { + "epoch": 0.05, + "learning_rate": 0.0004, + "loss": 0.7517, + "step": 78 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.5061, + "step": 79 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.7332, + "step": 80 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 1.6546, + "step": 81 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.4723, + "step": 82 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.2814, + "step": 83 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.4068, + "step": 84 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 1.1729, + "step": 85 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.698, + "step": 86 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.7746, + "step": 87 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.5065, + "step": 88 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.4654, + "step": 89 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.5724, + "step": 90 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 1.1535, + "step": 91 + }, + { + "epoch": 0.06, + "learning_rate": 0.0004, + "loss": 0.3854, + "step": 92 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 1.6848, + "step": 93 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.6667, + "step": 94 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.4526, + "step": 95 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.7112, + "step": 96 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.6816, + "step": 97 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.7496, + "step": 98 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 1.189, + "step": 99 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 1.4239, + "step": 100 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.4378, + "step": 101 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 2.3054, + "step": 102 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.1593, + "step": 103 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.2481, + "step": 104 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.6469, + "step": 105 + }, + { + "epoch": 0.07, + "learning_rate": 0.0004, + "loss": 0.7417, + "step": 106 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.7767, + "step": 107 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.831, + "step": 108 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.7954, + "step": 109 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.6376, + "step": 110 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.6208, + "step": 111 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.9038, + "step": 112 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.9994, + "step": 113 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 1.0233, + "step": 114 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.7573, + "step": 115 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.5333, + "step": 116 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.4886, + "step": 117 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.75, + "step": 118 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.6377, + "step": 119 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.6518, + "step": 120 + }, + { + "epoch": 0.08, + "learning_rate": 0.0004, + "loss": 0.6409, + "step": 121 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.5996, + "step": 122 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.7964, + "step": 123 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.5898, + "step": 124 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.8753, + "step": 125 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.6304, + "step": 126 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.5428, + "step": 127 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.8571, + "step": 128 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.8339, + "step": 129 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.6087, + "step": 130 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.3547, + "step": 131 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.3494, + "step": 132 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.5394, + "step": 133 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.4315, + "step": 134 + }, + { + "epoch": 0.09, + "learning_rate": 0.0004, + "loss": 0.3575, + "step": 135 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.4983, + "step": 136 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.4292, + "step": 137 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.6315, + "step": 138 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.9827, + "step": 139 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 1.4456, + "step": 140 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.5151, + "step": 141 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.4651, + "step": 142 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.8682, + "step": 143 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 1.1543, + "step": 144 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.3378, + "step": 145 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.5381, + "step": 146 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.5189, + "step": 147 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.3988, + "step": 148 + }, + { + "epoch": 0.1, + "learning_rate": 0.0004, + "loss": 0.9694, + "step": 149 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 1.2404, + "step": 150 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.3767, + "step": 151 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.5564, + "step": 152 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.2665, + "step": 153 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.172, + "step": 154 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.6584, + "step": 155 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.7588, + "step": 156 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.9431, + "step": 157 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.9456, + "step": 158 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.7471, + "step": 159 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.8477, + "step": 160 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.7679, + "step": 161 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 0.8937, + "step": 162 + }, + { + "epoch": 0.11, + "learning_rate": 0.0004, + "loss": 1.2858, + "step": 163 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.7806, + "step": 164 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.8034, + "step": 165 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.6372, + "step": 166 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.8217, + "step": 167 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.3706, + "step": 168 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 1.0851, + "step": 169 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.6719, + "step": 170 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 1.0366, + "step": 171 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.5603, + "step": 172 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.5651, + "step": 173 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.3804, + "step": 174 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.5946, + "step": 175 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.7649, + "step": 176 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.5035, + "step": 177 + }, + { + "epoch": 0.12, + "learning_rate": 0.0004, + "loss": 0.8066, + "step": 178 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.8046, + "step": 179 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.4233, + "step": 180 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.7945, + "step": 181 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.5722, + "step": 182 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.6088, + "step": 183 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.4229, + "step": 184 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.8723, + "step": 185 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.9287, + "step": 186 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.8082, + "step": 187 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.4144, + "step": 188 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.372, + "step": 189 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.5725, + "step": 190 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.3837, + "step": 191 + }, + { + "epoch": 0.13, + "learning_rate": 0.0004, + "loss": 0.5155, + "step": 192 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.5472, + "step": 193 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.6488, + "step": 194 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 1.4385, + "step": 195 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.3418, + "step": 196 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.8811, + "step": 197 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.8176, + "step": 198 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.5379, + "step": 199 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 1.0277, + "step": 200 + }, + { + "epoch": 0.14, + "eval_loss": 0.662463366985321, + "eval_runtime": 262.5709, + "eval_samples_per_second": 1.904, + "eval_steps_per_second": 0.952, + "step": 200 + }, + { + "epoch": 0.14, + "mmlu_eval_accuracy": 0.41276444254447825, + "mmlu_eval_accuracy_abstract_algebra": 0.09090909090909091, + "mmlu_eval_accuracy_anatomy": 0.42857142857142855, + "mmlu_eval_accuracy_astronomy": 0.5, + "mmlu_eval_accuracy_business_ethics": 0.2727272727272727, + "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655, + "mmlu_eval_accuracy_college_biology": 0.3125, + "mmlu_eval_accuracy_college_chemistry": 0.375, + "mmlu_eval_accuracy_college_computer_science": 0.5454545454545454, + "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727, + "mmlu_eval_accuracy_college_medicine": 0.6363636363636364, + "mmlu_eval_accuracy_college_physics": 0.2727272727272727, + "mmlu_eval_accuracy_computer_security": 0.36363636363636365, + "mmlu_eval_accuracy_conceptual_physics": 0.3076923076923077, + "mmlu_eval_accuracy_econometrics": 0.4166666666666667, + "mmlu_eval_accuracy_electrical_engineering": 0.5, + "mmlu_eval_accuracy_elementary_mathematics": 0.4146341463414634, + "mmlu_eval_accuracy_formal_logic": 0.5714285714285714, + "mmlu_eval_accuracy_global_facts": 0.2, + "mmlu_eval_accuracy_high_school_biology": 0.1875, + "mmlu_eval_accuracy_high_school_chemistry": 0.4090909090909091, + "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666, + "mmlu_eval_accuracy_high_school_european_history": 0.4444444444444444, + "mmlu_eval_accuracy_high_school_geography": 0.5909090909090909, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.32558139534883723, + "mmlu_eval_accuracy_high_school_mathematics": 0.4482758620689655, + "mmlu_eval_accuracy_high_school_microeconomics": 0.5714285714285714, + "mmlu_loss": 0.7435930466353893, + "step": 200 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 1.6342, + "step": 201 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.1179, + "step": 202 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.1083, + "step": 203 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.6355, + "step": 204 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.5694, + "step": 205 + }, + { + "epoch": 0.14, + "learning_rate": 0.0004, + "loss": 0.7707, + "step": 206 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.3314, + "step": 207 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 1.0069, + "step": 208 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.9113, + "step": 209 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.8692, + "step": 210 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.7362, + "step": 211 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.926, + "step": 212 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.7405, + "step": 213 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.9191, + "step": 214 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.9301, + "step": 215 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.8581, + "step": 216 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.4548, + "step": 217 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.6516, + "step": 218 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 0.7194, + "step": 219 + }, + { + "epoch": 0.15, + "learning_rate": 0.0004, + "loss": 1.0441, + "step": 220 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.9499, + "step": 221 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.5447, + "step": 222 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.8795, + "step": 223 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.5347, + "step": 224 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.5827, + "step": 225 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.311, + "step": 226 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.1071, + "step": 227 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.6336, + "step": 228 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.6212, + "step": 229 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.4345, + "step": 230 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.5373, + "step": 231 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.9612, + "step": 232 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 1.0251, + "step": 233 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.4267, + "step": 234 + }, + { + "epoch": 0.16, + "learning_rate": 0.0004, + "loss": 0.5767, + "step": 235 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.5123, + "step": 236 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.354, + "step": 237 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 1.1279, + "step": 238 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.6464, + "step": 239 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.7227, + "step": 240 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 1.3533, + "step": 241 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.3986, + "step": 242 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 1.0402, + "step": 243 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.4558, + "step": 244 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.5525, + "step": 245 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 1.5916, + "step": 246 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.9311, + "step": 247 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.3011, + "step": 248 + }, + { + "epoch": 0.17, + "learning_rate": 0.0004, + "loss": 0.5813, + "step": 249 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 1.134, + "step": 250 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.3062, + "step": 251 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.2848, + "step": 252 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.1995, + "step": 253 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.8426, + "step": 254 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.8062, + "step": 255 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 1.0046, + "step": 256 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.6235, + "step": 257 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 1.0021, + "step": 258 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.8788, + "step": 259 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.883, + "step": 260 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.8167, + "step": 261 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.8068, + "step": 262 + }, + { + "epoch": 0.18, + "learning_rate": 0.0004, + "loss": 0.4259, + "step": 263 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.8339, + "step": 264 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.8149, + "step": 265 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.6023, + "step": 266 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.8297, + "step": 267 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 1.0205, + "step": 268 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.5643, + "step": 269 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.719, + "step": 270 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.812, + "step": 271 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.5513, + "step": 272 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.7669, + "step": 273 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.6044, + "step": 274 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.7956, + "step": 275 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.569, + "step": 276 + }, + { + "epoch": 0.19, + "learning_rate": 0.0004, + "loss": 0.5728, + "step": 277 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.534, + "step": 278 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.4182, + "step": 279 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.9406, + "step": 280 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.8158, + "step": 281 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.674, + "step": 282 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.4463, + "step": 283 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.5038, + "step": 284 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.5103, + "step": 285 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.4692, + "step": 286 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.4955, + "step": 287 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 1.068, + "step": 288 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.5165, + "step": 289 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.6199, + "step": 290 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 1.1037, + "step": 291 + }, + { + "epoch": 0.2, + "learning_rate": 0.0004, + "loss": 0.155, + "step": 292 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 1.4112, + "step": 293 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.5906, + "step": 294 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.5932, + "step": 295 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.8826, + "step": 296 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.5626, + "step": 297 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.3403, + "step": 298 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 1.3782, + "step": 299 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 1.2619, + "step": 300 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.4879, + "step": 301 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.426, + "step": 302 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.2672, + "step": 303 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.4627, + "step": 304 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.8008, + "step": 305 + }, + { + "epoch": 0.21, + "learning_rate": 0.0004, + "loss": 0.8396, + "step": 306 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.826, + "step": 307 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.9369, + "step": 308 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.8878, + "step": 309 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.8131, + "step": 310 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.8733, + "step": 311 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.6789, + "step": 312 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.9527, + "step": 313 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.4861, + "step": 314 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.6534, + "step": 315 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.7253, + "step": 316 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.626, + "step": 317 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.7153, + "step": 318 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.5901, + "step": 319 + }, + { + "epoch": 0.22, + "learning_rate": 0.0004, + "loss": 0.5152, + "step": 320 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.6235, + "step": 321 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.7336, + "step": 322 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.8437, + "step": 323 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.8042, + "step": 324 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.524, + "step": 325 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.882, + "step": 326 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.5542, + "step": 327 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.5666, + "step": 328 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.7223, + "step": 329 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.4367, + "step": 330 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.6358, + "step": 331 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.6793, + "step": 332 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.3485, + "step": 333 + }, + { + "epoch": 0.23, + "learning_rate": 0.0004, + "loss": 0.6813, + "step": 334 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.8295, + "step": 335 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.4777, + "step": 336 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.8557, + "step": 337 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.7415, + "step": 338 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.5843, + "step": 339 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.4585, + "step": 340 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.2676, + "step": 341 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.2997, + "step": 342 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.5732, + "step": 343 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.3759, + "step": 344 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.4395, + "step": 345 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.4864, + "step": 346 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.3492, + "step": 347 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.5787, + "step": 348 + }, + { + "epoch": 0.24, + "learning_rate": 0.0004, + "loss": 0.8384, + "step": 349 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 1.2399, + "step": 350 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.248, + "step": 351 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.2635, + "step": 352 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.3762, + "step": 353 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.5572, + "step": 354 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 1.0953, + "step": 355 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.8513, + "step": 356 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.8504, + "step": 357 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.9141, + "step": 358 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.7751, + "step": 359 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.8659, + "step": 360 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.9634, + "step": 361 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.9354, + "step": 362 + }, + { + "epoch": 0.25, + "learning_rate": 0.0004, + "loss": 0.6382, + "step": 363 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.9403, + "step": 364 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.8036, + "step": 365 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.5506, + "step": 366 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.4639, + "step": 367 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.8315, + "step": 368 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.6634, + "step": 369 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.6388, + "step": 370 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.7629, + "step": 371 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.6548, + "step": 372 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.5165, + "step": 373 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.2858, + "step": 374 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.3376, + "step": 375 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.5373, + "step": 376 + }, + { + "epoch": 0.26, + "learning_rate": 0.0004, + "loss": 0.6229, + "step": 377 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.5867, + "step": 378 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.5533, + "step": 379 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.4804, + "step": 380 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.313, + "step": 381 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.7077, + "step": 382 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.5241, + "step": 383 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.3367, + "step": 384 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.8614, + "step": 385 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.4548, + "step": 386 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.6221, + "step": 387 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.8041, + "step": 388 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.2546, + "step": 389 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.5016, + "step": 390 + }, + { + "epoch": 0.27, + "learning_rate": 0.0004, + "loss": 0.4664, + "step": 391 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.2124, + "step": 392 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.4019, + "step": 393 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.8749, + "step": 394 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.2319, + "step": 395 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.9602, + "step": 396 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.9182, + "step": 397 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.2092, + "step": 398 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.2311, + "step": 399 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 1.6764, + "step": 400 + }, + { + "epoch": 0.28, + "eval_loss": 0.6333504915237427, + "eval_runtime": 261.9428, + "eval_samples_per_second": 1.909, + "eval_steps_per_second": 0.954, + "step": 400 + }, + { + "epoch": 0.28, + "mmlu_eval_accuracy": 0.4098082068017773, + "mmlu_eval_accuracy_abstract_algebra": 0.09090909090909091, + "mmlu_eval_accuracy_anatomy": 0.2857142857142857, + "mmlu_eval_accuracy_astronomy": 0.375, + "mmlu_eval_accuracy_business_ethics": 0.36363636363636365, + "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655, + "mmlu_eval_accuracy_college_biology": 0.4375, + "mmlu_eval_accuracy_college_chemistry": 0.25, + "mmlu_eval_accuracy_college_computer_science": 0.5454545454545454, + "mmlu_eval_accuracy_college_mathematics": 0.45454545454545453, + "mmlu_eval_accuracy_college_medicine": 0.5, + "mmlu_eval_accuracy_college_physics": 0.36363636363636365, + "mmlu_eval_accuracy_computer_security": 0.5454545454545454, + "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615, + "mmlu_eval_accuracy_econometrics": 0.16666666666666666, + "mmlu_eval_accuracy_electrical_engineering": 0.5625, + "mmlu_eval_accuracy_elementary_mathematics": 0.4146341463414634, + "mmlu_eval_accuracy_formal_logic": 0.35714285714285715, + "mmlu_eval_accuracy_global_facts": 0.2, + "mmlu_eval_accuracy_high_school_biology": 0.3125, + "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365, + "mmlu_eval_accuracy_high_school_computer_science": 0.7777777777777778, + "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112, + "mmlu_eval_accuracy_high_school_geography": 0.7272727272727273, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.3953488372093023, + "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276, + "mmlu_eval_accuracy_high_school_microeconomics": 0.35714285714285715, + "mmlu_loss": 0.6790264239534736, + "step": 400 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.3098, + "step": 401 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.0669, + "step": 402 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.3434, + "step": 403 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.6937, + "step": 404 + }, + { + "epoch": 0.28, + "learning_rate": 0.0004, + "loss": 0.6958, + "step": 405 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.7289, + "step": 406 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.804, + "step": 407 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.7795, + "step": 408 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.4095, + "step": 409 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.5514, + "step": 410 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.9134, + "step": 411 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.6661, + "step": 412 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.8238, + "step": 413 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.6895, + "step": 414 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.6924, + "step": 415 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.5246, + "step": 416 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.6762, + "step": 417 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.4997, + "step": 418 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.7983, + "step": 419 + }, + { + "epoch": 0.29, + "learning_rate": 0.0004, + "loss": 0.5747, + "step": 420 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 1.1613, + "step": 421 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.6814, + "step": 422 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.97, + "step": 423 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.4104, + "step": 424 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.5966, + "step": 425 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.7571, + "step": 426 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 1.2513, + "step": 427 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.594, + "step": 428 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.9536, + "step": 429 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.8351, + "step": 430 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.3044, + "step": 431 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.893, + "step": 432 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.8969, + "step": 433 + }, + { + "epoch": 0.3, + "learning_rate": 0.0004, + "loss": 0.7612, + "step": 434 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.2974, + "step": 435 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.5277, + "step": 436 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.3194, + "step": 437 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 1.1187, + "step": 438 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.5409, + "step": 439 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.4445, + "step": 440 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.3914, + "step": 441 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 2.116, + "step": 442 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.4477, + "step": 443 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 1.4711, + "step": 444 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.6186, + "step": 445 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.3651, + "step": 446 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.2939, + "step": 447 + }, + { + "epoch": 0.31, + "learning_rate": 0.0004, + "loss": 0.2754, + "step": 448 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.7015, + "step": 449 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 1.2934, + "step": 450 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.3308, + "step": 451 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.3002, + "step": 452 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.3875, + "step": 453 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.1527, + "step": 454 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.791, + "step": 455 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.6592, + "step": 456 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.8466, + "step": 457 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.7768, + "step": 458 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.7784, + "step": 459 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.8584, + "step": 460 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.8313, + "step": 461 + }, + { + "epoch": 0.32, + "learning_rate": 0.0004, + "loss": 0.57, + "step": 462 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.7299, + "step": 463 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.6277, + "step": 464 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.6326, + "step": 465 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.7089, + "step": 466 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.6943, + "step": 467 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.5197, + "step": 468 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.79, + "step": 469 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.5237, + "step": 470 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.5653, + "step": 471 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.5509, + "step": 472 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.9335, + "step": 473 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.816, + "step": 474 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.7155, + "step": 475 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.5283, + "step": 476 + }, + { + "epoch": 0.33, + "learning_rate": 0.0004, + "loss": 0.1808, + "step": 477 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.5369, + "step": 478 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.6086, + "step": 479 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.6205, + "step": 480 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 1.3518, + "step": 481 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.6041, + "step": 482 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.4709, + "step": 483 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.8976, + "step": 484 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.2544, + "step": 485 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 1.0326, + "step": 486 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.2799, + "step": 487 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 1.4241, + "step": 488 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.3467, + "step": 489 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.8382, + "step": 490 + }, + { + "epoch": 0.34, + "learning_rate": 0.0004, + "loss": 0.2025, + "step": 491 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.4382, + "step": 492 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.5643, + "step": 493 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.6231, + "step": 494 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 1.0425, + "step": 495 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.8847, + "step": 496 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.6748, + "step": 497 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.3893, + "step": 498 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.3734, + "step": 499 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 1.6683, + "step": 500 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.2681, + "step": 501 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.2262, + "step": 502 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.5853, + "step": 503 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.928, + "step": 504 + }, + { + "epoch": 0.35, + "learning_rate": 0.0004, + "loss": 0.8761, + "step": 505 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.8541, + "step": 506 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.4984, + "step": 507 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.9247, + "step": 508 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.8992, + "step": 509 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.8758, + "step": 510 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.6687, + "step": 511 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.7952, + "step": 512 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.8696, + "step": 513 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.761, + "step": 514 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.8351, + "step": 515 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.8167, + "step": 516 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.6166, + "step": 517 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.7103, + "step": 518 + }, + { + "epoch": 0.36, + "learning_rate": 0.0004, + "loss": 0.9162, + "step": 519 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.567, + "step": 520 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.9573, + "step": 521 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.4096, + "step": 522 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.6648, + "step": 523 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.6851, + "step": 524 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.4083, + "step": 525 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.8178, + "step": 526 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 1.1736, + "step": 527 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.5406, + "step": 528 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.552, + "step": 529 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.6497, + "step": 530 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.4467, + "step": 531 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.9224, + "step": 532 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.9014, + "step": 533 + }, + { + "epoch": 0.37, + "learning_rate": 0.0004, + "loss": 0.8595, + "step": 534 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.343, + "step": 535 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.6365, + "step": 536 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.8377, + "step": 537 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.5699, + "step": 538 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 1.0938, + "step": 539 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.2896, + "step": 540 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 1.2605, + "step": 541 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.5222, + "step": 542 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.327, + "step": 543 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.7968, + "step": 544 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.646, + "step": 545 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.8071, + "step": 546 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.3894, + "step": 547 + }, + { + "epoch": 0.38, + "learning_rate": 0.0004, + "loss": 0.4803, + "step": 548 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.3228, + "step": 549 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 1.3897, + "step": 550 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.1901, + "step": 551 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.321, + "step": 552 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.8044, + "step": 553 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.8365, + "step": 554 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.7818, + "step": 555 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.4648, + "step": 556 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.5499, + "step": 557 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.7847, + "step": 558 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.7892, + "step": 559 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.8517, + "step": 560 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.7914, + "step": 561 + }, + { + "epoch": 0.39, + "learning_rate": 0.0004, + "loss": 0.893, + "step": 562 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.8116, + "step": 563 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.8543, + "step": 564 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.8607, + "step": 565 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.6274, + "step": 566 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.7551, + "step": 567 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.7328, + "step": 568 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.7561, + "step": 569 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.7465, + "step": 570 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.697, + "step": 571 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.8938, + "step": 572 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.8216, + "step": 573 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.5859, + "step": 574 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.4194, + "step": 575 + }, + { + "epoch": 0.4, + "learning_rate": 0.0004, + "loss": 0.603, + "step": 576 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.6666, + "step": 577 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.4037, + "step": 578 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.6723, + "step": 579 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.5837, + "step": 580 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.534, + "step": 581 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.44, + "step": 582 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.8701, + "step": 583 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.6575, + "step": 584 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.3981, + "step": 585 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.4929, + "step": 586 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.6346, + "step": 587 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.88, + "step": 588 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.3083, + "step": 589 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.741, + "step": 590 + }, + { + "epoch": 0.41, + "learning_rate": 0.0004, + "loss": 0.3114, + "step": 591 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.1633, + "step": 592 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 1.5786, + "step": 593 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.9141, + "step": 594 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.4553, + "step": 595 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.4153, + "step": 596 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.8077, + "step": 597 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.5165, + "step": 598 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 1.2809, + "step": 599 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.1944, + "step": 600 + }, + { + "epoch": 0.42, + "eval_loss": 0.6178187727928162, + "eval_runtime": 262.0082, + "eval_samples_per_second": 1.908, + "eval_steps_per_second": 0.954, + "step": 600 + }, + { + "epoch": 0.42, + "mmlu_eval_accuracy": 0.3933105593085208, + "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.5, + "mmlu_eval_accuracy_business_ethics": 0.36363636363636365, + "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345, + "mmlu_eval_accuracy_college_biology": 0.4375, + "mmlu_eval_accuracy_college_chemistry": 0.25, + "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453, + "mmlu_eval_accuracy_college_mathematics": 0.09090909090909091, + "mmlu_eval_accuracy_college_medicine": 0.5, + "mmlu_eval_accuracy_college_physics": 0.6363636363636364, + "mmlu_eval_accuracy_computer_security": 0.36363636363636365, + "mmlu_eval_accuracy_conceptual_physics": 0.3076923076923077, + "mmlu_eval_accuracy_econometrics": 0.3333333333333333, + "mmlu_eval_accuracy_electrical_engineering": 0.5, + "mmlu_eval_accuracy_elementary_mathematics": 0.21951219512195122, + "mmlu_eval_accuracy_formal_logic": 0.35714285714285715, + "mmlu_eval_accuracy_global_facts": 0.3, + "mmlu_eval_accuracy_high_school_biology": 0.3125, + "mmlu_eval_accuracy_high_school_chemistry": 0.18181818181818182, + "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666, + "mmlu_eval_accuracy_high_school_european_history": 0.5555555555555556, + "mmlu_eval_accuracy_high_school_geography": 0.7272727272727273, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.42857142857142855, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.37209302325581395, + "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276, + "mmlu_eval_accuracy_high_school_microeconomics": 0.42857142857142855, + "mmlu_loss": 0.770003160238266, + "step": 600 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.2039, + "step": 601 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.3086, + "step": 602 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.4326, + "step": 603 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.8444, + "step": 604 + }, + { + "epoch": 0.42, + "learning_rate": 0.0004, + "loss": 0.6902, + "step": 605 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.8338, + "step": 606 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.9103, + "step": 607 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.9047, + "step": 608 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.82, + "step": 609 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.4912, + "step": 610 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.414, + "step": 611 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.735, + "step": 612 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.7284, + "step": 613 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.5756, + "step": 614 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.7305, + "step": 615 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.5331, + "step": 616 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.6001, + "step": 617 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.3613, + "step": 618 + }, + { + "epoch": 0.43, + "learning_rate": 0.0004, + "loss": 0.6111, + "step": 619 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.3517, + "step": 620 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.7228, + "step": 621 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.512, + "step": 622 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.4729, + "step": 623 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.6152, + "step": 624 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.6118, + "step": 625 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.9996, + "step": 626 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.2561, + "step": 627 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.375, + "step": 628 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.5303, + "step": 629 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.3277, + "step": 630 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.286, + "step": 631 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.4288, + "step": 632 + }, + { + "epoch": 0.44, + "learning_rate": 0.0004, + "loss": 0.652, + "step": 633 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.6116, + "step": 634 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.9324, + "step": 635 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.4822, + "step": 636 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.6417, + "step": 637 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.4586, + "step": 638 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.2927, + "step": 639 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.7887, + "step": 640 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.6607, + "step": 641 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.3745, + "step": 642 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.4677, + "step": 643 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.5891, + "step": 644 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.3206, + "step": 645 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.4792, + "step": 646 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.4851, + "step": 647 + }, + { + "epoch": 0.45, + "learning_rate": 0.0004, + "loss": 0.7103, + "step": 648 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.5212, + "step": 649 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.2019, + "step": 650 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.084, + "step": 651 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.4259, + "step": 652 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.7373, + "step": 653 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.8373, + "step": 654 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.887, + "step": 655 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.913, + "step": 656 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.9228, + "step": 657 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.7886, + "step": 658 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.8073, + "step": 659 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.7421, + "step": 660 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 1.0695, + "step": 661 + }, + { + "epoch": 0.46, + "learning_rate": 0.0004, + "loss": 0.719, + "step": 662 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.8423, + "step": 663 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.6688, + "step": 664 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.412, + "step": 665 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.7414, + "step": 666 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.7357, + "step": 667 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.5231, + "step": 668 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.6011, + "step": 669 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.4634, + "step": 670 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.2916, + "step": 671 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.9223, + "step": 672 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.8498, + "step": 673 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.6754, + "step": 674 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.5132, + "step": 675 + }, + { + "epoch": 0.47, + "learning_rate": 0.0004, + "loss": 0.6087, + "step": 676 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.7169, + "step": 677 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.4727, + "step": 678 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.8692, + "step": 679 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.1934, + "step": 680 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.2617, + "step": 681 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.6538, + "step": 682 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.3225, + "step": 683 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.7903, + "step": 684 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.6982, + "step": 685 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.3023, + "step": 686 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.5437, + "step": 687 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.2446, + "step": 688 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.4276, + "step": 689 + }, + { + "epoch": 0.48, + "learning_rate": 0.0004, + "loss": 0.5117, + "step": 690 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.7232, + "step": 691 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.2523, + "step": 692 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.4367, + "step": 693 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.2121, + "step": 694 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.384, + "step": 695 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.4618, + "step": 696 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.3786, + "step": 697 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.3153, + "step": 698 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 1.2427, + "step": 699 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 2.3096, + "step": 700 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.128, + "step": 701 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.2697, + "step": 702 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.3647, + "step": 703 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.7612, + "step": 704 + }, + { + "epoch": 0.49, + "learning_rate": 0.0004, + "loss": 0.9207, + "step": 705 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.6071, + "step": 706 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.7884, + "step": 707 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.9419, + "step": 708 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.7632, + "step": 709 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.7119, + "step": 710 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.696, + "step": 711 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.6528, + "step": 712 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.9004, + "step": 713 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.6225, + "step": 714 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.766, + "step": 715 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.8218, + "step": 716 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.6443, + "step": 717 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.6947, + "step": 718 + }, + { + "epoch": 0.5, + "learning_rate": 0.0004, + "loss": 0.5489, + "step": 719 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.5443, + "step": 720 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.3092, + "step": 721 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.6232, + "step": 722 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.473, + "step": 723 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.4059, + "step": 724 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.7054, + "step": 725 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.4207, + "step": 726 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.7167, + "step": 727 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.9212, + "step": 728 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.428, + "step": 729 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.5357, + "step": 730 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.5664, + "step": 731 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 0.9507, + "step": 732 + }, + { + "epoch": 0.51, + "learning_rate": 0.0004, + "loss": 1.1244, + "step": 733 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.5503, + "step": 734 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.259, + "step": 735 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 1.298, + "step": 736 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.3772, + "step": 737 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 1.0177, + "step": 738 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 1.0031, + "step": 739 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.8615, + "step": 740 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.4476, + "step": 741 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 1.4852, + "step": 742 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.5538, + "step": 743 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.5481, + "step": 744 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 1.0126, + "step": 745 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.5887, + "step": 746 + }, + { + "epoch": 0.52, + "learning_rate": 0.0004, + "loss": 0.4874, + "step": 747 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.685, + "step": 748 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.4105, + "step": 749 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.956, + "step": 750 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.1461, + "step": 751 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.1716, + "step": 752 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.2236, + "step": 753 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.5039, + "step": 754 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.5399, + "step": 755 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.8579, + "step": 756 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.8438, + "step": 757 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.6915, + "step": 758 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.9139, + "step": 759 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.7129, + "step": 760 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.7473, + "step": 761 + }, + { + "epoch": 0.53, + "learning_rate": 0.0004, + "loss": 0.8594, + "step": 762 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.5555, + "step": 763 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.9835, + "step": 764 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.5186, + "step": 765 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.8268, + "step": 766 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.7523, + "step": 767 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.5537, + "step": 768 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.5165, + "step": 769 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.6995, + "step": 770 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.785, + "step": 771 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.8304, + "step": 772 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.672, + "step": 773 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.785, + "step": 774 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.6968, + "step": 775 + }, + { + "epoch": 0.54, + "learning_rate": 0.0004, + "loss": 0.6106, + "step": 776 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.6045, + "step": 777 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.4182, + "step": 778 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 1.0079, + "step": 779 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.2465, + "step": 780 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.2605, + "step": 781 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 1.2797, + "step": 782 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.1785, + "step": 783 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.5563, + "step": 784 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.395, + "step": 785 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.256, + "step": 786 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 1.0868, + "step": 787 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.3505, + "step": 788 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.4305, + "step": 789 + }, + { + "epoch": 0.55, + "learning_rate": 0.0004, + "loss": 0.7121, + "step": 790 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.6441, + "step": 791 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 1.2003, + "step": 792 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.3511, + "step": 793 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.606, + "step": 794 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 1.3632, + "step": 795 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.5291, + "step": 796 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 1.0989, + "step": 797 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 1.6083, + "step": 798 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.4116, + "step": 799 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 2.924, + "step": 800 + }, + { + "epoch": 0.56, + "eval_loss": 0.6417471766471863, + "eval_runtime": 262.0155, + "eval_samples_per_second": 1.908, + "eval_steps_per_second": 0.954, + "step": 800 + }, + { + "epoch": 0.56, + "mmlu_eval_accuracy": 0.30538545442719256, + "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.25, + "mmlu_eval_accuracy_business_ethics": 0.5454545454545454, + "mmlu_eval_accuracy_clinical_knowledge": 0.41379310344827586, + "mmlu_eval_accuracy_college_biology": 0.375, + "mmlu_eval_accuracy_college_chemistry": 0.125, + "mmlu_eval_accuracy_college_computer_science": 0.09090909090909091, + "mmlu_eval_accuracy_college_mathematics": 0.36363636363636365, + "mmlu_eval_accuracy_college_medicine": 0.2727272727272727, + "mmlu_eval_accuracy_college_physics": 0.2727272727272727, + "mmlu_eval_accuracy_computer_security": 0.2727272727272727, + "mmlu_eval_accuracy_conceptual_physics": 0.23076923076923078, + "mmlu_eval_accuracy_econometrics": 0.25, + "mmlu_eval_accuracy_electrical_engineering": 0.1875, + "mmlu_eval_accuracy_elementary_mathematics": 0.2682926829268293, + "mmlu_eval_accuracy_formal_logic": 0.21428571428571427, + "mmlu_eval_accuracy_global_facts": 0.2, + "mmlu_eval_accuracy_high_school_biology": 0.25, + "mmlu_eval_accuracy_high_school_chemistry": 0.18181818181818182, + "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556, + "mmlu_eval_accuracy_high_school_european_history": 0.5, + "mmlu_eval_accuracy_high_school_geography": 0.6363636363636364, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.19047619047619047, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.27906976744186046, + "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276, + "mmlu_eval_accuracy_high_school_microeconomics": 0.35714285714285715, + "mmlu_loss": 1.06649634963274, + "step": 800 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.3342, + "step": 801 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.403, + "step": 802 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.525, + "step": 803 + }, + { + "epoch": 0.56, + "learning_rate": 0.0004, + "loss": 0.4384, + "step": 804 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 2.1165, + "step": 805 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.6357, + "step": 806 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8001, + "step": 807 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8867, + "step": 808 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.9335, + "step": 809 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.5799, + "step": 810 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8978, + "step": 811 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8002, + "step": 812 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8444, + "step": 813 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8727, + "step": 814 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.7608, + "step": 815 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.6937, + "step": 816 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.4899, + "step": 817 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.655, + "step": 818 + }, + { + "epoch": 0.57, + "learning_rate": 0.0004, + "loss": 0.8038, + "step": 819 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.5451, + "step": 820 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 1.2012, + "step": 821 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.3971, + "step": 822 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.3928, + "step": 823 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.2833, + "step": 824 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.4913, + "step": 825 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.3772, + "step": 826 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.714, + "step": 827 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.4023, + "step": 828 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.633, + "step": 829 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.3141, + "step": 830 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.4716, + "step": 831 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.3983, + "step": 832 + }, + { + "epoch": 0.58, + "learning_rate": 0.0004, + "loss": 0.2375, + "step": 833 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.375, + "step": 834 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.4347, + "step": 835 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 1.9594, + "step": 836 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.3025, + "step": 837 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.6105, + "step": 838 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.6737, + "step": 839 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.4223, + "step": 840 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.5873, + "step": 841 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.727, + "step": 842 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.5712, + "step": 843 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 1.0627, + "step": 844 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.3005, + "step": 845 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.4506, + "step": 846 + }, + { + "epoch": 0.59, + "learning_rate": 0.0004, + "loss": 0.4797, + "step": 847 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.4686, + "step": 848 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.2213, + "step": 849 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.8648, + "step": 850 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.2103, + "step": 851 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.4139, + "step": 852 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.1546, + "step": 853 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.1155, + "step": 854 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.625, + "step": 855 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.8956, + "step": 856 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.7037, + "step": 857 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.8418, + "step": 858 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.6672, + "step": 859 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.8433, + "step": 860 + }, + { + "epoch": 0.6, + "learning_rate": 0.0004, + "loss": 0.7895, + "step": 861 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.7541, + "step": 862 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.7261, + "step": 863 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.6541, + "step": 864 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.7801, + "step": 865 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.7632, + "step": 866 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.3367, + "step": 867 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.925, + "step": 868 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.5, + "step": 869 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.644, + "step": 870 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.5403, + "step": 871 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.7117, + "step": 872 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.5514, + "step": 873 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.5432, + "step": 874 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.6386, + "step": 875 + }, + { + "epoch": 0.61, + "learning_rate": 0.0004, + "loss": 0.8542, + "step": 876 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.5486, + "step": 877 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.4173, + "step": 878 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.5841, + "step": 879 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.48, + "step": 880 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.3887, + "step": 881 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.2361, + "step": 882 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.4635, + "step": 883 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.3724, + "step": 884 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.3429, + "step": 885 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.1864, + "step": 886 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 1.0825, + "step": 887 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.8833, + "step": 888 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 0.6354, + "step": 889 + }, + { + "epoch": 0.62, + "learning_rate": 0.0004, + "loss": 1.1362, + "step": 890 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.6212, + "step": 891 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 1.0072, + "step": 892 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.4585, + "step": 893 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.5292, + "step": 894 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.578, + "step": 895 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.5296, + "step": 896 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.1166, + "step": 897 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 1.0426, + "step": 898 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.3109, + "step": 899 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 1.5721, + "step": 900 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.3905, + "step": 901 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.1533, + "step": 902 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.214, + "step": 903 + }, + { + "epoch": 0.63, + "learning_rate": 0.0004, + "loss": 0.7675, + "step": 904 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.7776, + "step": 905 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.8972, + "step": 906 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.9331, + "step": 907 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.7464, + "step": 908 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.7094, + "step": 909 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.8942, + "step": 910 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.8436, + "step": 911 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.8331, + "step": 912 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.5505, + "step": 913 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.4096, + "step": 914 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.4566, + "step": 915 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.6529, + "step": 916 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.4911, + "step": 917 + }, + { + "epoch": 0.64, + "learning_rate": 0.0004, + "loss": 0.4905, + "step": 918 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.7721, + "step": 919 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.7032, + "step": 920 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.6081, + "step": 921 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.7841, + "step": 922 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.6164, + "step": 923 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.6198, + "step": 924 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.5534, + "step": 925 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.9022, + "step": 926 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.2717, + "step": 927 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.7298, + "step": 928 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.7003, + "step": 929 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.5087, + "step": 930 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.5762, + "step": 931 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.4049, + "step": 932 + }, + { + "epoch": 0.65, + "learning_rate": 0.0004, + "loss": 0.3152, + "step": 933 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.7488, + "step": 934 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.1635, + "step": 935 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.4921, + "step": 936 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.3392, + "step": 937 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.5501, + "step": 938 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.7825, + "step": 939 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.5922, + "step": 940 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 1.0295, + "step": 941 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.4026, + "step": 942 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.5983, + "step": 943 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 1.1622, + "step": 944 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.58, + "step": 945 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 0.3097, + "step": 946 + }, + { + "epoch": 0.66, + "learning_rate": 0.0004, + "loss": 1.2314, + "step": 947 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.404, + "step": 948 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.3989, + "step": 949 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.2975, + "step": 950 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.3987, + "step": 951 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.1746, + "step": 952 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 1.0329, + "step": 953 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.5963, + "step": 954 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.8642, + "step": 955 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.7199, + "step": 956 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.7995, + "step": 957 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.9483, + "step": 958 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.8768, + "step": 959 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.7032, + "step": 960 + }, + { + "epoch": 0.67, + "learning_rate": 0.0004, + "loss": 0.773, + "step": 961 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.7956, + "step": 962 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.7093, + "step": 963 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.7432, + "step": 964 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.8717, + "step": 965 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.472, + "step": 966 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.7872, + "step": 967 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.8238, + "step": 968 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.4689, + "step": 969 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.5136, + "step": 970 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.4863, + "step": 971 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.6755, + "step": 972 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.463, + "step": 973 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.5894, + "step": 974 + }, + { + "epoch": 0.68, + "learning_rate": 0.0004, + "loss": 0.4749, + "step": 975 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.644, + "step": 976 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.8176, + "step": 977 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.6255, + "step": 978 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.5182, + "step": 979 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.4552, + "step": 980 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.4505, + "step": 981 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.8373, + "step": 982 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.5285, + "step": 983 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 1.0748, + "step": 984 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.5467, + "step": 985 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.7294, + "step": 986 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.6341, + "step": 987 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.2867, + "step": 988 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 1.2924, + "step": 989 + }, + { + "epoch": 0.69, + "learning_rate": 0.0004, + "loss": 0.4195, + "step": 990 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.5777, + "step": 991 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 1.6128, + "step": 992 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.187, + "step": 993 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.3625, + "step": 994 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.2464, + "step": 995 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.6021, + "step": 996 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.5822, + "step": 997 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 1.3364, + "step": 998 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.2851, + "step": 999 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.5591, + "step": 1000 + }, + { + "epoch": 0.7, + "eval_loss": 0.6062986850738525, + "eval_runtime": 261.9741, + "eval_samples_per_second": 1.909, + "eval_steps_per_second": 0.954, + "step": 1000 + }, + { + "epoch": 0.7, + "mmlu_eval_accuracy": 0.2696631261532825, + "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365, + "mmlu_eval_accuracy_anatomy": 0.42857142857142855, + "mmlu_eval_accuracy_astronomy": 0.25, + "mmlu_eval_accuracy_business_ethics": 0.36363636363636365, + "mmlu_eval_accuracy_clinical_knowledge": 0.3793103448275862, + "mmlu_eval_accuracy_college_biology": 0.25, + "mmlu_eval_accuracy_college_chemistry": 0.125, + "mmlu_eval_accuracy_college_computer_science": 0.09090909090909091, + "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727, + "mmlu_eval_accuracy_college_medicine": 0.22727272727272727, + "mmlu_eval_accuracy_college_physics": 0.18181818181818182, + "mmlu_eval_accuracy_computer_security": 0.2727272727272727, + "mmlu_eval_accuracy_conceptual_physics": 0.23076923076923078, + "mmlu_eval_accuracy_econometrics": 0.25, + "mmlu_eval_accuracy_electrical_engineering": 0.1875, + "mmlu_eval_accuracy_elementary_mathematics": 0.24390243902439024, + "mmlu_eval_accuracy_formal_logic": 0.2857142857142857, + "mmlu_eval_accuracy_global_facts": 0.2, + "mmlu_eval_accuracy_high_school_biology": 0.1875, + "mmlu_eval_accuracy_high_school_chemistry": 0.13636363636363635, + "mmlu_eval_accuracy_high_school_computer_science": 0.4444444444444444, + "mmlu_eval_accuracy_high_school_european_history": 0.3333333333333333, + "mmlu_eval_accuracy_high_school_geography": 0.5909090909090909, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.09523809523809523, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.32558139534883723, + "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793, + "mmlu_eval_accuracy_high_school_microeconomics": 0.35714285714285715, + "mmlu_loss": 0.8371138982176781, + "step": 1000 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.2568, + "step": 1001 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.2256, + "step": 1002 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.1923, + "step": 1003 + }, + { + "epoch": 0.7, + "learning_rate": 0.0004, + "loss": 0.8292, + "step": 1004 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.7634, + "step": 1005 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.8455, + "step": 1006 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.9398, + "step": 1007 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.8622, + "step": 1008 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.6987, + "step": 1009 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.5708, + "step": 1010 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.889, + "step": 1011 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.754, + "step": 1012 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.7802, + "step": 1013 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.8057, + "step": 1014 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.7057, + "step": 1015 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.5939, + "step": 1016 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.9458, + "step": 1017 + }, + { + "epoch": 0.71, + "learning_rate": 0.0004, + "loss": 0.4199, + "step": 1018 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.6285, + "step": 1019 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.5437, + "step": 1020 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.7155, + "step": 1021 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.5762, + "step": 1022 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.6708, + "step": 1023 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.6386, + "step": 1024 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.6075, + "step": 1025 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.6806, + "step": 1026 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.5212, + "step": 1027 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.4784, + "step": 1028 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.2564, + "step": 1029 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.6152, + "step": 1030 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.3457, + "step": 1031 + }, + { + "epoch": 0.72, + "learning_rate": 0.0004, + "loss": 0.5407, + "step": 1032 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.4891, + "step": 1033 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.5085, + "step": 1034 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.2077, + "step": 1035 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.4229, + "step": 1036 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.6036, + "step": 1037 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.5033, + "step": 1038 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.3658, + "step": 1039 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.7938, + "step": 1040 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.4784, + "step": 1041 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.475, + "step": 1042 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.3961, + "step": 1043 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.5404, + "step": 1044 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.3528, + "step": 1045 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.3864, + "step": 1046 + }, + { + "epoch": 0.73, + "learning_rate": 0.0004, + "loss": 0.5713, + "step": 1047 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.9218, + "step": 1048 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.9482, + "step": 1049 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8584, + "step": 1050 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.2274, + "step": 1051 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.9419, + "step": 1052 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.6805, + "step": 1053 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8443, + "step": 1054 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8746, + "step": 1055 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8156, + "step": 1056 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.7459, + "step": 1057 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8752, + "step": 1058 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.7534, + "step": 1059 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8391, + "step": 1060 + }, + { + "epoch": 0.74, + "learning_rate": 0.0004, + "loss": 0.8513, + "step": 1061 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.8419, + "step": 1062 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.491, + "step": 1063 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.7671, + "step": 1064 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.912, + "step": 1065 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.3856, + "step": 1066 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.5201, + "step": 1067 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.54, + "step": 1068 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.4883, + "step": 1069 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.6604, + "step": 1070 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.8023, + "step": 1071 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.8695, + "step": 1072 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.7843, + "step": 1073 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.4502, + "step": 1074 + }, + { + "epoch": 0.75, + "learning_rate": 0.0004, + "loss": 0.6346, + "step": 1075 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.9306, + "step": 1076 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.5691, + "step": 1077 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.6568, + "step": 1078 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.7103, + "step": 1079 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.9179, + "step": 1080 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.3694, + "step": 1081 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.4531, + "step": 1082 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 1.2981, + "step": 1083 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 1.3878, + "step": 1084 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.5209, + "step": 1085 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 1.1522, + "step": 1086 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.2509, + "step": 1087 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 1.4895, + "step": 1088 + }, + { + "epoch": 0.76, + "learning_rate": 0.0004, + "loss": 0.8326, + "step": 1089 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.53, + "step": 1090 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.4127, + "step": 1091 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 1.2986, + "step": 1092 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.4335, + "step": 1093 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.7598, + "step": 1094 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.9294, + "step": 1095 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.7546, + "step": 1096 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.5667, + "step": 1097 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.444, + "step": 1098 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.4992, + "step": 1099 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 1.4161, + "step": 1100 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.1406, + "step": 1101 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.1881, + "step": 1102 + }, + { + "epoch": 0.77, + "learning_rate": 0.0004, + "loss": 0.1375, + "step": 1103 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.3047, + "step": 1104 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.1846, + "step": 1105 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.3801, + "step": 1106 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.9691, + "step": 1107 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.7683, + "step": 1108 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.8054, + "step": 1109 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.8484, + "step": 1110 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.84, + "step": 1111 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.6551, + "step": 1112 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.5611, + "step": 1113 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.7925, + "step": 1114 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.5565, + "step": 1115 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.5667, + "step": 1116 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.5563, + "step": 1117 + }, + { + "epoch": 0.78, + "learning_rate": 0.0004, + "loss": 0.5991, + "step": 1118 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.8101, + "step": 1119 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.5182, + "step": 1120 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.5591, + "step": 1121 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.8661, + "step": 1122 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.518, + "step": 1123 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.3308, + "step": 1124 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.2169, + "step": 1125 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.7888, + "step": 1126 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.9938, + "step": 1127 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.6998, + "step": 1128 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.3947, + "step": 1129 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.2432, + "step": 1130 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 1.0518, + "step": 1131 + }, + { + "epoch": 0.79, + "learning_rate": 0.0004, + "loss": 0.4578, + "step": 1132 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.6499, + "step": 1133 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.5507, + "step": 1134 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.5031, + "step": 1135 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.8217, + "step": 1136 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.9918, + "step": 1137 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.6194, + "step": 1138 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.3885, + "step": 1139 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.514, + "step": 1140 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.6231, + "step": 1141 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.3778, + "step": 1142 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.5501, + "step": 1143 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 1.0975, + "step": 1144 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.3143, + "step": 1145 + }, + { + "epoch": 0.8, + "learning_rate": 0.0004, + "loss": 0.5101, + "step": 1146 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.3943, + "step": 1147 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 1.0746, + "step": 1148 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.2978, + "step": 1149 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.342, + "step": 1150 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.1872, + "step": 1151 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.2124, + "step": 1152 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.2549, + "step": 1153 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.1347, + "step": 1154 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.8295, + "step": 1155 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.7941, + "step": 1156 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.3214, + "step": 1157 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.8334, + "step": 1158 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.7792, + "step": 1159 + }, + { + "epoch": 0.81, + "learning_rate": 0.0004, + "loss": 0.6914, + "step": 1160 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.9085, + "step": 1161 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.8467, + "step": 1162 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.7527, + "step": 1163 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.7065, + "step": 1164 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.5496, + "step": 1165 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.747, + "step": 1166 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.9068, + "step": 1167 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.5662, + "step": 1168 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.8689, + "step": 1169 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.8347, + "step": 1170 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.3442, + "step": 1171 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.375, + "step": 1172 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.6623, + "step": 1173 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.8362, + "step": 1174 + }, + { + "epoch": 0.82, + "learning_rate": 0.0004, + "loss": 0.7407, + "step": 1175 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.4577, + "step": 1176 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.4128, + "step": 1177 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.5529, + "step": 1178 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.385, + "step": 1179 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.7016, + "step": 1180 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.3416, + "step": 1181 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.3132, + "step": 1182 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.4332, + "step": 1183 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.4828, + "step": 1184 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.2299, + "step": 1185 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.8477, + "step": 1186 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.3425, + "step": 1187 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 1.0418, + "step": 1188 + }, + { + "epoch": 0.83, + "learning_rate": 0.0004, + "loss": 0.7957, + "step": 1189 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.2999, + "step": 1190 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.4034, + "step": 1191 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.6535, + "step": 1192 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.9691, + "step": 1193 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.559, + "step": 1194 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 1.8883, + "step": 1195 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.1952, + "step": 1196 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.4336, + "step": 1197 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.6651, + "step": 1198 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.7694, + "step": 1199 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.3228, + "step": 1200 + }, + { + "epoch": 0.84, + "eval_loss": 0.6253868341445923, + "eval_runtime": 261.8289, + "eval_samples_per_second": 1.91, + "eval_steps_per_second": 0.955, + "step": 1200 + }, + { + "epoch": 0.84, + "mmlu_eval_accuracy": 0.3628458989343948, + "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727, + "mmlu_eval_accuracy_anatomy": 0.42857142857142855, + "mmlu_eval_accuracy_astronomy": 0.3125, + "mmlu_eval_accuracy_business_ethics": 0.36363636363636365, + "mmlu_eval_accuracy_clinical_knowledge": 0.3793103448275862, + "mmlu_eval_accuracy_college_biology": 0.3125, + "mmlu_eval_accuracy_college_chemistry": 0.375, + "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453, + "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182, + "mmlu_eval_accuracy_college_medicine": 0.4090909090909091, + "mmlu_eval_accuracy_college_physics": 0.6363636363636364, + "mmlu_eval_accuracy_computer_security": 0.18181818181818182, + "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231, + "mmlu_eval_accuracy_econometrics": 0.25, + "mmlu_eval_accuracy_electrical_engineering": 0.5, + "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244, + "mmlu_eval_accuracy_formal_logic": 0.35714285714285715, + "mmlu_eval_accuracy_global_facts": 0.4, + "mmlu_eval_accuracy_high_school_biology": 0.40625, + "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727, + "mmlu_eval_accuracy_high_school_computer_science": 0.3333333333333333, + "mmlu_eval_accuracy_high_school_european_history": 0.2222222222222222, + "mmlu_eval_accuracy_high_school_geography": 0.6363636363636364, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.42857142857142855, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558, + "mmlu_eval_accuracy_high_school_mathematics": 0.06896551724137931, + "mmlu_eval_accuracy_high_school_microeconomics": 0.35714285714285715, + "mmlu_loss": 0.8975866470783949, + "step": 1200 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.1743, + "step": 1201 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.6816, + "step": 1202 + }, + { + "epoch": 0.84, + "learning_rate": 0.0004, + "loss": 0.5607, + "step": 1203 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.8083, + "step": 1204 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.7387, + "step": 1205 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.7967, + "step": 1206 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.827, + "step": 1207 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.7406, + "step": 1208 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.5886, + "step": 1209 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.6724, + "step": 1210 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.6053, + "step": 1211 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.9438, + "step": 1212 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.6901, + "step": 1213 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.4885, + "step": 1214 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.8167, + "step": 1215 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.6622, + "step": 1216 + }, + { + "epoch": 0.85, + "learning_rate": 0.0004, + "loss": 0.7829, + "step": 1217 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.8169, + "step": 1218 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.9069, + "step": 1219 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.6293, + "step": 1220 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.7289, + "step": 1221 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.3815, + "step": 1222 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.7074, + "step": 1223 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.4564, + "step": 1224 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.9812, + "step": 1225 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.5978, + "step": 1226 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.5474, + "step": 1227 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.1875, + "step": 1228 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.6129, + "step": 1229 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.4693, + "step": 1230 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.2681, + "step": 1231 + }, + { + "epoch": 0.86, + "learning_rate": 0.0004, + "loss": 0.3581, + "step": 1232 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.4277, + "step": 1233 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.2964, + "step": 1234 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.5602, + "step": 1235 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.3279, + "step": 1236 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.6986, + "step": 1237 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.291, + "step": 1238 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.4765, + "step": 1239 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.9116, + "step": 1240 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.3427, + "step": 1241 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.9017, + "step": 1242 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 1.1634, + "step": 1243 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.8345, + "step": 1244 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.2204, + "step": 1245 + }, + { + "epoch": 0.87, + "learning_rate": 0.0004, + "loss": 0.3808, + "step": 1246 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.4921, + "step": 1247 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.562, + "step": 1248 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.5004, + "step": 1249 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.4646, + "step": 1250 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.1831, + "step": 1251 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.3017, + "step": 1252 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.6987, + "step": 1253 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.2869, + "step": 1254 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.9395, + "step": 1255 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.8322, + "step": 1256 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.3549, + "step": 1257 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.8554, + "step": 1258 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.6045, + "step": 1259 + }, + { + "epoch": 0.88, + "learning_rate": 0.0004, + "loss": 0.7583, + "step": 1260 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.711, + "step": 1261 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.8744, + "step": 1262 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.7213, + "step": 1263 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.9005, + "step": 1264 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.7734, + "step": 1265 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.8033, + "step": 1266 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.4265, + "step": 1267 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.4348, + "step": 1268 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.6604, + "step": 1269 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.9578, + "step": 1270 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 1.1409, + "step": 1271 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.5768, + "step": 1272 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.6744, + "step": 1273 + }, + { + "epoch": 0.89, + "learning_rate": 0.0004, + "loss": 0.6866, + "step": 1274 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 1.0161, + "step": 1275 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.4803, + "step": 1276 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.4551, + "step": 1277 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.7931, + "step": 1278 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.6463, + "step": 1279 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.3938, + "step": 1280 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.2272, + "step": 1281 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.6787, + "step": 1282 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.5137, + "step": 1283 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.5202, + "step": 1284 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 1.2892, + "step": 1285 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.548, + "step": 1286 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.5782, + "step": 1287 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.4953, + "step": 1288 + }, + { + "epoch": 0.9, + "learning_rate": 0.0004, + "loss": 0.4502, + "step": 1289 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.5294, + "step": 1290 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 1.0318, + "step": 1291 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.6995, + "step": 1292 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.9352, + "step": 1293 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.4961, + "step": 1294 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.7605, + "step": 1295 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.3933, + "step": 1296 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.5961, + "step": 1297 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.7016, + "step": 1298 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.912, + "step": 1299 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.6619, + "step": 1300 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.2071, + "step": 1301 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.1415, + "step": 1302 + }, + { + "epoch": 0.91, + "learning_rate": 0.0004, + "loss": 0.6498, + "step": 1303 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.4005, + "step": 1304 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.8573, + "step": 1305 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.812, + "step": 1306 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.8208, + "step": 1307 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.7794, + "step": 1308 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.857, + "step": 1309 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.9111, + "step": 1310 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.8711, + "step": 1311 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.7179, + "step": 1312 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.7123, + "step": 1313 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.7376, + "step": 1314 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.4265, + "step": 1315 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.8858, + "step": 1316 + }, + { + "epoch": 0.92, + "learning_rate": 0.0004, + "loss": 0.5913, + "step": 1317 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.4801, + "step": 1318 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.3882, + "step": 1319 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.6328, + "step": 1320 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.3935, + "step": 1321 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.5347, + "step": 1322 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.286, + "step": 1323 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.533, + "step": 1324 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.5512, + "step": 1325 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.4157, + "step": 1326 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.8734, + "step": 1327 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.9141, + "step": 1328 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.9479, + "step": 1329 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.7035, + "step": 1330 + }, + { + "epoch": 0.93, + "learning_rate": 0.0004, + "loss": 0.5056, + "step": 1331 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.4605, + "step": 1332 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.5488, + "step": 1333 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 1.3185, + "step": 1334 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 1.4072, + "step": 1335 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.7788, + "step": 1336 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.1447, + "step": 1337 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.5009, + "step": 1338 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.4734, + "step": 1339 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.5158, + "step": 1340 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.2823, + "step": 1341 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 1.7664, + "step": 1342 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.3799, + "step": 1343 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.7159, + "step": 1344 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 0.4461, + "step": 1345 + }, + { + "epoch": 0.94, + "learning_rate": 0.0004, + "loss": 1.0385, + "step": 1346 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.9339, + "step": 1347 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.276, + "step": 1348 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.6796, + "step": 1349 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 2.0265, + "step": 1350 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.1967, + "step": 1351 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.2995, + "step": 1352 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.4029, + "step": 1353 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.7888, + "step": 1354 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.8218, + "step": 1355 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.733, + "step": 1356 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.8401, + "step": 1357 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.9573, + "step": 1358 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.7701, + "step": 1359 + }, + { + "epoch": 0.95, + "learning_rate": 0.0004, + "loss": 0.8522, + "step": 1360 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.6962, + "step": 1361 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.8918, + "step": 1362 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.8291, + "step": 1363 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.6959, + "step": 1364 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.7376, + "step": 1365 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.7284, + "step": 1366 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.8285, + "step": 1367 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.7935, + "step": 1368 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.1695, + "step": 1369 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.326, + "step": 1370 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.8143, + "step": 1371 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.7372, + "step": 1372 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.3449, + "step": 1373 + }, + { + "epoch": 0.96, + "learning_rate": 0.0004, + "loss": 0.6834, + "step": 1374 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.8138, + "step": 1375 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.4452, + "step": 1376 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.4034, + "step": 1377 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.4019, + "step": 1378 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.7407, + "step": 1379 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.3779, + "step": 1380 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.411, + "step": 1381 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.5773, + "step": 1382 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.6099, + "step": 1383 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.777, + "step": 1384 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.6538, + "step": 1385 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.3722, + "step": 1386 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 1.5937, + "step": 1387 + }, + { + "epoch": 0.97, + "learning_rate": 0.0004, + "loss": 0.327, + "step": 1388 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.6336, + "step": 1389 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.2354, + "step": 1390 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.2645, + "step": 1391 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.2908, + "step": 1392 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.4325, + "step": 1393 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.2703, + "step": 1394 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 1.1404, + "step": 1395 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.3881, + "step": 1396 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 1.0695, + "step": 1397 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.5117, + "step": 1398 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.4305, + "step": 1399 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 1.0432, + "step": 1400 + }, + { + "epoch": 0.98, + "eval_loss": 0.6039700508117676, + "eval_runtime": 261.9743, + "eval_samples_per_second": 1.909, + "eval_steps_per_second": 0.954, + "step": 1400 + }, + { + "epoch": 0.98, + "mmlu_eval_accuracy": 0.2685707889765786, + "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.25, + "mmlu_eval_accuracy_business_ethics": 0.18181818181818182, + "mmlu_eval_accuracy_clinical_knowledge": 0.3448275862068966, + "mmlu_eval_accuracy_college_biology": 0.375, + "mmlu_eval_accuracy_college_chemistry": 0.125, + "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727, + "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182, + "mmlu_eval_accuracy_college_medicine": 0.18181818181818182, + "mmlu_eval_accuracy_college_physics": 0.45454545454545453, + "mmlu_eval_accuracy_computer_security": 0.45454545454545453, + "mmlu_eval_accuracy_conceptual_physics": 0.3076923076923077, + "mmlu_eval_accuracy_econometrics": 0.25, + "mmlu_eval_accuracy_electrical_engineering": 0.1875, + "mmlu_eval_accuracy_elementary_mathematics": 0.2682926829268293, + "mmlu_eval_accuracy_formal_logic": 0.07142857142857142, + "mmlu_eval_accuracy_global_facts": 0.3, + "mmlu_eval_accuracy_high_school_biology": 0.40625, + "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727, + "mmlu_eval_accuracy_high_school_computer_science": 0.3333333333333333, + "mmlu_eval_accuracy_high_school_european_history": 0.2777777777777778, + "mmlu_eval_accuracy_high_school_geography": 0.45454545454545453, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.14285714285714285, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.20930232558139536, + "mmlu_eval_accuracy_high_school_mathematics": 0.10344827586206896, + "mmlu_eval_accuracy_high_school_microeconomics": 0.21428571428571427, + "mmlu_loss": 0.9329393486976624, + "step": 1400 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.1008, + "step": 1401 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.8622, + "step": 1402 + }, + { + "epoch": 0.98, + "learning_rate": 0.0004, + "loss": 0.824, + "step": 1403 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.7905, + "step": 1404 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.7081, + "step": 1405 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.6723, + "step": 1406 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.8331, + "step": 1407 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.4537, + "step": 1408 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 1.0326, + "step": 1409 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.5841, + "step": 1410 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.3543, + "step": 1411 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.5162, + "step": 1412 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.4088, + "step": 1413 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 1.0244, + "step": 1414 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.5694, + "step": 1415 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.5725, + "step": 1416 + }, + { + "epoch": 0.99, + "learning_rate": 0.0004, + "loss": 0.6169, + "step": 1417 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.21, + "step": 1418 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.3279, + "step": 1419 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.4013, + "step": 1420 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.7457, + "step": 1421 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 1.0264, + "step": 1422 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.6207, + "step": 1423 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.1758, + "step": 1424 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.6153, + "step": 1425 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.1339, + "step": 1426 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.1799, + "step": 1427 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.0921, + "step": 1428 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.5441, + "step": 1429 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.6633, + "step": 1430 + }, + { + "epoch": 1.0, + "learning_rate": 0.0004, + "loss": 0.5029, + "step": 1431 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.7309, + "step": 1432 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.3367, + "step": 1433 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.6515, + "step": 1434 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.5197, + "step": 1435 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.619, + "step": 1436 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.5078, + "step": 1437 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.694, + "step": 1438 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.4567, + "step": 1439 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.6259, + "step": 1440 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.3665, + "step": 1441 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.7823, + "step": 1442 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.5632, + "step": 1443 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.4752, + "step": 1444 + }, + { + "epoch": 1.01, + "learning_rate": 0.0004, + "loss": 0.144, + "step": 1445 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.1895, + "step": 1446 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.6465, + "step": 1447 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.3328, + "step": 1448 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.249, + "step": 1449 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.2817, + "step": 1450 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.1381, + "step": 1451 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.2751, + "step": 1452 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.1343, + "step": 1453 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.2988, + "step": 1454 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.3597, + "step": 1455 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.3396, + "step": 1456 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.3548, + "step": 1457 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.3077, + "step": 1458 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.3647, + "step": 1459 + }, + { + "epoch": 1.02, + "learning_rate": 0.0004, + "loss": 0.36, + "step": 1460 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.4379, + "step": 1461 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.2236, + "step": 1462 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.1498, + "step": 1463 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.1466, + "step": 1464 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.2215, + "step": 1465 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.6021, + "step": 1466 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.3759, + "step": 1467 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.1917, + "step": 1468 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.6682, + "step": 1469 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.3403, + "step": 1470 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.2112, + "step": 1471 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.191, + "step": 1472 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.1696, + "step": 1473 + }, + { + "epoch": 1.03, + "learning_rate": 0.0004, + "loss": 0.1381, + "step": 1474 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.5863, + "step": 1475 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.1988, + "step": 1476 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.118, + "step": 1477 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.1905, + "step": 1478 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.2293, + "step": 1479 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.7874, + "step": 1480 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.604, + "step": 1481 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.7774, + "step": 1482 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.6345, + "step": 1483 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.5432, + "step": 1484 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.4688, + "step": 1485 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.3003, + "step": 1486 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 0.5342, + "step": 1487 + }, + { + "epoch": 1.04, + "learning_rate": 0.0004, + "loss": 1.0727, + "step": 1488 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.5757, + "step": 1489 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.9071, + "step": 1490 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.3416, + "step": 1491 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.6551, + "step": 1492 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.5706, + "step": 1493 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.6314, + "step": 1494 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.3075, + "step": 1495 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.4266, + "step": 1496 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.2736, + "step": 1497 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.484, + "step": 1498 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.481, + "step": 1499 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.4488, + "step": 1500 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.4519, + "step": 1501 + }, + { + "epoch": 1.05, + "learning_rate": 0.0004, + "loss": 0.104, + "step": 1502 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.6287, + "step": 1503 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.8928, + "step": 1504 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.2621, + "step": 1505 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.1671, + "step": 1506 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.3018, + "step": 1507 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.2044, + "step": 1508 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.3761, + "step": 1509 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.6629, + "step": 1510 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.4316, + "step": 1511 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.604, + "step": 1512 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.2183, + "step": 1513 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.1533, + "step": 1514 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.2277, + "step": 1515 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.27, + "step": 1516 + }, + { + "epoch": 1.06, + "learning_rate": 0.0004, + "loss": 0.1822, + "step": 1517 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.7598, + "step": 1518 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.9697, + "step": 1519 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.4182, + "step": 1520 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.5511, + "step": 1521 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.254, + "step": 1522 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.1417, + "step": 1523 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.3965, + "step": 1524 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.7704, + "step": 1525 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.0824, + "step": 1526 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.2777, + "step": 1527 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.5894, + "step": 1528 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.3111, + "step": 1529 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.6391, + "step": 1530 + }, + { + "epoch": 1.07, + "learning_rate": 0.0004, + "loss": 0.6511, + "step": 1531 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.5499, + "step": 1532 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.5937, + "step": 1533 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.6998, + "step": 1534 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.4825, + "step": 1535 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.5036, + "step": 1536 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.5106, + "step": 1537 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.5183, + "step": 1538 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.4524, + "step": 1539 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.4344, + "step": 1540 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.6122, + "step": 1541 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.4406, + "step": 1542 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.4684, + "step": 1543 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 0.4944, + "step": 1544 + }, + { + "epoch": 1.08, + "learning_rate": 0.0004, + "loss": 1.0022, + "step": 1545 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.3016, + "step": 1546 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.2465, + "step": 1547 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.3819, + "step": 1548 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.3643, + "step": 1549 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.2682, + "step": 1550 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.4018, + "step": 1551 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.2784, + "step": 1552 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.2179, + "step": 1553 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.4126, + "step": 1554 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.263, + "step": 1555 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.4679, + "step": 1556 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.2468, + "step": 1557 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.2497, + "step": 1558 + }, + { + "epoch": 1.09, + "learning_rate": 0.0004, + "loss": 0.3128, + "step": 1559 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.4184, + "step": 1560 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.2205, + "step": 1561 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.2776, + "step": 1562 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.3908, + "step": 1563 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.3154, + "step": 1564 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.4719, + "step": 1565 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.4776, + "step": 1566 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.4879, + "step": 1567 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.3918, + "step": 1568 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.3047, + "step": 1569 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.2167, + "step": 1570 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.267, + "step": 1571 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.4316, + "step": 1572 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.2167, + "step": 1573 + }, + { + "epoch": 1.1, + "learning_rate": 0.0004, + "loss": 0.527, + "step": 1574 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.7904, + "step": 1575 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.1293, + "step": 1576 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.2587, + "step": 1577 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 1.0614, + "step": 1578 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.5228, + "step": 1579 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.6162, + "step": 1580 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.6659, + "step": 1581 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.6455, + "step": 1582 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.6458, + "step": 1583 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.5845, + "step": 1584 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.676, + "step": 1585 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.4458, + "step": 1586 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.5097, + "step": 1587 + }, + { + "epoch": 1.11, + "learning_rate": 0.0004, + "loss": 0.4247, + "step": 1588 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.5735, + "step": 1589 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.4008, + "step": 1590 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.599, + "step": 1591 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.6497, + "step": 1592 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.5961, + "step": 1593 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.5184, + "step": 1594 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.1147, + "step": 1595 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.492, + "step": 1596 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.4417, + "step": 1597 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.4014, + "step": 1598 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.2598, + "step": 1599 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.3197, + "step": 1600 + }, + { + "epoch": 1.12, + "eval_loss": 0.6235715746879578, + "eval_runtime": 261.9932, + "eval_samples_per_second": 1.908, + "eval_steps_per_second": 0.954, + "step": 1600 + }, + { + "epoch": 1.12, + "mmlu_eval_accuracy": 0.3525390650531232, + "mmlu_eval_accuracy_abstract_algebra": 0.0, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.375, + "mmlu_eval_accuracy_business_ethics": 0.09090909090909091, + "mmlu_eval_accuracy_clinical_knowledge": 0.3793103448275862, + "mmlu_eval_accuracy_college_biology": 0.375, + "mmlu_eval_accuracy_college_chemistry": 0.625, + "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453, + "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182, + "mmlu_eval_accuracy_college_medicine": 0.2727272727272727, + "mmlu_eval_accuracy_college_physics": 0.36363636363636365, + "mmlu_eval_accuracy_computer_security": 0.6363636363636364, + "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464, + "mmlu_eval_accuracy_econometrics": 0.3333333333333333, + "mmlu_eval_accuracy_electrical_engineering": 0.625, + "mmlu_eval_accuracy_elementary_mathematics": 0.24390243902439024, + "mmlu_eval_accuracy_formal_logic": 0.42857142857142855, + "mmlu_eval_accuracy_global_facts": 0.2, + "mmlu_eval_accuracy_high_school_biology": 0.21875, + "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727, + "mmlu_eval_accuracy_high_school_computer_science": 0.4444444444444444, + "mmlu_eval_accuracy_high_school_european_history": 0.2222222222222222, + "mmlu_eval_accuracy_high_school_geography": 0.5, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.37209302325581395, + "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793, + "mmlu_eval_accuracy_high_school_microeconomics": 0.42857142857142855, + "mmlu_loss": 0.7952154342532158, + "step": 1600 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.2261, + "step": 1601 + }, + { + "epoch": 1.12, + "learning_rate": 0.0004, + "loss": 0.238, + "step": 1602 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.4392, + "step": 1603 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.4089, + "step": 1604 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.4387, + "step": 1605 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.2984, + "step": 1606 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.6023, + "step": 1607 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.2376, + "step": 1608 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.1852, + "step": 1609 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.5266, + "step": 1610 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.259, + "step": 1611 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.4731, + "step": 1612 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.533, + "step": 1613 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.5364, + "step": 1614 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.2979, + "step": 1615 + }, + { + "epoch": 1.13, + "learning_rate": 0.0004, + "loss": 0.8257, + "step": 1616 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.6534, + "step": 1617 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.5292, + "step": 1618 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.4821, + "step": 1619 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.3823, + "step": 1620 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.2509, + "step": 1621 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1004, + "step": 1622 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1348, + "step": 1623 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1405, + "step": 1624 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.2882, + "step": 1625 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1414, + "step": 1626 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1747, + "step": 1627 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1746, + "step": 1628 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.1461, + "step": 1629 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.6798, + "step": 1630 + }, + { + "epoch": 1.14, + "learning_rate": 0.0004, + "loss": 0.766, + "step": 1631 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.6122, + "step": 1632 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5697, + "step": 1633 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.7384, + "step": 1634 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5936, + "step": 1635 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5243, + "step": 1636 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5446, + "step": 1637 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5542, + "step": 1638 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.395, + "step": 1639 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.59, + "step": 1640 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5137, + "step": 1641 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.831, + "step": 1642 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5179, + "step": 1643 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.4522, + "step": 1644 + }, + { + "epoch": 1.15, + "learning_rate": 0.0004, + "loss": 0.5093, + "step": 1645 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.2492, + "step": 1646 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.4021, + "step": 1647 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.2477, + "step": 1648 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.3649, + "step": 1649 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.2222, + "step": 1650 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.2333, + "step": 1651 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.3062, + "step": 1652 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.2826, + "step": 1653 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.4369, + "step": 1654 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.3068, + "step": 1655 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.1707, + "step": 1656 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.4054, + "step": 1657 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.5617, + "step": 1658 + }, + { + "epoch": 1.16, + "learning_rate": 0.0004, + "loss": 0.1718, + "step": 1659 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.5046, + "step": 1660 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.591, + "step": 1661 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.2547, + "step": 1662 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.6077, + "step": 1663 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.2695, + "step": 1664 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.4135, + "step": 1665 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.2086, + "step": 1666 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.8859, + "step": 1667 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.2759, + "step": 1668 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.1777, + "step": 1669 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.3034, + "step": 1670 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.2507, + "step": 1671 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.2771, + "step": 1672 + }, + { + "epoch": 1.17, + "learning_rate": 0.0004, + "loss": 0.1228, + "step": 1673 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.1244, + "step": 1674 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.1676, + "step": 1675 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.469, + "step": 1676 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.4767, + "step": 1677 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.0438, + "step": 1678 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.3189, + "step": 1679 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.6415, + "step": 1680 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.6347, + "step": 1681 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.7361, + "step": 1682 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.5118, + "step": 1683 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.3178, + "step": 1684 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.6868, + "step": 1685 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.6565, + "step": 1686 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.7181, + "step": 1687 + }, + { + "epoch": 1.18, + "learning_rate": 0.0004, + "loss": 0.8061, + "step": 1688 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.3408, + "step": 1689 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.6235, + "step": 1690 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.4925, + "step": 1691 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.5904, + "step": 1692 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.8302, + "step": 1693 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.3796, + "step": 1694 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.4523, + "step": 1695 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.4007, + "step": 1696 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.5336, + "step": 1697 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.3644, + "step": 1698 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.3232, + "step": 1699 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.5039, + "step": 1700 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.2815, + "step": 1701 + }, + { + "epoch": 1.19, + "learning_rate": 0.0004, + "loss": 0.4005, + "step": 1702 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.4543, + "step": 1703 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.3084, + "step": 1704 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.6492, + "step": 1705 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.4945, + "step": 1706 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.395, + "step": 1707 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.2501, + "step": 1708 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.1801, + "step": 1709 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.3444, + "step": 1710 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.0765, + "step": 1711 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.3827, + "step": 1712 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.3577, + "step": 1713 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 1.556, + "step": 1714 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.5528, + "step": 1715 + }, + { + "epoch": 1.2, + "learning_rate": 0.0004, + "loss": 0.2855, + "step": 1716 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.7544, + "step": 1717 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.1409, + "step": 1718 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.4003, + "step": 1719 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.2708, + "step": 1720 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.2607, + "step": 1721 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.317, + "step": 1722 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.1305, + "step": 1723 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.2528, + "step": 1724 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.8979, + "step": 1725 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.2475, + "step": 1726 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.0929, + "step": 1727 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.5937, + "step": 1728 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.343, + "step": 1729 + }, + { + "epoch": 1.21, + "learning_rate": 0.0004, + "loss": 0.3672, + "step": 1730 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.6094, + "step": 1731 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.703, + "step": 1732 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.6172, + "step": 1733 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.6706, + "step": 1734 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.5924, + "step": 1735 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.5584, + "step": 1736 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.4434, + "step": 1737 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.7197, + "step": 1738 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.4025, + "step": 1739 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.4315, + "step": 1740 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.3137, + "step": 1741 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.2386, + "step": 1742 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.3152, + "step": 1743 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.4141, + "step": 1744 + }, + { + "epoch": 1.22, + "learning_rate": 0.0004, + "loss": 0.294, + "step": 1745 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.5025, + "step": 1746 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.3873, + "step": 1747 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.3164, + "step": 1748 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.5181, + "step": 1749 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.4996, + "step": 1750 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.2139, + "step": 1751 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.3284, + "step": 1752 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.5692, + "step": 1753 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.2351, + "step": 1754 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.3178, + "step": 1755 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.7891, + "step": 1756 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.369, + "step": 1757 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.4885, + "step": 1758 + }, + { + "epoch": 1.23, + "learning_rate": 0.0004, + "loss": 0.2654, + "step": 1759 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.4451, + "step": 1760 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.3549, + "step": 1761 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.2437, + "step": 1762 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.49, + "step": 1763 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.3483, + "step": 1764 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.2153, + "step": 1765 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.4111, + "step": 1766 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.4723, + "step": 1767 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.2984, + "step": 1768 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.3834, + "step": 1769 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.4778, + "step": 1770 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.5302, + "step": 1771 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.3847, + "step": 1772 + }, + { + "epoch": 1.24, + "learning_rate": 0.0004, + "loss": 0.3156, + "step": 1773 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.5822, + "step": 1774 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.3504, + "step": 1775 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.4279, + "step": 1776 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.2788, + "step": 1777 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.2507, + "step": 1778 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.3883, + "step": 1779 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.7364, + "step": 1780 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.5108, + "step": 1781 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.6921, + "step": 1782 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.6624, + "step": 1783 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.6871, + "step": 1784 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.7601, + "step": 1785 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.5435, + "step": 1786 + }, + { + "epoch": 1.25, + "learning_rate": 0.0004, + "loss": 0.5029, + "step": 1787 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.6106, + "step": 1788 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 1.0211, + "step": 1789 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.4223, + "step": 1790 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.6167, + "step": 1791 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.3701, + "step": 1792 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.6277, + "step": 1793 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.3524, + "step": 1794 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.304, + "step": 1795 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.2555, + "step": 1796 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.5034, + "step": 1797 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.3424, + "step": 1798 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.6686, + "step": 1799 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.2169, + "step": 1800 + }, + { + "epoch": 1.26, + "eval_loss": 0.6171774864196777, + "eval_runtime": 261.9946, + "eval_samples_per_second": 1.908, + "eval_steps_per_second": 0.954, + "step": 1800 + }, + { + "epoch": 1.26, + "mmlu_eval_accuracy": 0.3966351568415356, + "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.375, + "mmlu_eval_accuracy_business_ethics": 0.5454545454545454, + "mmlu_eval_accuracy_clinical_knowledge": 0.3448275862068966, + "mmlu_eval_accuracy_college_biology": 0.375, + "mmlu_eval_accuracy_college_chemistry": 0.125, + "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365, + "mmlu_eval_accuracy_college_mathematics": 0.36363636363636365, + "mmlu_eval_accuracy_college_medicine": 0.5, + "mmlu_eval_accuracy_college_physics": 0.6363636363636364, + "mmlu_eval_accuracy_computer_security": 0.45454545454545453, + "mmlu_eval_accuracy_conceptual_physics": 0.5384615384615384, + "mmlu_eval_accuracy_econometrics": 0.25, + "mmlu_eval_accuracy_electrical_engineering": 0.3125, + "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244, + "mmlu_eval_accuracy_formal_logic": 0.2857142857142857, + "mmlu_eval_accuracy_global_facts": 0.3, + "mmlu_eval_accuracy_high_school_biology": 0.34375, + "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182, + "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666, + "mmlu_eval_accuracy_high_school_european_history": 0.3888888888888889, + "mmlu_eval_accuracy_high_school_geography": 0.6818181818181818, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.42857142857142855, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907, + "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483, + "mmlu_eval_accuracy_high_school_microeconomics": 0.5, + "mmlu_loss": 0.8713522016406059, + "step": 1800 + }, + { + "epoch": 1.26, + "learning_rate": 0.0004, + "loss": 0.4412, + "step": 1801 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2611, + "step": 1802 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.5278, + "step": 1803 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2195, + "step": 1804 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2161, + "step": 1805 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2203, + "step": 1806 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2626, + "step": 1807 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.1439, + "step": 1808 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.6514, + "step": 1809 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.3509, + "step": 1810 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2478, + "step": 1811 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.5442, + "step": 1812 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.382, + "step": 1813 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.3425, + "step": 1814 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.1316, + "step": 1815 + }, + { + "epoch": 1.27, + "learning_rate": 0.0004, + "loss": 0.2629, + "step": 1816 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.3028, + "step": 1817 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.3539, + "step": 1818 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.132, + "step": 1819 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.6408, + "step": 1820 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.399, + "step": 1821 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.3575, + "step": 1822 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.3284, + "step": 1823 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.2842, + "step": 1824 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.3397, + "step": 1825 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.1401, + "step": 1826 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.4233, + "step": 1827 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.3967, + "step": 1828 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.6983, + "step": 1829 + }, + { + "epoch": 1.28, + "learning_rate": 0.0004, + "loss": 0.6125, + "step": 1830 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.7198, + "step": 1831 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.719, + "step": 1832 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.7088, + "step": 1833 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.6215, + "step": 1834 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.7658, + "step": 1835 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.8701, + "step": 1836 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.5789, + "step": 1837 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.7612, + "step": 1838 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.7198, + "step": 1839 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.6812, + "step": 1840 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.5485, + "step": 1841 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.6622, + "step": 1842 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.5386, + "step": 1843 + }, + { + "epoch": 1.29, + "learning_rate": 0.0004, + "loss": 0.7649, + "step": 1844 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.4083, + "step": 1845 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.3639, + "step": 1846 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.4618, + "step": 1847 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.4848, + "step": 1848 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.306, + "step": 1849 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.2439, + "step": 1850 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.2792, + "step": 1851 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.4189, + "step": 1852 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.3237, + "step": 1853 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.442, + "step": 1854 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.4448, + "step": 1855 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.2863, + "step": 1856 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.3207, + "step": 1857 + }, + { + "epoch": 1.3, + "learning_rate": 0.0004, + "loss": 0.347, + "step": 1858 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.4951, + "step": 1859 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.2945, + "step": 1860 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.1332, + "step": 1861 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.2806, + "step": 1862 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.0788, + "step": 1863 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.325, + "step": 1864 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.6921, + "step": 1865 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.5997, + "step": 1866 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.9352, + "step": 1867 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.3313, + "step": 1868 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.4182, + "step": 1869 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.2932, + "step": 1870 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.3797, + "step": 1871 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.2001, + "step": 1872 + }, + { + "epoch": 1.31, + "learning_rate": 0.0004, + "loss": 0.4309, + "step": 1873 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.9557, + "step": 1874 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.4949, + "step": 1875 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.1564, + "step": 1876 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.2433, + "step": 1877 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.0373, + "step": 1878 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.8472, + "step": 1879 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.5403, + "step": 1880 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.794, + "step": 1881 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.7366, + "step": 1882 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.6495, + "step": 1883 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.5671, + "step": 1884 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.7946, + "step": 1885 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.6215, + "step": 1886 + }, + { + "epoch": 1.32, + "learning_rate": 0.0004, + "loss": 0.5827, + "step": 1887 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.525, + "step": 1888 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.5264, + "step": 1889 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.4308, + "step": 1890 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.5508, + "step": 1891 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.464, + "step": 1892 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.4226, + "step": 1893 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 1.0, + "step": 1894 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.307, + "step": 1895 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.426, + "step": 1896 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.2435, + "step": 1897 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.095, + "step": 1898 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.1996, + "step": 1899 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.1996, + "step": 1900 + }, + { + "epoch": 1.33, + "learning_rate": 0.0004, + "loss": 0.2877, + "step": 1901 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.2542, + "step": 1902 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.3194, + "step": 1903 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.2314, + "step": 1904 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.5016, + "step": 1905 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.2851, + "step": 1906 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.3719, + "step": 1907 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.2403, + "step": 1908 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.2757, + "step": 1909 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.9063, + "step": 1910 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.4526, + "step": 1911 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.3151, + "step": 1912 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.0797, + "step": 1913 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.9371, + "step": 1914 + }, + { + "epoch": 1.34, + "learning_rate": 0.0004, + "loss": 0.0895, + "step": 1915 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.6488, + "step": 1916 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.3337, + "step": 1917 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.4327, + "step": 1918 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.234, + "step": 1919 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.3792, + "step": 1920 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.855, + "step": 1921 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.3862, + "step": 1922 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.2604, + "step": 1923 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.14, + "step": 1924 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.7476, + "step": 1925 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.2392, + "step": 1926 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.1647, + "step": 1927 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.2199, + "step": 1928 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.3518, + "step": 1929 + }, + { + "epoch": 1.35, + "learning_rate": 0.0004, + "loss": 0.6243, + "step": 1930 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.3948, + "step": 1931 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.7569, + "step": 1932 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.1988, + "step": 1933 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.7297, + "step": 1934 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.7439, + "step": 1935 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.6155, + "step": 1936 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.5777, + "step": 1937 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.5955, + "step": 1938 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.5275, + "step": 1939 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.6974, + "step": 1940 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.462, + "step": 1941 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.7279, + "step": 1942 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.376, + "step": 1943 + }, + { + "epoch": 1.36, + "learning_rate": 0.0004, + "loss": 0.4693, + "step": 1944 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.3689, + "step": 1945 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.5774, + "step": 1946 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.4916, + "step": 1947 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.462, + "step": 1948 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.34, + "step": 1949 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.4286, + "step": 1950 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.2793, + "step": 1951 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.6458, + "step": 1952 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.4651, + "step": 1953 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.3107, + "step": 1954 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.3719, + "step": 1955 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.1667, + "step": 1956 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.3396, + "step": 1957 + }, + { + "epoch": 1.37, + "learning_rate": 0.0004, + "loss": 0.1771, + "step": 1958 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.2587, + "step": 1959 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.7324, + "step": 1960 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.2483, + "step": 1961 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.4789, + "step": 1962 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.1511, + "step": 1963 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.4942, + "step": 1964 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.1403, + "step": 1965 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.7503, + "step": 1966 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 1.3877, + "step": 1967 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.8641, + "step": 1968 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.3818, + "step": 1969 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.9843, + "step": 1970 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.5374, + "step": 1971 + }, + { + "epoch": 1.38, + "learning_rate": 0.0004, + "loss": 0.248, + "step": 1972 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.1829, + "step": 1973 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.654, + "step": 1974 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.2179, + "step": 1975 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.2443, + "step": 1976 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.1168, + "step": 1977 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.257, + "step": 1978 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.7514, + "step": 1979 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.7051, + "step": 1980 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.7264, + "step": 1981 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.6347, + "step": 1982 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.5677, + "step": 1983 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.6024, + "step": 1984 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.526, + "step": 1985 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.7371, + "step": 1986 + }, + { + "epoch": 1.39, + "learning_rate": 0.0004, + "loss": 0.6324, + "step": 1987 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.6033, + "step": 1988 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.3971, + "step": 1989 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.5322, + "step": 1990 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.6006, + "step": 1991 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.5509, + "step": 1992 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.4556, + "step": 1993 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.5473, + "step": 1994 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.5254, + "step": 1995 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.3377, + "step": 1996 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.641, + "step": 1997 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.1864, + "step": 1998 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.476, + "step": 1999 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.2446, + "step": 2000 + }, + { + "epoch": 1.4, + "eval_loss": 0.6020480990409851, + "eval_runtime": 261.9477, + "eval_samples_per_second": 1.909, + "eval_steps_per_second": 0.954, + "step": 2000 + }, + { + "epoch": 1.4, + "mmlu_eval_accuracy": 0.37121874323899023, + "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.3125, + "mmlu_eval_accuracy_business_ethics": 0.5454545454545454, + "mmlu_eval_accuracy_clinical_knowledge": 0.27586206896551724, + "mmlu_eval_accuracy_college_biology": 0.5, + "mmlu_eval_accuracy_college_chemistry": 0.25, + "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453, + "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727, + "mmlu_eval_accuracy_college_medicine": 0.18181818181818182, + "mmlu_eval_accuracy_college_physics": 0.45454545454545453, + "mmlu_eval_accuracy_computer_security": 0.5454545454545454, + "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231, + "mmlu_eval_accuracy_econometrics": 0.08333333333333333, + "mmlu_eval_accuracy_electrical_engineering": 0.3125, + "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073, + "mmlu_eval_accuracy_formal_logic": 0.2857142857142857, + "mmlu_eval_accuracy_global_facts": 0.2, + "mmlu_eval_accuracy_high_school_biology": 0.3125, + "mmlu_eval_accuracy_high_school_chemistry": 0.18181818181818182, + "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556, + "mmlu_eval_accuracy_high_school_european_history": 0.4444444444444444, + "mmlu_eval_accuracy_high_school_geography": 0.6363636363636364, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.32558139534883723, + "mmlu_eval_accuracy_high_school_mathematics": 0.3793103448275862, + "mmlu_eval_accuracy_high_school_microeconomics": 0.5714285714285714, + "mmlu_loss": 0.8414341709017754, + "step": 2000 + }, + { + "epoch": 1.4, + "learning_rate": 0.0004, + "loss": 0.2144, + "step": 2001 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.3005, + "step": 2002 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.1441, + "step": 2003 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.3636, + "step": 2004 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.1422, + "step": 2005 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.2546, + "step": 2006 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.4279, + "step": 2007 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.4146, + "step": 2008 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.3163, + "step": 2009 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.2806, + "step": 2010 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.3152, + "step": 2011 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.2931, + "step": 2012 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.7744, + "step": 2013 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.5543, + "step": 2014 + }, + { + "epoch": 1.41, + "learning_rate": 0.0004, + "loss": 0.2914, + "step": 2015 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.3962, + "step": 2016 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.5594, + "step": 2017 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.3175, + "step": 2018 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.7436, + "step": 2019 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 1.1536, + "step": 2020 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.4556, + "step": 2021 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.2425, + "step": 2022 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.3773, + "step": 2023 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.4466, + "step": 2024 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.2584, + "step": 2025 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.1396, + "step": 2026 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.1961, + "step": 2027 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.4203, + "step": 2028 + }, + { + "epoch": 1.42, + "learning_rate": 0.0004, + "loss": 0.5752, + "step": 2029 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.8026, + "step": 2030 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.6498, + "step": 2031 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.6984, + "step": 2032 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.8209, + "step": 2033 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.7935, + "step": 2034 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.4286, + "step": 2035 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.4898, + "step": 2036 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.632, + "step": 2037 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.4912, + "step": 2038 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.5579, + "step": 2039 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.66, + "step": 2040 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.4272, + "step": 2041 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.6665, + "step": 2042 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.3872, + "step": 2043 + }, + { + "epoch": 1.43, + "learning_rate": 0.0004, + "loss": 0.2888, + "step": 2044 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.1202, + "step": 2045 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.5074, + "step": 2046 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.5068, + "step": 2047 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.3681, + "step": 2048 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.3959, + "step": 2049 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.234, + "step": 2050 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.2518, + "step": 2051 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.368, + "step": 2052 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.0845, + "step": 2053 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.5733, + "step": 2054 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.4222, + "step": 2055 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.2822, + "step": 2056 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.1397, + "step": 2057 + }, + { + "epoch": 1.44, + "learning_rate": 0.0004, + "loss": 0.3823, + "step": 2058 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.2355, + "step": 2059 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.6346, + "step": 2060 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.4589, + "step": 2061 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.1446, + "step": 2062 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.5715, + "step": 2063 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.1982, + "step": 2064 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.8405, + "step": 2065 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.397, + "step": 2066 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.3097, + "step": 2067 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.5207, + "step": 2068 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.2528, + "step": 2069 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.2843, + "step": 2070 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.078, + "step": 2071 + }, + { + "epoch": 1.45, + "learning_rate": 0.0004, + "loss": 0.5684, + "step": 2072 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.3893, + "step": 2073 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.4531, + "step": 2074 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.7483, + "step": 2075 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.3853, + "step": 2076 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.4585, + "step": 2077 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.789, + "step": 2078 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.4455, + "step": 2079 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.4432, + "step": 2080 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.6246, + "step": 2081 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.781, + "step": 2082 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.7147, + "step": 2083 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.5783, + "step": 2084 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.6042, + "step": 2085 + }, + { + "epoch": 1.46, + "learning_rate": 0.0004, + "loss": 0.6984, + "step": 2086 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.5726, + "step": 2087 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.488, + "step": 2088 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.4764, + "step": 2089 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.8202, + "step": 2090 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.7303, + "step": 2091 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.5382, + "step": 2092 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.4465, + "step": 2093 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.3127, + "step": 2094 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.4182, + "step": 2095 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.3492, + "step": 2096 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.4503, + "step": 2097 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.4174, + "step": 2098 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.4187, + "step": 2099 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.5189, + "step": 2100 + }, + { + "epoch": 1.47, + "learning_rate": 0.0004, + "loss": 0.2724, + "step": 2101 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.4884, + "step": 2102 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.2851, + "step": 2103 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.2611, + "step": 2104 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.4673, + "step": 2105 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.2052, + "step": 2106 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.278, + "step": 2107 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.2143, + "step": 2108 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.3572, + "step": 2109 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.3034, + "step": 2110 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.4658, + "step": 2111 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.3368, + "step": 2112 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.6886, + "step": 2113 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.1493, + "step": 2114 + }, + { + "epoch": 1.48, + "learning_rate": 0.0004, + "loss": 0.2303, + "step": 2115 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.5624, + "step": 2116 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.676, + "step": 2117 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.2416, + "step": 2118 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.4938, + "step": 2119 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.2237, + "step": 2120 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.1452, + "step": 2121 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.6926, + "step": 2122 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.2817, + "step": 2123 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.1612, + "step": 2124 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.9952, + "step": 2125 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.3766, + "step": 2126 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.0753, + "step": 2127 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.1647, + "step": 2128 + }, + { + "epoch": 1.49, + "learning_rate": 0.0004, + "loss": 0.1259, + "step": 2129 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.329, + "step": 2130 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.5134, + "step": 2131 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.6215, + "step": 2132 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.7773, + "step": 2133 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.6366, + "step": 2134 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.7093, + "step": 2135 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.7674, + "step": 2136 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.6388, + "step": 2137 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.6402, + "step": 2138 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.6585, + "step": 2139 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.5551, + "step": 2140 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.2941, + "step": 2141 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.3575, + "step": 2142 + }, + { + "epoch": 1.5, + "learning_rate": 0.0004, + "loss": 0.3847, + "step": 2143 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.5263, + "step": 2144 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.56, + "step": 2145 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.784, + "step": 2146 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.478, + "step": 2147 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.1179, + "step": 2148 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.4275, + "step": 2149 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.7919, + "step": 2150 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.3718, + "step": 2151 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.3986, + "step": 2152 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.2648, + "step": 2153 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.3032, + "step": 2154 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.3529, + "step": 2155 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.2872, + "step": 2156 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.3681, + "step": 2157 + }, + { + "epoch": 1.51, + "learning_rate": 0.0004, + "loss": 0.3877, + "step": 2158 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.1434, + "step": 2159 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.2117, + "step": 2160 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.2389, + "step": 2161 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.3532, + "step": 2162 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.2162, + "step": 2163 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.2598, + "step": 2164 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.2051, + "step": 2165 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.4936, + "step": 2166 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.68, + "step": 2167 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.3412, + "step": 2168 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.149, + "step": 2169 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.4438, + "step": 2170 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.3842, + "step": 2171 + }, + { + "epoch": 1.52, + "learning_rate": 0.0004, + "loss": 0.6077, + "step": 2172 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.3149, + "step": 2173 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.2328, + "step": 2174 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.1124, + "step": 2175 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.183, + "step": 2176 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.2491, + "step": 2177 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.2271, + "step": 2178 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.7425, + "step": 2179 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.6153, + "step": 2180 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.7312, + "step": 2181 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.663, + "step": 2182 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.5814, + "step": 2183 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.6918, + "step": 2184 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.5355, + "step": 2185 + }, + { + "epoch": 1.53, + "learning_rate": 0.0004, + "loss": 0.7587, + "step": 2186 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.6402, + "step": 2187 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.3958, + "step": 2188 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.4737, + "step": 2189 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.9913, + "step": 2190 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.4609, + "step": 2191 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.114, + "step": 2192 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.5522, + "step": 2193 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.759, + "step": 2194 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.4682, + "step": 2195 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.6343, + "step": 2196 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.4365, + "step": 2197 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.5787, + "step": 2198 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.4957, + "step": 2199 + }, + { + "epoch": 1.54, + "learning_rate": 0.0004, + "loss": 0.3722, + "step": 2200 + }, + { + "epoch": 1.54, + "eval_loss": 0.6022484302520752, + "eval_runtime": 261.9297, + "eval_samples_per_second": 1.909, + "eval_steps_per_second": 0.954, + "step": 2200 + }, + { + "epoch": 1.54, + "mmlu_eval_accuracy": 0.32171567430705617, + "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.125, + "mmlu_eval_accuracy_business_ethics": 0.2727272727272727, + "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552, + "mmlu_eval_accuracy_college_biology": 0.1875, + "mmlu_eval_accuracy_college_chemistry": 0.25, + "mmlu_eval_accuracy_college_computer_science": 0.18181818181818182, + "mmlu_eval_accuracy_college_mathematics": 0.09090909090909091, + "mmlu_eval_accuracy_college_medicine": 0.36363636363636365, + "mmlu_eval_accuracy_college_physics": 0.45454545454545453, + "mmlu_eval_accuracy_computer_security": 0.6363636363636364, + "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231, + "mmlu_eval_accuracy_econometrics": 0.16666666666666666, + "mmlu_eval_accuracy_electrical_engineering": 0.3125, + "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683, + "mmlu_eval_accuracy_formal_logic": 0.21428571428571427, + "mmlu_eval_accuracy_global_facts": 0.3, + "mmlu_eval_accuracy_high_school_biology": 0.34375, + "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727, + "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666, + "mmlu_eval_accuracy_high_school_european_history": 0.3333333333333333, + "mmlu_eval_accuracy_high_school_geography": 0.5, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.14285714285714285, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.2558139534883721, + "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276, + "mmlu_eval_accuracy_high_school_microeconomics": 0.5, + "mmlu_loss": 0.8473919027745723, + "step": 2200 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.5794, + "step": 2201 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.3608, + "step": 2202 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.3758, + "step": 2203 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.4776, + "step": 2204 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.2293, + "step": 2205 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.4173, + "step": 2206 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.3153, + "step": 2207 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.3802, + "step": 2208 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.5872, + "step": 2209 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.2047, + "step": 2210 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.6362, + "step": 2211 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.3469, + "step": 2212 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.1011, + "step": 2213 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.2334, + "step": 2214 + }, + { + "epoch": 1.55, + "learning_rate": 0.0004, + "loss": 0.5445, + "step": 2215 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.4939, + "step": 2216 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.4606, + "step": 2217 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.6936, + "step": 2218 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.3738, + "step": 2219 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.4157, + "step": 2220 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.1422, + "step": 2221 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.7551, + "step": 2222 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.6505, + "step": 2223 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.6533, + "step": 2224 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.262, + "step": 2225 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.1619, + "step": 2226 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.0783, + "step": 2227 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.0897, + "step": 2228 + }, + { + "epoch": 1.56, + "learning_rate": 0.0004, + "loss": 0.5536, + "step": 2229 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.7113, + "step": 2230 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.7423, + "step": 2231 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.7676, + "step": 2232 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.5834, + "step": 2233 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.7186, + "step": 2234 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.6125, + "step": 2235 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.5336, + "step": 2236 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.6897, + "step": 2237 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.8054, + "step": 2238 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.4355, + "step": 2239 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.3831, + "step": 2240 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.3089, + "step": 2241 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.5363, + "step": 2242 + }, + { + "epoch": 1.57, + "learning_rate": 0.0004, + "loss": 0.234, + "step": 2243 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.5892, + "step": 2244 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.5374, + "step": 2245 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.2209, + "step": 2246 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.7054, + "step": 2247 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.4359, + "step": 2248 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.1997, + "step": 2249 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.3922, + "step": 2250 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.4776, + "step": 2251 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.4223, + "step": 2252 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.1897, + "step": 2253 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.1465, + "step": 2254 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.2998, + "step": 2255 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.2819, + "step": 2256 + }, + { + "epoch": 1.58, + "learning_rate": 0.0004, + "loss": 0.4949, + "step": 2257 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.263, + "step": 2258 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.316, + "step": 2259 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.2673, + "step": 2260 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.3976, + "step": 2261 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.3546, + "step": 2262 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.06, + "step": 2263 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.3699, + "step": 2264 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 2.3811, + "step": 2265 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.2166, + "step": 2266 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.39, + "step": 2267 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.8601, + "step": 2268 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.3761, + "step": 2269 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.7222, + "step": 2270 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.6224, + "step": 2271 + }, + { + "epoch": 1.59, + "learning_rate": 0.0004, + "loss": 0.3393, + "step": 2272 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.7647, + "step": 2273 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.2786, + "step": 2274 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.5096, + "step": 2275 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.1215, + "step": 2276 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.1043, + "step": 2277 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.3839, + "step": 2278 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.2799, + "step": 2279 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.5895, + "step": 2280 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.7284, + "step": 2281 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.8503, + "step": 2282 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.635, + "step": 2283 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.6922, + "step": 2284 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.7779, + "step": 2285 + }, + { + "epoch": 1.6, + "learning_rate": 0.0004, + "loss": 0.7469, + "step": 2286 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.5495, + "step": 2287 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.7706, + "step": 2288 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.8542, + "step": 2289 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.5416, + "step": 2290 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.5778, + "step": 2291 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.7231, + "step": 2292 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.5579, + "step": 2293 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.5263, + "step": 2294 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.6941, + "step": 2295 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.7623, + "step": 2296 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.2968, + "step": 2297 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.5879, + "step": 2298 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.4142, + "step": 2299 + }, + { + "epoch": 1.61, + "learning_rate": 0.0004, + "loss": 0.2731, + "step": 2300 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.3087, + "step": 2301 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.3355, + "step": 2302 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.1631, + "step": 2303 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.5026, + "step": 2304 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.6432, + "step": 2305 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.1377, + "step": 2306 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.1337, + "step": 2307 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.725, + "step": 2308 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.156, + "step": 2309 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.3484, + "step": 2310 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.2964, + "step": 2311 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.2421, + "step": 2312 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.2164, + "step": 2313 + }, + { + "epoch": 1.62, + "learning_rate": 0.0004, + "loss": 0.1569, + "step": 2314 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.2214, + "step": 2315 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1463, + "step": 2316 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.4592, + "step": 2317 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.598, + "step": 2318 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1551, + "step": 2319 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1532, + "step": 2320 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1785, + "step": 2321 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.8471, + "step": 2322 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.3809, + "step": 2323 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.3395, + "step": 2324 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.5324, + "step": 2325 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1795, + "step": 2326 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1887, + "step": 2327 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.1895, + "step": 2328 + }, + { + "epoch": 1.63, + "learning_rate": 0.0004, + "loss": 0.0834, + "step": 2329 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.4481, + "step": 2330 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.597, + "step": 2331 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.6873, + "step": 2332 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.6364, + "step": 2333 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.7961, + "step": 2334 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.7582, + "step": 2335 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.5907, + "step": 2336 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.4654, + "step": 2337 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.7886, + "step": 2338 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.4318, + "step": 2339 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.2774, + "step": 2340 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.6752, + "step": 2341 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.3075, + "step": 2342 + }, + { + "epoch": 1.64, + "learning_rate": 0.0004, + "loss": 0.4376, + "step": 2343 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.5908, + "step": 2344 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.2839, + "step": 2345 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.2, + "step": 2346 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.3703, + "step": 2347 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.4019, + "step": 2348 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.3931, + "step": 2349 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.3687, + "step": 2350 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.2171, + "step": 2351 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.1255, + "step": 2352 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.1423, + "step": 2353 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.427, + "step": 2354 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.3191, + "step": 2355 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 0.3793, + "step": 2356 + }, + { + "epoch": 1.65, + "learning_rate": 0.0004, + "loss": 1.4718, + "step": 2357 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.3295, + "step": 2358 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.6596, + "step": 2359 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.7055, + "step": 2360 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.3864, + "step": 2361 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.2441, + "step": 2362 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.2711, + "step": 2363 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.1584, + "step": 2364 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.476, + "step": 2365 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.5573, + "step": 2366 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.3922, + "step": 2367 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.4616, + "step": 2368 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.6054, + "step": 2369 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.2925, + "step": 2370 + }, + { + "epoch": 1.66, + "learning_rate": 0.0004, + "loss": 0.2729, + "step": 2371 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.3474, + "step": 2372 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.2859, + "step": 2373 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.1094, + "step": 2374 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.2757, + "step": 2375 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.58, + "step": 2376 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.9134, + "step": 2377 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.3713, + "step": 2378 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.7216, + "step": 2379 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.6031, + "step": 2380 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.652, + "step": 2381 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.7711, + "step": 2382 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.6787, + "step": 2383 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.7071, + "step": 2384 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.8117, + "step": 2385 + }, + { + "epoch": 1.67, + "learning_rate": 0.0004, + "loss": 0.7123, + "step": 2386 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.6345, + "step": 2387 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.3828, + "step": 2388 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.5207, + "step": 2389 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.5468, + "step": 2390 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.5053, + "step": 2391 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.6586, + "step": 2392 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.5375, + "step": 2393 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.6905, + "step": 2394 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.2198, + "step": 2395 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.486, + "step": 2396 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.3833, + "step": 2397 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.3549, + "step": 2398 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.1755, + "step": 2399 + }, + { + "epoch": 1.68, + "learning_rate": 0.0004, + "loss": 0.2511, + "step": 2400 + }, + { + "epoch": 1.68, + "eval_loss": 0.6019096970558167, + "eval_runtime": 261.859, + "eval_samples_per_second": 1.909, + "eval_steps_per_second": 0.955, + "step": 2400 + }, + { + "epoch": 1.68, + "mmlu_eval_accuracy": 0.3576617805741415, + "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727, + "mmlu_eval_accuracy_anatomy": 0.35714285714285715, + "mmlu_eval_accuracy_astronomy": 0.375, + "mmlu_eval_accuracy_business_ethics": 0.45454545454545453, + "mmlu_eval_accuracy_clinical_knowledge": 0.41379310344827586, + "mmlu_eval_accuracy_college_biology": 0.3125, + "mmlu_eval_accuracy_college_chemistry": 0.25, + "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453, + "mmlu_eval_accuracy_college_mathematics": 0.09090909090909091, + "mmlu_eval_accuracy_college_medicine": 0.5, + "mmlu_eval_accuracy_college_physics": 0.45454545454545453, + "mmlu_eval_accuracy_computer_security": 0.36363636363636365, + "mmlu_eval_accuracy_conceptual_physics": 0.2692307692307692, + "mmlu_eval_accuracy_econometrics": 0.4166666666666667, + "mmlu_eval_accuracy_electrical_engineering": 0.1875, + "mmlu_eval_accuracy_elementary_mathematics": 0.2682926829268293, + "mmlu_eval_accuracy_formal_logic": 0.42857142857142855, + "mmlu_eval_accuracy_global_facts": 0.1, + "mmlu_eval_accuracy_high_school_biology": 0.3125, + "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727, + "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556, + "mmlu_eval_accuracy_high_school_european_history": 0.3333333333333333, + "mmlu_eval_accuracy_high_school_geography": 0.6363636363636364, + "mmlu_eval_accuracy_high_school_government_and_politics": 0.42857142857142855, + "mmlu_eval_accuracy_high_school_macroeconomics": 0.3488372093023256, + "mmlu_eval_accuracy_high_school_mathematics": 0.3448275862068966, + "mmlu_eval_accuracy_high_school_microeconomics": 0.5, + "mmlu_loss": 0.6973872357606887, + "step": 2400 + } + ], + "max_steps": 30000, + "num_train_epochs": 22, + "total_flos": 4.768354997549138e+17, + "trial_name": null, + "trial_params": null +}