Upload model

.gitattributes

@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

config.json

@@ -0,0 +1,31 @@
+{
+  "_name_or_path": "google/mt5-small",
+  "architectures": [
+    "MT5ForConditionalGeneration"
+  ],
+  "d_ff": 1024,
+  "d_kv": 64,
+  "d_model": 512,
+  "decoder_start_token_id": 0,
+  "dense_act_fn": "gelu_new",
+  "dropout_rate": 0.1,
+  "eos_token_id": 1,
+  "feed_forward_proj": "gated-gelu",
+  "initializer_factor": 1.0,
+  "is_encoder_decoder": true,
+  "is_gated_act": true,
+  "layer_norm_epsilon": 1e-06,
+  "model_type": "mt5",
+  "num_decoder_layers": 8,
+  "num_heads": 6,
+  "num_layers": 8,
+  "pad_token_id": 0,
+  "relative_attention_max_distance": 128,
+  "relative_attention_num_buckets": 32,
+  "tie_word_embeddings": false,
+  "tokenizer_class": "T5Tokenizer",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.23.0.dev0",
+  "use_cache": true,
+  "vocab_size": 250100
+}

latest

@@ -0,0 +1 @@
+global_step11400

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e2995b5128e86f8cad57af7249371d9ac4d22bcdb290dc61313362a5d850344
+size 600356075

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3581f24b21aff97eddbb4e839feb42bc13bd5d0c5ea2970f39df72a3dc6d23e4
+size 14503

special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

spiece.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef78f86560d809067d12bac6c09f19a462cb3af3f54d2b8acbba26e1433125d6
+size 4309802

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93c3578052e1605d8332eb961bc08d72e246071974e4cc54aa6991826b802aa5
+size 16330369

tokenizer_config.json

@@ -0,0 +1,11 @@
+{
+  "additional_special_tokens": null,
+  "eos_token": "</s>",
+  "extra_ids": 0,
+  "name_or_path": "google/mt5-small",
+  "pad_token": "<pad>",
+  "sp_model_kwargs": {},
+  "special_tokens_map_file": "/home/txau/.cache/huggingface/hub/models--google--mt5-small/snapshots/f03a52d3eaa650878b6f52e443bc4d5b385e786e/special_tokens_map.json",
+  "tokenizer_class": "T5Tokenizer",
+  "unk_token": "<unk>"
+}

trainer_state.json

@@ -0,0 +1,1972 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.453662604578592,
+  "global_step": 11400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.02,
+      "eval_exact_match": 14.738231412230533,
+      "eval_f1": 19.992446610541194,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0005,
+      "loss": 3.6292,
+      "step": 200
+    },
+    {
+      "epoch": 0.04,
+      "eval_exact_match": 21.381434227892655,
+      "eval_f1": 31.43956554541397,
+      "step": 200
+    },
+    {
+      "epoch": 0.05,
+      "eval_exact_match": 24.175098988121427,
+      "eval_f1": 34.101042049094296,
+      "step": 300
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0005,
+      "loss": 2.7193,
+      "step": 400
+    },
+    {
+      "epoch": 0.07,
+      "eval_exact_match": 24.950592885375492,
+      "eval_f1": 38.217244544211226,
+      "step": 400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0005,
+      "loss": 2.4559,
+      "step": 600
+    },
+    {
+      "epoch": 0.11,
+      "eval_exact_match": 28.40909090909091,
+      "eval_f1": 44.64064089265683,
+      "step": 600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0005,
+      "loss": 2.2801,
+      "step": 800
+    },
+    {
+      "epoch": 0.14,
+      "eval_exact_match": 26.630434782608695,
+      "eval_f1": 45.504003307885625,
+      "step": 800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0005,
+      "loss": 2.1285,
+      "step": 1000
+    },
+    {
+      "epoch": 0.18,
+      "eval_exact_match": 30.8300395256917,
+      "eval_f1": 50.359736600917515,
+      "step": 1000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0005,
+      "loss": 2.026,
+      "step": 1200
+    },
+    {
+      "epoch": 0.21,
+      "eval_exact_match": 31.175889328063242,
+      "eval_f1": 51.623685774579855,
+      "step": 1200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0005,
+      "loss": 1.9281,
+      "step": 1400
+    },
+    {
+      "epoch": 0.25,
+      "eval_exact_match": 35.869565217391305,
+      "eval_f1": 56.21242269358832,
+      "step": 1400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0005,
+      "loss": 1.8836,
+      "step": 1600
+    },
+    {
+      "epoch": 0.28,
+      "eval_exact_match": 35.0296442687747,
+      "eval_f1": 57.21764925319378,
+      "step": 1600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0005,
+      "loss": 1.8377,
+      "step": 1800
+    },
+    {
+      "epoch": 0.32,
+      "eval_exact_match": 36.61067193675889,
+      "eval_f1": 58.4380904170952,
+      "step": 1800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0005,
+      "loss": 1.7926,
+      "step": 2000
+    },
+    {
+      "epoch": 0.35,
+      "eval_exact_match": 36.85770750988142,
+      "eval_f1": 59.677808943848156,
+      "step": 2000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0005,
+      "loss": 1.7356,
+      "step": 2200
+    },
+    {
+      "epoch": 0.39,
+      "eval_exact_match": 35.62252964426877,
+      "eval_f1": 59.02520911162799,
+      "step": 2200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0005,
+      "loss": 1.7001,
+      "step": 2400
+    },
+    {
+      "epoch": 0.42,
+      "eval_exact_match": 36.31422924901186,
+      "eval_f1": 60.89754377178488,
+      "step": 2400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0005,
+      "loss": 1.675,
+      "step": 2600
+    },
+    {
+      "epoch": 0.46,
+      "eval_exact_match": 36.26482213438735,
+      "eval_f1": 59.783069358423525,
+      "step": 2600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0005,
+      "loss": 1.6594,
+      "step": 2800
+    },
+    {
+      "epoch": 0.49,
+      "eval_exact_match": 36.56126482213439,
+      "eval_f1": 61.449547007432926,
+      "step": 2800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0005,
+      "loss": 1.6435,
+      "step": 3000
+    },
+    {
+      "epoch": 0.53,
+      "eval_exact_match": 39.92094861660079,
+      "eval_f1": 63.03510761568375,
+      "step": 3000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0005,
+      "loss": 1.6116,
+      "step": 3200
+    },
+    {
+      "epoch": 0.56,
+      "eval_exact_match": 37.30237154150198,
+      "eval_f1": 62.953985227470696,
+      "step": 3200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0005,
+      "loss": 1.5701,
+      "step": 3400
+    },
+    {
+      "epoch": 0.6,
+      "eval_exact_match": 36.6600790513834,
+      "eval_f1": 61.820603199741704,
+      "step": 3400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0005,
+      "loss": 1.5627,
+      "step": 3600
+    },
+    {
+      "epoch": 0.63,
+      "eval_exact_match": 38.04347826086956,
+      "eval_f1": 63.56696764038496,
+      "step": 3600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0005,
+      "loss": 1.5525,
+      "step": 3800
+    },
+    {
+      "epoch": 0.67,
+      "eval_exact_match": 39.03162055335969,
+      "eval_f1": 64.72028168259759,
+      "step": 3800
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0005,
+      "loss": 1.4518,
+      "step": 3850
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0005,
+      "loss": 1.4245,
+      "step": 3900
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0005,
+      "loss": 1.4164,
+      "step": 3950
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0005,
+      "loss": 1.4627,
+      "step": 4000
+    },
+    {
+      "epoch": 0.86,
+      "eval_exact_match": 42.5,
+      "eval_f1": 67.3159921257088,
+      "step": 4000
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0005,
+      "loss": 1.4369,
+      "step": 4050
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0005,
+      "loss": 1.4205,
+      "step": 4100
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0005,
+      "loss": 1.4189,
+      "step": 4150
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0005,
+      "loss": 1.4127,
+      "step": 4200
+    },
+    {
+      "epoch": 0.9,
+      "eval_exact_match": 44.95,
+      "eval_f1": 68.47203537029696,
+      "step": 4200
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0005,
+      "loss": 1.4329,
+      "step": 4250
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0005,
+      "loss": 1.4233,
+      "step": 4300
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0005,
+      "loss": 1.4426,
+      "step": 4350
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0005,
+      "loss": 1.3969,
+      "step": 4400
+    },
+    {
+      "epoch": 0.95,
+      "eval_exact_match": 46.9,
+      "eval_f1": 69.7906665651929,
+      "step": 4400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0005,
+      "loss": 1.4589,
+      "step": 4450
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0005,
+      "loss": 1.4199,
+      "step": 4500
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0005,
+      "loss": 1.3795,
+      "step": 4550
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0005,
+      "loss": 1.3794,
+      "step": 4600
+    },
+    {
+      "epoch": 0.99,
+      "eval_exact_match": 50.95,
+      "eval_f1": 71.26934663680022,
+      "step": 4600
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0005,
+      "loss": 1.3577,
+      "step": 4650
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0005,
+      "loss": 1.3633,
+      "step": 4700
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0005,
+      "loss": 1.3672,
+      "step": 4750
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0005,
+      "loss": 1.3568,
+      "step": 4800
+    },
+    {
+      "epoch": 1.03,
+      "eval_exact_match": 49.5,
+      "eval_f1": 71.44716194869687,
+      "step": 4800
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0005,
+      "loss": 1.334,
+      "step": 4850
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.0005,
+      "loss": 1.3173,
+      "step": 4900
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0005,
+      "loss": 1.3332,
+      "step": 4950
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0005,
+      "loss": 1.3858,
+      "step": 5000
+    },
+    {
+      "epoch": 1.08,
+      "eval_exact_match": 50.6,
+      "eval_f1": 71.4031953884389,
+      "step": 5000
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0005,
+      "loss": 1.3724,
+      "step": 5050
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0005,
+      "loss": 1.3356,
+      "step": 5100
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.0005,
+      "loss": 1.4173,
+      "step": 5150
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0005,
+      "loss": 1.3499,
+      "step": 5200
+    },
+    {
+      "epoch": 1.12,
+      "eval_exact_match": 50.45,
+      "eval_f1": 70.96365965036652,
+      "step": 5200
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0005,
+      "loss": 1.3853,
+      "step": 5250
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0005,
+      "loss": 1.337,
+      "step": 5300
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0005,
+      "loss": 1.3366,
+      "step": 5350
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0005,
+      "loss": 1.342,
+      "step": 5400
+    },
+    {
+      "epoch": 1.16,
+      "eval_exact_match": 51.25,
+      "eval_f1": 72.29202655492661,
+      "step": 5400
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0005,
+      "loss": 1.3289,
+      "step": 5450
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0005,
+      "loss": 1.3894,
+      "step": 5500
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0005,
+      "loss": 1.3393,
+      "step": 5550
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0005,
+      "loss": 1.2983,
+      "step": 5600
+    },
+    {
+      "epoch": 1.21,
+      "eval_exact_match": 52.15,
+      "eval_f1": 72.57033769389315,
+      "step": 5600
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0005,
+      "loss": 1.3181,
+      "step": 5650
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0005,
+      "loss": 1.3672,
+      "step": 5700
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0005,
+      "loss": 1.3064,
+      "step": 5750
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0005,
+      "loss": 1.3316,
+      "step": 5800
+    },
+    {
+      "epoch": 1.25,
+      "eval_exact_match": 52.6,
+      "eval_f1": 72.91182293082619,
+      "step": 5800
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.0005,
+      "loss": 1.3099,
+      "step": 5850
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.0005,
+      "loss": 1.293,
+      "step": 5900
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0005,
+      "loss": 1.3386,
+      "step": 5950
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.0005,
+      "loss": 1.3475,
+      "step": 6000
+    },
+    {
+      "epoch": 1.29,
+      "eval_exact_match": 50.7,
+      "eval_f1": 72.13671184582012,
+      "step": 6000
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0005,
+      "loss": 1.2548,
+      "step": 6050
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0005,
+      "loss": 1.3243,
+      "step": 6100
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0005,
+      "loss": 1.322,
+      "step": 6150
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.0005,
+      "loss": 1.3049,
+      "step": 6200
+    },
+    {
+      "epoch": 1.33,
+      "eval_exact_match": 51.25,
+      "eval_f1": 72.62424474108035,
+      "step": 6200
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.0005,
+      "loss": 1.274,
+      "step": 6250
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0005,
+      "loss": 1.3015,
+      "step": 6300
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0005,
+      "loss": 1.303,
+      "step": 6350
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0005,
+      "loss": 1.2727,
+      "step": 6400
+    },
+    {
+      "epoch": 1.38,
+      "eval_exact_match": 52.35,
+      "eval_f1": 73.1319111267331,
+      "step": 6400
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0005,
+      "loss": 1.2866,
+      "step": 6450
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0005,
+      "loss": 1.3302,
+      "step": 6500
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0005,
+      "loss": 1.2281,
+      "step": 6550
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0005,
+      "loss": 1.3037,
+      "step": 6600
+    },
+    {
+      "epoch": 1.42,
+      "eval_exact_match": 53.2,
+      "eval_f1": 73.01971722088678,
+      "step": 6600
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.0005,
+      "loss": 1.2822,
+      "step": 6650
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0005,
+      "loss": 1.2871,
+      "step": 6700
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0005,
+      "loss": 1.3176,
+      "step": 6750
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0005,
+      "loss": 1.2828,
+      "step": 6800
+    },
+    {
+      "epoch": 1.46,
+      "eval_exact_match": 52.85,
+      "eval_f1": 73.01360270382577,
+      "step": 6800
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0005,
+      "loss": 1.3448,
+      "step": 6850
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.0005,
+      "loss": 1.3102,
+      "step": 6900
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0005,
+      "loss": 1.3273,
+      "step": 6950
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0005,
+      "loss": 1.3016,
+      "step": 7000
+    },
+    {
+      "epoch": 1.51,
+      "eval_exact_match": 54.0,
+      "eval_f1": 73.98223052649728,
+      "step": 7000
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0005,
+      "loss": 1.2767,
+      "step": 7050
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0005,
+      "loss": 1.262,
+      "step": 7100
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0005,
+      "loss": 1.2846,
+      "step": 7150
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0005,
+      "loss": 1.2642,
+      "step": 7200
+    },
+    {
+      "epoch": 1.55,
+      "eval_exact_match": 52.45,
+      "eval_f1": 73.66687625952305,
+      "step": 7200
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0005,
+      "loss": 1.2875,
+      "step": 7250
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0005,
+      "loss": 1.247,
+      "step": 7300
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0005,
+      "loss": 1.2501,
+      "step": 7350
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0005,
+      "loss": 1.293,
+      "step": 7400
+    },
+    {
+      "epoch": 1.59,
+      "eval_exact_match": 52.3,
+      "eval_f1": 73.46398897242841,
+      "step": 7400
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0005,
+      "loss": 1.2373,
+      "step": 7450
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0005,
+      "loss": 1.2477,
+      "step": 7500
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0005,
+      "loss": 1.2647,
+      "step": 7550
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0005,
+      "loss": 1.2947,
+      "step": 7600
+    },
+    {
+      "epoch": 1.64,
+      "eval_exact_match": 53.05,
+      "eval_f1": 73.40471749318357,
+      "step": 7600
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0005,
+      "loss": 1.2237,
+      "step": 7650
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0005,
+      "loss": 1.2996,
+      "step": 7700
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0005,
+      "loss": 1.2833,
+      "step": 7750
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0005,
+      "loss": 1.2663,
+      "step": 7800
+    },
+    {
+      "epoch": 1.68,
+      "eval_exact_match": 53.95,
+      "eval_f1": 74.59089034170034,
+      "step": 7800
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0005,
+      "loss": 1.2519,
+      "step": 7850
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0005,
+      "loss": 1.2365,
+      "step": 7900
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0005,
+      "loss": 1.2152,
+      "step": 7950
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0005,
+      "loss": 1.2309,
+      "step": 8000
+    },
+    {
+      "epoch": 1.72,
+      "eval_exact_match": 53.35,
+      "eval_f1": 74.02883287579421,
+      "step": 8000
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0005,
+      "loss": 1.2532,
+      "step": 8050
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0005,
+      "loss": 1.2607,
+      "step": 8100
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0005,
+      "loss": 1.1885,
+      "step": 8150
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0005,
+      "loss": 1.2169,
+      "step": 8200
+    },
+    {
+      "epoch": 1.76,
+      "eval_exact_match": 54.35,
+      "eval_f1": 74.273655582778,
+      "step": 8200
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0005,
+      "loss": 1.2543,
+      "step": 8250
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0005,
+      "loss": 1.2352,
+      "step": 8300
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0005,
+      "loss": 1.2142,
+      "step": 8350
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0005,
+      "loss": 1.223,
+      "step": 8400
+    },
+    {
+      "epoch": 1.81,
+      "eval_exact_match": 52.7,
+      "eval_f1": 73.99393883664169,
+      "step": 8400
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0005,
+      "loss": 1.2145,
+      "step": 8450
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0005,
+      "loss": 1.1973,
+      "step": 8500
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0005,
+      "loss": 1.2277,
+      "step": 8550
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0005,
+      "loss": 1.2552,
+      "step": 8600
+    },
+    {
+      "epoch": 1.85,
+      "eval_exact_match": 53.8,
+      "eval_f1": 74.61861738151018,
+      "step": 8600
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0005,
+      "loss": 1.228,
+      "step": 8650
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0005,
+      "loss": 1.2117,
+      "step": 8700
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0005,
+      "loss": 1.2429,
+      "step": 8750
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0005,
+      "loss": 1.2279,
+      "step": 8800
+    },
+    {
+      "epoch": 1.89,
+      "eval_exact_match": 54.4,
+      "eval_f1": 74.93345914261613,
+      "step": 8800
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0005,
+      "loss": 1.2134,
+      "step": 8850
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0005,
+      "loss": 1.245,
+      "step": 8900
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0005,
+      "loss": 1.2179,
+      "step": 8950
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0005,
+      "loss": 1.2006,
+      "step": 9000
+    },
+    {
+      "epoch": 1.94,
+      "eval_exact_match": 55.4,
+      "eval_f1": 75.75427604493635,
+      "step": 9000
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0005,
+      "loss": 1.2037,
+      "step": 9050
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0005,
+      "loss": 1.1985,
+      "step": 9100
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0005,
+      "loss": 1.2012,
+      "step": 9150
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0005,
+      "loss": 1.2194,
+      "step": 9200
+    },
+    {
+      "epoch": 1.98,
+      "eval_exact_match": 54.0,
+      "eval_f1": 74.89545612203896,
+      "step": 9200
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0005,
+      "loss": 1.2173,
+      "step": 9250
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0005,
+      "loss": 1.2616,
+      "step": 9300
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0005,
+      "loss": 1.1785,
+      "step": 9350
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0005,
+      "loss": 1.1664,
+      "step": 9400
+    },
+    {
+      "epoch": 2.02,
+      "eval_exact_match": 55.2,
+      "eval_f1": 75.50544659346157,
+      "step": 9400
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0005,
+      "loss": 1.1271,
+      "step": 9450
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0005,
+      "loss": 1.1806,
+      "step": 9500
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0005,
+      "loss": 1.1865,
+      "step": 9550
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0005,
+      "loss": 1.198,
+      "step": 9600
+    },
+    {
+      "epoch": 2.07,
+      "eval_exact_match": 54.65,
+      "eval_f1": 75.02151439633785,
+      "step": 9600
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0005,
+      "loss": 1.1726,
+      "step": 9650
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0005,
+      "loss": 1.1799,
+      "step": 9700
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0005,
+      "loss": 1.1107,
+      "step": 9750
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0005,
+      "loss": 1.1585,
+      "step": 9800
+    },
+    {
+      "epoch": 2.11,
+      "eval_exact_match": 53.9,
+      "eval_f1": 74.15646190187827,
+      "step": 9800
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0005,
+      "loss": 1.1436,
+      "step": 9850
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0005,
+      "loss": 1.1806,
+      "step": 9900
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0005,
+      "loss": 1.1444,
+      "step": 9950
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0005,
+      "loss": 1.1738,
+      "step": 10000
+    },
+    {
+      "epoch": 2.15,
+      "eval_exact_match": 54.85,
+      "eval_f1": 75.11255802008874,
+      "step": 10000
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0005,
+      "loss": 1.1095,
+      "step": 10050
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0005,
+      "loss": 1.1511,
+      "step": 10100
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0005,
+      "loss": 1.1297,
+      "step": 10150
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0005,
+      "loss": 1.1505,
+      "step": 10200
+    },
+    {
+      "epoch": 2.2,
+      "eval_exact_match": 54.85,
+      "eval_f1": 75.52882358671673,
+      "step": 10200
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 5e-05,
+      "loss": 1.1081,
+      "step": 10210
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 5e-05,
+      "loss": 1.1572,
+      "step": 10220
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 5e-05,
+      "loss": 1.0839,
+      "step": 10230
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 5e-05,
+      "loss": 1.0971,
+      "step": 10240
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 5e-05,
+      "loss": 1.0737,
+      "step": 10250
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 5e-05,
+      "loss": 1.1049,
+      "step": 10260
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 5e-05,
+      "loss": 1.202,
+      "step": 10270
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 5e-05,
+      "loss": 1.1848,
+      "step": 10280
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 5e-05,
+      "loss": 1.1095,
+      "step": 10290
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 5e-05,
+      "loss": 1.2268,
+      "step": 10300
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 5e-05,
+      "loss": 1.1422,
+      "step": 10310
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 5e-05,
+      "loss": 1.1222,
+      "step": 10320
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 5e-05,
+      "loss": 1.1251,
+      "step": 10330
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5e-05,
+      "loss": 1.0694,
+      "step": 10340
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5e-05,
+      "loss": 1.1571,
+      "step": 10350
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5e-05,
+      "loss": 1.1546,
+      "step": 10360
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5e-05,
+      "loss": 1.2171,
+      "step": 10370
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5e-05,
+      "loss": 1.1283,
+      "step": 10380
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 5e-05,
+      "loss": 1.1215,
+      "step": 10390
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 5e-05,
+      "loss": 1.127,
+      "step": 10400
+    },
+    {
+      "epoch": 2.24,
+      "eval_exact_match": 55.35,
+      "eval_f1": 75.70498337751204,
+      "step": 10400
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 5e-05,
+      "loss": 1.0492,
+      "step": 10410
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 5e-05,
+      "loss": 1.1658,
+      "step": 10420
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 5e-05,
+      "loss": 1.1677,
+      "step": 10430
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 5e-05,
+      "loss": 1.1758,
+      "step": 10440
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 5e-05,
+      "loss": 1.1968,
+      "step": 10450
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 5e-05,
+      "loss": 1.1315,
+      "step": 10460
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 5e-05,
+      "loss": 1.1138,
+      "step": 10470
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5e-05,
+      "loss": 1.1375,
+      "step": 10480
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5e-05,
+      "loss": 1.1892,
+      "step": 10490
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5e-05,
+      "loss": 1.131,
+      "step": 10500
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5e-05,
+      "loss": 1.1195,
+      "step": 10510
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5e-05,
+      "loss": 1.1044,
+      "step": 10520
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 5e-05,
+      "loss": 1.1371,
+      "step": 10530
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 5e-05,
+      "loss": 1.1382,
+      "step": 10540
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 5e-05,
+      "loss": 1.1273,
+      "step": 10550
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 5e-05,
+      "loss": 1.1566,
+      "step": 10560
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 5e-05,
+      "loss": 1.209,
+      "step": 10570
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 5e-05,
+      "loss": 1.1336,
+      "step": 10580
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 5e-05,
+      "loss": 1.0522,
+      "step": 10590
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 5e-05,
+      "loss": 1.2019,
+      "step": 10600
+    },
+    {
+      "epoch": 2.28,
+      "eval_exact_match": 55.55,
+      "eval_f1": 76.02151178378499,
+      "step": 10600
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 5e-05,
+      "loss": 1.1227,
+      "step": 10610
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 5e-05,
+      "loss": 1.2202,
+      "step": 10620
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 5e-05,
+      "loss": 1.0762,
+      "step": 10630
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 5e-05,
+      "loss": 1.2074,
+      "step": 10640
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 5e-05,
+      "loss": 1.1283,
+      "step": 10650
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 5e-05,
+      "loss": 1.1631,
+      "step": 10660
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 5e-05,
+      "loss": 1.0954,
+      "step": 10670
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 5e-05,
+      "loss": 1.2253,
+      "step": 10680
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 5e-05,
+      "loss": 1.1141,
+      "step": 10690
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 5e-05,
+      "loss": 1.2454,
+      "step": 10700
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 5e-05,
+      "loss": 1.0997,
+      "step": 10710
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 5e-05,
+      "loss": 1.1408,
+      "step": 10720
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 5e-05,
+      "loss": 1.098,
+      "step": 10730
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 5e-05,
+      "loss": 1.2331,
+      "step": 10740
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 5e-05,
+      "loss": 1.1006,
+      "step": 10750
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 5e-05,
+      "loss": 1.1057,
+      "step": 10760
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 5e-05,
+      "loss": 1.2075,
+      "step": 10770
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 5e-05,
+      "loss": 1.0302,
+      "step": 10780
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 5e-05,
+      "loss": 1.1484,
+      "step": 10790
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 5e-05,
+      "loss": 1.1272,
+      "step": 10800
+    },
+    {
+      "epoch": 2.32,
+      "eval_exact_match": 55.6,
+      "eval_f1": 75.90896740958824,
+      "step": 10800
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 5e-05,
+      "loss": 1.1306,
+      "step": 10810
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 5e-05,
+      "loss": 1.1356,
+      "step": 10820
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 5e-05,
+      "loss": 1.1776,
+      "step": 10830
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 5e-05,
+      "loss": 1.1216,
+      "step": 10840
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 5e-05,
+      "loss": 1.1211,
+      "step": 10850
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 5e-05,
+      "loss": 1.2693,
+      "step": 10860
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 5e-05,
+      "loss": 1.1473,
+      "step": 10870
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 5e-05,
+      "loss": 1.0641,
+      "step": 10880
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 5e-05,
+      "loss": 1.2424,
+      "step": 10890
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 5e-05,
+      "loss": 1.1371,
+      "step": 10900
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 5e-05,
+      "loss": 1.1217,
+      "step": 10910
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 5e-05,
+      "loss": 1.2007,
+      "step": 10920
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 5e-05,
+      "loss": 1.1501,
+      "step": 10930
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 5e-05,
+      "loss": 1.1135,
+      "step": 10940
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 5e-05,
+      "loss": 1.137,
+      "step": 10950
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 5e-05,
+      "loss": 1.1422,
+      "step": 10960
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 5e-05,
+      "loss": 1.1248,
+      "step": 10970
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 5e-05,
+      "loss": 1.1899,
+      "step": 10980
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 5e-05,
+      "loss": 1.102,
+      "step": 10990
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 5e-05,
+      "loss": 1.1877,
+      "step": 11000
+    },
+    {
+      "epoch": 2.37,
+      "eval_exact_match": 55.95,
+      "eval_f1": 76.45373153349801,
+      "step": 11000
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 5e-05,
+      "loss": 1.1474,
+      "step": 11010
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 5e-05,
+      "loss": 1.1536,
+      "step": 11020
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 5e-05,
+      "loss": 1.2218,
+      "step": 11030
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 5e-05,
+      "loss": 1.0967,
+      "step": 11040
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 5e-05,
+      "loss": 1.1305,
+      "step": 11050
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 5e-05,
+      "loss": 1.087,
+      "step": 11060
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 5e-05,
+      "loss": 1.0908,
+      "step": 11070
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 5e-05,
+      "loss": 1.0134,
+      "step": 11080
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 5e-05,
+      "loss": 1.0997,
+      "step": 11090
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 5e-05,
+      "loss": 1.066,
+      "step": 11100
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 5e-05,
+      "loss": 1.2601,
+      "step": 11110
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 5e-05,
+      "loss": 1.1191,
+      "step": 11120
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 5e-05,
+      "loss": 1.1025,
+      "step": 11130
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 5e-05,
+      "loss": 1.112,
+      "step": 11140
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 5e-05,
+      "loss": 1.0794,
+      "step": 11150
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 5e-05,
+      "loss": 1.112,
+      "step": 11160
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 5e-05,
+      "loss": 1.1411,
+      "step": 11170
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 5e-05,
+      "loss": 1.1118,
+      "step": 11180
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 5e-05,
+      "loss": 1.1464,
+      "step": 11190
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 5e-05,
+      "loss": 1.1496,
+      "step": 11200
+    },
+    {
+      "epoch": 2.41,
+      "eval_exact_match": 56.15,
+      "eval_f1": 76.4356306603872,
+      "step": 11200
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 5e-05,
+      "loss": 1.1588,
+      "step": 11210
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 5e-05,
+      "loss": 1.1369,
+      "step": 11220
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 5e-05,
+      "loss": 1.1026,
+      "step": 11230
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 5e-05,
+      "loss": 1.1764,
+      "step": 11240
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 5e-05,
+      "loss": 1.1449,
+      "step": 11250
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 5e-05,
+      "loss": 1.1712,
+      "step": 11260
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 5e-05,
+      "loss": 1.0465,
+      "step": 11270
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 5e-05,
+      "loss": 1.1362,
+      "step": 11280
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 5e-05,
+      "loss": 1.1022,
+      "step": 11290
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 5e-05,
+      "loss": 1.1401,
+      "step": 11300
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 5e-05,
+      "loss": 1.0913,
+      "step": 11310
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 5e-05,
+      "loss": 1.1922,
+      "step": 11320
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 5e-05,
+      "loss": 1.101,
+      "step": 11330
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 5e-05,
+      "loss": 1.0261,
+      "step": 11340
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 5e-05,
+      "loss": 1.1109,
+      "step": 11350
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 5e-05,
+      "loss": 1.0958,
+      "step": 11360
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 5e-05,
+      "loss": 1.027,
+      "step": 11370
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 5e-05,
+      "loss": 1.117,
+      "step": 11380
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 5e-05,
+      "loss": 1.1295,
+      "step": 11390
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 5e-05,
+      "loss": 1.1344,
+      "step": 11400
+    },
+    {
+      "epoch": 2.45,
+      "eval_exact_match": 56.45,
+      "eval_f1": 76.81870981288014,
+      "step": 11400
+    }
+  ],
+  "max_steps": 92920,
+  "num_train_epochs": 20,
+  "total_flos": 3.858021871727411e+17,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a786dd68f2e8e10a738c3db218d056e17d76ec9e58bf6cc1a3e984cb86422ae3
+size 4591

zero_to_fp32.py

@@ -0,0 +1,482 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION,
+                                            OPTIMIZER_STATE_DICT,
+                                            SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS,
+                                            ZERO_STAGE,
+                                            PARTITION_COUNT,
+                                            PARAM_SHAPES,
+                                            BUFFER_NAMES)
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
+                                                "*_optim_states.pt")),
+                         key=natural_keys)
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    state_dict = torch.load(file, map_location=device)
+
+    if BUFFER_NAMES not in state_dict:
+        raise ValueError(f"{file} is not a model state checkpoint")
+    buffer_names = state_dict[BUFFER_NAMES]
+    if debug:
+        print("Found buffers:", buffer_names)
+
+    # recover just the buffers while restoring them to fp32 if they were saved in fp16
+    buffers = {
+        k: v.float()
+        for k,
+        v in state_dict["module"].items() if k in buffer_names
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(
+                    f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum([
+        full_single_fp32_vector.numel()
+        for full_single_fp32_vector in merged_single_partition_of_fp32_groups
+    ])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum(
+            [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(
+                    f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
+                )
+            state_dict[name] = full_single_fp32_vector.narrow(
+                0,
+                offset,
+                unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(
+                f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0,
+                                             offset,
+                                             partitioned_numel)
+                  for i in range(world_size)),
+            0).narrow(0,
+                      0,
+                      unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(
+            f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)