BobaZooba commited on
Commit
b6869f9
1 Parent(s): e12ae32

Training in progress, step 300, checkpoint

Browse files
last-checkpoint/README.md ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ ---
4
+ ## Training procedure
5
+
6
+
7
+ The following `bitsandbytes` quantization config was used during training:
8
+ - quant_method: bitsandbytes
9
+ - load_in_8bit: False
10
+ - load_in_4bit: True
11
+ - llm_int8_threshold: 6.0
12
+ - llm_int8_skip_modules: None
13
+ - llm_int8_enable_fp32_cpu_offload: False
14
+ - llm_int8_has_fp16_weight: True
15
+ - bnb_4bit_quant_type: nf4
16
+ - bnb_4bit_use_double_quant: True
17
+ - bnb_4bit_compute_dtype: bfloat16
18
+
19
+ The following `bitsandbytes` quantization config was used during training:
20
+ - quant_method: bitsandbytes
21
+ - load_in_8bit: False
22
+ - load_in_4bit: True
23
+ - llm_int8_threshold: 6.0
24
+ - llm_int8_skip_modules: None
25
+ - llm_int8_enable_fp32_cpu_offload: False
26
+ - llm_int8_has_fp16_weight: True
27
+ - bnb_4bit_quant_type: nf4
28
+ - bnb_4bit_use_double_quant: True
29
+ - bnb_4bit_compute_dtype: bfloat16
30
+
31
+ The following `bitsandbytes` quantization config was used during training:
32
+ - quant_method: bitsandbytes
33
+ - load_in_8bit: False
34
+ - load_in_4bit: True
35
+ - llm_int8_threshold: 6.0
36
+ - llm_int8_skip_modules: None
37
+ - llm_int8_enable_fp32_cpu_offload: False
38
+ - llm_int8_has_fp16_weight: True
39
+ - bnb_4bit_quant_type: nf4
40
+ - bnb_4bit_use_double_quant: True
41
+ - bnb_4bit_compute_dtype: bfloat16
42
+
43
+ The following `bitsandbytes` quantization config was used during training:
44
+ - quant_method: bitsandbytes
45
+ - load_in_8bit: False
46
+ - load_in_4bit: True
47
+ - llm_int8_threshold: 6.0
48
+ - llm_int8_skip_modules: None
49
+ - llm_int8_enable_fp32_cpu_offload: False
50
+ - llm_int8_has_fp16_weight: True
51
+ - bnb_4bit_quant_type: nf4
52
+ - bnb_4bit_use_double_quant: True
53
+ - bnb_4bit_compute_dtype: bfloat16
54
+ ### Framework versions
55
+
56
+ - PEFT 0.5.0
57
+ - PEFT 0.5.0
58
+ - PEFT 0.5.0
59
+
60
+ - PEFT 0.5.0
last-checkpoint/adapter_config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "auto_mapping": null,
3
+ "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
4
+ "bias": "none",
5
+ "fan_in_fan_out": false,
6
+ "inference_mode": true,
7
+ "init_lora_weights": true,
8
+ "layers_pattern": null,
9
+ "layers_to_transform": null,
10
+ "lora_alpha": 32,
11
+ "lora_dropout": 0.1,
12
+ "modules_to_save": null,
13
+ "peft_type": "LORA",
14
+ "r": 64,
15
+ "revision": null,
16
+ "target_modules": [
17
+ "q_proj",
18
+ "v_proj",
19
+ "up_proj",
20
+ "k_proj",
21
+ "o_proj",
22
+ "gate_proj",
23
+ "down_proj"
24
+ ],
25
+ "task_type": "CAUSAL_LM"
26
+ }
last-checkpoint/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3292b605913b97d05f8a16583a995dc94fc1588668a402e1784a8e0868dee3df
3
+ size 335605144
last-checkpoint/global_step300/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:70fe0a71c8f26897be2331c9786b68c4e3482c41ef6d0a297eaa3066626d6a24
3
+ size 1006661136
last-checkpoint/global_step300/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dd822321034aab707986931ca61ce44b4fb3a9646458d392113c239f8245c51c
3
+ size 1006661264
last-checkpoint/global_step300/mp_rank_00_model_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:603c77216377256c089cc0f1eac3f60042a6c86407f3a941e089fe1b31998d0b
3
+ size 8365038600
last-checkpoint/latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step300
last-checkpoint/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2981b7d07dbd40ffa300bdce2b2824c636c4c181ebf87cf28fe3945ccbe0c23b
3
+ size 14512
last-checkpoint/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04ee61073b659003784ef66df69f7c6a27137bf83502c0cfaa924c54bd6a6078
3
+ size 14512
last-checkpoint/trainer_state.json ADDED
@@ -0,0 +1,1827 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 1.3873900175094604,
3
+ "best_model_checkpoint": "./outputs/checkpoint-300",
4
+ "epoch": 1.6666666666666665,
5
+ "eval_steps": 300,
6
+ "global_step": 300,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.01,
13
+ "learning_rate": 0.0,
14
+ "loss": 2.8496,
15
+ "step": 1
16
+ },
17
+ {
18
+ "epoch": 0.01,
19
+ "learning_rate": 3.0102999566398115e-05,
20
+ "loss": 2.3809,
21
+ "step": 2
22
+ },
23
+ {
24
+ "epoch": 0.02,
25
+ "learning_rate": 4.771212547196624e-05,
26
+ "loss": 2.3604,
27
+ "step": 3
28
+ },
29
+ {
30
+ "epoch": 0.02,
31
+ "learning_rate": 6.020599913279623e-05,
32
+ "loss": 2.2425,
33
+ "step": 4
34
+ },
35
+ {
36
+ "epoch": 0.03,
37
+ "learning_rate": 6.989700043360187e-05,
38
+ "loss": 2.0888,
39
+ "step": 5
40
+ },
41
+ {
42
+ "epoch": 0.03,
43
+ "learning_rate": 7.781512503836436e-05,
44
+ "loss": 1.9669,
45
+ "step": 6
46
+ },
47
+ {
48
+ "epoch": 0.04,
49
+ "learning_rate": 8.450980400142567e-05,
50
+ "loss": 1.8589,
51
+ "step": 7
52
+ },
53
+ {
54
+ "epoch": 0.04,
55
+ "learning_rate": 9.030899869919434e-05,
56
+ "loss": 1.7722,
57
+ "step": 8
58
+ },
59
+ {
60
+ "epoch": 0.05,
61
+ "learning_rate": 9.542425094393248e-05,
62
+ "loss": 1.6765,
63
+ "step": 9
64
+ },
65
+ {
66
+ "epoch": 0.06,
67
+ "learning_rate": 9.999999999999999e-05,
68
+ "loss": 1.611,
69
+ "step": 10
70
+ },
71
+ {
72
+ "epoch": 0.06,
73
+ "learning_rate": 0.00010413926851582249,
74
+ "loss": 1.5547,
75
+ "step": 11
76
+ },
77
+ {
78
+ "epoch": 0.07,
79
+ "learning_rate": 0.00010791812460476247,
80
+ "loss": 1.5342,
81
+ "step": 12
82
+ },
83
+ {
84
+ "epoch": 0.07,
85
+ "learning_rate": 0.00011139433523068365,
86
+ "loss": 1.5394,
87
+ "step": 13
88
+ },
89
+ {
90
+ "epoch": 0.08,
91
+ "learning_rate": 0.00011461280356782378,
92
+ "loss": 1.5291,
93
+ "step": 14
94
+ },
95
+ {
96
+ "epoch": 0.08,
97
+ "learning_rate": 0.00011760912590556812,
98
+ "loss": 1.4864,
99
+ "step": 15
100
+ },
101
+ {
102
+ "epoch": 0.09,
103
+ "learning_rate": 0.00012041199826559246,
104
+ "loss": 1.4938,
105
+ "step": 16
106
+ },
107
+ {
108
+ "epoch": 0.09,
109
+ "learning_rate": 0.0001230448921378274,
110
+ "loss": 1.4965,
111
+ "step": 17
112
+ },
113
+ {
114
+ "epoch": 0.1,
115
+ "learning_rate": 0.00012552725051033058,
116
+ "loss": 1.4751,
117
+ "step": 18
118
+ },
119
+ {
120
+ "epoch": 0.11,
121
+ "learning_rate": 0.00012787536009528286,
122
+ "loss": 1.4768,
123
+ "step": 19
124
+ },
125
+ {
126
+ "epoch": 0.11,
127
+ "learning_rate": 0.0001301029995663981,
128
+ "loss": 1.4578,
129
+ "step": 20
130
+ },
131
+ {
132
+ "epoch": 0.12,
133
+ "learning_rate": 0.0001322219294733919,
134
+ "loss": 1.4627,
135
+ "step": 21
136
+ },
137
+ {
138
+ "epoch": 0.12,
139
+ "learning_rate": 0.00013424226808222061,
140
+ "loss": 1.4535,
141
+ "step": 22
142
+ },
143
+ {
144
+ "epoch": 0.13,
145
+ "learning_rate": 0.00013617278360175927,
146
+ "loss": 1.451,
147
+ "step": 23
148
+ },
149
+ {
150
+ "epoch": 0.13,
151
+ "learning_rate": 0.0001380211241711606,
152
+ "loss": 1.4446,
153
+ "step": 24
154
+ },
155
+ {
156
+ "epoch": 0.14,
157
+ "learning_rate": 0.00013979400086720374,
158
+ "loss": 1.4443,
159
+ "step": 25
160
+ },
161
+ {
162
+ "epoch": 0.14,
163
+ "learning_rate": 0.00014149733479708177,
164
+ "loss": 1.4401,
165
+ "step": 26
166
+ },
167
+ {
168
+ "epoch": 0.15,
169
+ "learning_rate": 0.00014313637641589873,
170
+ "loss": 1.4506,
171
+ "step": 27
172
+ },
173
+ {
174
+ "epoch": 0.16,
175
+ "learning_rate": 0.0001447158031342219,
176
+ "loss": 1.4289,
177
+ "step": 28
178
+ },
179
+ {
180
+ "epoch": 0.16,
181
+ "learning_rate": 0.0001462397997898956,
182
+ "loss": 1.433,
183
+ "step": 29
184
+ },
185
+ {
186
+ "epoch": 0.17,
187
+ "learning_rate": 0.00014771212547196622,
188
+ "loss": 1.4335,
189
+ "step": 30
190
+ },
191
+ {
192
+ "epoch": 0.17,
193
+ "learning_rate": 0.00014913616938342725,
194
+ "loss": 1.4263,
195
+ "step": 31
196
+ },
197
+ {
198
+ "epoch": 0.18,
199
+ "learning_rate": 0.0001505149978319906,
200
+ "loss": 1.4278,
201
+ "step": 32
202
+ },
203
+ {
204
+ "epoch": 0.18,
205
+ "learning_rate": 0.00015185139398778873,
206
+ "loss": 1.4219,
207
+ "step": 33
208
+ },
209
+ {
210
+ "epoch": 0.19,
211
+ "learning_rate": 0.0001531478917042255,
212
+ "loss": 1.4195,
213
+ "step": 34
214
+ },
215
+ {
216
+ "epoch": 0.19,
217
+ "learning_rate": 0.00015440680443502754,
218
+ "loss": 1.4314,
219
+ "step": 35
220
+ },
221
+ {
222
+ "epoch": 0.2,
223
+ "learning_rate": 0.00015563025007672872,
224
+ "loss": 1.4326,
225
+ "step": 36
226
+ },
227
+ {
228
+ "epoch": 0.21,
229
+ "learning_rate": 0.0001568201724066995,
230
+ "loss": 1.4224,
231
+ "step": 37
232
+ },
233
+ {
234
+ "epoch": 0.21,
235
+ "learning_rate": 0.000157978359661681,
236
+ "loss": 1.4233,
237
+ "step": 38
238
+ },
239
+ {
240
+ "epoch": 0.22,
241
+ "learning_rate": 0.0001591064607026499,
242
+ "loss": 1.417,
243
+ "step": 39
244
+ },
245
+ {
246
+ "epoch": 0.22,
247
+ "learning_rate": 0.0001602059991327962,
248
+ "loss": 1.417,
249
+ "step": 40
250
+ },
251
+ {
252
+ "epoch": 0.23,
253
+ "learning_rate": 0.00016127838567197352,
254
+ "loss": 1.41,
255
+ "step": 41
256
+ },
257
+ {
258
+ "epoch": 0.23,
259
+ "learning_rate": 0.00016232492903979004,
260
+ "loss": 1.4319,
261
+ "step": 42
262
+ },
263
+ {
264
+ "epoch": 0.24,
265
+ "learning_rate": 0.00016334684555795862,
266
+ "loss": 1.4218,
267
+ "step": 43
268
+ },
269
+ {
270
+ "epoch": 0.24,
271
+ "learning_rate": 0.00016434526764861872,
272
+ "loss": 1.4196,
273
+ "step": 44
274
+ },
275
+ {
276
+ "epoch": 0.25,
277
+ "learning_rate": 0.00016532125137753433,
278
+ "loss": 1.4157,
279
+ "step": 45
280
+ },
281
+ {
282
+ "epoch": 0.26,
283
+ "learning_rate": 0.0001662757831681574,
284
+ "loss": 1.4251,
285
+ "step": 46
286
+ },
287
+ {
288
+ "epoch": 0.26,
289
+ "learning_rate": 0.0001672097857935717,
290
+ "loss": 1.4248,
291
+ "step": 47
292
+ },
293
+ {
294
+ "epoch": 0.27,
295
+ "learning_rate": 0.0001681241237375587,
296
+ "loss": 1.4189,
297
+ "step": 48
298
+ },
299
+ {
300
+ "epoch": 0.27,
301
+ "learning_rate": 0.00016901960800285134,
302
+ "loss": 1.4165,
303
+ "step": 49
304
+ },
305
+ {
306
+ "epoch": 0.28,
307
+ "learning_rate": 0.00016989700043360185,
308
+ "loss": 1.4167,
309
+ "step": 50
310
+ },
311
+ {
312
+ "epoch": 0.28,
313
+ "learning_rate": 0.00017075701760979362,
314
+ "loss": 1.4146,
315
+ "step": 51
316
+ },
317
+ {
318
+ "epoch": 0.29,
319
+ "learning_rate": 0.0001716003343634799,
320
+ "loss": 1.4133,
321
+ "step": 52
322
+ },
323
+ {
324
+ "epoch": 0.29,
325
+ "learning_rate": 0.0001724275869600789,
326
+ "loss": 1.4074,
327
+ "step": 53
328
+ },
329
+ {
330
+ "epoch": 0.3,
331
+ "learning_rate": 0.00017323937598229684,
332
+ "loss": 1.4164,
333
+ "step": 54
334
+ },
335
+ {
336
+ "epoch": 0.31,
337
+ "learning_rate": 0.00017403626894942436,
338
+ "loss": 1.4178,
339
+ "step": 55
340
+ },
341
+ {
342
+ "epoch": 0.31,
343
+ "learning_rate": 0.00017481880270062004,
344
+ "loss": 1.4239,
345
+ "step": 56
346
+ },
347
+ {
348
+ "epoch": 0.32,
349
+ "learning_rate": 0.00017558748556724914,
350
+ "loss": 1.4074,
351
+ "step": 57
352
+ },
353
+ {
354
+ "epoch": 0.32,
355
+ "learning_rate": 0.0001763427993562937,
356
+ "loss": 1.4118,
357
+ "step": 58
358
+ },
359
+ {
360
+ "epoch": 0.33,
361
+ "learning_rate": 0.0001770852011642144,
362
+ "loss": 1.4088,
363
+ "step": 59
364
+ },
365
+ {
366
+ "epoch": 0.33,
367
+ "learning_rate": 0.00017781512503836432,
368
+ "loss": 1.4211,
369
+ "step": 60
370
+ },
371
+ {
372
+ "epoch": 0.34,
373
+ "learning_rate": 0.00017853298350107669,
374
+ "loss": 1.4019,
375
+ "step": 61
376
+ },
377
+ {
378
+ "epoch": 0.34,
379
+ "learning_rate": 0.00017923916894982539,
380
+ "loss": 1.4284,
381
+ "step": 62
382
+ },
383
+ {
384
+ "epoch": 0.35,
385
+ "learning_rate": 0.00017993405494535816,
386
+ "loss": 1.4092,
387
+ "step": 63
388
+ },
389
+ {
390
+ "epoch": 0.36,
391
+ "learning_rate": 0.00018061799739838867,
392
+ "loss": 1.415,
393
+ "step": 64
394
+ },
395
+ {
396
+ "epoch": 0.36,
397
+ "learning_rate": 0.00018129133566428552,
398
+ "loss": 1.4249,
399
+ "step": 65
400
+ },
401
+ {
402
+ "epoch": 0.37,
403
+ "learning_rate": 0.00018195439355418684,
404
+ "loss": 1.4114,
405
+ "step": 66
406
+ },
407
+ {
408
+ "epoch": 0.37,
409
+ "learning_rate": 0.0001826074802700826,
410
+ "loss": 1.4209,
411
+ "step": 67
412
+ },
413
+ {
414
+ "epoch": 0.38,
415
+ "learning_rate": 0.00018325089127062364,
416
+ "loss": 1.4141,
417
+ "step": 68
418
+ },
419
+ {
420
+ "epoch": 0.38,
421
+ "learning_rate": 0.00018388490907372552,
422
+ "loss": 1.4092,
423
+ "step": 69
424
+ },
425
+ {
426
+ "epoch": 0.39,
427
+ "learning_rate": 0.00018450980400142568,
428
+ "loss": 1.4177,
429
+ "step": 70
430
+ },
431
+ {
432
+ "epoch": 0.39,
433
+ "learning_rate": 0.00018512583487190751,
434
+ "loss": 1.4134,
435
+ "step": 71
436
+ },
437
+ {
438
+ "epoch": 0.4,
439
+ "learning_rate": 0.00018573324964312683,
440
+ "loss": 1.4015,
441
+ "step": 72
442
+ },
443
+ {
444
+ "epoch": 0.41,
445
+ "learning_rate": 0.00018633228601204555,
446
+ "loss": 1.4165,
447
+ "step": 73
448
+ },
449
+ {
450
+ "epoch": 0.41,
451
+ "learning_rate": 0.0001869231719730976,
452
+ "loss": 1.4043,
453
+ "step": 74
454
+ },
455
+ {
456
+ "epoch": 0.42,
457
+ "learning_rate": 0.00018750612633916996,
458
+ "loss": 1.4045,
459
+ "step": 75
460
+ },
461
+ {
462
+ "epoch": 0.42,
463
+ "learning_rate": 0.00018808135922807913,
464
+ "loss": 1.4063,
465
+ "step": 76
466
+ },
467
+ {
468
+ "epoch": 0.43,
469
+ "learning_rate": 0.00018864907251724816,
470
+ "loss": 1.4057,
471
+ "step": 77
472
+ },
473
+ {
474
+ "epoch": 0.43,
475
+ "learning_rate": 0.00018920946026904802,
476
+ "loss": 1.4009,
477
+ "step": 78
478
+ },
479
+ {
480
+ "epoch": 0.44,
481
+ "learning_rate": 0.00018976270912904412,
482
+ "loss": 1.4017,
483
+ "step": 79
484
+ },
485
+ {
486
+ "epoch": 0.44,
487
+ "learning_rate": 0.00019030899869919432,
488
+ "loss": 1.4172,
489
+ "step": 80
490
+ },
491
+ {
492
+ "epoch": 0.45,
493
+ "learning_rate": 0.00019084850188786495,
494
+ "loss": 1.4061,
495
+ "step": 81
496
+ },
497
+ {
498
+ "epoch": 0.46,
499
+ "learning_rate": 0.00019138138523837166,
500
+ "loss": 1.4047,
501
+ "step": 82
502
+ },
503
+ {
504
+ "epoch": 0.46,
505
+ "learning_rate": 0.00019190780923760738,
506
+ "loss": 1.408,
507
+ "step": 83
508
+ },
509
+ {
510
+ "epoch": 0.47,
511
+ "learning_rate": 0.00019242792860618813,
512
+ "loss": 1.4054,
513
+ "step": 84
514
+ },
515
+ {
516
+ "epoch": 0.47,
517
+ "learning_rate": 0.00019294189257142928,
518
+ "loss": 1.399,
519
+ "step": 85
520
+ },
521
+ {
522
+ "epoch": 0.48,
523
+ "learning_rate": 0.00019344984512435672,
524
+ "loss": 1.4015,
525
+ "step": 86
526
+ },
527
+ {
528
+ "epoch": 0.48,
529
+ "learning_rate": 0.00019395192526186183,
530
+ "loss": 1.4108,
531
+ "step": 87
532
+ },
533
+ {
534
+ "epoch": 0.49,
535
+ "learning_rate": 0.00019444826721501686,
536
+ "loss": 1.407,
537
+ "step": 88
538
+ },
539
+ {
540
+ "epoch": 0.49,
541
+ "learning_rate": 0.00019493900066449123,
542
+ "loss": 1.4018,
543
+ "step": 89
544
+ },
545
+ {
546
+ "epoch": 0.5,
547
+ "learning_rate": 0.00019542425094393247,
548
+ "loss": 1.4013,
549
+ "step": 90
550
+ },
551
+ {
552
+ "epoch": 0.51,
553
+ "learning_rate": 0.00019590413923210932,
554
+ "loss": 1.4123,
555
+ "step": 91
556
+ },
557
+ {
558
+ "epoch": 0.51,
559
+ "learning_rate": 0.0001963787827345555,
560
+ "loss": 1.4064,
561
+ "step": 92
562
+ },
563
+ {
564
+ "epoch": 0.52,
565
+ "learning_rate": 0.0001968482948553935,
566
+ "loss": 1.4009,
567
+ "step": 93
568
+ },
569
+ {
570
+ "epoch": 0.52,
571
+ "learning_rate": 0.00019731278535996984,
572
+ "loss": 1.4046,
573
+ "step": 94
574
+ },
575
+ {
576
+ "epoch": 0.53,
577
+ "learning_rate": 0.00019777236052888474,
578
+ "loss": 1.4029,
579
+ "step": 95
580
+ },
581
+ {
582
+ "epoch": 0.53,
583
+ "learning_rate": 0.00019822712330395682,
584
+ "loss": 1.4063,
585
+ "step": 96
586
+ },
587
+ {
588
+ "epoch": 0.54,
589
+ "learning_rate": 0.00019867717342662446,
590
+ "loss": 1.3971,
591
+ "step": 97
592
+ },
593
+ {
594
+ "epoch": 0.54,
595
+ "learning_rate": 0.00019912260756924948,
596
+ "loss": 1.3985,
597
+ "step": 98
598
+ },
599
+ {
600
+ "epoch": 0.55,
601
+ "learning_rate": 0.00019956351945975496,
602
+ "loss": 1.402,
603
+ "step": 99
604
+ },
605
+ {
606
+ "epoch": 0.56,
607
+ "learning_rate": 0.00019999999999999998,
608
+ "loss": 1.4014,
609
+ "step": 100
610
+ },
611
+ {
612
+ "epoch": 0.56,
613
+ "learning_rate": 0.0002,
614
+ "loss": 1.3979,
615
+ "step": 101
616
+ },
617
+ {
618
+ "epoch": 0.57,
619
+ "learning_rate": 0.00019975,
620
+ "loss": 1.4003,
621
+ "step": 102
622
+ },
623
+ {
624
+ "epoch": 0.57,
625
+ "learning_rate": 0.00019950000000000002,
626
+ "loss": 1.3969,
627
+ "step": 103
628
+ },
629
+ {
630
+ "epoch": 0.58,
631
+ "learning_rate": 0.00019925,
632
+ "loss": 1.4054,
633
+ "step": 104
634
+ },
635
+ {
636
+ "epoch": 0.58,
637
+ "learning_rate": 0.000199,
638
+ "loss": 1.3978,
639
+ "step": 105
640
+ },
641
+ {
642
+ "epoch": 0.59,
643
+ "learning_rate": 0.00019875,
644
+ "loss": 1.3996,
645
+ "step": 106
646
+ },
647
+ {
648
+ "epoch": 0.59,
649
+ "learning_rate": 0.00019850000000000003,
650
+ "loss": 1.3895,
651
+ "step": 107
652
+ },
653
+ {
654
+ "epoch": 0.6,
655
+ "learning_rate": 0.00019825,
656
+ "loss": 1.3965,
657
+ "step": 108
658
+ },
659
+ {
660
+ "epoch": 0.61,
661
+ "learning_rate": 0.00019800000000000002,
662
+ "loss": 1.4077,
663
+ "step": 109
664
+ },
665
+ {
666
+ "epoch": 0.61,
667
+ "learning_rate": 0.00019775,
668
+ "loss": 1.3999,
669
+ "step": 110
670
+ },
671
+ {
672
+ "epoch": 0.62,
673
+ "learning_rate": 0.00019750000000000003,
674
+ "loss": 1.3975,
675
+ "step": 111
676
+ },
677
+ {
678
+ "epoch": 0.62,
679
+ "learning_rate": 0.00019725,
680
+ "loss": 1.3985,
681
+ "step": 112
682
+ },
683
+ {
684
+ "epoch": 0.63,
685
+ "learning_rate": 0.00019700000000000002,
686
+ "loss": 1.3939,
687
+ "step": 113
688
+ },
689
+ {
690
+ "epoch": 0.63,
691
+ "learning_rate": 0.00019675,
692
+ "loss": 1.4033,
693
+ "step": 114
694
+ },
695
+ {
696
+ "epoch": 0.64,
697
+ "learning_rate": 0.0001965,
698
+ "loss": 1.4035,
699
+ "step": 115
700
+ },
701
+ {
702
+ "epoch": 0.64,
703
+ "learning_rate": 0.00019625,
704
+ "loss": 1.3979,
705
+ "step": 116
706
+ },
707
+ {
708
+ "epoch": 0.65,
709
+ "learning_rate": 0.000196,
710
+ "loss": 1.4002,
711
+ "step": 117
712
+ },
713
+ {
714
+ "epoch": 0.66,
715
+ "learning_rate": 0.00019575000000000001,
716
+ "loss": 1.3949,
717
+ "step": 118
718
+ },
719
+ {
720
+ "epoch": 0.66,
721
+ "learning_rate": 0.0001955,
722
+ "loss": 1.3906,
723
+ "step": 119
724
+ },
725
+ {
726
+ "epoch": 0.67,
727
+ "learning_rate": 0.00019525,
728
+ "loss": 1.4039,
729
+ "step": 120
730
+ },
731
+ {
732
+ "epoch": 0.67,
733
+ "learning_rate": 0.000195,
734
+ "loss": 1.4,
735
+ "step": 121
736
+ },
737
+ {
738
+ "epoch": 0.68,
739
+ "learning_rate": 0.00019475000000000002,
740
+ "loss": 1.3961,
741
+ "step": 122
742
+ },
743
+ {
744
+ "epoch": 0.68,
745
+ "learning_rate": 0.0001945,
746
+ "loss": 1.3997,
747
+ "step": 123
748
+ },
749
+ {
750
+ "epoch": 0.69,
751
+ "learning_rate": 0.00019425,
752
+ "loss": 1.414,
753
+ "step": 124
754
+ },
755
+ {
756
+ "epoch": 0.69,
757
+ "learning_rate": 0.000194,
758
+ "loss": 1.3942,
759
+ "step": 125
760
+ },
761
+ {
762
+ "epoch": 0.7,
763
+ "learning_rate": 0.00019375000000000002,
764
+ "loss": 1.406,
765
+ "step": 126
766
+ },
767
+ {
768
+ "epoch": 0.71,
769
+ "learning_rate": 0.00019350000000000001,
770
+ "loss": 1.4017,
771
+ "step": 127
772
+ },
773
+ {
774
+ "epoch": 0.71,
775
+ "learning_rate": 0.00019325,
776
+ "loss": 1.3917,
777
+ "step": 128
778
+ },
779
+ {
780
+ "epoch": 0.72,
781
+ "learning_rate": 0.000193,
782
+ "loss": 1.3945,
783
+ "step": 129
784
+ },
785
+ {
786
+ "epoch": 0.72,
787
+ "learning_rate": 0.00019275,
788
+ "loss": 1.3966,
789
+ "step": 130
790
+ },
791
+ {
792
+ "epoch": 0.73,
793
+ "learning_rate": 0.00019250000000000002,
794
+ "loss": 1.3942,
795
+ "step": 131
796
+ },
797
+ {
798
+ "epoch": 0.73,
799
+ "learning_rate": 0.00019225,
800
+ "loss": 1.3968,
801
+ "step": 132
802
+ },
803
+ {
804
+ "epoch": 0.74,
805
+ "learning_rate": 0.000192,
806
+ "loss": 1.3921,
807
+ "step": 133
808
+ },
809
+ {
810
+ "epoch": 0.74,
811
+ "learning_rate": 0.00019175,
812
+ "loss": 1.3931,
813
+ "step": 134
814
+ },
815
+ {
816
+ "epoch": 0.75,
817
+ "learning_rate": 0.00019150000000000002,
818
+ "loss": 1.3917,
819
+ "step": 135
820
+ },
821
+ {
822
+ "epoch": 0.76,
823
+ "learning_rate": 0.00019125000000000001,
824
+ "loss": 1.4025,
825
+ "step": 136
826
+ },
827
+ {
828
+ "epoch": 0.76,
829
+ "learning_rate": 0.000191,
830
+ "loss": 1.3997,
831
+ "step": 137
832
+ },
833
+ {
834
+ "epoch": 0.77,
835
+ "learning_rate": 0.00019075,
836
+ "loss": 1.4027,
837
+ "step": 138
838
+ },
839
+ {
840
+ "epoch": 0.77,
841
+ "learning_rate": 0.00019050000000000002,
842
+ "loss": 1.4068,
843
+ "step": 139
844
+ },
845
+ {
846
+ "epoch": 0.78,
847
+ "learning_rate": 0.00019025000000000002,
848
+ "loss": 1.4075,
849
+ "step": 140
850
+ },
851
+ {
852
+ "epoch": 0.78,
853
+ "learning_rate": 0.00019,
854
+ "loss": 1.3986,
855
+ "step": 141
856
+ },
857
+ {
858
+ "epoch": 0.79,
859
+ "learning_rate": 0.00018975,
860
+ "loss": 1.3996,
861
+ "step": 142
862
+ },
863
+ {
864
+ "epoch": 0.79,
865
+ "learning_rate": 0.0001895,
866
+ "loss": 1.3973,
867
+ "step": 143
868
+ },
869
+ {
870
+ "epoch": 0.8,
871
+ "learning_rate": 0.00018925000000000002,
872
+ "loss": 1.4034,
873
+ "step": 144
874
+ },
875
+ {
876
+ "epoch": 0.81,
877
+ "learning_rate": 0.00018899999999999999,
878
+ "loss": 1.4142,
879
+ "step": 145
880
+ },
881
+ {
882
+ "epoch": 0.81,
883
+ "learning_rate": 0.00018875,
884
+ "loss": 1.3989,
885
+ "step": 146
886
+ },
887
+ {
888
+ "epoch": 0.82,
889
+ "learning_rate": 0.0001885,
890
+ "loss": 1.3939,
891
+ "step": 147
892
+ },
893
+ {
894
+ "epoch": 0.82,
895
+ "learning_rate": 0.00018825000000000002,
896
+ "loss": 1.3948,
897
+ "step": 148
898
+ },
899
+ {
900
+ "epoch": 0.83,
901
+ "learning_rate": 0.000188,
902
+ "loss": 1.3951,
903
+ "step": 149
904
+ },
905
+ {
906
+ "epoch": 0.83,
907
+ "learning_rate": 0.00018775,
908
+ "loss": 1.3982,
909
+ "step": 150
910
+ },
911
+ {
912
+ "epoch": 0.84,
913
+ "learning_rate": 0.0001875,
914
+ "loss": 1.3982,
915
+ "step": 151
916
+ },
917
+ {
918
+ "epoch": 0.84,
919
+ "learning_rate": 0.00018725000000000002,
920
+ "loss": 1.404,
921
+ "step": 152
922
+ },
923
+ {
924
+ "epoch": 0.85,
925
+ "learning_rate": 0.00018700000000000002,
926
+ "loss": 1.3949,
927
+ "step": 153
928
+ },
929
+ {
930
+ "epoch": 0.86,
931
+ "learning_rate": 0.00018675,
932
+ "loss": 1.3955,
933
+ "step": 154
934
+ },
935
+ {
936
+ "epoch": 0.86,
937
+ "learning_rate": 0.0001865,
938
+ "loss": 1.3873,
939
+ "step": 155
940
+ },
941
+ {
942
+ "epoch": 0.87,
943
+ "learning_rate": 0.00018625,
944
+ "loss": 1.3995,
945
+ "step": 156
946
+ },
947
+ {
948
+ "epoch": 0.87,
949
+ "learning_rate": 0.00018600000000000002,
950
+ "loss": 1.389,
951
+ "step": 157
952
+ },
953
+ {
954
+ "epoch": 0.88,
955
+ "learning_rate": 0.00018575,
956
+ "loss": 1.3944,
957
+ "step": 158
958
+ },
959
+ {
960
+ "epoch": 0.88,
961
+ "learning_rate": 0.0001855,
962
+ "loss": 1.389,
963
+ "step": 159
964
+ },
965
+ {
966
+ "epoch": 0.89,
967
+ "learning_rate": 0.00018525,
968
+ "loss": 1.4016,
969
+ "step": 160
970
+ },
971
+ {
972
+ "epoch": 0.89,
973
+ "learning_rate": 0.00018500000000000002,
974
+ "loss": 1.3953,
975
+ "step": 161
976
+ },
977
+ {
978
+ "epoch": 0.9,
979
+ "learning_rate": 0.00018475,
980
+ "loss": 1.4005,
981
+ "step": 162
982
+ },
983
+ {
984
+ "epoch": 0.91,
985
+ "learning_rate": 0.0001845,
986
+ "loss": 1.3977,
987
+ "step": 163
988
+ },
989
+ {
990
+ "epoch": 0.91,
991
+ "learning_rate": 0.00018425,
992
+ "loss": 1.3914,
993
+ "step": 164
994
+ },
995
+ {
996
+ "epoch": 0.92,
997
+ "learning_rate": 0.00018400000000000003,
998
+ "loss": 1.391,
999
+ "step": 165
1000
+ },
1001
+ {
1002
+ "epoch": 0.92,
1003
+ "learning_rate": 0.00018375,
1004
+ "loss": 1.4055,
1005
+ "step": 166
1006
+ },
1007
+ {
1008
+ "epoch": 0.93,
1009
+ "learning_rate": 0.00018350000000000002,
1010
+ "loss": 1.3956,
1011
+ "step": 167
1012
+ },
1013
+ {
1014
+ "epoch": 0.93,
1015
+ "learning_rate": 0.00018325,
1016
+ "loss": 1.3845,
1017
+ "step": 168
1018
+ },
1019
+ {
1020
+ "epoch": 0.94,
1021
+ "learning_rate": 0.000183,
1022
+ "loss": 1.3882,
1023
+ "step": 169
1024
+ },
1025
+ {
1026
+ "epoch": 0.94,
1027
+ "learning_rate": 0.00018275,
1028
+ "loss": 1.3891,
1029
+ "step": 170
1030
+ },
1031
+ {
1032
+ "epoch": 0.95,
1033
+ "learning_rate": 0.0001825,
1034
+ "loss": 1.3868,
1035
+ "step": 171
1036
+ },
1037
+ {
1038
+ "epoch": 0.96,
1039
+ "learning_rate": 0.00018225,
1040
+ "loss": 1.3909,
1041
+ "step": 172
1042
+ },
1043
+ {
1044
+ "epoch": 0.96,
1045
+ "learning_rate": 0.000182,
1046
+ "loss": 1.392,
1047
+ "step": 173
1048
+ },
1049
+ {
1050
+ "epoch": 0.97,
1051
+ "learning_rate": 0.00018175,
1052
+ "loss": 1.3919,
1053
+ "step": 174
1054
+ },
1055
+ {
1056
+ "epoch": 0.97,
1057
+ "learning_rate": 0.0001815,
1058
+ "loss": 1.3952,
1059
+ "step": 175
1060
+ },
1061
+ {
1062
+ "epoch": 0.98,
1063
+ "learning_rate": 0.00018125000000000001,
1064
+ "loss": 1.3903,
1065
+ "step": 176
1066
+ },
1067
+ {
1068
+ "epoch": 0.98,
1069
+ "learning_rate": 0.000181,
1070
+ "loss": 1.4049,
1071
+ "step": 177
1072
+ },
1073
+ {
1074
+ "epoch": 0.99,
1075
+ "learning_rate": 0.00018075000000000003,
1076
+ "loss": 1.3926,
1077
+ "step": 178
1078
+ },
1079
+ {
1080
+ "epoch": 0.99,
1081
+ "learning_rate": 0.0001805,
1082
+ "loss": 1.3918,
1083
+ "step": 179
1084
+ },
1085
+ {
1086
+ "epoch": 1.0,
1087
+ "learning_rate": 0.00018025000000000002,
1088
+ "loss": 1.4015,
1089
+ "step": 180
1090
+ },
1091
+ {
1092
+ "epoch": 1.01,
1093
+ "learning_rate": 0.00018,
1094
+ "loss": 1.3907,
1095
+ "step": 181
1096
+ },
1097
+ {
1098
+ "epoch": 1.01,
1099
+ "learning_rate": 0.00017975,
1100
+ "loss": 1.3876,
1101
+ "step": 182
1102
+ },
1103
+ {
1104
+ "epoch": 1.02,
1105
+ "learning_rate": 0.0001795,
1106
+ "loss": 1.3907,
1107
+ "step": 183
1108
+ },
1109
+ {
1110
+ "epoch": 1.02,
1111
+ "learning_rate": 0.00017925000000000002,
1112
+ "loss": 1.3936,
1113
+ "step": 184
1114
+ },
1115
+ {
1116
+ "epoch": 1.03,
1117
+ "learning_rate": 0.00017900000000000001,
1118
+ "loss": 1.3907,
1119
+ "step": 185
1120
+ },
1121
+ {
1122
+ "epoch": 1.03,
1123
+ "learning_rate": 0.00017875,
1124
+ "loss": 1.4007,
1125
+ "step": 186
1126
+ },
1127
+ {
1128
+ "epoch": 1.04,
1129
+ "learning_rate": 0.0001785,
1130
+ "loss": 1.3944,
1131
+ "step": 187
1132
+ },
1133
+ {
1134
+ "epoch": 1.04,
1135
+ "learning_rate": 0.00017825,
1136
+ "loss": 1.3884,
1137
+ "step": 188
1138
+ },
1139
+ {
1140
+ "epoch": 1.05,
1141
+ "learning_rate": 0.00017800000000000002,
1142
+ "loss": 1.3892,
1143
+ "step": 189
1144
+ },
1145
+ {
1146
+ "epoch": 1.06,
1147
+ "learning_rate": 0.00017775,
1148
+ "loss": 1.3897,
1149
+ "step": 190
1150
+ },
1151
+ {
1152
+ "epoch": 1.06,
1153
+ "learning_rate": 0.0001775,
1154
+ "loss": 1.3958,
1155
+ "step": 191
1156
+ },
1157
+ {
1158
+ "epoch": 1.07,
1159
+ "learning_rate": 0.00017725,
1160
+ "loss": 1.3904,
1161
+ "step": 192
1162
+ },
1163
+ {
1164
+ "epoch": 1.07,
1165
+ "learning_rate": 0.00017700000000000002,
1166
+ "loss": 1.3907,
1167
+ "step": 193
1168
+ },
1169
+ {
1170
+ "epoch": 1.08,
1171
+ "learning_rate": 0.00017675000000000001,
1172
+ "loss": 1.395,
1173
+ "step": 194
1174
+ },
1175
+ {
1176
+ "epoch": 1.08,
1177
+ "learning_rate": 0.0001765,
1178
+ "loss": 1.3898,
1179
+ "step": 195
1180
+ },
1181
+ {
1182
+ "epoch": 1.09,
1183
+ "learning_rate": 0.00017625,
1184
+ "loss": 1.3949,
1185
+ "step": 196
1186
+ },
1187
+ {
1188
+ "epoch": 1.09,
1189
+ "learning_rate": 0.00017600000000000002,
1190
+ "loss": 1.3917,
1191
+ "step": 197
1192
+ },
1193
+ {
1194
+ "epoch": 1.1,
1195
+ "learning_rate": 0.00017575000000000002,
1196
+ "loss": 1.3944,
1197
+ "step": 198
1198
+ },
1199
+ {
1200
+ "epoch": 1.11,
1201
+ "learning_rate": 0.0001755,
1202
+ "loss": 1.3873,
1203
+ "step": 199
1204
+ },
1205
+ {
1206
+ "epoch": 1.11,
1207
+ "learning_rate": 0.00017525,
1208
+ "loss": 1.389,
1209
+ "step": 200
1210
+ },
1211
+ {
1212
+ "epoch": 1.12,
1213
+ "learning_rate": 0.000175,
1214
+ "loss": 1.3887,
1215
+ "step": 201
1216
+ },
1217
+ {
1218
+ "epoch": 1.12,
1219
+ "learning_rate": 0.00017475000000000002,
1220
+ "loss": 1.3971,
1221
+ "step": 202
1222
+ },
1223
+ {
1224
+ "epoch": 1.13,
1225
+ "learning_rate": 0.0001745,
1226
+ "loss": 1.3831,
1227
+ "step": 203
1228
+ },
1229
+ {
1230
+ "epoch": 1.13,
1231
+ "learning_rate": 0.00017425,
1232
+ "loss": 1.3864,
1233
+ "step": 204
1234
+ },
1235
+ {
1236
+ "epoch": 1.14,
1237
+ "learning_rate": 0.000174,
1238
+ "loss": 1.3874,
1239
+ "step": 205
1240
+ },
1241
+ {
1242
+ "epoch": 1.14,
1243
+ "learning_rate": 0.00017375000000000002,
1244
+ "loss": 1.3887,
1245
+ "step": 206
1246
+ },
1247
+ {
1248
+ "epoch": 1.15,
1249
+ "learning_rate": 0.00017350000000000002,
1250
+ "loss": 1.3884,
1251
+ "step": 207
1252
+ },
1253
+ {
1254
+ "epoch": 1.16,
1255
+ "learning_rate": 0.00017325,
1256
+ "loss": 1.3909,
1257
+ "step": 208
1258
+ },
1259
+ {
1260
+ "epoch": 1.16,
1261
+ "learning_rate": 0.000173,
1262
+ "loss": 1.392,
1263
+ "step": 209
1264
+ },
1265
+ {
1266
+ "epoch": 1.17,
1267
+ "learning_rate": 0.00017275000000000002,
1268
+ "loss": 1.3925,
1269
+ "step": 210
1270
+ },
1271
+ {
1272
+ "epoch": 1.17,
1273
+ "learning_rate": 0.00017250000000000002,
1274
+ "loss": 1.3947,
1275
+ "step": 211
1276
+ },
1277
+ {
1278
+ "epoch": 1.18,
1279
+ "learning_rate": 0.00017225,
1280
+ "loss": 1.3922,
1281
+ "step": 212
1282
+ },
1283
+ {
1284
+ "epoch": 1.18,
1285
+ "learning_rate": 0.000172,
1286
+ "loss": 1.3901,
1287
+ "step": 213
1288
+ },
1289
+ {
1290
+ "epoch": 1.19,
1291
+ "learning_rate": 0.00017175,
1292
+ "loss": 1.3882,
1293
+ "step": 214
1294
+ },
1295
+ {
1296
+ "epoch": 1.19,
1297
+ "learning_rate": 0.00017150000000000002,
1298
+ "loss": 1.3886,
1299
+ "step": 215
1300
+ },
1301
+ {
1302
+ "epoch": 1.2,
1303
+ "learning_rate": 0.00017125,
1304
+ "loss": 1.3953,
1305
+ "step": 216
1306
+ },
1307
+ {
1308
+ "epoch": 1.21,
1309
+ "learning_rate": 0.000171,
1310
+ "loss": 1.3935,
1311
+ "step": 217
1312
+ },
1313
+ {
1314
+ "epoch": 1.21,
1315
+ "learning_rate": 0.00017075,
1316
+ "loss": 1.3845,
1317
+ "step": 218
1318
+ },
1319
+ {
1320
+ "epoch": 1.22,
1321
+ "learning_rate": 0.00017050000000000002,
1322
+ "loss": 1.3986,
1323
+ "step": 219
1324
+ },
1325
+ {
1326
+ "epoch": 1.22,
1327
+ "learning_rate": 0.00017025,
1328
+ "loss": 1.3949,
1329
+ "step": 220
1330
+ },
1331
+ {
1332
+ "epoch": 1.23,
1333
+ "learning_rate": 0.00017,
1334
+ "loss": 1.3914,
1335
+ "step": 221
1336
+ },
1337
+ {
1338
+ "epoch": 1.23,
1339
+ "learning_rate": 0.00016975,
1340
+ "loss": 1.3975,
1341
+ "step": 222
1342
+ },
1343
+ {
1344
+ "epoch": 1.24,
1345
+ "learning_rate": 0.00016950000000000003,
1346
+ "loss": 1.3911,
1347
+ "step": 223
1348
+ },
1349
+ {
1350
+ "epoch": 1.24,
1351
+ "learning_rate": 0.00016925,
1352
+ "loss": 1.3945,
1353
+ "step": 224
1354
+ },
1355
+ {
1356
+ "epoch": 1.25,
1357
+ "learning_rate": 0.00016900000000000002,
1358
+ "loss": 1.394,
1359
+ "step": 225
1360
+ },
1361
+ {
1362
+ "epoch": 1.26,
1363
+ "learning_rate": 0.00016875,
1364
+ "loss": 1.385,
1365
+ "step": 226
1366
+ },
1367
+ {
1368
+ "epoch": 1.26,
1369
+ "learning_rate": 0.0001685,
1370
+ "loss": 1.3867,
1371
+ "step": 227
1372
+ },
1373
+ {
1374
+ "epoch": 1.27,
1375
+ "learning_rate": 0.00016825000000000002,
1376
+ "loss": 1.3976,
1377
+ "step": 228
1378
+ },
1379
+ {
1380
+ "epoch": 1.27,
1381
+ "learning_rate": 0.000168,
1382
+ "loss": 1.392,
1383
+ "step": 229
1384
+ },
1385
+ {
1386
+ "epoch": 1.28,
1387
+ "learning_rate": 0.00016775,
1388
+ "loss": 1.3883,
1389
+ "step": 230
1390
+ },
1391
+ {
1392
+ "epoch": 1.28,
1393
+ "learning_rate": 0.0001675,
1394
+ "loss": 1.3884,
1395
+ "step": 231
1396
+ },
1397
+ {
1398
+ "epoch": 1.29,
1399
+ "learning_rate": 0.00016725000000000003,
1400
+ "loss": 1.3893,
1401
+ "step": 232
1402
+ },
1403
+ {
1404
+ "epoch": 1.29,
1405
+ "learning_rate": 0.000167,
1406
+ "loss": 1.3888,
1407
+ "step": 233
1408
+ },
1409
+ {
1410
+ "epoch": 1.3,
1411
+ "learning_rate": 0.00016675000000000001,
1412
+ "loss": 1.3886,
1413
+ "step": 234
1414
+ },
1415
+ {
1416
+ "epoch": 1.31,
1417
+ "learning_rate": 0.0001665,
1418
+ "loss": 1.385,
1419
+ "step": 235
1420
+ },
1421
+ {
1422
+ "epoch": 1.31,
1423
+ "learning_rate": 0.00016625000000000003,
1424
+ "loss": 1.3939,
1425
+ "step": 236
1426
+ },
1427
+ {
1428
+ "epoch": 1.32,
1429
+ "learning_rate": 0.000166,
1430
+ "loss": 1.3952,
1431
+ "step": 237
1432
+ },
1433
+ {
1434
+ "epoch": 1.32,
1435
+ "learning_rate": 0.00016575000000000002,
1436
+ "loss": 1.3913,
1437
+ "step": 238
1438
+ },
1439
+ {
1440
+ "epoch": 1.33,
1441
+ "learning_rate": 0.0001655,
1442
+ "loss": 1.3863,
1443
+ "step": 239
1444
+ },
1445
+ {
1446
+ "epoch": 1.33,
1447
+ "learning_rate": 0.00016525,
1448
+ "loss": 1.3908,
1449
+ "step": 240
1450
+ },
1451
+ {
1452
+ "epoch": 1.34,
1453
+ "learning_rate": 0.000165,
1454
+ "loss": 1.3864,
1455
+ "step": 241
1456
+ },
1457
+ {
1458
+ "epoch": 1.34,
1459
+ "learning_rate": 0.00016475,
1460
+ "loss": 1.3878,
1461
+ "step": 242
1462
+ },
1463
+ {
1464
+ "epoch": 1.35,
1465
+ "learning_rate": 0.00016450000000000001,
1466
+ "loss": 1.3949,
1467
+ "step": 243
1468
+ },
1469
+ {
1470
+ "epoch": 1.36,
1471
+ "learning_rate": 0.00016425,
1472
+ "loss": 1.3902,
1473
+ "step": 244
1474
+ },
1475
+ {
1476
+ "epoch": 1.36,
1477
+ "learning_rate": 0.000164,
1478
+ "loss": 1.3879,
1479
+ "step": 245
1480
+ },
1481
+ {
1482
+ "epoch": 1.37,
1483
+ "learning_rate": 0.00016375,
1484
+ "loss": 1.3876,
1485
+ "step": 246
1486
+ },
1487
+ {
1488
+ "epoch": 1.37,
1489
+ "learning_rate": 0.00016350000000000002,
1490
+ "loss": 1.3838,
1491
+ "step": 247
1492
+ },
1493
+ {
1494
+ "epoch": 1.38,
1495
+ "learning_rate": 0.00016325,
1496
+ "loss": 1.3934,
1497
+ "step": 248
1498
+ },
1499
+ {
1500
+ "epoch": 1.38,
1501
+ "learning_rate": 0.000163,
1502
+ "loss": 1.3891,
1503
+ "step": 249
1504
+ },
1505
+ {
1506
+ "epoch": 1.39,
1507
+ "learning_rate": 0.00016275,
1508
+ "loss": 1.3939,
1509
+ "step": 250
1510
+ },
1511
+ {
1512
+ "epoch": 1.39,
1513
+ "learning_rate": 0.00016250000000000002,
1514
+ "loss": 1.3834,
1515
+ "step": 251
1516
+ },
1517
+ {
1518
+ "epoch": 1.4,
1519
+ "learning_rate": 0.00016225000000000001,
1520
+ "loss": 1.3885,
1521
+ "step": 252
1522
+ },
1523
+ {
1524
+ "epoch": 1.41,
1525
+ "learning_rate": 0.000162,
1526
+ "loss": 1.3857,
1527
+ "step": 253
1528
+ },
1529
+ {
1530
+ "epoch": 1.41,
1531
+ "learning_rate": 0.00016175,
1532
+ "loss": 1.389,
1533
+ "step": 254
1534
+ },
1535
+ {
1536
+ "epoch": 1.42,
1537
+ "learning_rate": 0.0001615,
1538
+ "loss": 1.3825,
1539
+ "step": 255
1540
+ },
1541
+ {
1542
+ "epoch": 1.42,
1543
+ "learning_rate": 0.00016125000000000002,
1544
+ "loss": 1.3851,
1545
+ "step": 256
1546
+ },
1547
+ {
1548
+ "epoch": 1.43,
1549
+ "learning_rate": 0.000161,
1550
+ "loss": 1.3811,
1551
+ "step": 257
1552
+ },
1553
+ {
1554
+ "epoch": 1.43,
1555
+ "learning_rate": 0.00016075,
1556
+ "loss": 1.3925,
1557
+ "step": 258
1558
+ },
1559
+ {
1560
+ "epoch": 1.44,
1561
+ "learning_rate": 0.0001605,
1562
+ "loss": 1.3881,
1563
+ "step": 259
1564
+ },
1565
+ {
1566
+ "epoch": 1.44,
1567
+ "learning_rate": 0.00016025000000000002,
1568
+ "loss": 1.3926,
1569
+ "step": 260
1570
+ },
1571
+ {
1572
+ "epoch": 1.45,
1573
+ "learning_rate": 0.00016,
1574
+ "loss": 1.3912,
1575
+ "step": 261
1576
+ },
1577
+ {
1578
+ "epoch": 1.46,
1579
+ "learning_rate": 0.00015975,
1580
+ "loss": 1.3906,
1581
+ "step": 262
1582
+ },
1583
+ {
1584
+ "epoch": 1.46,
1585
+ "learning_rate": 0.0001595,
1586
+ "loss": 1.3881,
1587
+ "step": 263
1588
+ },
1589
+ {
1590
+ "epoch": 1.47,
1591
+ "learning_rate": 0.00015925000000000002,
1592
+ "loss": 1.3869,
1593
+ "step": 264
1594
+ },
1595
+ {
1596
+ "epoch": 1.47,
1597
+ "learning_rate": 0.00015900000000000002,
1598
+ "loss": 1.3915,
1599
+ "step": 265
1600
+ },
1601
+ {
1602
+ "epoch": 1.48,
1603
+ "learning_rate": 0.00015875,
1604
+ "loss": 1.3899,
1605
+ "step": 266
1606
+ },
1607
+ {
1608
+ "epoch": 1.48,
1609
+ "learning_rate": 0.0001585,
1610
+ "loss": 1.387,
1611
+ "step": 267
1612
+ },
1613
+ {
1614
+ "epoch": 1.49,
1615
+ "learning_rate": 0.00015825,
1616
+ "loss": 1.3874,
1617
+ "step": 268
1618
+ },
1619
+ {
1620
+ "epoch": 1.49,
1621
+ "learning_rate": 0.00015800000000000002,
1622
+ "loss": 1.3911,
1623
+ "step": 269
1624
+ },
1625
+ {
1626
+ "epoch": 1.5,
1627
+ "learning_rate": 0.00015774999999999999,
1628
+ "loss": 1.3832,
1629
+ "step": 270
1630
+ },
1631
+ {
1632
+ "epoch": 1.51,
1633
+ "learning_rate": 0.0001575,
1634
+ "loss": 1.3916,
1635
+ "step": 271
1636
+ },
1637
+ {
1638
+ "epoch": 1.51,
1639
+ "learning_rate": 0.00015725,
1640
+ "loss": 1.3882,
1641
+ "step": 272
1642
+ },
1643
+ {
1644
+ "epoch": 1.52,
1645
+ "learning_rate": 0.00015700000000000002,
1646
+ "loss": 1.3844,
1647
+ "step": 273
1648
+ },
1649
+ {
1650
+ "epoch": 1.52,
1651
+ "learning_rate": 0.00015675,
1652
+ "loss": 1.391,
1653
+ "step": 274
1654
+ },
1655
+ {
1656
+ "epoch": 1.53,
1657
+ "learning_rate": 0.0001565,
1658
+ "loss": 1.3868,
1659
+ "step": 275
1660
+ },
1661
+ {
1662
+ "epoch": 1.53,
1663
+ "learning_rate": 0.00015625,
1664
+ "loss": 1.384,
1665
+ "step": 276
1666
+ },
1667
+ {
1668
+ "epoch": 1.54,
1669
+ "learning_rate": 0.00015600000000000002,
1670
+ "loss": 1.3825,
1671
+ "step": 277
1672
+ },
1673
+ {
1674
+ "epoch": 1.54,
1675
+ "learning_rate": 0.00015575000000000002,
1676
+ "loss": 1.3843,
1677
+ "step": 278
1678
+ },
1679
+ {
1680
+ "epoch": 1.55,
1681
+ "learning_rate": 0.0001555,
1682
+ "loss": 1.3842,
1683
+ "step": 279
1684
+ },
1685
+ {
1686
+ "epoch": 1.56,
1687
+ "learning_rate": 0.00015525,
1688
+ "loss": 1.3806,
1689
+ "step": 280
1690
+ },
1691
+ {
1692
+ "epoch": 1.56,
1693
+ "learning_rate": 0.000155,
1694
+ "loss": 1.382,
1695
+ "step": 281
1696
+ },
1697
+ {
1698
+ "epoch": 1.57,
1699
+ "learning_rate": 0.00015475000000000002,
1700
+ "loss": 1.3885,
1701
+ "step": 282
1702
+ },
1703
+ {
1704
+ "epoch": 1.57,
1705
+ "learning_rate": 0.0001545,
1706
+ "loss": 1.3863,
1707
+ "step": 283
1708
+ },
1709
+ {
1710
+ "epoch": 1.58,
1711
+ "learning_rate": 0.00015425,
1712
+ "loss": 1.3881,
1713
+ "step": 284
1714
+ },
1715
+ {
1716
+ "epoch": 1.58,
1717
+ "learning_rate": 0.000154,
1718
+ "loss": 1.3901,
1719
+ "step": 285
1720
+ },
1721
+ {
1722
+ "epoch": 1.59,
1723
+ "learning_rate": 0.00015375000000000002,
1724
+ "loss": 1.3802,
1725
+ "step": 286
1726
+ },
1727
+ {
1728
+ "epoch": 1.59,
1729
+ "learning_rate": 0.0001535,
1730
+ "loss": 1.3903,
1731
+ "step": 287
1732
+ },
1733
+ {
1734
+ "epoch": 1.6,
1735
+ "learning_rate": 0.00015325,
1736
+ "loss": 1.3905,
1737
+ "step": 288
1738
+ },
1739
+ {
1740
+ "epoch": 1.61,
1741
+ "learning_rate": 0.000153,
1742
+ "loss": 1.3858,
1743
+ "step": 289
1744
+ },
1745
+ {
1746
+ "epoch": 1.61,
1747
+ "learning_rate": 0.00015275000000000003,
1748
+ "loss": 1.3914,
1749
+ "step": 290
1750
+ },
1751
+ {
1752
+ "epoch": 1.62,
1753
+ "learning_rate": 0.0001525,
1754
+ "loss": 1.3874,
1755
+ "step": 291
1756
+ },
1757
+ {
1758
+ "epoch": 1.62,
1759
+ "learning_rate": 0.00015225000000000001,
1760
+ "loss": 1.3876,
1761
+ "step": 292
1762
+ },
1763
+ {
1764
+ "epoch": 1.63,
1765
+ "learning_rate": 0.000152,
1766
+ "loss": 1.385,
1767
+ "step": 293
1768
+ },
1769
+ {
1770
+ "epoch": 1.63,
1771
+ "learning_rate": 0.00015175,
1772
+ "loss": 1.3858,
1773
+ "step": 294
1774
+ },
1775
+ {
1776
+ "epoch": 1.64,
1777
+ "learning_rate": 0.0001515,
1778
+ "loss": 1.3849,
1779
+ "step": 295
1780
+ },
1781
+ {
1782
+ "epoch": 1.64,
1783
+ "learning_rate": 0.00015125,
1784
+ "loss": 1.39,
1785
+ "step": 296
1786
+ },
1787
+ {
1788
+ "epoch": 1.65,
1789
+ "learning_rate": 0.000151,
1790
+ "loss": 1.3869,
1791
+ "step": 297
1792
+ },
1793
+ {
1794
+ "epoch": 1.66,
1795
+ "learning_rate": 0.00015075,
1796
+ "loss": 1.3896,
1797
+ "step": 298
1798
+ },
1799
+ {
1800
+ "epoch": 1.66,
1801
+ "learning_rate": 0.0001505,
1802
+ "loss": 1.3926,
1803
+ "step": 299
1804
+ },
1805
+ {
1806
+ "epoch": 1.67,
1807
+ "learning_rate": 0.00015025,
1808
+ "loss": 1.3823,
1809
+ "step": 300
1810
+ },
1811
+ {
1812
+ "epoch": 1.67,
1813
+ "eval_loss": 1.3873900175094604,
1814
+ "eval_runtime": 1.5789,
1815
+ "eval_samples_per_second": 63.336,
1816
+ "eval_steps_per_second": 4.433,
1817
+ "step": 300
1818
+ }
1819
+ ],
1820
+ "logging_steps": 1,
1821
+ "max_steps": 900,
1822
+ "num_train_epochs": 5,
1823
+ "save_steps": 300,
1824
+ "total_flos": 6.755002364618342e+16,
1825
+ "trial_name": null,
1826
+ "trial_params": null
1827
+ }
last-checkpoint/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d0551ad96bb3a47f40b27101c8057f425b3b65b4a89dd1b9631ee395474864c
3
+ size 6200
last-checkpoint/zero_to_fp32.py ADDED
@@ -0,0 +1,587 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example: python zero_to_fp32.py . pytorch_model.bin
14
+
15
+ import argparse
16
+ import torch
17
+ import glob
18
+ import math
19
+ import os
20
+ import re
21
+ from collections import OrderedDict
22
+ from dataclasses import dataclass
23
+
24
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
25
+ # DeepSpeed data structures it has to be available in the current python environment.
26
+ from deepspeed.utils import logger
27
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
28
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
29
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
30
+
31
+
32
+ @dataclass
33
+ class zero_model_state:
34
+ buffers: dict()
35
+ param_shapes: dict()
36
+ shared_params: list
37
+ ds_version: int
38
+ frozen_param_shapes: dict()
39
+ frozen_param_fragments: dict()
40
+
41
+
42
+ debug = 0
43
+
44
+ # load to cpu
45
+ device = torch.device('cpu')
46
+
47
+
48
+ def atoi(text):
49
+ return int(text) if text.isdigit() else text
50
+
51
+
52
+ def natural_keys(text):
53
+ '''
54
+ alist.sort(key=natural_keys) sorts in human order
55
+ http://nedbatchelder.com/blog/200712/human_sorting.html
56
+ (See Toothy's implementation in the comments)
57
+ '''
58
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
59
+
60
+
61
+ def get_model_state_file(checkpoint_dir, zero_stage):
62
+ if not os.path.isdir(checkpoint_dir):
63
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
64
+
65
+ # there should be only one file
66
+ if zero_stage <= 2:
67
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
68
+ elif zero_stage == 3:
69
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
70
+
71
+ if not os.path.exists(file):
72
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
73
+
74
+ return file
75
+
76
+
77
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
78
+ # XXX: need to test that this simple glob rule works for multi-node setup too
79
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
80
+
81
+ if len(ckpt_files) == 0:
82
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
83
+
84
+ return ckpt_files
85
+
86
+
87
+ def get_optim_files(checkpoint_dir):
88
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
89
+
90
+
91
+ def get_model_state_files(checkpoint_dir):
92
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
93
+
94
+
95
+ def parse_model_states(files):
96
+ zero_model_states = []
97
+ for file in files:
98
+ state_dict = torch.load(file, map_location=device)
99
+
100
+ if BUFFER_NAMES not in state_dict:
101
+ raise ValueError(f"{file} is not a model state checkpoint")
102
+ buffer_names = state_dict[BUFFER_NAMES]
103
+ if debug:
104
+ print("Found buffers:", buffer_names)
105
+
106
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
107
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
108
+ param_shapes = state_dict[PARAM_SHAPES]
109
+
110
+ # collect parameters that are included in param_shapes
111
+ param_names = []
112
+ for s in param_shapes:
113
+ for name in s.keys():
114
+ param_names.append(name)
115
+
116
+ # update with frozen parameters
117
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
118
+ if frozen_param_shapes is not None:
119
+ if debug:
120
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
121
+ param_names += list(frozen_param_shapes.keys())
122
+
123
+ # handle shared params
124
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
125
+
126
+ ds_version = state_dict.get(DS_VERSION, None)
127
+
128
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
129
+
130
+ z_model_state = zero_model_state(buffers=buffers,
131
+ param_shapes=param_shapes,
132
+ shared_params=shared_params,
133
+ ds_version=ds_version,
134
+ frozen_param_shapes=frozen_param_shapes,
135
+ frozen_param_fragments=frozen_param_fragments)
136
+ zero_model_states.append(z_model_state)
137
+
138
+ return zero_model_states
139
+
140
+
141
+ def parse_optim_states(files, ds_checkpoint_dir):
142
+
143
+ total_files = len(files)
144
+ state_dicts = []
145
+ for f in files:
146
+ state_dict = torch.load(f, map_location=device)
147
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
148
+ # and also handle the case where it was already removed by another helper script
149
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
150
+ state_dicts.append(state_dict)
151
+
152
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
153
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
154
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
155
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
156
+
157
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
158
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
159
+ # use the max of the partition_count to get the dp world_size.
160
+
161
+ if type(world_size) is list:
162
+ world_size = max(world_size)
163
+
164
+ if world_size != total_files:
165
+ raise ValueError(
166
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
167
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
168
+ )
169
+
170
+ # the groups are named differently in each stage
171
+ if zero_stage <= 2:
172
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
173
+ elif zero_stage == 3:
174
+ fp32_groups_key = FP32_FLAT_GROUPS
175
+ else:
176
+ raise ValueError(f"unknown zero stage {zero_stage}")
177
+
178
+ if zero_stage <= 2:
179
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
180
+ elif zero_stage == 3:
181
+ # if there is more than one param group, there will be multiple flattened tensors - one
182
+ # flattened tensor per group - for simplicity merge them into a single tensor
183
+ #
184
+ # XXX: could make the script more memory efficient for when there are multiple groups - it
185
+ # will require matching the sub-lists of param_shapes for each param group flattened tensor
186
+
187
+ fp32_flat_groups = [
188
+ torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
189
+ ]
190
+
191
+ return zero_stage, world_size, fp32_flat_groups
192
+
193
+
194
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
195
+ """
196
+ Returns fp32 state_dict reconstructed from ds checkpoint
197
+
198
+ Args:
199
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
200
+
201
+ """
202
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
203
+
204
+ optim_files = get_optim_files(ds_checkpoint_dir)
205
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
206
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
207
+
208
+ model_files = get_model_state_files(ds_checkpoint_dir)
209
+
210
+ zero_model_states = parse_model_states(model_files)
211
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
212
+
213
+ if zero_stage <= 2:
214
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
215
+ elif zero_stage == 3:
216
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
217
+
218
+
219
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
220
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
221
+ return
222
+
223
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
224
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
225
+
226
+ if debug:
227
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
228
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
229
+
230
+ wanted_params = len(frozen_param_shapes)
231
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
232
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
233
+ print(f'Frozen params: Have {avail_numel} numels to process.')
234
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
235
+
236
+ total_params = 0
237
+ total_numel = 0
238
+ for name, shape in frozen_param_shapes.items():
239
+ total_params += 1
240
+ unpartitioned_numel = shape.numel()
241
+ total_numel += unpartitioned_numel
242
+
243
+ state_dict[name] = frozen_param_fragments[name]
244
+
245
+ if debug:
246
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
247
+
248
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
249
+
250
+
251
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
252
+ param_shapes = zero_model_states[0].param_shapes
253
+
254
+ # Reconstruction protocol:
255
+ #
256
+ # XXX: document this
257
+
258
+ if debug:
259
+ for i in range(world_size):
260
+ for j in range(len(fp32_flat_groups[0])):
261
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
262
+
263
+ # XXX: memory usage doubles here (zero2)
264
+ num_param_groups = len(fp32_flat_groups[0])
265
+ merged_single_partition_of_fp32_groups = []
266
+ for i in range(num_param_groups):
267
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
268
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
269
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
270
+ avail_numel = sum(
271
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
272
+
273
+ if debug:
274
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
275
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
276
+ # not asserting if there is a mismatch due to possible padding
277
+ print(f"Have {avail_numel} numels to process.")
278
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
279
+
280
+ # params
281
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
282
+ # out-of-core computing solution
283
+ total_numel = 0
284
+ total_params = 0
285
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
286
+ offset = 0
287
+ avail_numel = full_single_fp32_vector.numel()
288
+ for name, shape in shapes.items():
289
+
290
+ unpartitioned_numel = shape.numel()
291
+ total_numel += unpartitioned_numel
292
+ total_params += 1
293
+
294
+ if debug:
295
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
296
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
297
+ offset += unpartitioned_numel
298
+
299
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
300
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
301
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
302
+ # live optimizer object, so we are checking that the numbers are within the right range
303
+ align_to = 2 * world_size
304
+
305
+ def zero2_align(x):
306
+ return align_to * math.ceil(x / align_to)
307
+
308
+ if debug:
309
+ print(f"original offset={offset}, avail_numel={avail_numel}")
310
+
311
+ offset = zero2_align(offset)
312
+ avail_numel = zero2_align(avail_numel)
313
+
314
+ if debug:
315
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
316
+
317
+ # Sanity check
318
+ if offset != avail_numel:
319
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
320
+
321
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
322
+
323
+
324
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
325
+ state_dict = OrderedDict()
326
+
327
+ # buffers
328
+ buffers = zero_model_states[0].buffers
329
+ state_dict.update(buffers)
330
+ if debug:
331
+ print(f"added {len(buffers)} buffers")
332
+
333
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
334
+
335
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
336
+
337
+ # recover shared parameters
338
+ for pair in zero_model_states[0].shared_params:
339
+ if pair[1] in state_dict:
340
+ state_dict[pair[0]] = state_dict[pair[1]]
341
+
342
+ return state_dict
343
+
344
+
345
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
346
+ remainder = unpartitioned_numel % world_size
347
+ padding_numel = (world_size - remainder) if remainder else 0
348
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
349
+ return partitioned_numel, padding_numel
350
+
351
+
352
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
353
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
354
+ return
355
+
356
+ if debug:
357
+ for i in range(world_size):
358
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
359
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
360
+
361
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
362
+ wanted_params = len(frozen_param_shapes)
363
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
364
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
365
+ print(f'Frozen params: Have {avail_numel} numels to process.')
366
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
367
+
368
+ total_params = 0
369
+ total_numel = 0
370
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
371
+ total_params += 1
372
+ unpartitioned_numel = shape.numel()
373
+ total_numel += unpartitioned_numel
374
+
375
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
376
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
377
+
378
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
379
+
380
+ if debug:
381
+ print(
382
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
383
+ )
384
+
385
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
386
+
387
+
388
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
389
+ param_shapes = zero_model_states[0].param_shapes
390
+ avail_numel = fp32_flat_groups[0].numel() * world_size
391
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
392
+ # param, re-consolidating each param, while dealing with padding if any
393
+
394
+ # merge list of dicts, preserving order
395
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
396
+
397
+ if debug:
398
+ for i in range(world_size):
399
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
400
+
401
+ wanted_params = len(param_shapes)
402
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
403
+ # not asserting if there is a mismatch due to possible padding
404
+ avail_numel = fp32_flat_groups[0].numel() * world_size
405
+ print(f"Trainable params: Have {avail_numel} numels to process.")
406
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
407
+
408
+ # params
409
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
410
+ # out-of-core computing solution
411
+ offset = 0
412
+ total_numel = 0
413
+ total_params = 0
414
+ for name, shape in param_shapes.items():
415
+
416
+ unpartitioned_numel = shape.numel()
417
+ total_numel += unpartitioned_numel
418
+ total_params += 1
419
+
420
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
421
+
422
+ if debug:
423
+ print(
424
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
425
+ )
426
+
427
+ # XXX: memory usage doubles here
428
+ state_dict[name] = torch.cat(
429
+ tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
430
+ 0).narrow(0, 0, unpartitioned_numel).view(shape)
431
+ offset += partitioned_numel
432
+
433
+ offset *= world_size
434
+
435
+ # Sanity check
436
+ if offset != avail_numel:
437
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
438
+
439
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
440
+
441
+
442
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
443
+ state_dict = OrderedDict()
444
+
445
+ # buffers
446
+ buffers = zero_model_states[0].buffers
447
+ state_dict.update(buffers)
448
+ if debug:
449
+ print(f"added {len(buffers)} buffers")
450
+
451
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
452
+
453
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
454
+
455
+ # recover shared parameters
456
+ for pair in zero_model_states[0].shared_params:
457
+ if pair[1] in state_dict:
458
+ state_dict[pair[0]] = state_dict[pair[1]]
459
+
460
+ return state_dict
461
+
462
+
463
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
464
+ """
465
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
466
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
467
+ via a model hub.
468
+
469
+ Args:
470
+ - ``checkpoint_dir``: path to the desired checkpoint folder
471
+ - ``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``
472
+
473
+ Returns:
474
+ - pytorch ``state_dict``
475
+
476
+ Note: this approach may not work if your application doesn't have sufficient free CPU memory and
477
+ you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
478
+ the checkpoint.
479
+
480
+ A typical usage might be ::
481
+
482
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
483
+ # do the training and checkpoint saving
484
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
485
+ model = model.cpu() # move to cpu
486
+ model.load_state_dict(state_dict)
487
+ # submit to model hub or save the model to share with others
488
+
489
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
490
+ application. i.e. you will need to re-initialize the deepspeed engine, since
491
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
492
+
493
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
494
+
495
+ """
496
+ if tag is None:
497
+ latest_path = os.path.join(checkpoint_dir, 'latest')
498
+ if os.path.isfile(latest_path):
499
+ with open(latest_path, 'r') as fd:
500
+ tag = fd.read().strip()
501
+ else:
502
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
503
+
504
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
505
+
506
+ if not os.path.isdir(ds_checkpoint_dir):
507
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
508
+
509
+ return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
510
+
511
+
512
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
513
+ """
514
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
515
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
516
+
517
+ Args:
518
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
519
+ - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
520
+ - ``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``
521
+ """
522
+
523
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
524
+ print(f"Saving fp32 state dict to {output_file}")
525
+ torch.save(state_dict, output_file)
526
+
527
+
528
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
529
+ """
530
+ 1. Put the provided model to cpu
531
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
532
+ 3. Load it into the provided model
533
+
534
+ Args:
535
+ - ``model``: the model object to update
536
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
537
+ - ``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``
538
+
539
+ Returns:
540
+ - ``model`: modified model
541
+
542
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
543
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
544
+ conveniently placed for you in the checkpoint folder.
545
+
546
+ A typical usage might be ::
547
+
548
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
549
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
550
+ # submit to model hub or save the model to share with others
551
+
552
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
553
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
554
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
555
+
556
+ """
557
+ logger.info(f"Extracting fp32 weights")
558
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
559
+
560
+ logger.info(f"Overwriting model with fp32 weights")
561
+ model = model.cpu()
562
+ model.load_state_dict(state_dict, strict=False)
563
+
564
+ return model
565
+
566
+
567
+ if __name__ == "__main__":
568
+
569
+ parser = argparse.ArgumentParser()
570
+ parser.add_argument("checkpoint_dir",
571
+ type=str,
572
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
573
+ parser.add_argument(
574
+ "output_file",
575
+ type=str,
576
+ help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
577
+ parser.add_argument("-t",
578
+ "--tag",
579
+ type=str,
580
+ default=None,
581
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
582
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
583
+ args = parser.parse_args()
584
+
585
+ debug = args.debug
586
+
587
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)