hrw commited on
Commit
3dd812c
·
verified ·
1 Parent(s): 6514153

Upload folder using huggingface_hub

Browse files
README.md ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen2.5-Omni-7B
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:/workspace/haoran-cloud/models/Qwen2.5-Omni-7B/qwen/Qwen2___5-Omni-7B
7
+ - lora
8
+ - transformers
9
+ ---
10
+
11
+ # Model Card for Model ID
12
+
13
+ <!-- Provide a quick summary of what the model is/does. -->
14
+
15
+
16
+
17
+ ## Model Details
18
+
19
+ ### Model Description
20
+
21
+ <!-- Provide a longer summary of what this model is. -->
22
+
23
+
24
+
25
+ - **Developed by:** [More Information Needed]
26
+ - **Funded by [optional]:** [More Information Needed]
27
+ - **Shared by [optional]:** [More Information Needed]
28
+ - **Model type:** [More Information Needed]
29
+ - **Language(s) (NLP):** [More Information Needed]
30
+ - **License:** [More Information Needed]
31
+ - **Finetuned from model [optional]:** [More Information Needed]
32
+
33
+ ### Model Sources [optional]
34
+
35
+ <!-- Provide the basic links for the model. -->
36
+
37
+ - **Repository:** [More Information Needed]
38
+ - **Paper [optional]:** [More Information Needed]
39
+ - **Demo [optional]:** [More Information Needed]
40
+
41
+ ## Uses
42
+
43
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
44
+
45
+ ### Direct Use
46
+
47
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
48
+
49
+ [More Information Needed]
50
+
51
+ ### Downstream Use [optional]
52
+
53
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
54
+
55
+ [More Information Needed]
56
+
57
+ ### Out-of-Scope Use
58
+
59
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
60
+
61
+ [More Information Needed]
62
+
63
+ ## Bias, Risks, and Limitations
64
+
65
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
66
+
67
+ [More Information Needed]
68
+
69
+ ### Recommendations
70
+
71
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
72
+
73
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
74
+
75
+ ## How to Get Started with the Model
76
+
77
+ Use the code below to get started with the model.
78
+
79
+ [More Information Needed]
80
+
81
+ ## Training Details
82
+
83
+ ### Training Data
84
+
85
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
86
+
87
+ [More Information Needed]
88
+
89
+ ### Training Procedure
90
+
91
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
92
+
93
+ #### Preprocessing [optional]
94
+
95
+ [More Information Needed]
96
+
97
+
98
+ #### Training Hyperparameters
99
+
100
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
101
+
102
+ #### Speeds, Sizes, Times [optional]
103
+
104
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
105
+
106
+ [More Information Needed]
107
+
108
+ ## Evaluation
109
+
110
+ <!-- This section describes the evaluation protocols and provides the results. -->
111
+
112
+ ### Testing Data, Factors & Metrics
113
+
114
+ #### Testing Data
115
+
116
+ <!-- This should link to a Dataset Card if possible. -->
117
+
118
+ [More Information Needed]
119
+
120
+ #### Factors
121
+
122
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
123
+
124
+ [More Information Needed]
125
+
126
+ #### Metrics
127
+
128
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
129
+
130
+ [More Information Needed]
131
+
132
+ ### Results
133
+
134
+ [More Information Needed]
135
+
136
+ #### Summary
137
+
138
+
139
+
140
+ ## Model Examination [optional]
141
+
142
+ <!-- Relevant interpretability work for the model goes here -->
143
+
144
+ [More Information Needed]
145
+
146
+ ## Environmental Impact
147
+
148
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
149
+
150
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
151
+
152
+ - **Hardware Type:** [More Information Needed]
153
+ - **Hours used:** [More Information Needed]
154
+ - **Cloud Provider:** [More Information Needed]
155
+ - **Compute Region:** [More Information Needed]
156
+ - **Carbon Emitted:** [More Information Needed]
157
+
158
+ ## Technical Specifications [optional]
159
+
160
+ ### Model Architecture and Objective
161
+
162
+ [More Information Needed]
163
+
164
+ ### Compute Infrastructure
165
+
166
+ [More Information Needed]
167
+
168
+ #### Hardware
169
+
170
+ [More Information Needed]
171
+
172
+ #### Software
173
+
174
+ [More Information Needed]
175
+
176
+ ## Citation [optional]
177
+
178
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
179
+
180
+ **BibTeX:**
181
+
182
+ [More Information Needed]
183
+
184
+ **APA:**
185
+
186
+ [More Information Needed]
187
+
188
+ ## Glossary [optional]
189
+
190
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
191
+
192
+ [More Information Needed]
193
+
194
+ ## More Information [optional]
195
+
196
+ [More Information Needed]
197
+
198
+ ## Model Card Authors [optional]
199
+
200
+ [More Information Needed]
201
+
202
+ ## Model Card Contact
203
+
204
+ [More Information Needed]
205
+ ### Framework versions
206
+
207
+ - PEFT 0.16.0
adapter_config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/workspace/haoran-cloud/models/Qwen2.5-Omni-7B/qwen/Qwen2___5-Omni-7B",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 32,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": [],
22
+ "peft_type": "LORA",
23
+ "qalora_group_size": 16,
24
+ "r": 8,
25
+ "rank_pattern": {},
26
+ "revision": null,
27
+ "target_modules": "^(thinker.model.*\\.(v_proj|gate_proj|o_proj|up_proj|down_proj|k_proj|q_proj))$",
28
+ "task_type": "CAUSAL_LM",
29
+ "trainable_token_indices": null,
30
+ "use_dora": false,
31
+ "use_qalora": false,
32
+ "use_rslora": false
33
+ }
adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbe70527e61de1778638f21c96914388e75884dfd71bced5df5ac87d0979f8be
3
+ size 40425344
additional_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"lora_dtype": null, "lorap_lr_ratio": null, "lorap_emb_lr": 1e-06}
args.json ADDED
@@ -0,0 +1,367 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "output_dir": "/workspace/haoran-cloud/omni/OmniCritic/nothink-outputs/v1-20250918-084101",
3
+ "overwrite_output_dir": false,
4
+ "do_train": true,
5
+ "do_eval": false,
6
+ "do_predict": false,
7
+ "eval_strategy": "no",
8
+ "prediction_loss_only": false,
9
+ "per_device_train_batch_size": 1,
10
+ "per_device_eval_batch_size": 1,
11
+ "per_gpu_train_batch_size": null,
12
+ "per_gpu_eval_batch_size": null,
13
+ "gradient_accumulation_steps": 12,
14
+ "eval_accumulation_steps": null,
15
+ "eval_delay": 0,
16
+ "torch_empty_cache_steps": null,
17
+ "learning_rate": 5e-06,
18
+ "weight_decay": 0.1,
19
+ "adam_beta1": 0.9,
20
+ "adam_beta2": 0.95,
21
+ "adam_epsilon": 1e-08,
22
+ "max_grad_norm": 1.0,
23
+ "num_train_epochs": 2.0,
24
+ "max_steps": -1,
25
+ "lr_scheduler_type": "cosine",
26
+ "lr_scheduler_kwargs": null,
27
+ "warmup_ratio": 0.03,
28
+ "warmup_steps": 0,
29
+ "log_level": "passive",
30
+ "log_level_replica": "warning",
31
+ "log_on_each_node": true,
32
+ "logging_dir": "/workspace/haoran-cloud/omni/OmniCritic/nothink-outputs/v1-20250918-084101/runs",
33
+ "logging_strategy": "steps",
34
+ "logging_first_step": true,
35
+ "logging_steps": 5,
36
+ "logging_nan_inf_filter": true,
37
+ "save_strategy": "steps",
38
+ "save_steps": 100.0,
39
+ "save_total_limit": 4,
40
+ "save_safetensors": true,
41
+ "save_on_each_node": false,
42
+ "save_only_model": false,
43
+ "restore_callback_states_from_checkpoint": false,
44
+ "no_cuda": false,
45
+ "use_cpu": false,
46
+ "use_mps_device": false,
47
+ "seed": 42,
48
+ "data_seed": 42,
49
+ "jit_mode_eval": false,
50
+ "use_ipex": false,
51
+ "bf16": true,
52
+ "fp16": false,
53
+ "fp16_opt_level": "O1",
54
+ "half_precision_backend": "auto",
55
+ "bf16_full_eval": false,
56
+ "fp16_full_eval": false,
57
+ "tf32": null,
58
+ "local_rank": 0,
59
+ "ddp_backend": null,
60
+ "tpu_num_cores": null,
61
+ "tpu_metrics_debug": false,
62
+ "debug": null,
63
+ "dataloader_drop_last": false,
64
+ "eval_steps": 100.0,
65
+ "dataloader_num_workers": 2,
66
+ "dataloader_prefetch_factor": null,
67
+ "past_index": -1,
68
+ "run_name": "/workspace/haoran-cloud/omni/OmniCritic/nothink-outputs/v1-20250918-084101",
69
+ "disable_tqdm": null,
70
+ "remove_unused_columns": true,
71
+ "label_names": null,
72
+ "load_best_model_at_end": false,
73
+ "metric_for_best_model": "loss",
74
+ "greater_is_better": false,
75
+ "ignore_data_skip": false,
76
+ "fsdp": "",
77
+ "fsdp_min_num_params": 0,
78
+ "fsdp_config": null,
79
+ "fsdp_transformer_layer_cls_to_wrap": null,
80
+ "accelerator_config": {
81
+ "dispatch_batches": false
82
+ },
83
+ "deepspeed": {
84
+ "fp16": {
85
+ "enabled": "auto",
86
+ "loss_scale": 0,
87
+ "loss_scale_window": 1000,
88
+ "initial_scale_power": 16,
89
+ "hysteresis": 2,
90
+ "min_loss_scale": 1
91
+ },
92
+ "bf16": {
93
+ "enabled": "auto"
94
+ },
95
+ "zero_optimization": {
96
+ "stage": 2,
97
+ "offload_optimizer": {
98
+ "device": "none",
99
+ "pin_memory": true
100
+ },
101
+ "allgather_partitions": true,
102
+ "allgather_bucket_size": 200000000.0,
103
+ "overlap_comm": false,
104
+ "reduce_scatter": true,
105
+ "reduce_bucket_size": 200000000.0,
106
+ "contiguous_gradients": true
107
+ },
108
+ "gradient_accumulation_steps": "auto",
109
+ "gradient_clipping": "auto",
110
+ "steps_per_print": 2000,
111
+ "train_batch_size": "auto",
112
+ "train_micro_batch_size_per_gpu": "auto",
113
+ "wall_clock_breakdown": false
114
+ },
115
+ "label_smoothing_factor": 0.0,
116
+ "optim": "adamw_torch",
117
+ "optim_args": null,
118
+ "adafactor": false,
119
+ "group_by_length": false,
120
+ "length_column_name": "length",
121
+ "report_to": [
122
+ "tensorboard"
123
+ ],
124
+ "ddp_find_unused_parameters": null,
125
+ "ddp_bucket_cap_mb": null,
126
+ "ddp_broadcast_buffers": null,
127
+ "dataloader_pin_memory": true,
128
+ "dataloader_persistent_workers": false,
129
+ "skip_memory_metrics": true,
130
+ "use_legacy_prediction_loop": false,
131
+ "push_to_hub": false,
132
+ "resume_from_checkpoint": null,
133
+ "hub_model_id": null,
134
+ "hub_strategy": "every_save",
135
+ "hub_token": null,
136
+ "hub_private_repo": null,
137
+ "hub_always_push": false,
138
+ "hub_revision": null,
139
+ "gradient_checkpointing": true,
140
+ "gradient_checkpointing_kwargs": null,
141
+ "include_inputs_for_metrics": false,
142
+ "include_for_metrics": [],
143
+ "eval_do_concat_batches": true,
144
+ "fp16_backend": "auto",
145
+ "push_to_hub_model_id": null,
146
+ "push_to_hub_organization": null,
147
+ "push_to_hub_token": null,
148
+ "mp_parameters": "",
149
+ "auto_find_batch_size": false,
150
+ "full_determinism": false,
151
+ "torchdynamo": null,
152
+ "ray_scope": "last",
153
+ "ddp_timeout": 18000000,
154
+ "torch_compile": false,
155
+ "torch_compile_backend": null,
156
+ "torch_compile_mode": null,
157
+ "include_tokens_per_second": false,
158
+ "include_num_input_tokens_seen": false,
159
+ "neftune_noise_alpha": null,
160
+ "optim_target_modules": null,
161
+ "batch_eval_metrics": false,
162
+ "eval_on_start": false,
163
+ "use_liger_kernel": false,
164
+ "liger_kernel_config": null,
165
+ "eval_use_gather_object": false,
166
+ "average_tokens_across_devices": false,
167
+ "sortish_sampler": false,
168
+ "predict_with_generate": false,
169
+ "generation_max_length": null,
170
+ "generation_num_beams": null,
171
+ "generation_config": null,
172
+ "vit_gradient_checkpointing": null,
173
+ "check_model": true,
174
+ "acc_strategy": "token",
175
+ "train_dataloader_shuffle": true,
176
+ "max_epochs": null,
177
+ "aligner_lr": null,
178
+ "vit_lr": null,
179
+ "optimizer": null,
180
+ "use_logits_to_keep": null,
181
+ "channels": null,
182
+ "ds3_gather_for_generation": true,
183
+ "resume_only_model": false,
184
+ "metric_warmup_step": 0,
185
+ "fsdp_num": 1,
186
+ "acc_steps": 1,
187
+ "eval_use_evalscope": false,
188
+ "eval_dataset": [],
189
+ "eval_dataset_args": null,
190
+ "eval_limit": null,
191
+ "eval_generation_config": null,
192
+ "model": "/workspace/haoran-cloud/models/Qwen2.5-Omni-7B/qwen/Qwen2___5-Omni-7B/",
193
+ "model_type": "qwen2_5_omni",
194
+ "model_revision": null,
195
+ "task_type": "causal_lm",
196
+ "torch_dtype": "bfloat16",
197
+ "attn_impl": null,
198
+ "new_special_tokens": [],
199
+ "num_labels": null,
200
+ "problem_type": null,
201
+ "rope_scaling": null,
202
+ "device_map": null,
203
+ "max_memory": {},
204
+ "local_repo_path": null,
205
+ "init_strategy": null,
206
+ "template": "qwen2_5_omni",
207
+ "system": null,
208
+ "max_length": 3072,
209
+ "truncation_strategy": "delete",
210
+ "max_pixels": null,
211
+ "agent_template": null,
212
+ "norm_bbox": null,
213
+ "use_chat_template": true,
214
+ "padding_free": false,
215
+ "padding_side": "right",
216
+ "loss_scale": "default",
217
+ "sequence_parallel_size": 1,
218
+ "response_prefix": null,
219
+ "template_backend": "swift",
220
+ "dataset": [
221
+ "/workspace/haoran-cloud/omni/nothinking-training/dataset/audio/new_final_sft_data.jsonl",
222
+ "/workspace/haoran-cloud/omni/nothinking-training/dataset/video/new_final_sft_data.jsonl",
223
+ "/workspace/haoran-cloud/omni/nothinking-training/dataset/image/new_final_sft_data.jsonl"
224
+ ],
225
+ "val_dataset": [],
226
+ "split_dataset_ratio": 0.0,
227
+ "dataset_num_proc": 1,
228
+ "load_from_cache_file": true,
229
+ "dataset_shuffle": true,
230
+ "val_dataset_shuffle": false,
231
+ "streaming": false,
232
+ "interleave_prob": null,
233
+ "stopping_strategy": "first_exhausted",
234
+ "shuffle_buffer_size": 1000,
235
+ "download_mode": "reuse_dataset_if_exists",
236
+ "columns": {},
237
+ "strict": false,
238
+ "model_name": null,
239
+ "model_author": null,
240
+ "custom_dataset_info": [],
241
+ "quant_method": null,
242
+ "quant_bits": null,
243
+ "hqq_axis": null,
244
+ "bnb_4bit_compute_dtype": "bfloat16",
245
+ "bnb_4bit_quant_type": "nf4",
246
+ "bnb_4bit_use_double_quant": true,
247
+ "bnb_4bit_quant_storage": null,
248
+ "max_new_tokens": 64,
249
+ "temperature": 0.0,
250
+ "top_k": null,
251
+ "top_p": null,
252
+ "repetition_penalty": null,
253
+ "num_beams": 1,
254
+ "stream": false,
255
+ "stop_words": [],
256
+ "logprobs": false,
257
+ "top_logprobs": null,
258
+ "ckpt_dir": null,
259
+ "lora_modules": [],
260
+ "tuner_backend": "peft",
261
+ "train_type": "lora",
262
+ "adapters": [],
263
+ "external_plugins": [],
264
+ "model_kwargs": {},
265
+ "load_args": false,
266
+ "load_data_args": false,
267
+ "packing": false,
268
+ "custom_register_path": [],
269
+ "use_hf": false,
270
+ "ignore_args_error": false,
271
+ "use_swift_lora": false,
272
+ "freeze_parameters": [],
273
+ "freeze_parameters_regex": null,
274
+ "freeze_parameters_ratio": 0.0,
275
+ "trainable_parameters": [],
276
+ "trainable_parameters_regex": null,
277
+ "freeze_llm": false,
278
+ "freeze_vit": true,
279
+ "freeze_aligner": true,
280
+ "target_modules": [
281
+ "all-linear"
282
+ ],
283
+ "target_regex": null,
284
+ "modules_to_save": [],
285
+ "lora_rank": 8,
286
+ "lora_alpha": 32,
287
+ "lora_dropout": 0.05,
288
+ "lora_bias": "none",
289
+ "lora_dtype": null,
290
+ "lorap_lr_ratio": null,
291
+ "use_rslora": false,
292
+ "use_dora": false,
293
+ "lora_ga_batch_size": 2,
294
+ "lora_ga_iters": 2,
295
+ "lora_ga_max_length": 1024,
296
+ "lora_ga_direction": "ArB2r",
297
+ "lora_ga_scale": "stable",
298
+ "lora_ga_stable_gamma": 16,
299
+ "init_weights": true,
300
+ "fourier_n_frequency": 2000,
301
+ "fourier_scaling": 300.0,
302
+ "boft_block_size": 4,
303
+ "boft_block_num": 0,
304
+ "boft_n_butterfly_factor": 1,
305
+ "boft_dropout": 0.0,
306
+ "vera_rank": 256,
307
+ "vera_projection_prng_key": 0,
308
+ "vera_dropout": 0.0,
309
+ "vera_d_initial": 0.1,
310
+ "adapter_act": "gelu",
311
+ "adapter_length": 128,
312
+ "use_galore": false,
313
+ "galore_target_modules": null,
314
+ "galore_rank": 128,
315
+ "galore_update_proj_gap": 50,
316
+ "galore_scale": 1.0,
317
+ "galore_proj_type": "std",
318
+ "galore_optim_per_parameter": false,
319
+ "galore_with_embedding": false,
320
+ "galore_quantization": false,
321
+ "galore_proj_quant": false,
322
+ "galore_proj_bits": 4,
323
+ "galore_proj_group_size": 256,
324
+ "galore_cos_threshold": 0.4,
325
+ "galore_gamma_proj": 2,
326
+ "galore_queue_size": 5,
327
+ "adalora_target_r": 8,
328
+ "adalora_init_r": 12,
329
+ "adalora_tinit": 0,
330
+ "adalora_tfinal": 0,
331
+ "adalora_deltaT": 1,
332
+ "adalora_beta1": 0.85,
333
+ "adalora_beta2": 0.85,
334
+ "adalora_orth_reg_weight": 0.5,
335
+ "llamapro_num_new_blocks": 4,
336
+ "llamapro_num_groups": null,
337
+ "lisa_activated_layers": 0,
338
+ "lisa_step_interval": 20,
339
+ "reft_layer_key": null,
340
+ "reft_layers": null,
341
+ "reft_rank": 4,
342
+ "reft_intervention_type": "LoreftIntervention",
343
+ "reft_args": null,
344
+ "swanlab_token": null,
345
+ "swanlab_project": null,
346
+ "swanlab_workspace": null,
347
+ "swanlab_exp_name": null,
348
+ "swanlab_lark_webhook_url": null,
349
+ "swanlab_lark_secret": null,
350
+ "swanlab_mode": "cloud",
351
+ "add_version": true,
352
+ "create_checkpoint_symlink": false,
353
+ "lazy_tokenize": true,
354
+ "loss_type": null,
355
+ "metric": null,
356
+ "zero_hpz_partition_size": null,
357
+ "rank": 0,
358
+ "global_world_size": 4,
359
+ "local_world_size": 4,
360
+ "model_suffix": "Qwen2.5-Omni-7B",
361
+ "model_info": "ModelInfo(model_type='qwen2_5_omni', model_dir='/workspace/haoran-cloud/models/Qwen2.5-Omni-7B/qwen/Qwen2___5-Omni-7B', torch_dtype=torch.bfloat16, max_model_len=32768, quant_method=None, quant_bits=None, rope_scaling={'mrope_section': [16, 24, 24], 'rope_type': 'default', 'type': 'default'}, is_moe_model=False, config=None, task_type='causal_lm', num_labels=None)",
362
+ "model_meta": "ModelMeta(model_type='qwen2_5_omni', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen2.5-Omni-3B', hf_model_id='Qwen/Qwen2.5-Omni-3B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen2.5-Omni-7B', hf_model_id='Qwen/Qwen2.5-Omni-7B', model_path=None, ms_revision=None, hf_revision=None)], ignore_patterns=None, requires=None, tags=[])], template='qwen2_5_omni', get_function=<function get_model_tokenizer_qwen2_5_omni at 0x7f6750ba7420>, model_arch='qwen2_5_omni', architectures=['Qwen2_5OmniModel', 'Qwen2_5OmniForConditionalGeneration'], additional_saved_files=['spk_dict.pt'], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=[], requires=['transformers>=4.50', 'soundfile', 'qwen_omni_utils', 'decord'], tags=['vision', 'video', 'audio'])",
363
+ "model_dir": "/workspace/haoran-cloud/models/Qwen2.5-Omni-7B/qwen/Qwen2___5-Omni-7B",
364
+ "hub": "<class 'swift.hub.hub.MSHub'>",
365
+ "evaluation_strategy": "steps",
366
+ "training_args": "Seq2SeqTrainingArguments(output_dir='/workspace/haoran-cloud/omni/OmniCritic/nothink-outputs/v1-20250918-084101', overwrite_output_dir=False, do_train=True, do_eval=False, do_predict=False, eval_strategy=<IntervalStrategy.NO: 'no'>, prediction_loss_only=False, per_device_train_batch_size=1, per_device_eval_batch_size=1, per_gpu_train_batch_size=None, per_gpu_eval_batch_size=None, gradient_accumulation_steps=12, eval_accumulation_steps=None, eval_delay=0, torch_empty_cache_steps=None, learning_rate=5e-06, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, max_grad_norm=1.0, num_train_epochs=2.0, max_steps=-1, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_ratio=0.03, warmup_steps=0, log_level='passive', log_level_replica='warning', log_on_each_node=True, logging_dir='/workspace/haoran-cloud/omni/OmniCritic/nothink-outputs/v1-20250918-084101/runs', logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_first_step=True, logging_steps=5, logging_nan_inf_filter=True, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=100, save_total_limit=4, save_safetensors=True, save_on_each_node=False, save_only_model=False, restore_callback_states_from_checkpoint=False, no_cuda=False, use_cpu=False, use_mps_device=False, seed=42, data_seed=42, jit_mode_eval=False, use_ipex=False, bf16=True, fp16=False, fp16_opt_level='O1', half_precision_backend='auto', bf16_full_eval=False, fp16_full_eval=False, tf32=None, local_rank=0, ddp_backend=None, tpu_num_cores=None, tpu_metrics_debug=False, debug=[], dataloader_drop_last=False, eval_steps=100.0, dataloader_num_workers=2, dataloader_prefetch_factor=10, past_index=-1, run_name='/workspace/haoran-cloud/omni/OmniCritic/nothink-outputs/v1-20250918-084101', disable_tqdm=False, remove_unused_columns=False, label_names=None, load_best_model_at_end=False, metric_for_best_model='loss', greater_is_better=False, ignore_data_skip=False, fsdp=[], fsdp_min_num_params=0, fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, fsdp_transformer_layer_cls_to_wrap=None, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': False, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, label_smoothing_factor=0.0, optim=<OptimizerNames.ADAMW_TORCH: 'adamw_torch'>, optim_args=None, adafactor=False, group_by_length=False, length_column_name='length', report_to=['tensorboard'], ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, dataloader_pin_memory=True, dataloader_persistent_workers=False, skip_memory_metrics=True, use_legacy_prediction_loop=False, push_to_hub=False, resume_from_checkpoint=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_token=None, hub_private_repo=None, hub_always_push=False, hub_revision=None, gradient_checkpointing=True, gradient_checkpointing_kwargs=None, include_inputs_for_metrics=False, include_for_metrics=[], eval_do_concat_batches=True, fp16_backend='auto', push_to_hub_model_id=None, push_to_hub_organization=None, push_to_hub_token=None, mp_parameters='', auto_find_batch_size=False, full_determinism=False, torchdynamo=None, ray_scope='last', ddp_timeout=18000000, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, include_tokens_per_second=None, include_num_input_tokens_seen=None, neftune_noise_alpha=None, optim_target_modules=None, batch_eval_metrics=False, eval_on_start=False, use_liger_kernel=False, liger_kernel_config=None, eval_use_gather_object=False, average_tokens_across_devices=None, sortish_sampler=False, predict_with_generate=False, generation_max_length=None, generation_num_beams=None, generation_config=None, vit_gradient_checkpointing=True, check_model=True, acc_strategy='token', train_dataloader_shuffle=True, max_epochs=None, aligner_lr=None, vit_lr=None, optimizer=None, use_logits_to_keep=None, channels=None, ds3_gather_for_generation=True, resume_only_model=False, metric_warmup_step=0, fsdp_num=1, acc_steps=1, eval_use_evalscope=False, eval_dataset=[], eval_dataset_args=None, eval_limit=None, eval_generation_config=None, sft_alpha=0, train_type='lora', local_repo_path=None, galore_config=None)"
367
+ }
global_step1098/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:fa2800f7cb28f5a6e5bd08ccfa247ec1ad4bb8425c85bac7adc1cf0086e4822d
3
+ size 60570224
global_step1098/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:dbfbb2cc207d9b4a4ec27e5ddd806c7cc922f16eaeae3aac40965398bbf0779f
3
+ size 60570288
global_step1098/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57f7273127a5090701c8853e1462719576db19fd017481d16d1dfa4063b64ac6
3
+ size 60570352
global_step1098/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78d88c5efd388646ad06842420291df0dbf1ad1093b54fcfe19eea6437a02032
3
+ size 60570352
global_step1098/mp_rank_00_model_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecc5e2020b9048676daf8ae5fa8f3d2eb3d6eeb8cac020f18579f52ae40c8ded
3
+ size 40901816
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step1098
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:44db698130de8a15418c674ee00a25c86d0d4dec58e9bf7a888def28ea3b2d4a
3
+ size 15024
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:557232deb7c22baf07e93e20a562ff1bcf79026611e8896b43bee52d0169c848
3
+ size 15024
rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:98629a99a0bbea55ffd2edbbbec950d83e2411b407c75bb13562ff478b4d61cd
3
+ size 15024
rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3cf86bde132c134fc8c349ff9821df3c09d776e7ff9b0384bf545f0473ea45e3
3
+ size 15024
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dd94944a325240c3c366378ddb559791cf9fe3903dde65a230985e02116cab81
3
+ size 1064
trainer_state.json ADDED
@@ -0,0 +1,2244 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 2.0,
6
+ "eval_steps": 100.0,
7
+ "global_step": 1100,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.0018203883495145632,
14
+ "grad_norm": 3.6015546321868896,
15
+ "learning_rate": 1.5151515151515152e-07,
16
+ "loss": 0.5787061452865601,
17
+ "memory(GiB)": 40.37,
18
+ "step": 1,
19
+ "token_acc": 0.8690476190476191,
20
+ "train_speed(iter/s)": 0.029162
21
+ },
22
+ {
23
+ "epoch": 0.009101941747572815,
24
+ "grad_norm": 3.435741424560547,
25
+ "learning_rate": 7.575757575757576e-07,
26
+ "loss": 0.5881168842315674,
27
+ "memory(GiB)": 40.37,
28
+ "step": 5,
29
+ "token_acc": 0.8701684836471755,
30
+ "train_speed(iter/s)": 0.06174
31
+ },
32
+ {
33
+ "epoch": 0.01820388349514563,
34
+ "grad_norm": 3.448568105697632,
35
+ "learning_rate": 1.5151515151515152e-06,
36
+ "loss": 0.5851926326751709,
37
+ "memory(GiB)": 40.39,
38
+ "step": 10,
39
+ "token_acc": 0.8622327790973872,
40
+ "train_speed(iter/s)": 0.067399
41
+ },
42
+ {
43
+ "epoch": 0.027305825242718445,
44
+ "grad_norm": 3.405535936355591,
45
+ "learning_rate": 2.2727272727272728e-06,
46
+ "loss": 0.6001698970794678,
47
+ "memory(GiB)": 40.39,
48
+ "step": 15,
49
+ "token_acc": 0.8716323296354992,
50
+ "train_speed(iter/s)": 0.068625
51
+ },
52
+ {
53
+ "epoch": 0.03640776699029126,
54
+ "grad_norm": 3.6892027854919434,
55
+ "learning_rate": 3.0303030303030305e-06,
56
+ "loss": 0.5676679611206055,
57
+ "memory(GiB)": 40.39,
58
+ "step": 20,
59
+ "token_acc": 0.8715305313243458,
60
+ "train_speed(iter/s)": 0.069219
61
+ },
62
+ {
63
+ "epoch": 0.04550970873786408,
64
+ "grad_norm": 3.9115183353424072,
65
+ "learning_rate": 3.7878787878787882e-06,
66
+ "loss": 0.5411659717559815,
67
+ "memory(GiB)": 40.39,
68
+ "step": 25,
69
+ "token_acc": 0.8685669041963578,
70
+ "train_speed(iter/s)": 0.071459
71
+ },
72
+ {
73
+ "epoch": 0.05461165048543689,
74
+ "grad_norm": 3.357640027999878,
75
+ "learning_rate": 4.5454545454545455e-06,
76
+ "loss": 0.4885613441467285,
77
+ "memory(GiB)": 40.39,
78
+ "step": 30,
79
+ "token_acc": 0.8682539682539683,
80
+ "train_speed(iter/s)": 0.072567
81
+ },
82
+ {
83
+ "epoch": 0.06371359223300971,
84
+ "grad_norm": 3.3015964031219482,
85
+ "learning_rate": 4.999956654935265e-06,
86
+ "loss": 0.4215705871582031,
87
+ "memory(GiB)": 40.39,
88
+ "step": 35,
89
+ "token_acc": 0.8692551505546752,
90
+ "train_speed(iter/s)": 0.07282
91
+ },
92
+ {
93
+ "epoch": 0.07281553398058252,
94
+ "grad_norm": 2.447498321533203,
95
+ "learning_rate": 4.999469040218251e-06,
96
+ "loss": 0.2957149982452393,
97
+ "memory(GiB)": 40.39,
98
+ "step": 40,
99
+ "token_acc": 0.8690476190476191,
100
+ "train_speed(iter/s)": 0.074291
101
+ },
102
+ {
103
+ "epoch": 0.08191747572815535,
104
+ "grad_norm": 0.8801060914993286,
105
+ "learning_rate": 4.9984397354824345e-06,
106
+ "loss": 0.21340658664703369,
107
+ "memory(GiB)": 40.39,
108
+ "step": 45,
109
+ "token_acc": 0.9135606661379857,
110
+ "train_speed(iter/s)": 0.074508
111
+ },
112
+ {
113
+ "epoch": 0.09101941747572816,
114
+ "grad_norm": 0.7226303815841675,
115
+ "learning_rate": 4.996868963800831e-06,
116
+ "loss": 0.1777859926223755,
117
+ "memory(GiB)": 40.39,
118
+ "step": 50,
119
+ "token_acc": 0.9239904988123515,
120
+ "train_speed(iter/s)": 0.075304
121
+ },
122
+ {
123
+ "epoch": 0.10012135922330097,
124
+ "grad_norm": 0.7329442501068115,
125
+ "learning_rate": 4.99475706559428e-06,
126
+ "loss": 0.17081427574157715,
127
+ "memory(GiB)": 40.39,
128
+ "step": 55,
129
+ "token_acc": 0.9238699444885012,
130
+ "train_speed(iter/s)": 0.075664
131
+ },
132
+ {
133
+ "epoch": 0.10922330097087378,
134
+ "grad_norm": 0.48636820912361145,
135
+ "learning_rate": 4.992104498557657e-06,
136
+ "loss": 0.15634163618087768,
137
+ "memory(GiB)": 40.39,
138
+ "step": 60,
139
+ "token_acc": 0.9262490087232356,
140
+ "train_speed(iter/s)": 0.076641
141
+ },
142
+ {
143
+ "epoch": 0.1183252427184466,
144
+ "grad_norm": 0.44267499446868896,
145
+ "learning_rate": 4.988911837560691e-06,
146
+ "loss": 0.1444383144378662,
147
+ "memory(GiB)": 40.39,
148
+ "step": 65,
149
+ "token_acc": 0.9350237717908082,
150
+ "train_speed(iter/s)": 0.074923
151
+ },
152
+ {
153
+ "epoch": 0.12742718446601942,
154
+ "grad_norm": 0.4311356544494629,
155
+ "learning_rate": 4.985179774523375e-06,
156
+ "loss": 0.14677078723907472,
157
+ "memory(GiB)": 40.39,
158
+ "step": 70,
159
+ "token_acc": 0.9444444444444444,
160
+ "train_speed(iter/s)": 0.075087
161
+ },
162
+ {
163
+ "epoch": 0.13652912621359223,
164
+ "grad_norm": 0.3981742858886719,
165
+ "learning_rate": 4.980909118266006e-06,
166
+ "loss": 0.13511970043182372,
167
+ "memory(GiB)": 40.39,
168
+ "step": 75,
169
+ "token_acc": 0.9484944532488114,
170
+ "train_speed(iter/s)": 0.074414
171
+ },
172
+ {
173
+ "epoch": 0.14563106796116504,
174
+ "grad_norm": 0.4317033290863037,
175
+ "learning_rate": 4.976100794333903e-06,
176
+ "loss": 0.12185637950897217,
177
+ "memory(GiB)": 40.39,
178
+ "step": 80,
179
+ "token_acc": 0.9627279936558287,
180
+ "train_speed(iter/s)": 0.074464
181
+ },
182
+ {
183
+ "epoch": 0.15473300970873785,
184
+ "grad_norm": 0.3179706335067749,
185
+ "learning_rate": 4.970755844796817e-06,
186
+ "loss": 0.12840776443481444,
187
+ "memory(GiB)": 40.39,
188
+ "step": 85,
189
+ "token_acc": 0.9492063492063492,
190
+ "train_speed(iter/s)": 0.074996
191
+ },
192
+ {
193
+ "epoch": 0.1638349514563107,
194
+ "grad_norm": 0.3189823031425476,
195
+ "learning_rate": 4.964875428023093e-06,
196
+ "loss": 0.12376663684844971,
197
+ "memory(GiB)": 40.39,
198
+ "step": 90,
199
+ "token_acc": 0.957936507936508,
200
+ "train_speed(iter/s)": 0.075144
201
+ },
202
+ {
203
+ "epoch": 0.1729368932038835,
204
+ "grad_norm": 0.33377909660339355,
205
+ "learning_rate": 4.958460818428627e-06,
206
+ "loss": 0.11574116945266724,
207
+ "memory(GiB)": 40.39,
208
+ "step": 95,
209
+ "token_acc": 0.9563492063492064,
210
+ "train_speed(iter/s)": 0.075617
211
+ },
212
+ {
213
+ "epoch": 0.1820388349514563,
214
+ "grad_norm": 0.4928111433982849,
215
+ "learning_rate": 4.951513406200667e-06,
216
+ "loss": 0.1149444341659546,
217
+ "memory(GiB)": 40.39,
218
+ "step": 100,
219
+ "token_acc": 0.9508716323296355,
220
+ "train_speed(iter/s)": 0.075828
221
+ },
222
+ {
223
+ "epoch": 0.19114077669902912,
224
+ "grad_norm": 0.3134707808494568,
225
+ "learning_rate": 4.944034696996534e-06,
226
+ "loss": 0.11119295358657837,
227
+ "memory(GiB)": 40.39,
228
+ "step": 105,
229
+ "token_acc": 0.9595238095238096,
230
+ "train_speed(iter/s)": 0.075066
231
+ },
232
+ {
233
+ "epoch": 0.20024271844660194,
234
+ "grad_norm": 0.2365858554840088,
235
+ "learning_rate": 4.936026311617316e-06,
236
+ "loss": 0.11442217826843262,
237
+ "memory(GiB)": 40.39,
238
+ "step": 110,
239
+ "token_acc": 0.9588281868566905,
240
+ "train_speed(iter/s)": 0.075061
241
+ },
242
+ {
243
+ "epoch": 0.20934466019417475,
244
+ "grad_norm": 0.3145173490047455,
245
+ "learning_rate": 4.927489985656591e-06,
246
+ "loss": 0.10322239398956298,
247
+ "memory(GiB)": 40.39,
248
+ "step": 115,
249
+ "token_acc": 0.9658730158730159,
250
+ "train_speed(iter/s)": 0.074479
251
+ },
252
+ {
253
+ "epoch": 0.21844660194174756,
254
+ "grad_norm": 0.33202633261680603,
255
+ "learning_rate": 4.918427569124302e-06,
256
+ "loss": 0.10661822557449341,
257
+ "memory(GiB)": 40.39,
258
+ "step": 120,
259
+ "token_acc": 0.9556259904912837,
260
+ "train_speed(iter/s)": 0.074637
261
+ },
262
+ {
263
+ "epoch": 0.2275485436893204,
264
+ "grad_norm": 0.3093946874141693,
265
+ "learning_rate": 4.908841026045809e-06,
266
+ "loss": 0.10065805912017822,
267
+ "memory(GiB)": 40.39,
268
+ "step": 125,
269
+ "token_acc": 0.9540412044374009,
270
+ "train_speed(iter/s)": 0.074905
271
+ },
272
+ {
273
+ "epoch": 0.2366504854368932,
274
+ "grad_norm": 0.39363232254981995,
275
+ "learning_rate": 4.8987324340362445e-06,
276
+ "loss": 0.114447021484375,
277
+ "memory(GiB)": 40.39,
278
+ "step": 130,
279
+ "token_acc": 0.9571428571428572,
280
+ "train_speed(iter/s)": 0.075072
281
+ },
282
+ {
283
+ "epoch": 0.24575242718446602,
284
+ "grad_norm": 0.37065446376800537,
285
+ "learning_rate": 4.888103983850245e-06,
286
+ "loss": 0.10610785484313964,
287
+ "memory(GiB)": 40.39,
288
+ "step": 135,
289
+ "token_acc": 0.9565217391304348,
290
+ "train_speed(iter/s)": 0.075167
291
+ },
292
+ {
293
+ "epoch": 0.25485436893203883,
294
+ "grad_norm": 0.542117714881897,
295
+ "learning_rate": 4.876957978907176e-06,
296
+ "loss": 0.0954114019870758,
297
+ "memory(GiB)": 40.39,
298
+ "step": 140,
299
+ "token_acc": 0.9666666666666667,
300
+ "train_speed(iter/s)": 0.075346
301
+ },
302
+ {
303
+ "epoch": 0.26395631067961167,
304
+ "grad_norm": 0.3225058913230896,
305
+ "learning_rate": 4.865296834791918e-06,
306
+ "loss": 0.0959049105644226,
307
+ "memory(GiB)": 40.39,
308
+ "step": 145,
309
+ "token_acc": 0.9587955625990491,
310
+ "train_speed(iter/s)": 0.075467
311
+ },
312
+ {
313
+ "epoch": 0.27305825242718446,
314
+ "grad_norm": 0.3421016037464142,
315
+ "learning_rate": 4.853123078731363e-06,
316
+ "loss": 0.09874246120452881,
317
+ "memory(GiB)": 40.39,
318
+ "step": 150,
319
+ "token_acc": 0.9650793650793651,
320
+ "train_speed(iter/s)": 0.075618
321
+ },
322
+ {
323
+ "epoch": 0.2821601941747573,
324
+ "grad_norm": 0.3102968633174896,
325
+ "learning_rate": 4.8404393490467085e-06,
326
+ "loss": 0.09461469650268554,
327
+ "memory(GiB)": 40.39,
328
+ "step": 155,
329
+ "token_acc": 0.9547977795400476,
330
+ "train_speed(iter/s)": 0.075855
331
+ },
332
+ {
333
+ "epoch": 0.2912621359223301,
334
+ "grad_norm": 0.4729763865470886,
335
+ "learning_rate": 4.827248394581672e-06,
336
+ "loss": 0.10038878917694091,
337
+ "memory(GiB)": 40.39,
338
+ "step": 160,
339
+ "token_acc": 0.9650793650793651,
340
+ "train_speed(iter/s)": 0.075945
341
+ },
342
+ {
343
+ "epoch": 0.3003640776699029,
344
+ "grad_norm": 0.3695836365222931,
345
+ "learning_rate": 4.813553074106761e-06,
346
+ "loss": 0.09139147400856018,
347
+ "memory(GiB)": 40.39,
348
+ "step": 165,
349
+ "token_acc": 0.9627279936558287,
350
+ "train_speed(iter/s)": 0.075756
351
+ },
352
+ {
353
+ "epoch": 0.3094660194174757,
354
+ "grad_norm": 0.47110962867736816,
355
+ "learning_rate": 4.799356355699708e-06,
356
+ "loss": 0.09496045112609863,
357
+ "memory(GiB)": 40.39,
358
+ "step": 170,
359
+ "token_acc": 0.9698412698412698,
360
+ "train_speed(iter/s)": 0.075898
361
+ },
362
+ {
363
+ "epoch": 0.31856796116504854,
364
+ "grad_norm": 0.3773088753223419,
365
+ "learning_rate": 4.784661316102229e-06,
366
+ "loss": 0.09658662080764771,
367
+ "memory(GiB)": 40.4,
368
+ "step": 175,
369
+ "token_acc": 0.96513470681458,
370
+ "train_speed(iter/s)": 0.075914
371
+ },
372
+ {
373
+ "epoch": 0.3276699029126214,
374
+ "grad_norm": 0.3394829034805298,
375
+ "learning_rate": 4.769471140053221e-06,
376
+ "loss": 0.08639374971389771,
377
+ "memory(GiB)": 40.4,
378
+ "step": 180,
379
+ "token_acc": 0.969047619047619,
380
+ "train_speed(iter/s)": 0.076076
381
+ },
382
+ {
383
+ "epoch": 0.33677184466019416,
384
+ "grad_norm": 0.4525506794452667,
385
+ "learning_rate": 4.753789119598563e-06,
386
+ "loss": 0.09742268323898315,
387
+ "memory(GiB)": 40.4,
388
+ "step": 185,
389
+ "token_acc": 0.9587301587301588,
390
+ "train_speed(iter/s)": 0.076177
391
+ },
392
+ {
393
+ "epoch": 0.345873786407767,
394
+ "grad_norm": 0.3789404332637787,
395
+ "learning_rate": 4.737618653377651e-06,
396
+ "loss": 0.09391134977340698,
397
+ "memory(GiB)": 40.4,
398
+ "step": 190,
399
+ "token_acc": 0.9651070578905631,
400
+ "train_speed(iter/s)": 0.07649
401
+ },
402
+ {
403
+ "epoch": 0.3549757281553398,
404
+ "grad_norm": 0.5464370250701904,
405
+ "learning_rate": 4.720963245886846e-06,
406
+ "loss": 0.0969527006149292,
407
+ "memory(GiB)": 40.4,
408
+ "step": 195,
409
+ "token_acc": 0.9659270998415214,
410
+ "train_speed(iter/s)": 0.076513
411
+ },
412
+ {
413
+ "epoch": 0.3640776699029126,
414
+ "grad_norm": 0.3459813892841339,
415
+ "learning_rate": 4.703826506719964e-06,
416
+ "loss": 0.08732333183288574,
417
+ "memory(GiB)": 40.4,
418
+ "step": 200,
419
+ "token_acc": 0.96513470681458,
420
+ "train_speed(iter/s)": 0.076587
421
+ },
422
+ {
423
+ "epoch": 0.3731796116504854,
424
+ "grad_norm": 0.3549191653728485,
425
+ "learning_rate": 4.686212149786007e-06,
426
+ "loss": 0.08515737056732178,
427
+ "memory(GiB)": 40.4,
428
+ "step": 205,
429
+ "token_acc": 0.96513470681458,
430
+ "train_speed(iter/s)": 0.076344
431
+ },
432
+ {
433
+ "epoch": 0.38228155339805825,
434
+ "grad_norm": 0.7434160709381104,
435
+ "learning_rate": 4.668123992504267e-06,
436
+ "loss": 0.09526927471160888,
437
+ "memory(GiB)": 40.4,
438
+ "step": 210,
439
+ "token_acc": 0.9666666666666667,
440
+ "train_speed(iter/s)": 0.076513
441
+ },
442
+ {
443
+ "epoch": 0.3913834951456311,
444
+ "grad_norm": 0.464631587266922,
445
+ "learning_rate": 4.649565954977015e-06,
446
+ "loss": 0.09264343380928039,
447
+ "memory(GiB)": 40.4,
448
+ "step": 215,
449
+ "token_acc": 0.9620253164556962,
450
+ "train_speed(iter/s)": 0.076143
451
+ },
452
+ {
453
+ "epoch": 0.40048543689320387,
454
+ "grad_norm": 0.5145648121833801,
455
+ "learning_rate": 4.630542059139923e-06,
456
+ "loss": 0.09688866138458252,
457
+ "memory(GiB)": 40.4,
458
+ "step": 220,
459
+ "token_acc": 0.9667458432304038,
460
+ "train_speed(iter/s)": 0.076292
461
+ },
462
+ {
463
+ "epoch": 0.4095873786407767,
464
+ "grad_norm": 0.33657485246658325,
465
+ "learning_rate": 4.611056427890428e-06,
466
+ "loss": 0.09414277076721192,
467
+ "memory(GiB)": 40.4,
468
+ "step": 225,
469
+ "token_acc": 0.9587301587301588,
470
+ "train_speed(iter/s)": 0.076275
471
+ },
472
+ {
473
+ "epoch": 0.4186893203883495,
474
+ "grad_norm": 0.47585147619247437,
475
+ "learning_rate": 4.5911132841942e-06,
476
+ "loss": 0.08656486272811889,
477
+ "memory(GiB)": 40.4,
478
+ "step": 230,
479
+ "token_acc": 0.9698651863600317,
480
+ "train_speed(iter/s)": 0.076342
481
+ },
482
+ {
483
+ "epoch": 0.42779126213592233,
484
+ "grad_norm": 0.3516729176044464,
485
+ "learning_rate": 4.570716950169944e-06,
486
+ "loss": 0.08657894730567932,
487
+ "memory(GiB)": 40.4,
488
+ "step": 235,
489
+ "token_acc": 0.9642857142857143,
490
+ "train_speed(iter/s)": 0.076493
491
+ },
492
+ {
493
+ "epoch": 0.4368932038834951,
494
+ "grad_norm": 0.48757559061050415,
495
+ "learning_rate": 4.5498718461526895e-06,
496
+ "loss": 0.09453780055046082,
497
+ "memory(GiB)": 40.4,
498
+ "step": 240,
499
+ "token_acc": 0.9643705463182898,
500
+ "train_speed(iter/s)": 0.07656
501
+ },
502
+ {
503
+ "epoch": 0.44599514563106796,
504
+ "grad_norm": 0.5283713936805725,
505
+ "learning_rate": 4.528582489735818e-06,
506
+ "loss": 0.08740494847297668,
507
+ "memory(GiB)": 40.4,
508
+ "step": 245,
509
+ "token_acc": 0.9587628865979382,
510
+ "train_speed(iter/s)": 0.07663
511
+ },
512
+ {
513
+ "epoch": 0.4550970873786408,
514
+ "grad_norm": 0.3577844500541687,
515
+ "learning_rate": 4.506853494791992e-06,
516
+ "loss": 0.08014656901359558,
517
+ "memory(GiB)": 40.4,
518
+ "step": 250,
519
+ "token_acc": 0.971473851030111,
520
+ "train_speed(iter/s)": 0.076543
521
+ },
522
+ {
523
+ "epoch": 0.4641990291262136,
524
+ "grad_norm": 0.5026013851165771,
525
+ "learning_rate": 4.484689570473232e-06,
526
+ "loss": 0.08635783195495605,
527
+ "memory(GiB)": 40.4,
528
+ "step": 255,
529
+ "token_acc": 0.9682791435368755,
530
+ "train_speed(iter/s)": 0.076578
531
+ },
532
+ {
533
+ "epoch": 0.4733009708737864,
534
+ "grad_norm": 0.45232078433036804,
535
+ "learning_rate": 4.462095520190336e-06,
536
+ "loss": 0.08593440055847168,
537
+ "memory(GiB)": 40.4,
538
+ "step": 260,
539
+ "token_acc": 0.9699367088607594,
540
+ "train_speed(iter/s)": 0.076538
541
+ },
542
+ {
543
+ "epoch": 0.4824029126213592,
544
+ "grad_norm": 0.47390663623809814,
545
+ "learning_rate": 4.43907624057188e-06,
546
+ "loss": 0.08747667074203491,
547
+ "memory(GiB)": 40.4,
548
+ "step": 265,
549
+ "token_acc": 0.9619047619047619,
550
+ "train_speed(iter/s)": 0.076588
551
+ },
552
+ {
553
+ "epoch": 0.49150485436893204,
554
+ "grad_norm": 0.43587085604667664,
555
+ "learning_rate": 4.415636720403005e-06,
556
+ "loss": 0.08902972340583801,
557
+ "memory(GiB)": 40.4,
558
+ "step": 270,
559
+ "token_acc": 0.9619349722442506,
560
+ "train_speed(iter/s)": 0.076484
561
+ },
562
+ {
563
+ "epoch": 0.5006067961165048,
564
+ "grad_norm": 0.41671204566955566,
565
+ "learning_rate": 4.391782039544239e-06,
566
+ "loss": 0.08426393270492553,
567
+ "memory(GiB)": 40.4,
568
+ "step": 275,
569
+ "token_acc": 0.9603489294210944,
570
+ "train_speed(iter/s)": 0.076586
571
+ },
572
+ {
573
+ "epoch": 0.5097087378640777,
574
+ "grad_norm": 0.3852890133857727,
575
+ "learning_rate": 4.367517367830581e-06,
576
+ "loss": 0.08224607706069946,
577
+ "memory(GiB)": 40.4,
578
+ "step": 280,
579
+ "token_acc": 0.9730372720063442,
580
+ "train_speed(iter/s)": 0.0767
581
+ },
582
+ {
583
+ "epoch": 0.5188106796116505,
584
+ "grad_norm": 0.5980095863342285,
585
+ "learning_rate": 4.342847963951085e-06,
586
+ "loss": 0.09114923477172851,
587
+ "memory(GiB)": 40.4,
588
+ "step": 285,
589
+ "token_acc": 0.9642857142857143,
590
+ "train_speed(iter/s)": 0.076804
591
+ },
592
+ {
593
+ "epoch": 0.5279126213592233,
594
+ "grad_norm": 0.5370866656303406,
595
+ "learning_rate": 4.317779174309179e-06,
596
+ "loss": 0.09176770448684693,
597
+ "memory(GiB)": 40.4,
598
+ "step": 290,
599
+ "token_acc": 0.9595879556259905,
600
+ "train_speed(iter/s)": 0.076902
601
+ },
602
+ {
603
+ "epoch": 0.5370145631067961,
604
+ "grad_norm": 0.5857056975364685,
605
+ "learning_rate": 4.292316431863991e-06,
606
+ "loss": 0.08232347965240479,
607
+ "memory(GiB)": 40.4,
608
+ "step": 295,
609
+ "token_acc": 0.9635210150674068,
610
+ "train_speed(iter/s)": 0.076861
611
+ },
612
+ {
613
+ "epoch": 0.5461165048543689,
614
+ "grad_norm": 0.45398032665252686,
615
+ "learning_rate": 4.2664652549528995e-06,
616
+ "loss": 0.0860186755657196,
617
+ "memory(GiB)": 40.4,
618
+ "step": 300,
619
+ "token_acc": 0.9603174603174603,
620
+ "train_speed(iter/s)": 0.076918
621
+ },
622
+ {
623
+ "epoch": 0.5552184466019418,
624
+ "grad_norm": 0.4008013904094696,
625
+ "learning_rate": 4.240231246095593e-06,
626
+ "loss": 0.08663930892944335,
627
+ "memory(GiB)": 40.4,
628
+ "step": 305,
629
+ "token_acc": 0.9698651863600317,
630
+ "train_speed(iter/s)": 0.076723
631
+ },
632
+ {
633
+ "epoch": 0.5643203883495146,
634
+ "grad_norm": 0.6199547052383423,
635
+ "learning_rate": 4.213620090779877e-06,
636
+ "loss": 0.08223216533660889,
637
+ "memory(GiB)": 40.4,
638
+ "step": 310,
639
+ "token_acc": 0.9674861221252974,
640
+ "train_speed(iter/s)": 0.076805
641
+ },
642
+ {
643
+ "epoch": 0.5734223300970874,
644
+ "grad_norm": 0.37448298931121826,
645
+ "learning_rate": 4.186637556229508e-06,
646
+ "loss": 0.08296606540679932,
647
+ "memory(GiB)": 40.4,
648
+ "step": 315,
649
+ "token_acc": 0.9666931007137193,
650
+ "train_speed(iter/s)": 0.076708
651
+ },
652
+ {
653
+ "epoch": 0.5825242718446602,
654
+ "grad_norm": 0.4003507196903229,
655
+ "learning_rate": 4.159289490154305e-06,
656
+ "loss": 0.07931501269340516,
657
+ "memory(GiB)": 40.4,
658
+ "step": 320,
659
+ "token_acc": 0.9642857142857143,
660
+ "train_speed(iter/s)": 0.076845
661
+ },
662
+ {
663
+ "epoch": 0.591626213592233,
664
+ "grad_norm": 0.49439844489097595,
665
+ "learning_rate": 4.1315818194828196e-06,
666
+ "loss": 0.08067693710327148,
667
+ "memory(GiB)": 40.4,
668
+ "step": 325,
669
+ "token_acc": 0.9698412698412698,
670
+ "train_speed(iter/s)": 0.076875
671
+ },
672
+ {
673
+ "epoch": 0.6007281553398058,
674
+ "grad_norm": 0.584017813205719,
675
+ "learning_rate": 4.1035205490778505e-06,
676
+ "loss": 0.09277031421661378,
677
+ "memory(GiB)": 40.4,
678
+ "step": 330,
679
+ "token_acc": 0.9595879556259905,
680
+ "train_speed(iter/s)": 0.076692
681
+ },
682
+ {
683
+ "epoch": 0.6098300970873787,
684
+ "grad_norm": 0.47020280361175537,
685
+ "learning_rate": 4.075111760435045e-06,
686
+ "loss": 0.07749168276786804,
687
+ "memory(GiB)": 40.4,
688
+ "step": 335,
689
+ "token_acc": 0.96513470681458,
690
+ "train_speed(iter/s)": 0.076884
691
+ },
692
+ {
693
+ "epoch": 0.6189320388349514,
694
+ "grad_norm": 0.4876089096069336,
695
+ "learning_rate": 4.046361610364913e-06,
696
+ "loss": 0.07796428203582764,
697
+ "memory(GiB)": 40.4,
698
+ "step": 340,
699
+ "token_acc": 0.9691699604743083,
700
+ "train_speed(iter/s)": 0.076913
701
+ },
702
+ {
703
+ "epoch": 0.6280339805825242,
704
+ "grad_norm": 0.5511714220046997,
705
+ "learning_rate": 4.017276329658506e-06,
706
+ "loss": 0.08419817090034484,
707
+ "memory(GiB)": 40.4,
708
+ "step": 345,
709
+ "token_acc": 0.9707278481012658,
710
+ "train_speed(iter/s)": 0.07696
711
+ },
712
+ {
713
+ "epoch": 0.6371359223300971,
714
+ "grad_norm": 0.5659735798835754,
715
+ "learning_rate": 3.987862221737072e-06,
716
+ "loss": 0.0797402322292328,
717
+ "memory(GiB)": 40.4,
718
+ "step": 350,
719
+ "token_acc": 0.9659270998415214,
720
+ "train_speed(iter/s)": 0.076995
721
+ },
722
+ {
723
+ "epoch": 0.6462378640776699,
724
+ "grad_norm": 0.5157150030136108,
725
+ "learning_rate": 3.958125661285959e-06,
726
+ "loss": 0.0838176965713501,
727
+ "memory(GiB)": 40.4,
728
+ "step": 355,
729
+ "token_acc": 0.9690721649484536,
730
+ "train_speed(iter/s)": 0.076909
731
+ },
732
+ {
733
+ "epoch": 0.6553398058252428,
734
+ "grad_norm": 0.5069080591201782,
735
+ "learning_rate": 3.928073092873088e-06,
736
+ "loss": 0.07343612313270569,
737
+ "memory(GiB)": 40.4,
738
+ "step": 360,
739
+ "token_acc": 0.9746233148295004,
740
+ "train_speed(iter/s)": 0.076991
741
+ },
742
+ {
743
+ "epoch": 0.6644417475728155,
744
+ "grad_norm": 0.49923259019851685,
745
+ "learning_rate": 3.897711029552264e-06,
746
+ "loss": 0.07626074552536011,
747
+ "memory(GiB)": 40.4,
748
+ "step": 365,
749
+ "token_acc": 0.9683544303797469,
750
+ "train_speed(iter/s)": 0.076983
751
+ },
752
+ {
753
+ "epoch": 0.6735436893203883,
754
+ "grad_norm": 0.35883885622024536,
755
+ "learning_rate": 3.8670460514516615e-06,
756
+ "loss": 0.08405499458312989,
757
+ "memory(GiB)": 40.4,
758
+ "step": 370,
759
+ "token_acc": 0.9635499207606973,
760
+ "train_speed(iter/s)": 0.077013
761
+ },
762
+ {
763
+ "epoch": 0.6826456310679612,
764
+ "grad_norm": 0.4520786702632904,
765
+ "learning_rate": 3.836084804347763e-06,
766
+ "loss": 0.07998884916305542,
767
+ "memory(GiB)": 40.4,
768
+ "step": 375,
769
+ "token_acc": 0.9698412698412698,
770
+ "train_speed(iter/s)": 0.07694
771
+ },
772
+ {
773
+ "epoch": 0.691747572815534,
774
+ "grad_norm": 0.47654658555984497,
775
+ "learning_rate": 3.8048339982250705e-06,
776
+ "loss": 0.08119775056838989,
777
+ "memory(GiB)": 40.4,
778
+ "step": 380,
779
+ "token_acc": 0.9667194928684627,
780
+ "train_speed(iter/s)": 0.077002
781
+ },
782
+ {
783
+ "epoch": 0.7008495145631068,
784
+ "grad_norm": 0.5640057325363159,
785
+ "learning_rate": 3.773300405821908e-06,
786
+ "loss": 0.08841820359230042,
787
+ "memory(GiB)": 40.4,
788
+ "step": 385,
789
+ "token_acc": 0.9595559080095163,
790
+ "train_speed(iter/s)": 0.077061
791
+ },
792
+ {
793
+ "epoch": 0.7099514563106796,
794
+ "grad_norm": 0.42381900548934937,
795
+ "learning_rate": 3.7414908611626162e-06,
796
+ "loss": 0.08166542053222656,
797
+ "memory(GiB)": 40.4,
798
+ "step": 390,
799
+ "token_acc": 0.969047619047619,
800
+ "train_speed(iter/s)": 0.077092
801
+ },
802
+ {
803
+ "epoch": 0.7190533980582524,
804
+ "grad_norm": 0.510867714881897,
805
+ "learning_rate": 3.709412258076471e-06,
806
+ "loss": 0.08081957101821899,
807
+ "memory(GiB)": 40.4,
808
+ "step": 395,
809
+ "token_acc": 0.9699129057798892,
810
+ "train_speed(iter/s)": 0.077233
811
+ },
812
+ {
813
+ "epoch": 0.7281553398058253,
814
+ "grad_norm": 0.5211343169212341,
815
+ "learning_rate": 3.6770715487036413e-06,
816
+ "loss": 0.08312466740608215,
817
+ "memory(GiB)": 40.4,
818
+ "step": 400,
819
+ "token_acc": 0.9611419508326725,
820
+ "train_speed(iter/s)": 0.077264
821
+ },
822
+ {
823
+ "epoch": 0.7372572815533981,
824
+ "grad_norm": 0.46672672033309937,
825
+ "learning_rate": 3.644475741988499e-06,
826
+ "loss": 0.08163590431213379,
827
+ "memory(GiB)": 40.4,
828
+ "step": 405,
829
+ "token_acc": 0.9666666666666667,
830
+ "train_speed(iter/s)": 0.07706
831
+ },
832
+ {
833
+ "epoch": 0.7463592233009708,
834
+ "grad_norm": 0.4190872013568878,
835
+ "learning_rate": 3.6116319021606345e-06,
836
+ "loss": 0.08278034925460816,
837
+ "memory(GiB)": 40.4,
838
+ "step": 410,
839
+ "token_acc": 0.9603803486529319,
840
+ "train_speed(iter/s)": 0.077071
841
+ },
842
+ {
843
+ "epoch": 0.7554611650485437,
844
+ "grad_norm": 0.4177815318107605,
845
+ "learning_rate": 3.5785471472038784e-06,
846
+ "loss": 0.07709290385246277,
847
+ "memory(GiB)": 40.4,
848
+ "step": 415,
849
+ "token_acc": 0.9714512291831879,
850
+ "train_speed(iter/s)": 0.077076
851
+ },
852
+ {
853
+ "epoch": 0.7645631067961165,
854
+ "grad_norm": 0.7115554213523865,
855
+ "learning_rate": 3.545228647313679e-06,
856
+ "loss": 0.08126543164253235,
857
+ "memory(GiB)": 40.4,
858
+ "step": 420,
859
+ "token_acc": 0.9674861221252974,
860
+ "train_speed(iter/s)": 0.07706
861
+ },
862
+ {
863
+ "epoch": 0.7736650485436893,
864
+ "grad_norm": 0.43985486030578613,
865
+ "learning_rate": 3.5116836233431616e-06,
866
+ "loss": 0.08477982282638549,
867
+ "memory(GiB)": 40.4,
868
+ "step": 425,
869
+ "token_acc": 0.9628164556962026,
870
+ "train_speed(iter/s)": 0.077154
871
+ },
872
+ {
873
+ "epoch": 0.7827669902912622,
874
+ "grad_norm": 0.48275941610336304,
875
+ "learning_rate": 3.477919345238213e-06,
876
+ "loss": 0.07978797554969788,
877
+ "memory(GiB)": 40.4,
878
+ "step": 430,
879
+ "token_acc": 0.9627279936558287,
880
+ "train_speed(iter/s)": 0.077173
881
+ },
882
+ {
883
+ "epoch": 0.7918689320388349,
884
+ "grad_norm": 0.5005500912666321,
885
+ "learning_rate": 3.4439431304619207e-06,
886
+ "loss": 0.07624109983444213,
887
+ "memory(GiB)": 40.4,
888
+ "step": 435,
889
+ "token_acc": 0.9659270998415214,
890
+ "train_speed(iter/s)": 0.077238
891
+ },
892
+ {
893
+ "epoch": 0.8009708737864077,
894
+ "grad_norm": 0.5146210789680481,
895
+ "learning_rate": 3.4097623424087196e-06,
896
+ "loss": 0.080259507894516,
897
+ "memory(GiB)": 40.4,
898
+ "step": 440,
899
+ "token_acc": 0.9706582077716098,
900
+ "train_speed(iter/s)": 0.077241
901
+ },
902
+ {
903
+ "epoch": 0.8100728155339806,
904
+ "grad_norm": 0.558778703212738,
905
+ "learning_rate": 3.3753843888085806e-06,
906
+ "loss": 0.07813260555267335,
907
+ "memory(GiB)": 40.4,
908
+ "step": 445,
909
+ "token_acc": 0.9658730158730159,
910
+ "train_speed(iter/s)": 0.077226
911
+ },
912
+ {
913
+ "epoch": 0.8191747572815534,
914
+ "grad_norm": 0.574676513671875,
915
+ "learning_rate": 3.340816720121597e-06,
916
+ "loss": 0.0761204183101654,
917
+ "memory(GiB)": 40.4,
918
+ "step": 450,
919
+ "token_acc": 0.9691699604743083,
920
+ "train_speed(iter/s)": 0.077059
921
+ },
922
+ {
923
+ "epoch": 0.8282766990291263,
924
+ "grad_norm": 0.5359216332435608,
925
+ "learning_rate": 3.3060668279232964e-06,
926
+ "loss": 0.07063559293746949,
927
+ "memory(GiB)": 40.4,
928
+ "step": 455,
929
+ "token_acc": 0.9746233148295004,
930
+ "train_speed(iter/s)": 0.077103
931
+ },
932
+ {
933
+ "epoch": 0.837378640776699,
934
+ "grad_norm": 0.5926820635795593,
935
+ "learning_rate": 3.2711422432810624e-06,
936
+ "loss": 0.07327613830566407,
937
+ "memory(GiB)": 40.4,
938
+ "step": 460,
939
+ "token_acc": 0.9666666666666667,
940
+ "train_speed(iter/s)": 0.077136
941
+ },
942
+ {
943
+ "epoch": 0.8464805825242718,
944
+ "grad_norm": 0.4923359155654907,
945
+ "learning_rate": 3.236050535121976e-06,
946
+ "loss": 0.0849435031414032,
947
+ "memory(GiB)": 40.4,
948
+ "step": 465,
949
+ "token_acc": 0.9628164556962026,
950
+ "train_speed(iter/s)": 0.077175
951
+ },
952
+ {
953
+ "epoch": 0.8555825242718447,
954
+ "grad_norm": 0.5079782605171204,
955
+ "learning_rate": 3.2007993085924694e-06,
956
+ "loss": 0.07131590843200683,
957
+ "memory(GiB)": 40.4,
958
+ "step": 470,
959
+ "token_acc": 0.9603489294210944,
960
+ "train_speed(iter/s)": 0.077219
961
+ },
962
+ {
963
+ "epoch": 0.8646844660194175,
964
+ "grad_norm": 0.47359853982925415,
965
+ "learning_rate": 3.165396203410121e-06,
966
+ "loss": 0.08230514526367187,
967
+ "memory(GiB)": 40.4,
968
+ "step": 475,
969
+ "token_acc": 0.9603489294210944,
970
+ "train_speed(iter/s)": 0.077276
971
+ },
972
+ {
973
+ "epoch": 0.8737864077669902,
974
+ "grad_norm": 0.5094448328018188,
975
+ "learning_rate": 3.1298488922079597e-06,
976
+ "loss": 0.07572669386863709,
977
+ "memory(GiB)": 40.4,
978
+ "step": 480,
979
+ "token_acc": 0.9683042789223455,
980
+ "train_speed(iter/s)": 0.077301
981
+ },
982
+ {
983
+ "epoch": 0.8828883495145631,
984
+ "grad_norm": 0.6144260764122009,
985
+ "learning_rate": 3.094165078871634e-06,
986
+ "loss": 0.07770437002182007,
987
+ "memory(GiB)": 40.4,
988
+ "step": 485,
989
+ "token_acc": 0.9674603174603175,
990
+ "train_speed(iter/s)": 0.077291
991
+ },
992
+ {
993
+ "epoch": 0.8919902912621359,
994
+ "grad_norm": 0.7166838049888611,
995
+ "learning_rate": 3.0583524968698176e-06,
996
+ "loss": 0.07593016624450684,
997
+ "memory(GiB)": 40.4,
998
+ "step": 490,
999
+ "token_acc": 0.9706582077716098,
1000
+ "train_speed(iter/s)": 0.077337
1001
+ },
1002
+ {
1003
+ "epoch": 0.9010922330097088,
1004
+ "grad_norm": 0.5843172073364258,
1005
+ "learning_rate": 3.0224189075781886e-06,
1006
+ "loss": 0.0753251850605011,
1007
+ "memory(GiB)": 40.4,
1008
+ "step": 495,
1009
+ "token_acc": 0.9675889328063241,
1010
+ "train_speed(iter/s)": 0.077398
1011
+ },
1012
+ {
1013
+ "epoch": 0.9101941747572816,
1014
+ "grad_norm": 0.4273771643638611,
1015
+ "learning_rate": 2.9863720985973697e-06,
1016
+ "loss": 0.07616569995880126,
1017
+ "memory(GiB)": 40.4,
1018
+ "step": 500,
1019
+ "token_acc": 0.9746031746031746,
1020
+ "train_speed(iter/s)": 0.077368
1021
+ },
1022
+ {
1023
+ "epoch": 0.9192961165048543,
1024
+ "grad_norm": 0.5440679788589478,
1025
+ "learning_rate": 2.9502198820651903e-06,
1026
+ "loss": 0.07991842031478882,
1027
+ "memory(GiB)": 40.4,
1028
+ "step": 505,
1029
+ "token_acc": 0.9642857142857143,
1030
+ "train_speed(iter/s)": 0.077195
1031
+ },
1032
+ {
1033
+ "epoch": 0.9283980582524272,
1034
+ "grad_norm": 0.6545736789703369,
1035
+ "learning_rate": 2.9139700929636134e-06,
1036
+ "loss": 0.07855194211006164,
1037
+ "memory(GiB)": 40.4,
1038
+ "step": 510,
1039
+ "token_acc": 0.9587301587301588,
1040
+ "train_speed(iter/s)": 0.077178
1041
+ },
1042
+ {
1043
+ "epoch": 0.9375,
1044
+ "grad_norm": 0.5470529794692993,
1045
+ "learning_rate": 2.8776305874207305e-06,
1046
+ "loss": 0.07507063150405884,
1047
+ "memory(GiB)": 40.4,
1048
+ "step": 515,
1049
+ "token_acc": 0.9675376088677752,
1050
+ "train_speed(iter/s)": 0.077176
1051
+ },
1052
+ {
1053
+ "epoch": 0.9466019417475728,
1054
+ "grad_norm": 0.5262081623077393,
1055
+ "learning_rate": 2.8412092410081645e-06,
1056
+ "loss": 0.08568469285964966,
1057
+ "memory(GiB)": 40.4,
1058
+ "step": 520,
1059
+ "token_acc": 0.9659270998415214,
1060
+ "train_speed(iter/s)": 0.077164
1061
+ },
1062
+ {
1063
+ "epoch": 0.9557038834951457,
1064
+ "grad_norm": 0.48101773858070374,
1065
+ "learning_rate": 2.804713947034254e-06,
1066
+ "loss": 0.07408897280693054,
1067
+ "memory(GiB)": 40.4,
1068
+ "step": 525,
1069
+ "token_acc": 0.9715189873417721,
1070
+ "train_speed(iter/s)": 0.077248
1071
+ },
1072
+ {
1073
+ "epoch": 0.9648058252427184,
1074
+ "grad_norm": 0.7088754773139954,
1075
+ "learning_rate": 2.7681526148334074e-06,
1076
+ "loss": 0.07859846353530883,
1077
+ "memory(GiB)": 40.4,
1078
+ "step": 530,
1079
+ "token_acc": 0.9651070578905631,
1080
+ "train_speed(iter/s)": 0.077348
1081
+ },
1082
+ {
1083
+ "epoch": 0.9739077669902912,
1084
+ "grad_norm": 0.5357980728149414,
1085
+ "learning_rate": 2.73153316805197e-06,
1086
+ "loss": 0.07618768811225891,
1087
+ "memory(GiB)": 40.4,
1088
+ "step": 535,
1089
+ "token_acc": 0.9683042789223455,
1090
+ "train_speed(iter/s)": 0.077388
1091
+ },
1092
+ {
1093
+ "epoch": 0.9830097087378641,
1094
+ "grad_norm": 0.4719216823577881,
1095
+ "learning_rate": 2.6948635429309984e-06,
1096
+ "loss": 0.08283294439315796,
1097
+ "memory(GiB)": 40.4,
1098
+ "step": 540,
1099
+ "token_acc": 0.9666666666666667,
1100
+ "train_speed(iter/s)": 0.077404
1101
+ },
1102
+ {
1103
+ "epoch": 0.9921116504854369,
1104
+ "grad_norm": 0.4105032980442047,
1105
+ "learning_rate": 2.6581516865863006e-06,
1106
+ "loss": 0.07635112404823304,
1107
+ "memory(GiB)": 40.4,
1108
+ "step": 545,
1109
+ "token_acc": 0.9666666666666667,
1110
+ "train_speed(iter/s)": 0.077461
1111
+ },
1112
+ {
1113
+ "epoch": 1.0,
1114
+ "grad_norm": 0.4639950096607208,
1115
+ "learning_rate": 2.6214055552861213e-06,
1116
+ "loss": 0.07352917194366455,
1117
+ "memory(GiB)": 40.4,
1118
+ "step": 550,
1119
+ "token_acc": 0.9652014652014652,
1120
+ "train_speed(iter/s)": 0.077567
1121
+ },
1122
+ {
1123
+ "epoch": 1.0091019417475728,
1124
+ "grad_norm": 0.5708960294723511,
1125
+ "learning_rate": 2.5846331127268432e-06,
1126
+ "loss": 0.06939817667007446,
1127
+ "memory(GiB)": 40.4,
1128
+ "step": 555,
1129
+ "token_acc": 0.9746634996041171,
1130
+ "train_speed(iter/s)": 0.077516
1131
+ },
1132
+ {
1133
+ "epoch": 1.0182038834951457,
1134
+ "grad_norm": 0.5500112771987915,
1135
+ "learning_rate": 2.5478423283070797e-06,
1136
+ "loss": 0.08004761338233948,
1137
+ "memory(GiB)": 40.4,
1138
+ "step": 560,
1139
+ "token_acc": 0.9666666666666667,
1140
+ "train_speed(iter/s)": 0.077461
1141
+ },
1142
+ {
1143
+ "epoch": 1.0273058252427185,
1144
+ "grad_norm": 0.6031087040901184,
1145
+ "learning_rate": 2.5110411754005277e-06,
1146
+ "loss": 0.07369757890701294,
1147
+ "memory(GiB)": 40.4,
1148
+ "step": 565,
1149
+ "token_acc": 0.9675118858954042,
1150
+ "train_speed(iter/s)": 0.077479
1151
+ },
1152
+ {
1153
+ "epoch": 1.0364077669902914,
1154
+ "grad_norm": 0.6123142242431641,
1155
+ "learning_rate": 2.4742376296279656e-06,
1156
+ "loss": 0.07673358917236328,
1157
+ "memory(GiB)": 40.4,
1158
+ "step": 570,
1159
+ "token_acc": 0.96513470681458,
1160
+ "train_speed(iter/s)": 0.077492
1161
+ },
1162
+ {
1163
+ "epoch": 1.045509708737864,
1164
+ "grad_norm": 0.4750412404537201,
1165
+ "learning_rate": 2.437439667128757e-06,
1166
+ "loss": 0.07482797503471375,
1167
+ "memory(GiB)": 40.4,
1168
+ "step": 575,
1169
+ "token_acc": 0.9722222222222222,
1170
+ "train_speed(iter/s)": 0.077462
1171
+ },
1172
+ {
1173
+ "epoch": 1.0546116504854368,
1174
+ "grad_norm": 0.6936323642730713,
1175
+ "learning_rate": 2.4006552628322495e-06,
1176
+ "loss": 0.07669172286987305,
1177
+ "memory(GiB)": 40.4,
1178
+ "step": 580,
1179
+ "token_acc": 0.9698890649762282,
1180
+ "train_speed(iter/s)": 0.077497
1181
+ },
1182
+ {
1183
+ "epoch": 1.0637135922330097,
1184
+ "grad_norm": 0.5415986180305481,
1185
+ "learning_rate": 2.3638923887294252e-06,
1186
+ "loss": 0.07764337062835694,
1187
+ "memory(GiB)": 40.4,
1188
+ "step": 585,
1189
+ "token_acc": 0.9722662440570523,
1190
+ "train_speed(iter/s)": 0.077534
1191
+ },
1192
+ {
1193
+ "epoch": 1.0728155339805825,
1194
+ "grad_norm": 0.5562268495559692,
1195
+ "learning_rate": 2.3271590121452034e-06,
1196
+ "loss": 0.07850711941719055,
1197
+ "memory(GiB)": 40.4,
1198
+ "step": 590,
1199
+ "token_acc": 0.9627575277337559,
1200
+ "train_speed(iter/s)": 0.077312
1201
+ },
1202
+ {
1203
+ "epoch": 1.0819174757281553,
1204
+ "grad_norm": 0.5438592433929443,
1205
+ "learning_rate": 2.2904630940117383e-06,
1206
+ "loss": 0.07206880450248718,
1207
+ "memory(GiB)": 40.4,
1208
+ "step": 595,
1209
+ "token_acc": 0.9706582077716098,
1210
+ "train_speed(iter/s)": 0.077329
1211
+ },
1212
+ {
1213
+ "epoch": 1.0910194174757282,
1214
+ "grad_norm": 0.7570096254348755,
1215
+ "learning_rate": 2.253812587143113e-06,
1216
+ "loss": 0.07922015190124512,
1217
+ "memory(GiB)": 40.4,
1218
+ "step": 600,
1219
+ "token_acc": 0.9675632911392406,
1220
+ "train_speed(iter/s)": 0.077373
1221
+ },
1222
+ {
1223
+ "epoch": 1.100121359223301,
1224
+ "grad_norm": 0.44248196482658386,
1225
+ "learning_rate": 2.2172154345117896e-06,
1226
+ "loss": 0.07421438097953796,
1227
+ "memory(GiB)": 40.4,
1228
+ "step": 605,
1229
+ "token_acc": 0.969047619047619,
1230
+ "train_speed(iter/s)": 0.077227
1231
+ },
1232
+ {
1233
+ "epoch": 1.1092233009708738,
1234
+ "grad_norm": 0.8693225383758545,
1235
+ "learning_rate": 2.18067956752719e-06,
1236
+ "loss": 0.07179425954818726,
1237
+ "memory(GiB)": 40.4,
1238
+ "step": 610,
1239
+ "token_acc": 0.9738302934179223,
1240
+ "train_speed(iter/s)": 0.077227
1241
+ },
1242
+ {
1243
+ "epoch": 1.1183252427184467,
1244
+ "grad_norm": 0.6093197464942932,
1245
+ "learning_rate": 2.1442129043167877e-06,
1246
+ "loss": 0.07261105179786682,
1247
+ "memory(GiB)": 40.4,
1248
+ "step": 615,
1249
+ "token_acc": 0.972244250594766,
1250
+ "train_speed(iter/s)": 0.077265
1251
+ },
1252
+ {
1253
+ "epoch": 1.1274271844660193,
1254
+ "grad_norm": 0.47732552886009216,
1255
+ "learning_rate": 2.1078233480100708e-06,
1256
+ "loss": 0.07763968706130982,
1257
+ "memory(GiB)": 40.4,
1258
+ "step": 620,
1259
+ "token_acc": 0.9746233148295004,
1260
+ "train_speed(iter/s)": 0.077083
1261
+ },
1262
+ {
1263
+ "epoch": 1.1365291262135921,
1264
+ "grad_norm": 0.6436070799827576,
1265
+ "learning_rate": 2.0715187850257645e-06,
1266
+ "loss": 0.07869491577148438,
1267
+ "memory(GiB)": 40.4,
1268
+ "step": 625,
1269
+ "token_acc": 0.9675632911392406,
1270
+ "train_speed(iter/s)": 0.077031
1271
+ },
1272
+ {
1273
+ "epoch": 1.145631067961165,
1274
+ "grad_norm": 0.6669154167175293,
1275
+ "learning_rate": 2.0353070833626684e-06,
1276
+ "loss": 0.07925596237182617,
1277
+ "memory(GiB)": 40.4,
1278
+ "step": 630,
1279
+ "token_acc": 0.964314036478985,
1280
+ "train_speed(iter/s)": 0.077048
1281
+ },
1282
+ {
1283
+ "epoch": 1.1547330097087378,
1284
+ "grad_norm": 0.6365996599197388,
1285
+ "learning_rate": 1.999196090894485e-06,
1286
+ "loss": 0.06456078886985779,
1287
+ "memory(GiB)": 40.4,
1288
+ "step": 635,
1289
+ "token_acc": 0.9667194928684627,
1290
+ "train_speed(iter/s)": 0.077101
1291
+ },
1292
+ {
1293
+ "epoch": 1.1638349514563107,
1294
+ "grad_norm": 0.5614244341850281,
1295
+ "learning_rate": 1.963193633669018e-06,
1296
+ "loss": 0.07243520021438599,
1297
+ "memory(GiB)": 40.4,
1298
+ "step": 640,
1299
+ "token_acc": 0.9666931007137193,
1300
+ "train_speed(iter/s)": 0.077155
1301
+ },
1302
+ {
1303
+ "epoch": 1.1729368932038835,
1304
+ "grad_norm": 0.8191459774971008,
1305
+ "learning_rate": 1.927307514212089e-06,
1306
+ "loss": 0.0762752890586853,
1307
+ "memory(GiB)": 40.4,
1308
+ "step": 645,
1309
+ "token_acc": 0.9698412698412698,
1310
+ "train_speed(iter/s)": 0.077153
1311
+ },
1312
+ {
1313
+ "epoch": 1.1820388349514563,
1314
+ "grad_norm": 0.523980438709259,
1315
+ "learning_rate": 1.8915455098365651e-06,
1316
+ "loss": 0.0773351550102234,
1317
+ "memory(GiB)": 40.4,
1318
+ "step": 650,
1319
+ "token_acc": 0.9675118858954042,
1320
+ "train_speed(iter/s)": 0.077211
1321
+ },
1322
+ {
1323
+ "epoch": 1.1911407766990292,
1324
+ "grad_norm": 0.5650423169136047,
1325
+ "learning_rate": 1.8559153709568393e-06,
1326
+ "loss": 0.07858687043190002,
1327
+ "memory(GiB)": 40.4,
1328
+ "step": 655,
1329
+ "token_acc": 0.9635499207606973,
1330
+ "train_speed(iter/s)": 0.077253
1331
+ },
1332
+ {
1333
+ "epoch": 1.200242718446602,
1334
+ "grad_norm": 0.3905327022075653,
1335
+ "learning_rate": 1.8204248194091429e-06,
1336
+ "loss": 0.07570682168006897,
1337
+ "memory(GiB)": 40.4,
1338
+ "step": 660,
1339
+ "token_acc": 0.9674861221252974,
1340
+ "train_speed(iter/s)": 0.077222
1341
+ },
1342
+ {
1343
+ "epoch": 1.2093446601941746,
1344
+ "grad_norm": 0.6456849575042725,
1345
+ "learning_rate": 1.7850815467780616e-06,
1346
+ "loss": 0.06978952884674072,
1347
+ "memory(GiB)": 40.4,
1348
+ "step": 665,
1349
+ "token_acc": 0.976984126984127,
1350
+ "train_speed(iter/s)": 0.077238
1351
+ },
1352
+ {
1353
+ "epoch": 1.2184466019417475,
1354
+ "grad_norm": 0.49169182777404785,
1355
+ "learning_rate": 1.7498932127295892e-06,
1356
+ "loss": 0.06932756900787354,
1357
+ "memory(GiB)": 40.4,
1358
+ "step": 670,
1359
+ "token_acc": 0.9674603174603175,
1360
+ "train_speed(iter/s)": 0.077305
1361
+ },
1362
+ {
1363
+ "epoch": 1.2275485436893203,
1364
+ "grad_norm": 0.8174545764923096,
1365
+ "learning_rate": 1.7148674433511176e-06,
1366
+ "loss": 0.07247714400291443,
1367
+ "memory(GiB)": 40.4,
1368
+ "step": 675,
1369
+ "token_acc": 0.9785714285714285,
1370
+ "train_speed(iter/s)": 0.077358
1371
+ },
1372
+ {
1373
+ "epoch": 1.2366504854368932,
1374
+ "grad_norm": 0.5874563455581665,
1375
+ "learning_rate": 1.6800118294986936e-06,
1376
+ "loss": 0.08156619668006897,
1377
+ "memory(GiB)": 40.4,
1378
+ "step": 680,
1379
+ "token_acc": 0.9619952494061758,
1380
+ "train_speed(iter/s)": 0.077379
1381
+ },
1382
+ {
1383
+ "epoch": 1.245752427184466,
1384
+ "grad_norm": 0.7023929357528687,
1385
+ "learning_rate": 1.645333925151908e-06,
1386
+ "loss": 0.0740778088569641,
1387
+ "memory(GiB)": 40.4,
1388
+ "step": 685,
1389
+ "token_acc": 0.9643423137876387,
1390
+ "train_speed(iter/s)": 0.077282
1391
+ },
1392
+ {
1393
+ "epoch": 1.2548543689320388,
1394
+ "grad_norm": 0.6284681558609009,
1395
+ "learning_rate": 1.610841245776789e-06,
1396
+ "loss": 0.07937963008880615,
1397
+ "memory(GiB)": 40.4,
1398
+ "step": 690,
1399
+ "token_acc": 0.9682791435368755,
1400
+ "train_speed(iter/s)": 0.077267
1401
+ },
1402
+ {
1403
+ "epoch": 1.2639563106796117,
1404
+ "grad_norm": 0.4900761544704437,
1405
+ "learning_rate": 1.5765412666970302e-06,
1406
+ "loss": 0.07481481432914734,
1407
+ "memory(GiB)": 40.4,
1408
+ "step": 695,
1409
+ "token_acc": 0.9714512291831879,
1410
+ "train_speed(iter/s)": 0.077241
1411
+ },
1412
+ {
1413
+ "epoch": 1.2730582524271845,
1414
+ "grad_norm": 0.7159978747367859,
1415
+ "learning_rate": 1.5424414214739258e-06,
1416
+ "loss": 0.07213735580444336,
1417
+ "memory(GiB)": 40.4,
1418
+ "step": 700,
1419
+ "token_acc": 0.9738302934179223,
1420
+ "train_speed(iter/s)": 0.077237
1421
+ },
1422
+ {
1423
+ "epoch": 1.2821601941747574,
1424
+ "grad_norm": 0.6261754631996155,
1425
+ "learning_rate": 1.5085491002953535e-06,
1426
+ "loss": 0.07179176211357116,
1427
+ "memory(GiB)": 40.4,
1428
+ "step": 705,
1429
+ "token_acc": 0.969047619047619,
1430
+ "train_speed(iter/s)": 0.077083
1431
+ },
1432
+ {
1433
+ "epoch": 1.29126213592233,
1434
+ "grad_norm": 0.9063695073127747,
1435
+ "learning_rate": 1.4748716483741562e-06,
1436
+ "loss": 0.07754602432250976,
1437
+ "memory(GiB)": 40.4,
1438
+ "step": 710,
1439
+ "token_acc": 0.96513470681458,
1440
+ "train_speed(iter/s)": 0.077061
1441
+ },
1442
+ {
1443
+ "epoch": 1.300364077669903,
1444
+ "grad_norm": 0.6574028134346008,
1445
+ "learning_rate": 1.4414163643562755e-06,
1446
+ "loss": 0.07884335517883301,
1447
+ "memory(GiB)": 40.4,
1448
+ "step": 715,
1449
+ "token_acc": 0.9675376088677752,
1450
+ "train_speed(iter/s)": 0.077069
1451
+ },
1452
+ {
1453
+ "epoch": 1.3094660194174756,
1454
+ "grad_norm": 0.5524230599403381,
1455
+ "learning_rate": 1.4081904987389701e-06,
1456
+ "loss": 0.07660083174705505,
1457
+ "memory(GiB)": 40.4,
1458
+ "step": 720,
1459
+ "token_acc": 0.9635210150674068,
1460
+ "train_speed(iter/s)": 0.077072
1461
+ },
1462
+ {
1463
+ "epoch": 1.3185679611650485,
1464
+ "grad_norm": 0.5381263494491577,
1465
+ "learning_rate": 1.375201252299479e-06,
1466
+ "loss": 0.07187164425849915,
1467
+ "memory(GiB)": 40.4,
1468
+ "step": 725,
1469
+ "token_acc": 0.9690966719492868,
1470
+ "train_speed(iter/s)": 0.077084
1471
+ },
1472
+ {
1473
+ "epoch": 1.3276699029126213,
1474
+ "grad_norm": 0.6094266176223755,
1475
+ "learning_rate": 1.3424557745344508e-06,
1476
+ "loss": 0.07152368426322937,
1477
+ "memory(GiB)": 40.4,
1478
+ "step": 730,
1479
+ "token_acc": 0.9690966719492868,
1480
+ "train_speed(iter/s)": 0.07712
1481
+ },
1482
+ {
1483
+ "epoch": 1.3367718446601942,
1484
+ "grad_norm": 0.37662273645401,
1485
+ "learning_rate": 1.3099611621104875e-06,
1486
+ "loss": 0.07852091193199158,
1487
+ "memory(GiB)": 40.4,
1488
+ "step": 735,
1489
+ "token_acc": 0.9698412698412698,
1490
+ "train_speed(iter/s)": 0.077111
1491
+ },
1492
+ {
1493
+ "epoch": 1.345873786407767,
1494
+ "grad_norm": 0.8660151958465576,
1495
+ "learning_rate": 1.2777244573261479e-06,
1496
+ "loss": 0.0761515736579895,
1497
+ "memory(GiB)": 40.4,
1498
+ "step": 740,
1499
+ "token_acc": 0.9650793650793651,
1500
+ "train_speed(iter/s)": 0.077083
1501
+ },
1502
+ {
1503
+ "epoch": 1.3549757281553398,
1504
+ "grad_norm": 0.8635317087173462,
1505
+ "learning_rate": 1.245752646585719e-06,
1506
+ "loss": 0.07429265975952148,
1507
+ "memory(GiB)": 40.4,
1508
+ "step": 745,
1509
+ "token_acc": 0.9706582077716098,
1510
+ "train_speed(iter/s)": 0.077017
1511
+ },
1512
+ {
1513
+ "epoch": 1.3640776699029127,
1514
+ "grad_norm": 0.6921953558921814,
1515
+ "learning_rate": 1.214052658885113e-06,
1516
+ "loss": 0.08055119514465332,
1517
+ "memory(GiB)": 40.4,
1518
+ "step": 750,
1519
+ "token_acc": 0.9659000793021412,
1520
+ "train_speed(iter/s)": 0.07705
1521
+ },
1522
+ {
1523
+ "epoch": 1.3731796116504853,
1524
+ "grad_norm": 0.512025773525238,
1525
+ "learning_rate": 1.182631364310199e-06,
1526
+ "loss": 0.07414981126785278,
1527
+ "memory(GiB)": 40.4,
1528
+ "step": 755,
1529
+ "token_acc": 0.9738095238095238,
1530
+ "train_speed(iter/s)": 0.077125
1531
+ },
1532
+ {
1533
+ "epoch": 1.3822815533980584,
1534
+ "grad_norm": 0.47374847531318665,
1535
+ "learning_rate": 1.1514955725479057e-06,
1536
+ "loss": 0.07829545140266418,
1537
+ "memory(GiB)": 40.4,
1538
+ "step": 760,
1539
+ "token_acc": 0.9675118858954042,
1540
+ "train_speed(iter/s)": 0.077061
1541
+ },
1542
+ {
1543
+ "epoch": 1.391383495145631,
1544
+ "grad_norm": 0.5193628072738647,
1545
+ "learning_rate": 1.1206520314104083e-06,
1546
+ "loss": 0.06979748606681824,
1547
+ "memory(GiB)": 40.4,
1548
+ "step": 765,
1549
+ "token_acc": 0.9730799683293745,
1550
+ "train_speed(iter/s)": 0.077097
1551
+ },
1552
+ {
1553
+ "epoch": 1.4004854368932038,
1554
+ "grad_norm": 0.5398116707801819,
1555
+ "learning_rate": 1.0901074253727338e-06,
1556
+ "loss": 0.07316485643386841,
1557
+ "memory(GiB)": 40.4,
1558
+ "step": 770,
1559
+ "token_acc": 0.9674861221252974,
1560
+ "train_speed(iter/s)": 0.077134
1561
+ },
1562
+ {
1563
+ "epoch": 1.4095873786407767,
1564
+ "grad_norm": 0.9198482036590576,
1565
+ "learning_rate": 1.0598683741240861e-06,
1566
+ "loss": 0.0778656005859375,
1567
+ "memory(GiB)": 40.4,
1568
+ "step": 775,
1569
+ "token_acc": 0.9714512291831879,
1570
+ "train_speed(iter/s)": 0.077187
1571
+ },
1572
+ {
1573
+ "epoch": 1.4186893203883495,
1574
+ "grad_norm": 0.5479600429534912,
1575
+ "learning_rate": 1.0299414311332107e-06,
1576
+ "loss": 0.0758398413658142,
1577
+ "memory(GiB)": 40.4,
1578
+ "step": 780,
1579
+ "token_acc": 0.9706582077716098,
1580
+ "train_speed(iter/s)": 0.077204
1581
+ },
1582
+ {
1583
+ "epoch": 1.4277912621359223,
1584
+ "grad_norm": 0.562239944934845,
1585
+ "learning_rate": 1.0003330822281188e-06,
1586
+ "loss": 0.08118345737457275,
1587
+ "memory(GiB)": 40.4,
1588
+ "step": 785,
1589
+ "token_acc": 0.9658730158730159,
1590
+ "train_speed(iter/s)": 0.077197
1591
+ },
1592
+ {
1593
+ "epoch": 1.4368932038834952,
1594
+ "grad_norm": 0.608139157295227,
1595
+ "learning_rate": 9.710497441904614e-07,
1596
+ "loss": 0.07277892231941223,
1597
+ "memory(GiB)": 40.4,
1598
+ "step": 790,
1599
+ "token_acc": 0.9739130434782609,
1600
+ "train_speed(iter/s)": 0.077169
1601
+ },
1602
+ {
1603
+ "epoch": 1.445995145631068,
1604
+ "grad_norm": 0.6108372807502747,
1605
+ "learning_rate": 9.420977633648739e-07,
1606
+ "loss": 0.0743071436882019,
1607
+ "memory(GiB)": 40.4,
1608
+ "step": 795,
1609
+ "token_acc": 0.9651070578905631,
1610
+ "train_speed(iter/s)": 0.077195
1611
+ },
1612
+ {
1613
+ "epoch": 1.4550970873786409,
1614
+ "grad_norm": 0.5900782346725464,
1615
+ "learning_rate": 9.134834142835794e-07,
1616
+ "loss": 0.07513993978500366,
1617
+ "memory(GiB)": 40.4,
1618
+ "step": 800,
1619
+ "token_acc": 0.9738302934179223,
1620
+ "train_speed(iter/s)": 0.07724
1621
+ },
1622
+ {
1623
+ "epoch": 1.4641990291262137,
1624
+ "grad_norm": 0.5346866846084595,
1625
+ "learning_rate": 8.852128983065653e-07,
1626
+ "loss": 0.07092651724815369,
1627
+ "memory(GiB)": 40.4,
1628
+ "step": 805,
1629
+ "token_acc": 0.9722662440570523,
1630
+ "train_speed(iter/s)": 0.077133
1631
+ },
1632
+ {
1633
+ "epoch": 1.4733009708737863,
1634
+ "grad_norm": 0.504199743270874,
1635
+ "learning_rate": 8.572923422776055e-07,
1636
+ "loss": 0.07900516986846924,
1637
+ "memory(GiB)": 40.4,
1638
+ "step": 810,
1639
+ "token_acc": 0.9524564183835182,
1640
+ "train_speed(iter/s)": 0.077129
1641
+ },
1642
+ {
1643
+ "epoch": 1.4824029126213591,
1644
+ "grad_norm": 0.5348660349845886,
1645
+ "learning_rate": 8.297277971964443e-07,
1646
+ "loss": 0.07192928791046142,
1647
+ "memory(GiB)": 40.4,
1648
+ "step": 815,
1649
+ "token_acc": 0.9706349206349206,
1650
+ "train_speed(iter/s)": 0.077153
1651
+ },
1652
+ {
1653
+ "epoch": 1.491504854368932,
1654
+ "grad_norm": 0.7142664194107056,
1655
+ "learning_rate": 8.025252369074077e-07,
1656
+ "loss": 0.07966341972351074,
1657
+ "memory(GiB)": 40.4,
1658
+ "step": 820,
1659
+ "token_acc": 0.9714285714285714,
1660
+ "train_speed(iter/s)": 0.077158
1661
+ },
1662
+ {
1663
+ "epoch": 1.5006067961165048,
1664
+ "grad_norm": 0.670011579990387,
1665
+ "learning_rate": 7.756905568047393e-07,
1666
+ "loss": 0.07460339069366455,
1667
+ "memory(GiB)": 40.4,
1668
+ "step": 825,
1669
+ "token_acc": 0.9698412698412698,
1670
+ "train_speed(iter/s)": 0.077072
1671
+ },
1672
+ {
1673
+ "epoch": 1.5097087378640777,
1674
+ "grad_norm": 0.9091220498085022,
1675
+ "learning_rate": 7.492295725549423e-07,
1676
+ "loss": 0.07916736602783203,
1677
+ "memory(GiB)": 40.4,
1678
+ "step": 830,
1679
+ "token_acc": 0.9714512291831879,
1680
+ "train_speed(iter/s)": 0.077125
1681
+ },
1682
+ {
1683
+ "epoch": 1.5188106796116505,
1684
+ "grad_norm": 0.5154448747634888,
1685
+ "learning_rate": 7.231480188363906e-07,
1686
+ "loss": 0.07609822750091552,
1687
+ "memory(GiB)": 40.4,
1688
+ "step": 835,
1689
+ "token_acc": 0.9619047619047619,
1690
+ "train_speed(iter/s)": 0.077151
1691
+ },
1692
+ {
1693
+ "epoch": 1.5279126213592233,
1694
+ "grad_norm": 0.5767259001731873,
1695
+ "learning_rate": 6.974515480965038e-07,
1696
+ "loss": 0.07642306089401245,
1697
+ "memory(GiB)": 40.4,
1698
+ "step": 840,
1699
+ "token_acc": 0.9635499207606973,
1700
+ "train_speed(iter/s)": 0.077169
1701
+ },
1702
+ {
1703
+ "epoch": 1.537014563106796,
1704
+ "grad_norm": 0.559921145439148,
1705
+ "learning_rate": 6.721457293267344e-07,
1706
+ "loss": 0.07739580273628235,
1707
+ "memory(GiB)": 40.4,
1708
+ "step": 845,
1709
+ "token_acc": 0.9659540775930324,
1710
+ "train_speed(iter/s)": 0.077202
1711
+ },
1712
+ {
1713
+ "epoch": 1.546116504854369,
1714
+ "grad_norm": 0.5525022745132446,
1715
+ "learning_rate": 6.472360468556419e-07,
1716
+ "loss": 0.07661284804344178,
1717
+ "memory(GiB)": 40.4,
1718
+ "step": 850,
1719
+ "token_acc": 0.9690966719492868,
1720
+ "train_speed(iter/s)": 0.077223
1721
+ },
1722
+ {
1723
+ "epoch": 1.5552184466019416,
1724
+ "grad_norm": 0.7156991958618164,
1725
+ "learning_rate": 6.227278991603239e-07,
1726
+ "loss": 0.07607601881027222,
1727
+ "memory(GiB)": 40.4,
1728
+ "step": 855,
1729
+ "token_acc": 0.9738924050632911,
1730
+ "train_speed(iter/s)": 0.077263
1731
+ },
1732
+ {
1733
+ "epoch": 1.5643203883495147,
1734
+ "grad_norm": 0.578790009021759,
1735
+ "learning_rate": 5.986265976964412e-07,
1736
+ "loss": 0.07703717947006225,
1737
+ "memory(GiB)": 40.4,
1738
+ "step": 860,
1739
+ "token_acc": 0.9627575277337559,
1740
+ "train_speed(iter/s)": 0.077321
1741
+ },
1742
+ {
1743
+ "epoch": 1.5734223300970873,
1744
+ "grad_norm": 0.41067153215408325,
1745
+ "learning_rate": 5.749373657471127e-07,
1746
+ "loss": 0.07262166738510131,
1747
+ "memory(GiB)": 40.4,
1748
+ "step": 865,
1749
+ "token_acc": 0.9666931007137193,
1750
+ "train_speed(iter/s)": 0.077313
1751
+ },
1752
+ {
1753
+ "epoch": 1.5825242718446602,
1754
+ "grad_norm": 0.6594594120979309,
1755
+ "learning_rate": 5.516653372909142e-07,
1756
+ "loss": 0.07546203732490539,
1757
+ "memory(GiB)": 40.4,
1758
+ "step": 870,
1759
+ "token_acc": 0.9730799683293745,
1760
+ "train_speed(iter/s)": 0.077321
1761
+ },
1762
+ {
1763
+ "epoch": 1.591626213592233,
1764
+ "grad_norm": 0.6693688035011292,
1765
+ "learning_rate": 5.28815555889228e-07,
1766
+ "loss": 0.07242462635040284,
1767
+ "memory(GiB)": 40.4,
1768
+ "step": 875,
1769
+ "token_acc": 0.9714964370546318,
1770
+ "train_speed(iter/s)": 0.077315
1771
+ },
1772
+ {
1773
+ "epoch": 1.6007281553398058,
1774
+ "grad_norm": 0.5314414501190186,
1775
+ "learning_rate": 5.063929735931985e-07,
1776
+ "loss": 0.07621661424636841,
1777
+ "memory(GiB)": 40.4,
1778
+ "step": 880,
1779
+ "token_acc": 0.9746634996041171,
1780
+ "train_speed(iter/s)": 0.077305
1781
+ },
1782
+ {
1783
+ "epoch": 1.6098300970873787,
1784
+ "grad_norm": 0.39022502303123474,
1785
+ "learning_rate": 4.844024498705072e-07,
1786
+ "loss": 0.07379111647605896,
1787
+ "memory(GiB)": 40.4,
1788
+ "step": 885,
1789
+ "token_acc": 0.9770023790642347,
1790
+ "train_speed(iter/s)": 0.077319
1791
+ },
1792
+ {
1793
+ "epoch": 1.6189320388349513,
1794
+ "grad_norm": 0.5611955523490906,
1795
+ "learning_rate": 4.6284875055222415e-07,
1796
+ "loss": 0.07641223073005676,
1797
+ "memory(GiB)": 40.4,
1798
+ "step": 890,
1799
+ "token_acc": 0.969047619047619,
1800
+ "train_speed(iter/s)": 0.07736
1801
+ },
1802
+ {
1803
+ "epoch": 1.6280339805825244,
1804
+ "grad_norm": 0.5914463996887207,
1805
+ "learning_rate": 4.4173654679994543e-07,
1806
+ "loss": 0.07118785977363587,
1807
+ "memory(GiB)": 40.4,
1808
+ "step": 895,
1809
+ "token_acc": 0.9666931007137193,
1810
+ "train_speed(iter/s)": 0.077387
1811
+ },
1812
+ {
1813
+ "epoch": 1.637135922330097,
1814
+ "grad_norm": 0.6131768226623535,
1815
+ "learning_rate": 4.2107041409344686e-07,
1816
+ "loss": 0.06656063199043274,
1817
+ "memory(GiB)": 40.4,
1818
+ "step": 900,
1819
+ "token_acc": 0.9730586370839936,
1820
+ "train_speed(iter/s)": 0.077393
1821
+ },
1822
+ {
1823
+ "epoch": 1.64623786407767,
1824
+ "grad_norm": 0.6083477139472961,
1825
+ "learning_rate": 4.00854831239082e-07,
1826
+ "loss": 0.07548041343688965,
1827
+ "memory(GiB)": 40.4,
1828
+ "step": 905,
1829
+ "token_acc": 0.9706814580031695,
1830
+ "train_speed(iter/s)": 0.07732
1831
+ },
1832
+ {
1833
+ "epoch": 1.6553398058252426,
1834
+ "grad_norm": 0.5123993158340454,
1835
+ "learning_rate": 3.8109417939912044e-07,
1836
+ "loss": 0.07632001638412475,
1837
+ "memory(GiB)": 40.4,
1838
+ "step": 910,
1839
+ "token_acc": 0.9651070578905631,
1840
+ "train_speed(iter/s)": 0.07734
1841
+ },
1842
+ {
1843
+ "epoch": 1.6644417475728155,
1844
+ "grad_norm": 0.6305170655250549,
1845
+ "learning_rate": 3.617927411422584e-07,
1846
+ "loss": 0.07312512397766113,
1847
+ "memory(GiB)": 40.4,
1848
+ "step": 915,
1849
+ "token_acc": 0.9675376088677752,
1850
+ "train_speed(iter/s)": 0.077345
1851
+ },
1852
+ {
1853
+ "epoch": 1.6735436893203883,
1854
+ "grad_norm": 0.5339434742927551,
1855
+ "learning_rate": 3.4295469951548894e-07,
1856
+ "loss": 0.06849889755249024,
1857
+ "memory(GiB)": 40.4,
1858
+ "step": 920,
1859
+ "token_acc": 0.9674861221252974,
1860
+ "train_speed(iter/s)": 0.077349
1861
+ },
1862
+ {
1863
+ "epoch": 1.6826456310679612,
1864
+ "grad_norm": 0.532629132270813,
1865
+ "learning_rate": 3.24584137137543e-07,
1866
+ "loss": 0.07681695818901062,
1867
+ "memory(GiB)": 40.4,
1868
+ "step": 925,
1869
+ "token_acc": 0.9722222222222222,
1870
+ "train_speed(iter/s)": 0.077356
1871
+ },
1872
+ {
1873
+ "epoch": 1.691747572815534,
1874
+ "grad_norm": 0.4466962516307831,
1875
+ "learning_rate": 3.0668503531409876e-07,
1876
+ "loss": 0.06994915008544922,
1877
+ "memory(GiB)": 40.4,
1878
+ "step": 930,
1879
+ "token_acc": 0.9714964370546318,
1880
+ "train_speed(iter/s)": 0.077371
1881
+ },
1882
+ {
1883
+ "epoch": 1.7008495145631068,
1884
+ "grad_norm": 0.586765706539154,
1885
+ "learning_rate": 2.892612731749414e-07,
1886
+ "loss": 0.07494070529937744,
1887
+ "memory(GiB)": 40.4,
1888
+ "step": 935,
1889
+ "token_acc": 0.969047619047619,
1890
+ "train_speed(iter/s)": 0.077342
1891
+ },
1892
+ {
1893
+ "epoch": 1.7099514563106797,
1894
+ "grad_norm": 0.5412377715110779,
1895
+ "learning_rate": 2.723166268332733e-07,
1896
+ "loss": 0.07770473957061767,
1897
+ "memory(GiB)": 40.4,
1898
+ "step": 940,
1899
+ "token_acc": 0.9676145339652449,
1900
+ "train_speed(iter/s)": 0.077329
1901
+ },
1902
+ {
1903
+ "epoch": 1.7190533980582523,
1904
+ "grad_norm": 0.911586582660675,
1905
+ "learning_rate": 2.55854768567346e-07,
1906
+ "loss": 0.07914371490478515,
1907
+ "memory(GiB)": 40.4,
1908
+ "step": 945,
1909
+ "token_acc": 0.9674861221252974,
1910
+ "train_speed(iter/s)": 0.077298
1911
+ },
1912
+ {
1913
+ "epoch": 1.7281553398058254,
1914
+ "grad_norm": 0.6137750148773193,
1915
+ "learning_rate": 2.3987926602459465e-07,
1916
+ "loss": 0.08327807188034057,
1917
+ "memory(GiB)": 40.4,
1918
+ "step": 950,
1919
+ "token_acc": 0.9706349206349206,
1920
+ "train_speed(iter/s)": 0.077305
1921
+ },
1922
+ {
1923
+ "epoch": 1.737257281553398,
1924
+ "grad_norm": 0.576627790927887,
1925
+ "learning_rate": 2.2439358144845464e-07,
1926
+ "loss": 0.08012324571609497,
1927
+ "memory(GiB)": 40.4,
1928
+ "step": 955,
1929
+ "token_acc": 0.9643423137876387,
1930
+ "train_speed(iter/s)": 0.077328
1931
+ },
1932
+ {
1933
+ "epoch": 1.7463592233009708,
1934
+ "grad_norm": 0.6456671953201294,
1935
+ "learning_rate": 2.09401070928012e-07,
1936
+ "loss": 0.06627861261367798,
1937
+ "memory(GiB)": 40.4,
1938
+ "step": 960,
1939
+ "token_acc": 0.9714285714285714,
1940
+ "train_speed(iter/s)": 0.077243
1941
+ },
1942
+ {
1943
+ "epoch": 1.7554611650485437,
1944
+ "grad_norm": 0.6002473831176758,
1945
+ "learning_rate": 1.9490498367066817e-07,
1946
+ "loss": 0.071403968334198,
1947
+ "memory(GiB)": 40.4,
1948
+ "step": 965,
1949
+ "token_acc": 0.9682791435368755,
1950
+ "train_speed(iter/s)": 0.077258
1951
+ },
1952
+ {
1953
+ "epoch": 1.7645631067961165,
1954
+ "grad_norm": 0.7518230080604553,
1955
+ "learning_rate": 1.8090846129796586e-07,
1956
+ "loss": 0.07573525905609131,
1957
+ "memory(GiB)": 40.4,
1958
+ "step": 970,
1959
+ "token_acc": 0.9722222222222222,
1960
+ "train_speed(iter/s)": 0.077252
1961
+ },
1962
+ {
1963
+ "epoch": 1.7736650485436893,
1964
+ "grad_norm": 0.41464531421661377,
1965
+ "learning_rate": 1.6741453716472677e-07,
1966
+ "loss": 0.07870721817016602,
1967
+ "memory(GiB)": 40.4,
1968
+ "step": 975,
1969
+ "token_acc": 0.9627870150435471,
1970
+ "train_speed(iter/s)": 0.077259
1971
+ },
1972
+ {
1973
+ "epoch": 1.7827669902912622,
1974
+ "grad_norm": 0.7254371643066406,
1975
+ "learning_rate": 1.5442613570165993e-07,
1976
+ "loss": 0.08646805882453919,
1977
+ "memory(GiB)": 40.4,
1978
+ "step": 980,
1979
+ "token_acc": 0.9611419508326725,
1980
+ "train_speed(iter/s)": 0.077274
1981
+ },
1982
+ {
1983
+ "epoch": 1.791868932038835,
1984
+ "grad_norm": 0.7164713740348816,
1985
+ "learning_rate": 1.4194607178157237e-07,
1986
+ "loss": 0.07055433988571166,
1987
+ "memory(GiB)": 40.4,
1988
+ "step": 985,
1989
+ "token_acc": 0.9706349206349206,
1990
+ "train_speed(iter/s)": 0.077341
1991
+ },
1992
+ {
1993
+ "epoch": 1.8009708737864076,
1994
+ "grad_norm": 0.5821430087089539,
1995
+ "learning_rate": 1.2997705010932394e-07,
1996
+ "loss": 0.07743188142776489,
1997
+ "memory(GiB)": 40.4,
1998
+ "step": 990,
1999
+ "token_acc": 0.9674861221252974,
2000
+ "train_speed(iter/s)": 0.077362
2001
+ },
2002
+ {
2003
+ "epoch": 1.8100728155339807,
2004
+ "grad_norm": 0.766345739364624,
2005
+ "learning_rate": 1.1852166463565767e-07,
2006
+ "loss": 0.07668507099151611,
2007
+ "memory(GiB)": 40.4,
2008
+ "step": 995,
2009
+ "token_acc": 0.9770023790642347,
2010
+ "train_speed(iter/s)": 0.077362
2011
+ },
2012
+ {
2013
+ "epoch": 1.8191747572815533,
2014
+ "grad_norm": 0.5379170179367065,
2015
+ "learning_rate": 1.0758239799503412e-07,
2016
+ "loss": 0.06778880357742309,
2017
+ "memory(GiB)": 40.4,
2018
+ "step": 1000,
2019
+ "token_acc": 0.9746233148295004,
2020
+ "train_speed(iter/s)": 0.077358
2021
+ },
2022
+ {
2023
+ "epoch": 1.8282766990291264,
2024
+ "grad_norm": 0.587326169013977,
2025
+ "learning_rate": 9.716162096759019e-08,
2026
+ "loss": 0.07784827947616577,
2027
+ "memory(GiB)": 40.4,
2028
+ "step": 1005,
2029
+ "token_acc": 0.9770206022187005,
2030
+ "train_speed(iter/s)": 0.077289
2031
+ },
2032
+ {
2033
+ "epoch": 1.837378640776699,
2034
+ "grad_norm": 0.5790999531745911,
2035
+ "learning_rate": 8.726159196533718e-08,
2036
+ "loss": 0.07364106178283691,
2037
+ "memory(GiB)": 40.4,
2038
+ "step": 1010,
2039
+ "token_acc": 0.9730372720063442,
2040
+ "train_speed(iter/s)": 0.077306
2041
+ },
2042
+ {
2043
+ "epoch": 1.8464805825242718,
2044
+ "grad_norm": 0.5765237808227539,
2045
+ "learning_rate": 7.788445654271532e-08,
2046
+ "loss": 0.07042239308357238,
2047
+ "memory(GiB)": 40.4,
2048
+ "step": 1015,
2049
+ "token_acc": 0.9682539682539683,
2050
+ "train_speed(iter/s)": 0.077338
2051
+ },
2052
+ {
2053
+ "epoch": 1.8555825242718447,
2054
+ "grad_norm": 0.4627252221107483,
2055
+ "learning_rate": 6.903224693160348e-08,
2056
+ "loss": 0.06837155222892762,
2057
+ "memory(GiB)": 40.4,
2058
+ "step": 1020,
2059
+ "token_acc": 0.9754358161648178,
2060
+ "train_speed(iter/s)": 0.077366
2061
+ },
2062
+ {
2063
+ "epoch": 1.8646844660194175,
2064
+ "grad_norm": 0.5963551998138428,
2065
+ "learning_rate": 6.070688160088961e-08,
2066
+ "loss": 0.0674078106880188,
2067
+ "memory(GiB)": 40.4,
2068
+ "step": 1025,
2069
+ "token_acc": 0.9659270998415214,
2070
+ "train_speed(iter/s)": 0.077385
2071
+ },
2072
+ {
2073
+ "epoch": 1.8737864077669903,
2074
+ "grad_norm": 0.6391610503196716,
2075
+ "learning_rate": 5.291016484069683e-08,
2076
+ "loss": 0.07277075052261353,
2077
+ "memory(GiB)": 40.4,
2078
+ "step": 1030,
2079
+ "token_acc": 0.9659540775930324,
2080
+ "train_speed(iter/s)": 0.077401
2081
+ },
2082
+ {
2083
+ "epoch": 1.882888349514563,
2084
+ "grad_norm": 0.5019727945327759,
2085
+ "learning_rate": 4.564378637135408e-08,
2086
+ "loss": 0.0752260446548462,
2087
+ "memory(GiB)": 40.4,
2088
+ "step": 1035,
2089
+ "token_acc": 0.9682791435368755,
2090
+ "train_speed(iter/s)": 0.077434
2091
+ },
2092
+ {
2093
+ "epoch": 1.891990291262136,
2094
+ "grad_norm": 0.4186345040798187,
2095
+ "learning_rate": 3.890932097719624e-08,
2096
+ "loss": 0.06725120544433594,
2097
+ "memory(GiB)": 40.4,
2098
+ "step": 1040,
2099
+ "token_acc": 0.9730799683293745,
2100
+ "train_speed(iter/s)": 0.077451
2101
+ },
2102
+ {
2103
+ "epoch": 1.9010922330097086,
2104
+ "grad_norm": 0.6359046697616577,
2105
+ "learning_rate": 3.270822816527325e-08,
2106
+ "loss": 0.07682465314865113,
2107
+ "memory(GiB)": 40.4,
2108
+ "step": 1045,
2109
+ "token_acc": 0.969047619047619,
2110
+ "train_speed(iter/s)": 0.077498
2111
+ },
2112
+ {
2113
+ "epoch": 1.9101941747572817,
2114
+ "grad_norm": 0.5813617706298828,
2115
+ "learning_rate": 2.7041851849043678e-08,
2116
+ "loss": 0.0773731827735901,
2117
+ "memory(GiB)": 40.4,
2118
+ "step": 1050,
2119
+ "token_acc": 0.9674861221252974,
2120
+ "train_speed(iter/s)": 0.077486
2121
+ },
2122
+ {
2123
+ "epoch": 1.9192961165048543,
2124
+ "grad_norm": 0.4645262062549591,
2125
+ "learning_rate": 2.1911420057117994e-08,
2126
+ "loss": 0.07277056574821472,
2127
+ "memory(GiB)": 40.4,
2128
+ "step": 1055,
2129
+ "token_acc": 0.9690721649484536,
2130
+ "train_speed(iter/s)": 0.077447
2131
+ },
2132
+ {
2133
+ "epoch": 1.9283980582524272,
2134
+ "grad_norm": 0.8828046917915344,
2135
+ "learning_rate": 1.7318044667119226e-08,
2136
+ "loss": 0.07312785387039185,
2137
+ "memory(GiB)": 40.4,
2138
+ "step": 1060,
2139
+ "token_acc": 0.9675118858954042,
2140
+ "train_speed(iter/s)": 0.077476
2141
+ },
2142
+ {
2143
+ "epoch": 1.9375,
2144
+ "grad_norm": 0.8438335657119751,
2145
+ "learning_rate": 1.3262721164712667e-08,
2146
+ "loss": 0.07410634756088257,
2147
+ "memory(GiB)": 40.4,
2148
+ "step": 1065,
2149
+ "token_acc": 0.9698651863600317,
2150
+ "train_speed(iter/s)": 0.077482
2151
+ },
2152
+ {
2153
+ "epoch": 1.9466019417475728,
2154
+ "grad_norm": 0.6822603344917297,
2155
+ "learning_rate": 9.746328427863993e-09,
2156
+ "loss": 0.0720213532447815,
2157
+ "memory(GiB)": 40.4,
2158
+ "step": 1070,
2159
+ "token_acc": 0.9666666666666667,
2160
+ "train_speed(iter/s)": 0.077488
2161
+ },
2162
+ {
2163
+ "epoch": 1.9557038834951457,
2164
+ "grad_norm": 0.5685479640960693,
2165
+ "learning_rate": 6.769628536364981e-09,
2166
+ "loss": 0.07333976030349731,
2167
+ "memory(GiB)": 40.4,
2168
+ "step": 1075,
2169
+ "token_acc": 0.973015873015873,
2170
+ "train_speed(iter/s)": 0.077502
2171
+ },
2172
+ {
2173
+ "epoch": 1.9648058252427183,
2174
+ "grad_norm": 0.5445531606674194,
2175
+ "learning_rate": 4.333266606676711e-09,
2176
+ "loss": 0.07253679037094116,
2177
+ "memory(GiB)": 40.4,
2178
+ "step": 1080,
2179
+ "token_acc": 0.9730586370839936,
2180
+ "train_speed(iter/s)": 0.077494
2181
+ },
2182
+ {
2183
+ "epoch": 1.9739077669902914,
2184
+ "grad_norm": 0.6113319993019104,
2185
+ "learning_rate": 2.4377706521164224e-09,
2186
+ "loss": 0.07309662699699401,
2187
+ "memory(GiB)": 40.4,
2188
+ "step": 1085,
2189
+ "token_acc": 0.9722222222222222,
2190
+ "train_speed(iter/s)": 0.077475
2191
+ },
2192
+ {
2193
+ "epoch": 1.983009708737864,
2194
+ "grad_norm": 0.5483999252319336,
2195
+ "learning_rate": 1.0835514684262583e-09,
2196
+ "loss": 0.07428893446922302,
2197
+ "memory(GiB)": 40.4,
2198
+ "step": 1090,
2199
+ "token_acc": 0.9690966719492868,
2200
+ "train_speed(iter/s)": 0.077464
2201
+ },
2202
+ {
2203
+ "epoch": 1.992111650485437,
2204
+ "grad_norm": 0.6084752082824707,
2205
+ "learning_rate": 2.7090254474421154e-10,
2206
+ "loss": 0.07023123502731324,
2207
+ "memory(GiB)": 40.4,
2208
+ "step": 1095,
2209
+ "token_acc": 0.9786223277909739,
2210
+ "train_speed(iter/s)": 0.077453
2211
+ },
2212
+ {
2213
+ "epoch": 2.0,
2214
+ "grad_norm": 0.5853410363197327,
2215
+ "learning_rate": 0.0,
2216
+ "loss": 0.0724187433719635,
2217
+ "memory(GiB)": 40.4,
2218
+ "step": 1100,
2219
+ "token_acc": 0.9679780420860018,
2220
+ "train_speed(iter/s)": 0.077495
2221
+ }
2222
+ ],
2223
+ "logging_steps": 5,
2224
+ "max_steps": 1100,
2225
+ "num_input_tokens_seen": 0,
2226
+ "num_train_epochs": 2,
2227
+ "save_steps": 100,
2228
+ "stateful_callbacks": {
2229
+ "TrainerControl": {
2230
+ "args": {
2231
+ "should_epoch_stop": false,
2232
+ "should_evaluate": false,
2233
+ "should_log": false,
2234
+ "should_save": true,
2235
+ "should_training_stop": true
2236
+ },
2237
+ "attributes": {}
2238
+ }
2239
+ },
2240
+ "total_flos": 3.488531281539498e+18,
2241
+ "train_batch_size": 1,
2242
+ "trial_name": null,
2243
+ "trial_params": null
2244
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39ee48a7bb7b7a5bb370c36f96dc579ae71d79b3b404edce79e93b50b1f18da2
3
+ size 8120
zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``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``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``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``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``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``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info("Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info("Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)