Hugging Face's logo

Asap7772
/
Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0-ckpt-135

Safetensors
qwen3
trl
dpo
rlhf
alignment
Model card Files
xet
Community
Asap7772 commited on
Commit
884089d
·
verified ·
1 Parent(s): 51f0400

Upload checkpoint from checkpoint-135

Browse files
Files changed (27) hide show
  1. .gitattributes +1 -0
  2. README.md +219 -0
  3. added_tokens.json +28 -0
  4. chat_template.jinja +86 -0
  5. config.json +68 -0
  6. generation_config.json +12 -0
  7. latest +1 -0
  8. merges.txt +0 -0
  9. model-00001-of-00002.safetensors +3 -0
  10. model-00002-of-00002.safetensors +3 -0
  11. model.safetensors.index.json +406 -0
  12. rng_state_0.pth +3 -0
  13. rng_state_1.pth +3 -0
  14. rng_state_2.pth +3 -0
  15. rng_state_3.pth +3 -0
  16. rng_state_4.pth +3 -0
  17. rng_state_5.pth +3 -0
  18. rng_state_6.pth +3 -0
  19. rng_state_7.pth +3 -0
  20. scheduler.pt +3 -0
  21. special_tokens_map.json +31 -0
  22. tokenizer.json +3 -0
  23. tokenizer_config.json +239 -0
  24. trainer_state.json +2279 -0
  25. training_args.bin +3 -0
  26. vocab.json +0 -0
  27. zero_to_fp32.py +760 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: Asap7772/qwen3-4b-impabs-warmstart-sft-1e-5-1epoch-0828
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - rlhf
8
+ - alignment
9
+ ---
10
+
11
+ # Asap7772/Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0-ckpt-135
12
+
13
+ This model is a fine-tuned version of [Asap7772/qwen3-4b-impabs-warmstart-sft-1e-5-1epoch-0828](https://huggingface.co/Asap7772/qwen3-4b-impabs-warmstart-sft-1e-5-1epoch-0828) using TRL (Transformer Reinforcement Learning).
14
+
15
+ ## Model Details
16
+
17
+ - **Base Model**: Asap7772/qwen3-4b-impabs-warmstart-sft-1e-5-1epoch-0828
18
+ - **Checkpoint**: checkpoint-135
19
+ - **Fine-tuning Method**: DPO (Direct Preference Optimization)
20
+ - **Framework**: TRL
21
+
22
+ ## Usage
23
+
24
+ ```python
25
+ from transformers import AutoModelForCausalLM, AutoTokenizer
26
+
27
+ model = AutoModelForCausalLM.from_pretrained("Asap7772/Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0-ckpt-135")
28
+ tokenizer = AutoTokenizer.from_pretrained("Asap7772/Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0-ckpt-135")
29
+
30
+ # Your inference code here
31
+ ```
32
+
33
+ ## Training Details
34
+
35
+ This model was trained using the TRL library with DPO (Direct Preference Optimization).
36
+
37
+ ### Training Arguments
38
+
39
+ ```json
40
+ {
41
+ "output_dir": "/iris/u/asap7772/trl/checkpoints/Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0",
42
+ "overwrite_output_dir": false,
43
+ "do_train": false,
44
+ "do_eval": true,
45
+ "do_predict": false,
46
+ "eval_strategy": "epoch",
47
+ "prediction_loss_only": false,
48
+ "per_device_train_batch_size": 1,
49
+ "per_device_eval_batch_size": 8,
50
+ "per_gpu_train_batch_size": null,
51
+ "per_gpu_eval_batch_size": null,
52
+ "gradient_accumulation_steps": 16,
53
+ "eval_accumulation_steps": null,
54
+ "eval_delay": 0,
55
+ "torch_empty_cache_steps": null,
56
+ "learning_rate": 1e-07,
57
+ "weight_decay": 0.01,
58
+ "adam_beta1": 0.9,
59
+ "adam_beta2": 0.999,
60
+ "adam_epsilon": 1e-08,
61
+ "max_grad_norm": 1.0,
62
+ "num_train_epochs": 5,
63
+ "max_steps": -1,
64
+ "lr_scheduler_type": "linear",
65
+ "lr_scheduler_kwargs": {},
66
+ "warmup_ratio": 0.05,
67
+ "warmup_steps": 0,
68
+ "log_level": "passive",
69
+ "log_level_replica": "warning",
70
+ "log_on_each_node": true,
71
+ "logging_dir": "/iris/u/asap7772/trl/checkpoints/Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0/runs/Sep15_09-27-29_iris-hgx-2.stanford.edu",
72
+ "logging_strategy": "steps",
73
+ "logging_first_step": true,
74
+ "logging_steps": 1,
75
+ "logging_nan_inf_filter": true,
76
+ "save_strategy": "epoch",
77
+ "save_steps": 500,
78
+ "save_total_limit": null,
79
+ "save_safetensors": true,
80
+ "save_on_each_node": false,
81
+ "save_only_model": false,
82
+ "restore_callback_states_from_checkpoint": false,
83
+ "no_cuda": false,
84
+ "use_cpu": false,
85
+ "use_mps_device": false,
86
+ "seed": 42,
87
+ "data_seed": null,
88
+ "jit_mode_eval": false,
89
+ "use_ipex": false,
90
+ "bf16": true,
91
+ "fp16": false,
92
+ "fp16_opt_level": "O1",
93
+ "half_precision_backend": "auto",
94
+ "bf16_full_eval": false,
95
+ "fp16_full_eval": false,
96
+ "tf32": null,
97
+ "local_rank": 0,
98
+ "ddp_backend": null,
99
+ "tpu_num_cores": null,
100
+ "tpu_metrics_debug": false,
101
+ "debug": [],
102
+ "dataloader_drop_last": true,
103
+ "eval_steps": 1,
104
+ "dataloader_num_workers": 0,
105
+ "dataloader_prefetch_factor": null,
106
+ "past_index": -1,
107
+ "run_name": "Qwen3-4B-second-stage-DPO-lr-1e-7-beta-0.1-loss-sigmoid-rpo-1.0",
108
+ "disable_tqdm": false,
109
+ "remove_unused_columns": false,
110
+ "label_names": null,
111
+ "load_best_model_at_end": false,
112
+ "metric_for_best_model": null,
113
+ "greater_is_better": null,
114
+ "ignore_data_skip": false,
115
+ "fsdp": [],
116
+ "fsdp_min_num_params": 0,
117
+ "fsdp_config": {
118
+ "min_num_params": 0,
119
+ "xla": false,
120
+ "xla_fsdp_v2": false,
121
+ "xla_fsdp_grad_ckpt": false
122
+ },
123
+ "fsdp_transformer_layer_cls_to_wrap": null,
124
+ "accelerator_config": "AcceleratorConfig(split_batches=False, dispatch_batches=None, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False)",
125
+ "parallelism_config": null,
126
+ "deepspeed": null,
127
+ "label_smoothing_factor": 0.0,
128
+ "optim": "adamw_torch",
129
+ "optim_args": null,
130
+ "adafactor": false,
131
+ "group_by_length": false,
132
+ "length_column_name": "length",
133
+ "report_to": [
134
+ "wandb"
135
+ ],
136
+ "ddp_find_unused_parameters": null,
137
+ "ddp_bucket_cap_mb": null,
138
+ "ddp_broadcast_buffers": null,
139
+ "dataloader_pin_memory": false,
140
+ "dataloader_persistent_workers": false,
141
+ "skip_memory_metrics": true,
142
+ "use_legacy_prediction_loop": false,
143
+ "push_to_hub": false,
144
+ "resume_from_checkpoint": null,
145
+ "hub_model_id": null,
146
+ "hub_strategy": "every_save",
147
+ "hub_token": null,
148
+ "hub_private_repo": null,
149
+ "hub_always_push": false,
150
+ "hub_revision": null,
151
+ "gradient_checkpointing": true,
152
+ "gradient_checkpointing_kwargs": null,
153
+ "include_inputs_for_metrics": false,
154
+ "include_for_metrics": [],
155
+ "eval_do_concat_batches": true,
156
+ "fp16_backend": "auto",
157
+ "push_to_hub_model_id": null,
158
+ "push_to_hub_organization": null,
159
+ "push_to_hub_token": null,
160
+ "mp_parameters": "",
161
+ "auto_find_batch_size": false,
162
+ "full_determinism": false,
163
+ "torchdynamo": null,
164
+ "ray_scope": "last",
165
+ "ddp_timeout": 1800,
166
+ "torch_compile": false,
167
+ "torch_compile_backend": null,
168
+ "torch_compile_mode": null,
169
+ "include_tokens_per_second": false,
170
+ "include_num_input_tokens_seen": false,
171
+ "neftune_noise_alpha": null,
172
+ "optim_target_modules": null,
173
+ "batch_eval_metrics": false,
174
+ "eval_on_start": false,
175
+ "use_liger_kernel": false,
176
+ "liger_kernel_config": null,
177
+ "eval_use_gather_object": false,
178
+ "average_tokens_across_devices": true,
179
+ "model_init_kwargs": null,
180
+ "ref_model_init_kwargs": null,
181
+ "model_adapter_name": null,
182
+ "ref_adapter_name": null,
183
+ "force_use_ref_model": false,
184
+ "disable_dropout": true,
185
+ "use_logits_to_keep": false,
186
+ "dataset_num_proc": null,
187
+ "padding_value": null,
188
+ "label_pad_token_id": -100,
189
+ "max_prompt_length": 8192,
190
+ "max_completion_length": 1024,
191
+ "max_length": 9216,
192
+ "truncation_mode": "keep_end",
193
+ "padding_free": false,
194
+ "precompute_ref_log_probs": false,
195
+ "precompute_ref_batch_size": null,
196
+ "tools": null,
197
+ "loss_type": "sigmoid",
198
+ "use_liger_loss": false,
199
+ "base_model_attribute_name": "model",
200
+ "beta": 0.1,
201
+ "f_divergence_type": "FDivergenceType.REVERSE_KL",
202
+ "f_alpha_divergence_coef": 1.0,
203
+ "reference_free": false,
204
+ "label_smoothing": 0.0,
205
+ "use_weighting": false,
206
+ "rpo_alpha": 1.0,
207
+ "ld_alpha": null,
208
+ "discopop_tau": 0.05,
209
+ "loss_weights": null,
210
+ "sync_ref_model": false,
211
+ "ref_model_mixup_alpha": 0.6,
212
+ "ref_model_sync_steps": 512,
213
+ "generate_during_eval": false,
214
+ "distributed_state": "Distributed environment: DEEPSPEED Backend: nccl\nNum processes: 8\nProcess index: 0\nLocal process index: 0\nDevice: cuda:0\n",
215
+ "_n_gpu": 1,
216
+ "__cached__setup_devices": "cuda:0",
217
+ "deepspeed_plugin": "DeepSpeedPlugin(hf_ds_config=<transformers.integrations.deepspeed.HfTrainerDeepSpeedConfig object at 0x7f06dacf49d0>, gradient_accumulation_steps='auto', gradient_clipping='auto', zero_stage=3, is_train_batch_min=True, offload_optimizer_device='cpu', offload_param_device='cpu', offload_optimizer_nvme_path='none', offload_param_nvme_path='none', zero3_init_flag=True, zero3_save_16bit_model=True, transformer_moe_cls_names=None, enable_msamp=False, msamp_opt_level='O1')"
218
+ }
219
+ ```
added_tokens.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</think>": 151668,
3
+ "</tool_call>": 151658,
4
+ "</tool_response>": 151666,
5
+ "<think>": 151667,
6
+ "<tool_call>": 151657,
7
+ "<tool_response>": 151665,
8
+ "<|box_end|>": 151649,
9
+ "<|box_start|>": 151648,
10
+ "<|endoftext|>": 151643,
11
+ "<|file_sep|>": 151664,
12
+ "<|fim_middle|>": 151660,
13
+ "<|fim_pad|>": 151662,
14
+ "<|fim_prefix|>": 151659,
15
+ "<|fim_suffix|>": 151661,
16
+ "<|im_end|>": 151645,
17
+ "<|im_start|>": 151644,
18
+ "<|image_pad|>": 151655,
19
+ "<|object_ref_end|>": 151647,
20
+ "<|object_ref_start|>": 151646,
21
+ "<|quad_end|>": 151651,
22
+ "<|quad_start|>": 151650,
23
+ "<|repo_name|>": 151663,
24
+ "<|video_pad|>": 151656,
25
+ "<|vision_end|>": 151653,
26
+ "<|vision_pad|>": 151654,
27
+ "<|vision_start|>": 151652
28
+ }
chat_template.jinja ADDED
@@ -0,0 +1,86 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if message.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "dtype": "bfloat16",
8
+ "eos_token_id": 151645,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2560,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 9728,
14
+ "layer_types": [
15
+ "full_attention",
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention"
51
+ ],
52
+ "max_position_embeddings": 262144,
53
+ "max_window_layers": 36,
54
+ "model_type": "qwen3",
55
+ "num_attention_heads": 32,
56
+ "num_hidden_layers": 36,
57
+ "num_key_value_heads": 8,
58
+ "pad_token_id": 151643,
59
+ "rms_norm_eps": 1e-06,
60
+ "rope_scaling": null,
61
+ "rope_theta": 5000000,
62
+ "sliding_window": null,
63
+ "tie_word_embeddings": true,
64
+ "transformers_version": "4.56.1",
65
+ "use_cache": true,
66
+ "use_sliding_window": false,
67
+ "vocab_size": 151936
68
+ }
generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "temperature": 0.7,
9
+ "top_k": 20,
10
+ "top_p": 0.8,
11
+ "transformers_version": "4.56.1"
12
+ }
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step135
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:107094cfafebdfcc548258dece36eab7f75c4872b9e9e4e906cf1130d65f802b
3
+ size 4967215360
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd07ecba75303c70ebf6b161d984c36801a5953ef26852e8f50923153e717a38
3
+ size 3077766632
model.safetensors.index.json ADDED
@@ -0,0 +1,406 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_parameters": 196096,
4
+ "total_size": 8044936192
5
+ },
6
+ "weight_map": {
7
+ "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
8
+ "model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
9
+ "model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
10
+ "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
11
+ "model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
12
+ "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
13
+ "model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
14
+ "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
15
+ "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
16
+ "model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
17
+ "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
18
+ "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
19
+ "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
20
+ "model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
21
+ "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
22
+ "model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
23
+ "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
24
+ "model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
25
+ "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
26
+ "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
27
+ "model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
28
+ "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
29
+ "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
30
+ "model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
31
+ "model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
32
+ "model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
33
+ "model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
34
+ "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
35
+ "model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
36
+ "model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
37
+ "model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
38
+ "model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
39
+ "model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
40
+ "model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
41
+ "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
42
+ "model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
43
+ "model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
44
+ "model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
45
+ "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
46
+ "model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
47
+ "model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
48
+ "model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
49
+ "model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
50
+ "model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
51
+ "model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
52
+ "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
53
+ "model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
54
+ "model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
55
+ "model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
56
+ "model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
57
+ "model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
58
+ "model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
59
+ "model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
60
+ "model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
61
+ "model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
62
+ "model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
63
+ "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
64
+ "model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
65
+ "model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
66
+ "model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
67
+ "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
68
+ "model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
69
+ "model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
70
+ "model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
71
+ "model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
72
+ "model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
73
+ "model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
74
+ "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
75
+ "model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
76
+ "model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
77
+ "model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
78
+ "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
79
+ "model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
80
+ "model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
81
+ "model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
82
+ "model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
83
+ "model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
84
+ "model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
85
+ "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
86
+ "model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
87
+ "model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
88
+ "model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
89
+ "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
90
+ "model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
91
+ "model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
92
+ "model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
93
+ "model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
94
+ "model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
95
+ "model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
96
+ "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
97
+ "model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
98
+ "model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
99
+ "model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
100
+ "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
101
+ "model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
102
+ "model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
103
+ "model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
104
+ "model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
105
+ "model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
106
+ "model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
107
+ "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
108
+ "model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
109
+ "model.layers.17.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
110
+ "model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
111
+ "model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
112
+ "model.layers.17.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
113
+ "model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
114
+ "model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
115
+ "model.layers.17.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
116
+ "model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
117
+ "model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
118
+ "model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
119
+ "model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
120
+ "model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
121
+ "model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
122
+ "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
123
+ "model.layers.18.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
124
+ "model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
125
+ "model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
126
+ "model.layers.18.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
127
+ "model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
128
+ "model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
129
+ "model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
130
+ "model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
131
+ "model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
132
+ "model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
133
+ "model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
134
+ "model.layers.19.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
135
+ "model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
136
+ "model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
137
+ "model.layers.19.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
138
+ "model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
139
+ "model.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
140
+ "model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
141
+ "model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
142
+ "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
143
+ "model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
144
+ "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
145
+ "model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
146
+ "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
147
+ "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
148
+ "model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
149
+ "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
150
+ "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
151
+ "model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
152
+ "model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
153
+ "model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
154
+ "model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
155
+ "model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
156
+ "model.layers.20.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
157
+ "model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
158
+ "model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
159
+ "model.layers.20.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
160
+ "model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
161
+ "model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
162
+ "model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
163
+ "model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
164
+ "model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
165
+ "model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
166
+ "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
167
+ "model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
168
+ "model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
169
+ "model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
170
+ "model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
171
+ "model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
172
+ "model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
173
+ "model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
174
+ "model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
175
+ "model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
176
+ "model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
177
+ "model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
178
+ "model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
179
+ "model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
180
+ "model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
181
+ "model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
182
+ "model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
183
+ "model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
184
+ "model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
185
+ "model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
186
+ "model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
187
+ "model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
188
+ "model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
189
+ "model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
190
+ "model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
191
+ "model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
192
+ "model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
193
+ "model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
194
+ "model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
195
+ "model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
196
+ "model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
197
+ "model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
198
+ "model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
199
+ "model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
200
+ "model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
201
+ "model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
202
+ "model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
203
+ "model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
204
+ "model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
205
+ "model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
206
+ "model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
207
+ "model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
208
+ "model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
209
+ "model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
210
+ "model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
211
+ "model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
212
+ "model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
213
+ "model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
214
+ "model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
215
+ "model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
216
+ "model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
217
+ "model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
218
+ "model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
219
+ "model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
220
+ "model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
221
+ "model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
222
+ "model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
223
+ "model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
224
+ "model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
225
+ "model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
226
+ "model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
227
+ "model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
228
+ "model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
229
+ "model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
230
+ "model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
231
+ "model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
232
+ "model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
233
+ "model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
234
+ "model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
235
+ "model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
236
+ "model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
237
+ "model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
238
+ "model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
239
+ "model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
240
+ "model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
241
+ "model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
242
+ "model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
243
+ "model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
244
+ "model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
245
+ "model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
246
+ "model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
247
+ "model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
248
+ "model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
249
+ "model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
250
+ "model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
251
+ "model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
252
+ "model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
253
+ "model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
254
+ "model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
255
+ "model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
256
+ "model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
257
+ "model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
258
+ "model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
259
+ "model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
260
+ "model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
261
+ "model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
262
+ "model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
263
+ "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
264
+ "model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
265
+ "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
266
+ "model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
267
+ "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
268
+ "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
269
+ "model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
270
+ "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
271
+ "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
272
+ "model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
273
+ "model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
274
+ "model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
275
+ "model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
276
+ "model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
277
+ "model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
278
+ "model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
279
+ "model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
280
+ "model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
281
+ "model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
282
+ "model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
283
+ "model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
284
+ "model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
285
+ "model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
286
+ "model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
287
+ "model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
288
+ "model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
289
+ "model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
290
+ "model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
291
+ "model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
292
+ "model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
293
+ "model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
294
+ "model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
295
+ "model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
296
+ "model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
297
+ "model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
298
+ "model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
299
+ "model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
300
+ "model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
301
+ "model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
302
+ "model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
303
+ "model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
304
+ "model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
305
+ "model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
306
+ "model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
307
+ "model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
308
+ "model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
309
+ "model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
310
+ "model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
311
+ "model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
312
+ "model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
313
+ "model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
314
+ "model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
315
+ "model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
316
+ "model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
317
+ "model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
318
+ "model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
319
+ "model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
320
+ "model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
321
+ "model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
322
+ "model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
323
+ "model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
324
+ "model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
325
+ "model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
326
+ "model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
327
+ "model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
328
+ "model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
329
+ "model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
330
+ "model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
331
+ "model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
332
+ "model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
333
+ "model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
334
+ "model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
335
+ "model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
336
+ "model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
337
+ "model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
338
+ "model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
339
+ "model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
340
+ "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
341
+ "model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
342
+ "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
343
+ "model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
344
+ "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
345
+ "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
346
+ "model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
347
+ "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
348
+ "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
349
+ "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
350
+ "model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
351
+ "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
352
+ "model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
353
+ "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
354
+ "model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
355
+ "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
356
+ "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
357
+ "model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
358
+ "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
359
+ "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
360
+ "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
361
+ "model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
362
+ "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
363
+ "model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
364
+ "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
365
+ "model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
366
+ "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
367
+ "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
368
+ "model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
369
+ "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
370
+ "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
371
+ "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
372
+ "model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
373
+ "model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
374
+ "model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
375
+ "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
376
+ "model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
377
+ "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
378
+ "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
379
+ "model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
380
+ "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
381
+ "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
382
+ "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
383
+ "model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
384
+ "model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
385
+ "model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
386
+ "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
387
+ "model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
388
+ "model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
389
+ "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
390
+ "model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
391
+ "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
392
+ "model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
393
+ "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
394
+ "model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
395
+ "model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
396
+ "model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
397
+ "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
398
+ "model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
399
+ "model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
400
+ "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
401
+ "model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
402
+ "model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
403
+ "model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
404
+ "model.norm.weight": "model-00002-of-00002.safetensors"
405
+ }
406
+ }
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72e1b7d09d327da58dbf4f6ec91c25f8266fe9a398376fbb9f55466fba682801
3
+ size 16389
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b97114733f7d9f71d4fe5cf9c672395a21afc8c2d9910bc8f8d1aec20ef50fe7
3
+ size 16389
rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0eb9f1668c686a559f86c6a7d79fbf9c53dde5e1386ed17acd9234f519c7d6f4
3
+ size 16389
rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7802066665f429dc81402261bdd2fe5a2b86f195839dcaafee252a7936c58395
3
+ size 16389
rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19ab5c5e0552eff918c411c59be2a0f8003290ae5b74428a65debca9250295fd
3
+ size 16389
rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e15ece2216eb800331569258df1ca68a32b091e22dc07b06537dd3e7abecaec9
3
+ size 16389
rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f78f2e6557578841f88d7d977a66dcc5ccc8e45c0dea5ff3d7e07de7a27de3f
3
+ size 16389
rng_state_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:020cdda2a92333ab58d00856b156214e46b7374245507464250503080f07d9f0
3
+ size 16389
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dd4f970a43642d01289826979c401878bee63bc6647c3ab462cecf4c26cab0b3
3
+ size 1465
special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "extra_special_tokens": {},
234
+ "model_max_length": 262144,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }
trainer_state.json ADDED
@@ -0,0 +1,2279 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 5.0,
6
+ "eval_steps": 1,
7
+ "global_step": 135,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.037122969837587005,
14
+ "grad_norm": 37.7234992980957,
15
+ "learning_rate": 0.0,
16
+ "logits/chosen": 0.43707275390625,
17
+ "logits/rejected": 0.40472412109375,
18
+ "logps/chosen": -176.875,
19
+ "logps/rejected": -148.3125,
20
+ "loss": 3.0214,
21
+ "nll_loss": 2.3330078125,
22
+ "rewards/accuracies": 0.0,
23
+ "rewards/chosen": 0.0,
24
+ "rewards/margins": 0.0,
25
+ "rewards/rejected": 0.0,
26
+ "step": 1
27
+ },
28
+ {
29
+ "epoch": 0.07424593967517401,
30
+ "grad_norm": 33.89510726928711,
31
+ "learning_rate": 1.4285714285714284e-08,
32
+ "logits/chosen": 0.40576171875,
33
+ "logits/rejected": 0.33673095703125,
34
+ "logps/chosen": -167.6875,
35
+ "logps/rejected": -161.625,
36
+ "loss": 2.9501,
37
+ "nll_loss": 2.26171875,
38
+ "rewards/accuracies": 0.0,
39
+ "rewards/chosen": 0.0,
40
+ "rewards/margins": 0.0,
41
+ "rewards/rejected": 0.0,
42
+ "step": 2
43
+ },
44
+ {
45
+ "epoch": 0.11136890951276102,
46
+ "grad_norm": 37.10581970214844,
47
+ "learning_rate": 2.857142857142857e-08,
48
+ "logits/chosen": 0.38861083984375,
49
+ "logits/rejected": 0.3741455078125,
50
+ "logps/chosen": -175.125,
51
+ "logps/rejected": -155.5625,
52
+ "loss": 3.0054,
53
+ "nll_loss": 2.310546875,
54
+ "rewards/accuracies": 0.2578125,
55
+ "rewards/chosen": 0.001953125,
56
+ "rewards/margins": -0.0054779052734375,
57
+ "rewards/rejected": 0.007439613342285156,
58
+ "step": 3
59
+ },
60
+ {
61
+ "epoch": 0.14849187935034802,
62
+ "grad_norm": 35.035152435302734,
63
+ "learning_rate": 4.285714285714285e-08,
64
+ "logits/chosen": 0.4200439453125,
65
+ "logits/rejected": 0.41497802734375,
66
+ "logps/chosen": -172.125,
67
+ "logps/rejected": -158.4375,
68
+ "loss": 2.9678,
69
+ "nll_loss": 2.2705078125,
70
+ "rewards/accuracies": 0.2265625,
71
+ "rewards/chosen": -0.009188652038574219,
72
+ "rewards/margins": -0.010580062866210938,
73
+ "rewards/rejected": 0.0013666152954101562,
74
+ "step": 4
75
+ },
76
+ {
77
+ "epoch": 0.18561484918793503,
78
+ "grad_norm": 36.488365173339844,
79
+ "learning_rate": 5.714285714285714e-08,
80
+ "logits/chosen": 0.3895263671875,
81
+ "logits/rejected": 0.38287353515625,
82
+ "logps/chosen": -175.125,
83
+ "logps/rejected": -152.21875,
84
+ "loss": 2.9839,
85
+ "nll_loss": 2.2900390625,
86
+ "rewards/accuracies": 0.234375,
87
+ "rewards/chosen": -0.0039196014404296875,
88
+ "rewards/margins": -0.007814407348632812,
89
+ "rewards/rejected": 0.0039081573486328125,
90
+ "step": 5
91
+ },
92
+ {
93
+ "epoch": 0.22273781902552203,
94
+ "grad_norm": 35.085018157958984,
95
+ "learning_rate": 7.142857142857142e-08,
96
+ "logits/chosen": 0.417724609375,
97
+ "logits/rejected": 0.39959716796875,
98
+ "logps/chosen": -168.4375,
99
+ "logps/rejected": -156.59375,
100
+ "loss": 2.9844,
101
+ "nll_loss": 2.2919921875,
102
+ "rewards/accuracies": 0.28125,
103
+ "rewards/chosen": 0.002349853515625,
104
+ "rewards/margins": 0.0003871917724609375,
105
+ "rewards/rejected": 0.0019512176513671875,
106
+ "step": 6
107
+ },
108
+ {
109
+ "epoch": 0.25986078886310904,
110
+ "grad_norm": 38.6133918762207,
111
+ "learning_rate": 8.57142857142857e-08,
112
+ "logits/chosen": 0.43389892578125,
113
+ "logits/rejected": 0.4083251953125,
114
+ "logps/chosen": -172.0625,
115
+ "logps/rejected": -149.53125,
116
+ "loss": 2.9744,
117
+ "nll_loss": 2.28662109375,
118
+ "rewards/accuracies": 0.296875,
119
+ "rewards/chosen": 0.00312042236328125,
120
+ "rewards/margins": 0.0039272308349609375,
121
+ "rewards/rejected": -0.000789642333984375,
122
+ "step": 7
123
+ },
124
+ {
125
+ "epoch": 0.29698375870069604,
126
+ "grad_norm": 38.21547317504883,
127
+ "learning_rate": 1e-07,
128
+ "logits/chosen": 0.381500244140625,
129
+ "logits/rejected": 0.3905029296875,
130
+ "logps/chosen": -176.875,
131
+ "logps/rejected": -152.5625,
132
+ "loss": 3.0114,
133
+ "nll_loss": 2.31640625,
134
+ "rewards/accuracies": 0.25,
135
+ "rewards/chosen": -0.0007686614990234375,
136
+ "rewards/margins": -0.004500389099121094,
137
+ "rewards/rejected": 0.003711700439453125,
138
+ "step": 8
139
+ },
140
+ {
141
+ "epoch": 0.33410672853828305,
142
+ "grad_norm": 39.607139587402344,
143
+ "learning_rate": 9.921874999999999e-08,
144
+ "logits/chosen": 0.37255859375,
145
+ "logits/rejected": 0.38885498046875,
146
+ "logps/chosen": -181.625,
147
+ "logps/rejected": -158.75,
148
+ "loss": 3.0305,
149
+ "nll_loss": 2.3447265625,
150
+ "rewards/accuracies": 0.328125,
151
+ "rewards/chosen": 0.008989334106445312,
152
+ "rewards/margins": 0.008792877197265625,
153
+ "rewards/rejected": 0.000186920166015625,
154
+ "step": 9
155
+ },
156
+ {
157
+ "epoch": 0.37122969837587005,
158
+ "grad_norm": 35.63096237182617,
159
+ "learning_rate": 9.84375e-08,
160
+ "logits/chosen": 0.4375,
161
+ "logits/rejected": 0.39324951171875,
162
+ "logps/chosen": -170.6875,
163
+ "logps/rejected": -157.5625,
164
+ "loss": 2.9717,
165
+ "nll_loss": 2.2802734375,
166
+ "rewards/accuracies": 0.3046875,
167
+ "rewards/chosen": 0.001560211181640625,
168
+ "rewards/margins": -0.0023441314697265625,
169
+ "rewards/rejected": 0.0039081573486328125,
170
+ "step": 10
171
+ },
172
+ {
173
+ "epoch": 0.40835266821345706,
174
+ "grad_norm": 39.5410041809082,
175
+ "learning_rate": 9.765624999999999e-08,
176
+ "logits/chosen": 0.38726806640625,
177
+ "logits/rejected": 0.34307861328125,
178
+ "logps/chosen": -174.0625,
179
+ "logps/rejected": -150.40625,
180
+ "loss": 2.9924,
181
+ "nll_loss": 2.302734375,
182
+ "rewards/accuracies": 0.3125,
183
+ "rewards/chosen": 0.014478683471679688,
184
+ "rewards/margins": 0.005096435546875,
185
+ "rewards/rejected": 0.0093841552734375,
186
+ "step": 11
187
+ },
188
+ {
189
+ "epoch": 0.44547563805104406,
190
+ "grad_norm": 37.51298522949219,
191
+ "learning_rate": 9.6875e-08,
192
+ "logits/chosen": 0.3577880859375,
193
+ "logits/rejected": 0.359130859375,
194
+ "logps/chosen": -177.625,
195
+ "logps/rejected": -156.8125,
196
+ "loss": 3.0106,
197
+ "nll_loss": 2.3212890625,
198
+ "rewards/accuracies": 0.296875,
199
+ "rewards/chosen": 0.015645980834960938,
200
+ "rewards/margins": 0.004897117614746094,
201
+ "rewards/rejected": 0.010748863220214844,
202
+ "step": 12
203
+ },
204
+ {
205
+ "epoch": 0.48259860788863107,
206
+ "grad_norm": 34.97195053100586,
207
+ "learning_rate": 9.609374999999999e-08,
208
+ "logits/chosen": 0.447021484375,
209
+ "logits/rejected": 0.3858642578125,
210
+ "logps/chosen": -175.625,
211
+ "logps/rejected": -155.25,
212
+ "loss": 2.9915,
213
+ "nll_loss": 2.29833984375,
214
+ "rewards/accuracies": 0.2265625,
215
+ "rewards/chosen": 0.0062618255615234375,
216
+ "rewards/margins": -0.009004592895507812,
217
+ "rewards/rejected": 0.01526641845703125,
218
+ "step": 13
219
+ },
220
+ {
221
+ "epoch": 0.5197215777262181,
222
+ "grad_norm": 39.796241760253906,
223
+ "learning_rate": 9.53125e-08,
224
+ "logits/chosen": 0.43328857421875,
225
+ "logits/rejected": 0.4168701171875,
226
+ "logps/chosen": -168.4375,
227
+ "logps/rejected": -146.90625,
228
+ "loss": 2.9739,
229
+ "nll_loss": 2.27734375,
230
+ "rewards/accuracies": 0.2421875,
231
+ "rewards/chosen": 0.01877593994140625,
232
+ "rewards/margins": -0.006046295166015625,
233
+ "rewards/rejected": 0.02483844757080078,
234
+ "step": 14
235
+ },
236
+ {
237
+ "epoch": 0.5568445475638051,
238
+ "grad_norm": 38.428070068359375,
239
+ "learning_rate": 9.453125e-08,
240
+ "logits/chosen": 0.3983154296875,
241
+ "logits/rejected": 0.394775390625,
242
+ "logps/chosen": -172.0625,
243
+ "logps/rejected": -147.90625,
244
+ "loss": 2.9338,
245
+ "nll_loss": 2.244140625,
246
+ "rewards/accuracies": 0.3046875,
247
+ "rewards/chosen": 0.030483245849609375,
248
+ "rewards/margins": 0.0029296875,
249
+ "rewards/rejected": 0.027555465698242188,
250
+ "step": 15
251
+ },
252
+ {
253
+ "epoch": 0.5939675174013921,
254
+ "grad_norm": 39.628963470458984,
255
+ "learning_rate": 9.375e-08,
256
+ "logits/chosen": 0.39715576171875,
257
+ "logits/rejected": 0.414306640625,
258
+ "logps/chosen": -178.0,
259
+ "logps/rejected": -159.875,
260
+ "loss": 3.0548,
261
+ "nll_loss": 2.3701171875,
262
+ "rewards/accuracies": 0.3828125,
263
+ "rewards/chosen": 0.05942535400390625,
264
+ "rewards/margins": 0.013860702514648438,
265
+ "rewards/rejected": 0.04556083679199219,
266
+ "step": 16
267
+ },
268
+ {
269
+ "epoch": 0.6310904872389791,
270
+ "grad_norm": 35.322757720947266,
271
+ "learning_rate": 9.296875e-08,
272
+ "logits/chosen": 0.38311767578125,
273
+ "logits/rejected": 0.3482666015625,
274
+ "logps/chosen": -173.21875,
275
+ "logps/rejected": -158.6875,
276
+ "loss": 2.9951,
277
+ "nll_loss": 2.3076171875,
278
+ "rewards/accuracies": 0.3828125,
279
+ "rewards/chosen": 0.05277252197265625,
280
+ "rewards/margins": 0.0072422027587890625,
281
+ "rewards/rejected": 0.0455322265625,
282
+ "step": 17
283
+ },
284
+ {
285
+ "epoch": 0.6682134570765661,
286
+ "grad_norm": 34.979583740234375,
287
+ "learning_rate": 9.218749999999999e-08,
288
+ "logits/chosen": 0.427734375,
289
+ "logits/rejected": 0.4029541015625,
290
+ "logps/chosen": -173.9375,
291
+ "logps/rejected": -152.125,
292
+ "loss": 3.002,
293
+ "nll_loss": 2.31689453125,
294
+ "rewards/accuracies": 0.3125,
295
+ "rewards/chosen": 0.0594482421875,
296
+ "rewards/margins": 0.007232666015625,
297
+ "rewards/rejected": 0.05219459533691406,
298
+ "step": 18
299
+ },
300
+ {
301
+ "epoch": 0.7053364269141531,
302
+ "grad_norm": 34.32613754272461,
303
+ "learning_rate": 9.140625e-08,
304
+ "logits/chosen": 0.32867431640625,
305
+ "logits/rejected": 0.3881378173828125,
306
+ "logps/chosen": -172.875,
307
+ "logps/rejected": -163.375,
308
+ "loss": 2.9933,
309
+ "nll_loss": 2.30517578125,
310
+ "rewards/accuracies": 0.3359375,
311
+ "rewards/chosen": 0.07083892822265625,
312
+ "rewards/margins": 0.007622718811035156,
313
+ "rewards/rejected": 0.06317138671875,
314
+ "step": 19
315
+ },
316
+ {
317
+ "epoch": 0.7424593967517401,
318
+ "grad_norm": 41.6298828125,
319
+ "learning_rate": 9.062499999999999e-08,
320
+ "logits/chosen": 0.4229736328125,
321
+ "logits/rejected": 0.375244140625,
322
+ "logps/chosen": -176.0,
323
+ "logps/rejected": -154.9375,
324
+ "loss": 2.9649,
325
+ "nll_loss": 2.279296875,
326
+ "rewards/accuracies": 0.3359375,
327
+ "rewards/chosen": 0.07122039794921875,
328
+ "rewards/margins": 0.010354995727539062,
329
+ "rewards/rejected": 0.0608367919921875,
330
+ "step": 20
331
+ },
332
+ {
333
+ "epoch": 0.7795823665893271,
334
+ "grad_norm": 37.83958053588867,
335
+ "learning_rate": 8.984375e-08,
336
+ "logits/chosen": 0.4124755859375,
337
+ "logits/rejected": 0.36822509765625,
338
+ "logps/chosen": -171.4375,
339
+ "logps/rejected": -147.25,
340
+ "loss": 2.9553,
341
+ "nll_loss": 2.263671875,
342
+ "rewards/accuracies": 0.3359375,
343
+ "rewards/chosen": 0.0637664794921875,
344
+ "rewards/margins": -0.0027446746826171875,
345
+ "rewards/rejected": 0.06647491455078125,
346
+ "step": 21
347
+ },
348
+ {
349
+ "epoch": 0.8167053364269141,
350
+ "grad_norm": 36.95314407348633,
351
+ "learning_rate": 8.90625e-08,
352
+ "logits/chosen": 0.4412841796875,
353
+ "logits/rejected": 0.4178466796875,
354
+ "logps/chosen": -172.3125,
355
+ "logps/rejected": -151.78125,
356
+ "loss": 2.9086,
357
+ "nll_loss": 2.22705078125,
358
+ "rewards/accuracies": 0.3828125,
359
+ "rewards/chosen": 0.083648681640625,
360
+ "rewards/margins": 0.020734786987304688,
361
+ "rewards/rejected": 0.06292724609375,
362
+ "step": 22
363
+ },
364
+ {
365
+ "epoch": 0.8538283062645011,
366
+ "grad_norm": 37.84122085571289,
367
+ "learning_rate": 8.828125e-08,
368
+ "logits/chosen": 0.43475341796875,
369
+ "logits/rejected": 0.39007568359375,
370
+ "logps/chosen": -174.5625,
371
+ "logps/rejected": -146.96875,
372
+ "loss": 2.9311,
373
+ "nll_loss": 2.248046875,
374
+ "rewards/accuracies": 0.4140625,
375
+ "rewards/chosen": 0.0770416259765625,
376
+ "rewards/margins": 0.017015457153320312,
377
+ "rewards/rejected": 0.06001091003417969,
378
+ "step": 23
379
+ },
380
+ {
381
+ "epoch": 0.8909512761020881,
382
+ "grad_norm": 34.304290771484375,
383
+ "learning_rate": 8.75e-08,
384
+ "logits/chosen": 0.455078125,
385
+ "logits/rejected": 0.4635009765625,
386
+ "logps/chosen": -161.5625,
387
+ "logps/rejected": -152.21875,
388
+ "loss": 2.9568,
389
+ "nll_loss": 2.2685546875,
390
+ "rewards/accuracies": 0.3359375,
391
+ "rewards/chosen": 0.06918716430664062,
392
+ "rewards/margins": 0.009189605712890625,
393
+ "rewards/rejected": 0.05999183654785156,
394
+ "step": 24
395
+ },
396
+ {
397
+ "epoch": 0.9280742459396751,
398
+ "grad_norm": 34.58511734008789,
399
+ "learning_rate": 8.671874999999999e-08,
400
+ "logits/chosen": 0.4024658203125,
401
+ "logits/rejected": 0.35870361328125,
402
+ "logps/chosen": -167.84375,
403
+ "logps/rejected": -155.6875,
404
+ "loss": 2.9536,
405
+ "nll_loss": 2.26953125,
406
+ "rewards/accuracies": 0.3359375,
407
+ "rewards/chosen": 0.084014892578125,
408
+ "rewards/margins": 0.009586334228515625,
409
+ "rewards/rejected": 0.0745086669921875,
410
+ "step": 25
411
+ },
412
+ {
413
+ "epoch": 0.9651972157772621,
414
+ "grad_norm": 32.51408386230469,
415
+ "learning_rate": 8.59375e-08,
416
+ "logits/chosen": 0.387939453125,
417
+ "logits/rejected": 0.41217041015625,
418
+ "logps/chosen": -171.625,
419
+ "logps/rejected": -156.21875,
420
+ "loss": 2.9408,
421
+ "nll_loss": 2.2529296875,
422
+ "rewards/accuracies": 0.3359375,
423
+ "rewards/chosen": 0.131378173828125,
424
+ "rewards/margins": 0.0060577392578125,
425
+ "rewards/rejected": 0.1253662109375,
426
+ "step": 26
427
+ },
428
+ {
429
+ "epoch": 1.0,
430
+ "grad_norm": 37.21652603149414,
431
+ "learning_rate": 8.515624999999999e-08,
432
+ "logits/chosen": 0.36829426884651184,
433
+ "logits/rejected": 0.33606770634651184,
434
+ "logps/chosen": -174.1999969482422,
435
+ "logps/rejected": -151.1666717529297,
436
+ "loss": 2.9704,
437
+ "nll_loss": 2.2901041507720947,
438
+ "rewards/accuracies": 0.4166666567325592,
439
+ "rewards/chosen": 0.16803385317325592,
440
+ "rewards/margins": 0.02587076835334301,
441
+ "rewards/rejected": 0.1422526091337204,
442
+ "step": 27
443
+ },
444
+ {
445
+ "epoch": 1.0,
446
+ "eval_logits/chosen": 0.3971354067325592,
447
+ "eval_logits/rejected": 0.3619791567325592,
448
+ "eval_logps/chosen": -169.8333282470703,
449
+ "eval_logps/rejected": -148.8333282470703,
450
+ "eval_loss": 2.9026691913604736,
451
+ "eval_nll_loss": 2.2213542461395264,
452
+ "eval_rewards/accuracies": 0.4010416567325592,
453
+ "eval_rewards/chosen": 0.1822916716337204,
454
+ "eval_rewards/margins": 0.02267964743077755,
455
+ "eval_rewards/rejected": 0.15966796875,
456
+ "eval_runtime": 109.0244,
457
+ "eval_samples_per_second": 3.522,
458
+ "eval_steps_per_second": 0.055,
459
+ "step": 27
460
+ },
461
+ {
462
+ "epoch": 1.037122969837587,
463
+ "grad_norm": 33.881961822509766,
464
+ "learning_rate": 8.4375e-08,
465
+ "logits/chosen": 0.34295654296875,
466
+ "logits/rejected": 0.30322265625,
467
+ "logps/chosen": -169.875,
468
+ "logps/rejected": -156.375,
469
+ "loss": 2.9593,
470
+ "nll_loss": 2.28125,
471
+ "rewards/accuracies": 0.453125,
472
+ "rewards/chosen": 0.19097900390625,
473
+ "rewards/margins": 0.030971527099609375,
474
+ "rewards/rejected": 0.160064697265625,
475
+ "step": 28
476
+ },
477
+ {
478
+ "epoch": 1.074245939675174,
479
+ "grad_norm": 35.8636474609375,
480
+ "learning_rate": 8.359375e-08,
481
+ "logits/chosen": 0.406494140625,
482
+ "logits/rejected": 0.4139404296875,
483
+ "logps/chosen": -173.4375,
484
+ "logps/rejected": -149.90625,
485
+ "loss": 2.9845,
486
+ "nll_loss": 2.3017578125,
487
+ "rewards/accuracies": 0.4296875,
488
+ "rewards/chosen": 0.1990966796875,
489
+ "rewards/margins": 0.017164230346679688,
490
+ "rewards/rejected": 0.1820068359375,
491
+ "step": 29
492
+ },
493
+ {
494
+ "epoch": 1.111368909512761,
495
+ "grad_norm": 36.001094818115234,
496
+ "learning_rate": 8.28125e-08,
497
+ "logits/chosen": 0.43310546875,
498
+ "logits/rejected": 0.40234375,
499
+ "logps/chosen": -169.9375,
500
+ "logps/rejected": -149.40625,
501
+ "loss": 2.8939,
502
+ "nll_loss": 2.21533203125,
503
+ "rewards/accuracies": 0.421875,
504
+ "rewards/chosen": 0.20184326171875,
505
+ "rewards/margins": 0.026065826416015625,
506
+ "rewards/rejected": 0.17584228515625,
507
+ "step": 30
508
+ },
509
+ {
510
+ "epoch": 1.148491879350348,
511
+ "grad_norm": 35.54602813720703,
512
+ "learning_rate": 8.203125e-08,
513
+ "logits/chosen": 0.39208984375,
514
+ "logits/rejected": 0.39398193359375,
515
+ "logps/chosen": -173.4375,
516
+ "logps/rejected": -151.4375,
517
+ "loss": 2.9575,
518
+ "nll_loss": 2.28125,
519
+ "rewards/accuracies": 0.3984375,
520
+ "rewards/chosen": 0.228515625,
521
+ "rewards/margins": 0.028167724609375,
522
+ "rewards/rejected": 0.20025634765625,
523
+ "step": 31
524
+ },
525
+ {
526
+ "epoch": 1.185614849187935,
527
+ "grad_norm": 34.40830993652344,
528
+ "learning_rate": 8.124999999999999e-08,
529
+ "logits/chosen": 0.41754150390625,
530
+ "logits/rejected": 0.380126953125,
531
+ "logps/chosen": -165.4375,
532
+ "logps/rejected": -148.84375,
533
+ "loss": 2.9465,
534
+ "nll_loss": 2.2705078125,
535
+ "rewards/accuracies": 0.5,
536
+ "rewards/chosen": 0.208984375,
537
+ "rewards/margins": 0.03348350524902344,
538
+ "rewards/rejected": 0.17547607421875,
539
+ "step": 32
540
+ },
541
+ {
542
+ "epoch": 1.222737819025522,
543
+ "grad_norm": 36.301544189453125,
544
+ "learning_rate": 8.046875e-08,
545
+ "logits/chosen": 0.4114990234375,
546
+ "logits/rejected": 0.3790283203125,
547
+ "logps/chosen": -173.6875,
548
+ "logps/rejected": -149.3125,
549
+ "loss": 2.9736,
550
+ "nll_loss": 2.291015625,
551
+ "rewards/accuracies": 0.4296875,
552
+ "rewards/chosen": 0.23309326171875,
553
+ "rewards/margins": 0.022678375244140625,
554
+ "rewards/rejected": 0.210205078125,
555
+ "step": 33
556
+ },
557
+ {
558
+ "epoch": 1.259860788863109,
559
+ "grad_norm": 32.567413330078125,
560
+ "learning_rate": 7.968749999999999e-08,
561
+ "logits/chosen": 0.395172119140625,
562
+ "logits/rejected": 0.4110107421875,
563
+ "logps/chosen": -171.75,
564
+ "logps/rejected": -154.0,
565
+ "loss": 2.9779,
566
+ "nll_loss": 2.2890625,
567
+ "rewards/accuracies": 0.390625,
568
+ "rewards/chosen": 0.23516845703125,
569
+ "rewards/margins": 0.01137542724609375,
570
+ "rewards/rejected": 0.223876953125,
571
+ "step": 34
572
+ },
573
+ {
574
+ "epoch": 1.296983758700696,
575
+ "grad_norm": 33.108497619628906,
576
+ "learning_rate": 7.890625e-08,
577
+ "logits/chosen": 0.37713623046875,
578
+ "logits/rejected": 0.37548828125,
579
+ "logps/chosen": -171.125,
580
+ "logps/rejected": -158.125,
581
+ "loss": 2.9559,
582
+ "nll_loss": 2.2744140625,
583
+ "rewards/accuracies": 0.4140625,
584
+ "rewards/chosen": 0.25018310546875,
585
+ "rewards/margins": 0.023487091064453125,
586
+ "rewards/rejected": 0.22705078125,
587
+ "step": 35
588
+ },
589
+ {
590
+ "epoch": 1.334106728538283,
591
+ "grad_norm": 33.0572624206543,
592
+ "learning_rate": 7.812499999999999e-08,
593
+ "logits/chosen": 0.430419921875,
594
+ "logits/rejected": 0.409912109375,
595
+ "logps/chosen": -170.0625,
596
+ "logps/rejected": -156.5,
597
+ "loss": 2.9622,
598
+ "nll_loss": 2.279296875,
599
+ "rewards/accuracies": 0.359375,
600
+ "rewards/chosen": 0.24749755859375,
601
+ "rewards/margins": 0.016834259033203125,
602
+ "rewards/rejected": 0.23046875,
603
+ "step": 36
604
+ },
605
+ {
606
+ "epoch": 1.37122969837587,
607
+ "grad_norm": 34.27482604980469,
608
+ "learning_rate": 7.734375e-08,
609
+ "logits/chosen": 0.3834228515625,
610
+ "logits/rejected": 0.3140869140625,
611
+ "logps/chosen": -174.0625,
612
+ "logps/rejected": -158.4375,
613
+ "loss": 2.9689,
614
+ "nll_loss": 2.296875,
615
+ "rewards/accuracies": 0.5078125,
616
+ "rewards/chosen": 0.25750732421875,
617
+ "rewards/margins": 0.03867340087890625,
618
+ "rewards/rejected": 0.21893310546875,
619
+ "step": 37
620
+ },
621
+ {
622
+ "epoch": 1.408352668213457,
623
+ "grad_norm": 33.35799789428711,
624
+ "learning_rate": 7.65625e-08,
625
+ "logits/chosen": 0.350555419921875,
626
+ "logits/rejected": 0.335540771484375,
627
+ "logps/chosen": -174.3125,
628
+ "logps/rejected": -162.6875,
629
+ "loss": 2.9839,
630
+ "nll_loss": 2.30078125,
631
+ "rewards/accuracies": 0.421875,
632
+ "rewards/chosen": 0.2733154296875,
633
+ "rewards/margins": 0.0215606689453125,
634
+ "rewards/rejected": 0.251953125,
635
+ "step": 38
636
+ },
637
+ {
638
+ "epoch": 1.445475638051044,
639
+ "grad_norm": 34.38155746459961,
640
+ "learning_rate": 7.578125e-08,
641
+ "logits/chosen": 0.4075927734375,
642
+ "logits/rejected": 0.3931884765625,
643
+ "logps/chosen": -176.25,
644
+ "logps/rejected": -158.875,
645
+ "loss": 2.9568,
646
+ "nll_loss": 2.2841796875,
647
+ "rewards/accuracies": 0.453125,
648
+ "rewards/chosen": 0.279541015625,
649
+ "rewards/margins": 0.04210662841796875,
650
+ "rewards/rejected": 0.23712158203125,
651
+ "step": 39
652
+ },
653
+ {
654
+ "epoch": 1.482598607888631,
655
+ "grad_norm": 36.7269401550293,
656
+ "learning_rate": 7.5e-08,
657
+ "logits/chosen": 0.39630126953125,
658
+ "logits/rejected": 0.34423828125,
659
+ "logps/chosen": -171.1875,
660
+ "logps/rejected": -147.5625,
661
+ "loss": 2.9215,
662
+ "nll_loss": 2.2470703125,
663
+ "rewards/accuracies": 0.5390625,
664
+ "rewards/chosen": 0.271240234375,
665
+ "rewards/margins": 0.03932952880859375,
666
+ "rewards/rejected": 0.23175048828125,
667
+ "step": 40
668
+ },
669
+ {
670
+ "epoch": 1.519721577726218,
671
+ "grad_norm": 36.291969299316406,
672
+ "learning_rate": 7.421874999999999e-08,
673
+ "logits/chosen": 0.399261474609375,
674
+ "logits/rejected": 0.35772705078125,
675
+ "logps/chosen": -167.75,
676
+ "logps/rejected": -142.875,
677
+ "loss": 2.9056,
678
+ "nll_loss": 2.2353515625,
679
+ "rewards/accuracies": 0.4140625,
680
+ "rewards/chosen": 0.287353515625,
681
+ "rewards/margins": 0.04146575927734375,
682
+ "rewards/rejected": 0.2457275390625,
683
+ "step": 41
684
+ },
685
+ {
686
+ "epoch": 1.556844547563805,
687
+ "grad_norm": 33.512359619140625,
688
+ "learning_rate": 7.34375e-08,
689
+ "logits/chosen": 0.4334716796875,
690
+ "logits/rejected": 0.4254150390625,
691
+ "logps/chosen": -170.75,
692
+ "logps/rejected": -156.75,
693
+ "loss": 2.8999,
694
+ "nll_loss": 2.220703125,
695
+ "rewards/accuracies": 0.4453125,
696
+ "rewards/chosen": 0.2911376953125,
697
+ "rewards/margins": 0.02759552001953125,
698
+ "rewards/rejected": 0.263427734375,
699
+ "step": 42
700
+ },
701
+ {
702
+ "epoch": 1.593967517401392,
703
+ "grad_norm": 33.475074768066406,
704
+ "learning_rate": 7.265624999999999e-08,
705
+ "logits/chosen": 0.4449462890625,
706
+ "logits/rejected": 0.39874267578125,
707
+ "logps/chosen": -167.1875,
708
+ "logps/rejected": -146.78125,
709
+ "loss": 2.8644,
710
+ "nll_loss": 2.18994140625,
711
+ "rewards/accuracies": 0.4609375,
712
+ "rewards/chosen": 0.291259765625,
713
+ "rewards/margins": 0.03582000732421875,
714
+ "rewards/rejected": 0.25537109375,
715
+ "step": 43
716
+ },
717
+ {
718
+ "epoch": 1.631090487238979,
719
+ "grad_norm": 35.19614791870117,
720
+ "learning_rate": 7.1875e-08,
721
+ "logits/chosen": 0.385009765625,
722
+ "logits/rejected": 0.39849853515625,
723
+ "logps/chosen": -169.3125,
724
+ "logps/rejected": -144.96875,
725
+ "loss": 2.8976,
726
+ "nll_loss": 2.22900390625,
727
+ "rewards/accuracies": 0.5234375,
728
+ "rewards/chosen": 0.3013916015625,
729
+ "rewards/margins": 0.04638671875,
730
+ "rewards/rejected": 0.2550048828125,
731
+ "step": 44
732
+ },
733
+ {
734
+ "epoch": 1.668213457076566,
735
+ "grad_norm": 34.17641067504883,
736
+ "learning_rate": 7.109375e-08,
737
+ "logits/chosen": 0.388427734375,
738
+ "logits/rejected": 0.35906982421875,
739
+ "logps/chosen": -177.625,
740
+ "logps/rejected": -156.375,
741
+ "loss": 2.9559,
742
+ "nll_loss": 2.2822265625,
743
+ "rewards/accuracies": 0.484375,
744
+ "rewards/chosen": 0.3175048828125,
745
+ "rewards/margins": 0.04502105712890625,
746
+ "rewards/rejected": 0.2720947265625,
747
+ "step": 45
748
+ },
749
+ {
750
+ "epoch": 1.705336426914153,
751
+ "grad_norm": 32.45627975463867,
752
+ "learning_rate": 7.03125e-08,
753
+ "logits/chosen": 0.378936767578125,
754
+ "logits/rejected": 0.37091064453125,
755
+ "logps/chosen": -166.4375,
756
+ "logps/rejected": -153.34375,
757
+ "loss": 2.9161,
758
+ "nll_loss": 2.2353515625,
759
+ "rewards/accuracies": 0.40625,
760
+ "rewards/chosen": 0.3055419921875,
761
+ "rewards/margins": 0.026824951171875,
762
+ "rewards/rejected": 0.27874755859375,
763
+ "step": 46
764
+ },
765
+ {
766
+ "epoch": 1.74245939675174,
767
+ "grad_norm": 32.74002456665039,
768
+ "learning_rate": 6.953125e-08,
769
+ "logits/chosen": 0.39892578125,
770
+ "logits/rejected": 0.36236572265625,
771
+ "logps/chosen": -169.375,
772
+ "logps/rejected": -153.375,
773
+ "loss": 2.8928,
774
+ "nll_loss": 2.21044921875,
775
+ "rewards/accuracies": 0.4609375,
776
+ "rewards/chosen": 0.297607421875,
777
+ "rewards/margins": 0.02545928955078125,
778
+ "rewards/rejected": 0.2724609375,
779
+ "step": 47
780
+ },
781
+ {
782
+ "epoch": 1.7795823665893271,
783
+ "grad_norm": 34.69235610961914,
784
+ "learning_rate": 6.875e-08,
785
+ "logits/chosen": 0.3896484375,
786
+ "logits/rejected": 0.32672119140625,
787
+ "logps/chosen": -167.625,
788
+ "logps/rejected": -154.90625,
789
+ "loss": 2.9146,
790
+ "nll_loss": 2.234375,
791
+ "rewards/accuracies": 0.40625,
792
+ "rewards/chosen": 0.31982421875,
793
+ "rewards/margins": 0.027408599853515625,
794
+ "rewards/rejected": 0.29248046875,
795
+ "step": 48
796
+ },
797
+ {
798
+ "epoch": 1.8167053364269141,
799
+ "grad_norm": 34.722328186035156,
800
+ "learning_rate": 6.796875e-08,
801
+ "logits/chosen": 0.39080810546875,
802
+ "logits/rejected": 0.3468017578125,
803
+ "logps/chosen": -170.5,
804
+ "logps/rejected": -151.96875,
805
+ "loss": 2.9396,
806
+ "nll_loss": 2.25341796875,
807
+ "rewards/accuracies": 0.4453125,
808
+ "rewards/chosen": 0.3145751953125,
809
+ "rewards/margins": 0.01688385009765625,
810
+ "rewards/rejected": 0.2972412109375,
811
+ "step": 49
812
+ },
813
+ {
814
+ "epoch": 1.8538283062645011,
815
+ "grad_norm": 35.193389892578125,
816
+ "learning_rate": 6.718749999999999e-08,
817
+ "logits/chosen": 0.3734130859375,
818
+ "logits/rejected": 0.354248046875,
819
+ "logps/chosen": -169.4375,
820
+ "logps/rejected": -148.875,
821
+ "loss": 2.9095,
822
+ "nll_loss": 2.2333984375,
823
+ "rewards/accuracies": 0.4296875,
824
+ "rewards/chosen": 0.341552734375,
825
+ "rewards/margins": 0.03308868408203125,
826
+ "rewards/rejected": 0.3084716796875,
827
+ "step": 50
828
+ },
829
+ {
830
+ "epoch": 1.8909512761020881,
831
+ "grad_norm": 33.2057991027832,
832
+ "learning_rate": 6.640625e-08,
833
+ "logits/chosen": 0.37799072265625,
834
+ "logits/rejected": 0.3985595703125,
835
+ "logps/chosen": -168.3125,
836
+ "logps/rejected": -147.34375,
837
+ "loss": 2.9147,
838
+ "nll_loss": 2.24365234375,
839
+ "rewards/accuracies": 0.4609375,
840
+ "rewards/chosen": 0.4085693359375,
841
+ "rewards/margins": 0.0473785400390625,
842
+ "rewards/rejected": 0.3614501953125,
843
+ "step": 51
844
+ },
845
+ {
846
+ "epoch": 1.9280742459396751,
847
+ "grad_norm": 32.10712432861328,
848
+ "learning_rate": 6.5625e-08,
849
+ "logits/chosen": 0.31439208984375,
850
+ "logits/rejected": 0.35784912109375,
851
+ "logps/chosen": -166.6875,
852
+ "logps/rejected": -148.40625,
853
+ "loss": 2.8973,
854
+ "nll_loss": 2.22412109375,
855
+ "rewards/accuracies": 0.5,
856
+ "rewards/chosen": 0.4617919921875,
857
+ "rewards/margins": 0.04436492919921875,
858
+ "rewards/rejected": 0.417724609375,
859
+ "step": 52
860
+ },
861
+ {
862
+ "epoch": 1.9651972157772621,
863
+ "grad_norm": 32.033958435058594,
864
+ "learning_rate": 6.484375e-08,
865
+ "logits/chosen": 0.34063720703125,
866
+ "logits/rejected": 0.35675048828125,
867
+ "logps/chosen": -167.9375,
868
+ "logps/rejected": -147.875,
869
+ "loss": 2.9434,
870
+ "nll_loss": 2.2705078125,
871
+ "rewards/accuracies": 0.4609375,
872
+ "rewards/chosen": 0.519775390625,
873
+ "rewards/margins": 0.0435333251953125,
874
+ "rewards/rejected": 0.47607421875,
875
+ "step": 53
876
+ },
877
+ {
878
+ "epoch": 2.0,
879
+ "grad_norm": 34.35704040527344,
880
+ "learning_rate": 6.40625e-08,
881
+ "logits/chosen": 0.3671875,
882
+ "logits/rejected": 0.3578124940395355,
883
+ "logps/chosen": -168.73333740234375,
884
+ "logps/rejected": -142.5,
885
+ "loss": 2.8677,
886
+ "nll_loss": 2.211458444595337,
887
+ "rewards/accuracies": 0.5,
888
+ "rewards/chosen": 0.5489583611488342,
889
+ "rewards/margins": 0.07861328125,
890
+ "rewards/rejected": 0.470703125,
891
+ "step": 54
892
+ },
893
+ {
894
+ "epoch": 2.0,
895
+ "eval_logits/chosen": 0.3727213442325592,
896
+ "eval_logits/rejected": 0.3382161557674408,
897
+ "eval_logps/chosen": -166.0,
898
+ "eval_logps/rejected": -145.1666717529297,
899
+ "eval_loss": 2.8258464336395264,
900
+ "eval_nll_loss": 2.1666667461395264,
901
+ "eval_rewards/accuracies": 0.5338541865348816,
902
+ "eval_rewards/chosen": 0.5930989384651184,
903
+ "eval_rewards/margins": 0.0773213729262352,
904
+ "eval_rewards/rejected": 0.5166015625,
905
+ "eval_runtime": 103.3619,
906
+ "eval_samples_per_second": 3.715,
907
+ "eval_steps_per_second": 0.058,
908
+ "step": 54
909
+ },
910
+ {
911
+ "epoch": 2.0371229698375872,
912
+ "grad_norm": 32.25424575805664,
913
+ "learning_rate": 6.328125e-08,
914
+ "logits/chosen": 0.37457275390625,
915
+ "logits/rejected": 0.35791015625,
916
+ "logps/chosen": -168.6875,
917
+ "logps/rejected": -154.3125,
918
+ "loss": 2.8842,
919
+ "nll_loss": 2.220703125,
920
+ "rewards/accuracies": 0.515625,
921
+ "rewards/chosen": 0.600830078125,
922
+ "rewards/margins": 0.0645904541015625,
923
+ "rewards/rejected": 0.5364990234375,
924
+ "step": 55
925
+ },
926
+ {
927
+ "epoch": 2.074245939675174,
928
+ "grad_norm": 30.806289672851562,
929
+ "learning_rate": 6.25e-08,
930
+ "logits/chosen": 0.33087158203125,
931
+ "logits/rejected": 0.30120849609375,
932
+ "logps/chosen": -164.5,
933
+ "logps/rejected": -147.65625,
934
+ "loss": 2.8889,
935
+ "nll_loss": 2.21923828125,
936
+ "rewards/accuracies": 0.453125,
937
+ "rewards/chosen": 0.616455078125,
938
+ "rewards/margins": 0.05657958984375,
939
+ "rewards/rejected": 0.5601806640625,
940
+ "step": 56
941
+ },
942
+ {
943
+ "epoch": 2.111368909512761,
944
+ "grad_norm": 33.29100799560547,
945
+ "learning_rate": 6.171874999999999e-08,
946
+ "logits/chosen": 0.330352783203125,
947
+ "logits/rejected": 0.3253173828125,
948
+ "logps/chosen": -168.125,
949
+ "logps/rejected": -142.21875,
950
+ "loss": 2.8463,
951
+ "nll_loss": 2.19287109375,
952
+ "rewards/accuracies": 0.53125,
953
+ "rewards/chosen": 0.65087890625,
954
+ "rewards/margins": 0.0902557373046875,
955
+ "rewards/rejected": 0.5606689453125,
956
+ "step": 57
957
+ },
958
+ {
959
+ "epoch": 2.148491879350348,
960
+ "grad_norm": 30.142637252807617,
961
+ "learning_rate": 6.09375e-08,
962
+ "logits/chosen": 0.3507080078125,
963
+ "logits/rejected": 0.34368896484375,
964
+ "logps/chosen": -163.625,
965
+ "logps/rejected": -147.8125,
966
+ "loss": 2.8784,
967
+ "nll_loss": 2.20654296875,
968
+ "rewards/accuracies": 0.4765625,
969
+ "rewards/chosen": 0.662841796875,
970
+ "rewards/margins": 0.0572052001953125,
971
+ "rewards/rejected": 0.60546875,
972
+ "step": 58
973
+ },
974
+ {
975
+ "epoch": 2.1856148491879352,
976
+ "grad_norm": 31.264293670654297,
977
+ "learning_rate": 6.015624999999999e-08,
978
+ "logits/chosen": 0.3809814453125,
979
+ "logits/rejected": 0.39666748046875,
980
+ "logps/chosen": -167.9375,
981
+ "logps/rejected": -149.28125,
982
+ "loss": 2.8646,
983
+ "nll_loss": 2.2001953125,
984
+ "rewards/accuracies": 0.5234375,
985
+ "rewards/chosen": 0.68896484375,
986
+ "rewards/margins": 0.0715789794921875,
987
+ "rewards/rejected": 0.6170654296875,
988
+ "step": 59
989
+ },
990
+ {
991
+ "epoch": 2.222737819025522,
992
+ "grad_norm": 28.420000076293945,
993
+ "learning_rate": 5.9375e-08,
994
+ "logits/chosen": 0.3916015625,
995
+ "logits/rejected": 0.34088134765625,
996
+ "logps/chosen": -161.5,
997
+ "logps/rejected": -152.5625,
998
+ "loss": 2.8021,
999
+ "nll_loss": 2.13037109375,
1000
+ "rewards/accuracies": 0.453125,
1001
+ "rewards/chosen": 0.707275390625,
1002
+ "rewards/margins": 0.0574188232421875,
1003
+ "rewards/rejected": 0.64990234375,
1004
+ "step": 60
1005
+ },
1006
+ {
1007
+ "epoch": 2.259860788863109,
1008
+ "grad_norm": 31.354408264160156,
1009
+ "learning_rate": 5.8593749999999995e-08,
1010
+ "logits/chosen": 0.373992919921875,
1011
+ "logits/rejected": 0.327880859375,
1012
+ "logps/chosen": -166.3125,
1013
+ "logps/rejected": -153.5625,
1014
+ "loss": 2.9203,
1015
+ "nll_loss": 2.2626953125,
1016
+ "rewards/accuracies": 0.53125,
1017
+ "rewards/chosen": 0.742431640625,
1018
+ "rewards/margins": 0.082183837890625,
1019
+ "rewards/rejected": 0.660400390625,
1020
+ "step": 61
1021
+ },
1022
+ {
1023
+ "epoch": 2.296983758700696,
1024
+ "grad_norm": 30.681615829467773,
1025
+ "learning_rate": 5.7812499999999996e-08,
1026
+ "logits/chosen": 0.35882568359375,
1027
+ "logits/rejected": 0.33074951171875,
1028
+ "logps/chosen": -162.6875,
1029
+ "logps/rejected": -145.03125,
1030
+ "loss": 2.8244,
1031
+ "nll_loss": 2.16943359375,
1032
+ "rewards/accuracies": 0.5,
1033
+ "rewards/chosen": 0.709228515625,
1034
+ "rewards/margins": 0.0884552001953125,
1035
+ "rewards/rejected": 0.620849609375,
1036
+ "step": 62
1037
+ },
1038
+ {
1039
+ "epoch": 2.3341067285382833,
1040
+ "grad_norm": 28.197145462036133,
1041
+ "learning_rate": 5.7031249999999997e-08,
1042
+ "logits/chosen": 0.3367919921875,
1043
+ "logits/rejected": 0.3353271484375,
1044
+ "logps/chosen": -166.125,
1045
+ "logps/rejected": -157.625,
1046
+ "loss": 2.8934,
1047
+ "nll_loss": 2.21533203125,
1048
+ "rewards/accuracies": 0.46875,
1049
+ "rewards/chosen": 0.7568359375,
1050
+ "rewards/margins": 0.0417938232421875,
1051
+ "rewards/rejected": 0.714599609375,
1052
+ "step": 63
1053
+ },
1054
+ {
1055
+ "epoch": 2.37122969837587,
1056
+ "grad_norm": 33.402793884277344,
1057
+ "learning_rate": 5.625e-08,
1058
+ "logits/chosen": 0.35845947265625,
1059
+ "logits/rejected": 0.3208465576171875,
1060
+ "logps/chosen": -168.9375,
1061
+ "logps/rejected": -147.6875,
1062
+ "loss": 2.8792,
1063
+ "nll_loss": 2.22900390625,
1064
+ "rewards/accuracies": 0.5234375,
1065
+ "rewards/chosen": 0.7763671875,
1066
+ "rewards/margins": 0.102294921875,
1067
+ "rewards/rejected": 0.673828125,
1068
+ "step": 64
1069
+ },
1070
+ {
1071
+ "epoch": 2.408352668213457,
1072
+ "grad_norm": 31.952545166015625,
1073
+ "learning_rate": 5.546875e-08,
1074
+ "logits/chosen": 0.32257080078125,
1075
+ "logits/rejected": 0.34716796875,
1076
+ "logps/chosen": -174.25,
1077
+ "logps/rejected": -149.625,
1078
+ "loss": 2.8956,
1079
+ "nll_loss": 2.25,
1080
+ "rewards/accuracies": 0.5625,
1081
+ "rewards/chosen": 0.796875,
1082
+ "rewards/margins": 0.10577392578125,
1083
+ "rewards/rejected": 0.691162109375,
1084
+ "step": 65
1085
+ },
1086
+ {
1087
+ "epoch": 2.445475638051044,
1088
+ "grad_norm": 28.59982681274414,
1089
+ "learning_rate": 5.46875e-08,
1090
+ "logits/chosen": 0.3751220703125,
1091
+ "logits/rejected": 0.35809326171875,
1092
+ "logps/chosen": -161.25,
1093
+ "logps/rejected": -143.8125,
1094
+ "loss": 2.8065,
1095
+ "nll_loss": 2.14111328125,
1096
+ "rewards/accuracies": 0.4765625,
1097
+ "rewards/chosen": 0.77685546875,
1098
+ "rewards/margins": 0.0706024169921875,
1099
+ "rewards/rejected": 0.705810546875,
1100
+ "step": 66
1101
+ },
1102
+ {
1103
+ "epoch": 2.4825986078886313,
1104
+ "grad_norm": 32.72235107421875,
1105
+ "learning_rate": 5.390625e-08,
1106
+ "logits/chosen": 0.37261962890625,
1107
+ "logits/rejected": 0.33990478515625,
1108
+ "logps/chosen": -162.9375,
1109
+ "logps/rejected": -138.4375,
1110
+ "loss": 2.8118,
1111
+ "nll_loss": 2.1708984375,
1112
+ "rewards/accuracies": 0.5234375,
1113
+ "rewards/chosen": 0.792236328125,
1114
+ "rewards/margins": 0.1248931884765625,
1115
+ "rewards/rejected": 0.6671142578125,
1116
+ "step": 67
1117
+ },
1118
+ {
1119
+ "epoch": 2.519721577726218,
1120
+ "grad_norm": 29.726709365844727,
1121
+ "learning_rate": 5.3124999999999994e-08,
1122
+ "logits/chosen": 0.40264892578125,
1123
+ "logits/rejected": 0.37481689453125,
1124
+ "logps/chosen": -164.6875,
1125
+ "logps/rejected": -145.59375,
1126
+ "loss": 2.8351,
1127
+ "nll_loss": 2.173828125,
1128
+ "rewards/accuracies": 0.515625,
1129
+ "rewards/chosen": 0.793701171875,
1130
+ "rewards/margins": 0.0774993896484375,
1131
+ "rewards/rejected": 0.715576171875,
1132
+ "step": 68
1133
+ },
1134
+ {
1135
+ "epoch": 2.556844547563805,
1136
+ "grad_norm": 30.94222068786621,
1137
+ "learning_rate": 5.2343749999999995e-08,
1138
+ "logits/chosen": 0.3359375,
1139
+ "logits/rejected": 0.34417724609375,
1140
+ "logps/chosen": -172.875,
1141
+ "logps/rejected": -156.71875,
1142
+ "loss": 2.8849,
1143
+ "nll_loss": 2.22607421875,
1144
+ "rewards/accuracies": 0.4453125,
1145
+ "rewards/chosen": 0.8564453125,
1146
+ "rewards/margins": 0.0876007080078125,
1147
+ "rewards/rejected": 0.7685546875,
1148
+ "step": 69
1149
+ },
1150
+ {
1151
+ "epoch": 2.593967517401392,
1152
+ "grad_norm": 27.713054656982422,
1153
+ "learning_rate": 5.1562499999999996e-08,
1154
+ "logits/chosen": 0.413330078125,
1155
+ "logits/rejected": 0.35443115234375,
1156
+ "logps/chosen": -167.6875,
1157
+ "logps/rejected": -155.5625,
1158
+ "loss": 2.8779,
1159
+ "nll_loss": 2.2021484375,
1160
+ "rewards/accuracies": 0.4296875,
1161
+ "rewards/chosen": 0.825927734375,
1162
+ "rewards/margins": 0.0465087890625,
1163
+ "rewards/rejected": 0.77978515625,
1164
+ "step": 70
1165
+ },
1166
+ {
1167
+ "epoch": 2.6310904872389793,
1168
+ "grad_norm": 30.74073600769043,
1169
+ "learning_rate": 5.078125e-08,
1170
+ "logits/chosen": 0.3853759765625,
1171
+ "logits/rejected": 0.38409423828125,
1172
+ "logps/chosen": -168.625,
1173
+ "logps/rejected": -143.5625,
1174
+ "loss": 2.8197,
1175
+ "nll_loss": 2.1552734375,
1176
+ "rewards/accuracies": 0.484375,
1177
+ "rewards/chosen": 0.820556640625,
1178
+ "rewards/margins": 0.0794830322265625,
1179
+ "rewards/rejected": 0.74169921875,
1180
+ "step": 71
1181
+ },
1182
+ {
1183
+ "epoch": 2.668213457076566,
1184
+ "grad_norm": 28.98983383178711,
1185
+ "learning_rate": 5e-08,
1186
+ "logits/chosen": 0.3699951171875,
1187
+ "logits/rejected": 0.36859130859375,
1188
+ "logps/chosen": -168.25,
1189
+ "logps/rejected": -150.84375,
1190
+ "loss": 2.8354,
1191
+ "nll_loss": 2.18115234375,
1192
+ "rewards/accuracies": 0.53125,
1193
+ "rewards/chosen": 0.835693359375,
1194
+ "rewards/margins": 0.093719482421875,
1195
+ "rewards/rejected": 0.742431640625,
1196
+ "step": 72
1197
+ },
1198
+ {
1199
+ "epoch": 2.705336426914153,
1200
+ "grad_norm": 29.943038940429688,
1201
+ "learning_rate": 4.921875e-08,
1202
+ "logits/chosen": 0.32293701171875,
1203
+ "logits/rejected": 0.337158203125,
1204
+ "logps/chosen": -167.4375,
1205
+ "logps/rejected": -145.90625,
1206
+ "loss": 2.8192,
1207
+ "nll_loss": 2.16748046875,
1208
+ "rewards/accuracies": 0.5234375,
1209
+ "rewards/chosen": 0.83203125,
1210
+ "rewards/margins": 0.094451904296875,
1211
+ "rewards/rejected": 0.73876953125,
1212
+ "step": 73
1213
+ },
1214
+ {
1215
+ "epoch": 2.74245939675174,
1216
+ "grad_norm": 32.40934753417969,
1217
+ "learning_rate": 4.84375e-08,
1218
+ "logits/chosen": 0.38946533203125,
1219
+ "logits/rejected": 0.38507080078125,
1220
+ "logps/chosen": -165.25,
1221
+ "logps/rejected": -144.0,
1222
+ "loss": 2.8231,
1223
+ "nll_loss": 2.18310546875,
1224
+ "rewards/accuracies": 0.578125,
1225
+ "rewards/chosen": 0.861328125,
1226
+ "rewards/margins": 0.1257476806640625,
1227
+ "rewards/rejected": 0.7353515625,
1228
+ "step": 74
1229
+ },
1230
+ {
1231
+ "epoch": 2.7795823665893273,
1232
+ "grad_norm": 31.82373809814453,
1233
+ "learning_rate": 4.765625e-08,
1234
+ "logits/chosen": 0.41070556640625,
1235
+ "logits/rejected": 0.36328125,
1236
+ "logps/chosen": -164.1875,
1237
+ "logps/rejected": -143.71875,
1238
+ "loss": 2.8237,
1239
+ "nll_loss": 2.181640625,
1240
+ "rewards/accuracies": 0.546875,
1241
+ "rewards/chosen": 0.864990234375,
1242
+ "rewards/margins": 0.121978759765625,
1243
+ "rewards/rejected": 0.742919921875,
1244
+ "step": 75
1245
+ },
1246
+ {
1247
+ "epoch": 2.816705336426914,
1248
+ "grad_norm": 30.668813705444336,
1249
+ "learning_rate": 4.6875e-08,
1250
+ "logits/chosen": 0.30828857421875,
1251
+ "logits/rejected": 0.283935546875,
1252
+ "logps/chosen": -165.0,
1253
+ "logps/rejected": -147.0,
1254
+ "loss": 2.8801,
1255
+ "nll_loss": 2.23095703125,
1256
+ "rewards/accuracies": 0.5703125,
1257
+ "rewards/chosen": 0.84814453125,
1258
+ "rewards/margins": 0.110992431640625,
1259
+ "rewards/rejected": 0.737548828125,
1260
+ "step": 76
1261
+ },
1262
+ {
1263
+ "epoch": 2.853828306264501,
1264
+ "grad_norm": 30.72187614440918,
1265
+ "learning_rate": 4.6093749999999995e-08,
1266
+ "logits/chosen": 0.31927490234375,
1267
+ "logits/rejected": 0.308197021484375,
1268
+ "logps/chosen": -164.75,
1269
+ "logps/rejected": -146.1875,
1270
+ "loss": 2.8469,
1271
+ "nll_loss": 2.203125,
1272
+ "rewards/accuracies": 0.5546875,
1273
+ "rewards/chosen": 0.867919921875,
1274
+ "rewards/margins": 0.1165924072265625,
1275
+ "rewards/rejected": 0.7509765625,
1276
+ "step": 77
1277
+ },
1278
+ {
1279
+ "epoch": 2.890951276102088,
1280
+ "grad_norm": 29.66461944580078,
1281
+ "learning_rate": 4.5312499999999996e-08,
1282
+ "logits/chosen": 0.3563232421875,
1283
+ "logits/rejected": 0.3353271484375,
1284
+ "logps/chosen": -162.4375,
1285
+ "logps/rejected": -141.90625,
1286
+ "loss": 2.8171,
1287
+ "nll_loss": 2.1650390625,
1288
+ "rewards/accuracies": 0.5703125,
1289
+ "rewards/chosen": 0.865966796875,
1290
+ "rewards/margins": 0.0980224609375,
1291
+ "rewards/rejected": 0.76806640625,
1292
+ "step": 78
1293
+ },
1294
+ {
1295
+ "epoch": 2.9280742459396754,
1296
+ "grad_norm": 28.587329864501953,
1297
+ "learning_rate": 4.453125e-08,
1298
+ "logits/chosen": 0.34588623046875,
1299
+ "logits/rejected": 0.31683349609375,
1300
+ "logps/chosen": -159.9375,
1301
+ "logps/rejected": -141.90625,
1302
+ "loss": 2.8468,
1303
+ "nll_loss": 2.18505859375,
1304
+ "rewards/accuracies": 0.5,
1305
+ "rewards/chosen": 0.8525390625,
1306
+ "rewards/margins": 0.0854949951171875,
1307
+ "rewards/rejected": 0.76708984375,
1308
+ "step": 79
1309
+ },
1310
+ {
1311
+ "epoch": 2.965197215777262,
1312
+ "grad_norm": 29.01342010498047,
1313
+ "learning_rate": 4.375e-08,
1314
+ "logits/chosen": 0.4154052734375,
1315
+ "logits/rejected": 0.372314453125,
1316
+ "logps/chosen": -161.375,
1317
+ "logps/rejected": -145.96875,
1318
+ "loss": 2.7547,
1319
+ "nll_loss": 2.10107421875,
1320
+ "rewards/accuracies": 0.5546875,
1321
+ "rewards/chosen": 0.854248046875,
1322
+ "rewards/margins": 0.0937042236328125,
1323
+ "rewards/rejected": 0.7607421875,
1324
+ "step": 80
1325
+ },
1326
+ {
1327
+ "epoch": 3.0,
1328
+ "grad_norm": 30.070417404174805,
1329
+ "learning_rate": 4.296875e-08,
1330
+ "logits/chosen": 0.38977864384651184,
1331
+ "logits/rejected": 0.3753906190395355,
1332
+ "logps/chosen": -162.86666870117188,
1333
+ "logps/rejected": -145.36666870117188,
1334
+ "loss": 2.8077,
1335
+ "nll_loss": 2.152083396911621,
1336
+ "rewards/accuracies": 0.5583333373069763,
1337
+ "rewards/chosen": 0.8833333253860474,
1338
+ "rewards/margins": 0.10224609076976776,
1339
+ "rewards/rejected": 0.7809895873069763,
1340
+ "step": 81
1341
+ },
1342
+ {
1343
+ "epoch": 3.0,
1344
+ "eval_logits/chosen": 0.3587239682674408,
1345
+ "eval_logits/rejected": 0.3251953125,
1346
+ "eval_logps/chosen": -163.1666717529297,
1347
+ "eval_logps/rejected": -142.8333282470703,
1348
+ "eval_loss": 2.7760417461395264,
1349
+ "eval_nll_loss": 2.1302082538604736,
1350
+ "eval_rewards/accuracies": 0.5651041865348816,
1351
+ "eval_rewards/chosen": 0.8802083134651184,
1352
+ "eval_rewards/margins": 0.1203206405043602,
1353
+ "eval_rewards/rejected": 0.7591145634651184,
1354
+ "eval_runtime": 102.88,
1355
+ "eval_samples_per_second": 3.733,
1356
+ "eval_steps_per_second": 0.058,
1357
+ "step": 81
1358
+ },
1359
+ {
1360
+ "epoch": 3.0371229698375872,
1361
+ "grad_norm": 27.77758026123047,
1362
+ "learning_rate": 4.21875e-08,
1363
+ "logits/chosen": 0.3466796875,
1364
+ "logits/rejected": 0.3443603515625,
1365
+ "logps/chosen": -167.625,
1366
+ "logps/rejected": -153.53125,
1367
+ "loss": 2.8614,
1368
+ "nll_loss": 2.1962890625,
1369
+ "rewards/accuracies": 0.53125,
1370
+ "rewards/chosen": 0.88720703125,
1371
+ "rewards/margins": 0.0737762451171875,
1372
+ "rewards/rejected": 0.813720703125,
1373
+ "step": 82
1374
+ },
1375
+ {
1376
+ "epoch": 3.074245939675174,
1377
+ "grad_norm": 30.16431427001953,
1378
+ "learning_rate": 4.140625e-08,
1379
+ "logits/chosen": 0.33489990234375,
1380
+ "logits/rejected": 0.3492431640625,
1381
+ "logps/chosen": -169.6875,
1382
+ "logps/rejected": -151.375,
1383
+ "loss": 2.8557,
1384
+ "nll_loss": 2.205078125,
1385
+ "rewards/accuracies": 0.5703125,
1386
+ "rewards/chosen": 0.91748046875,
1387
+ "rewards/margins": 0.1014862060546875,
1388
+ "rewards/rejected": 0.815673828125,
1389
+ "step": 83
1390
+ },
1391
+ {
1392
+ "epoch": 3.111368909512761,
1393
+ "grad_norm": 28.804847717285156,
1394
+ "learning_rate": 4.0624999999999995e-08,
1395
+ "logits/chosen": 0.350799560546875,
1396
+ "logits/rejected": 0.3480224609375,
1397
+ "logps/chosen": -168.75,
1398
+ "logps/rejected": -152.9375,
1399
+ "loss": 2.8726,
1400
+ "nll_loss": 2.216796875,
1401
+ "rewards/accuracies": 0.484375,
1402
+ "rewards/chosen": 0.9228515625,
1403
+ "rewards/margins": 0.0991668701171875,
1404
+ "rewards/rejected": 0.82373046875,
1405
+ "step": 84
1406
+ },
1407
+ {
1408
+ "epoch": 3.148491879350348,
1409
+ "grad_norm": 27.3331298828125,
1410
+ "learning_rate": 3.9843749999999996e-08,
1411
+ "logits/chosen": 0.4110107421875,
1412
+ "logits/rejected": 0.35845947265625,
1413
+ "logps/chosen": -160.6875,
1414
+ "logps/rejected": -147.5625,
1415
+ "loss": 2.7726,
1416
+ "nll_loss": 2.107421875,
1417
+ "rewards/accuracies": 0.515625,
1418
+ "rewards/chosen": 0.869384765625,
1419
+ "rewards/margins": 0.07293701171875,
1420
+ "rewards/rejected": 0.796630859375,
1421
+ "step": 85
1422
+ },
1423
+ {
1424
+ "epoch": 3.1856148491879352,
1425
+ "grad_norm": 29.69388198852539,
1426
+ "learning_rate": 3.9062499999999997e-08,
1427
+ "logits/chosen": 0.39044189453125,
1428
+ "logits/rejected": 0.318359375,
1429
+ "logps/chosen": -166.1875,
1430
+ "logps/rejected": -146.90625,
1431
+ "loss": 2.8218,
1432
+ "nll_loss": 2.166015625,
1433
+ "rewards/accuracies": 0.546875,
1434
+ "rewards/chosen": 0.915283203125,
1435
+ "rewards/margins": 0.0997161865234375,
1436
+ "rewards/rejected": 0.815185546875,
1437
+ "step": 86
1438
+ },
1439
+ {
1440
+ "epoch": 3.222737819025522,
1441
+ "grad_norm": 28.804176330566406,
1442
+ "learning_rate": 3.828125e-08,
1443
+ "logits/chosen": 0.36474609375,
1444
+ "logits/rejected": 0.36553955078125,
1445
+ "logps/chosen": -163.3125,
1446
+ "logps/rejected": -148.8125,
1447
+ "loss": 2.8217,
1448
+ "nll_loss": 2.16162109375,
1449
+ "rewards/accuracies": 0.515625,
1450
+ "rewards/chosen": 0.906982421875,
1451
+ "rewards/margins": 0.08441162109375,
1452
+ "rewards/rejected": 0.822509765625,
1453
+ "step": 87
1454
+ },
1455
+ {
1456
+ "epoch": 3.259860788863109,
1457
+ "grad_norm": 27.526470184326172,
1458
+ "learning_rate": 3.75e-08,
1459
+ "logits/chosen": 0.3748779296875,
1460
+ "logits/rejected": 0.36016845703125,
1461
+ "logps/chosen": -162.0,
1462
+ "logps/rejected": -154.125,
1463
+ "loss": 2.8472,
1464
+ "nll_loss": 2.18017578125,
1465
+ "rewards/accuracies": 0.515625,
1466
+ "rewards/chosen": 0.91943359375,
1467
+ "rewards/margins": 0.069366455078125,
1468
+ "rewards/rejected": 0.849853515625,
1469
+ "step": 88
1470
+ },
1471
+ {
1472
+ "epoch": 3.296983758700696,
1473
+ "grad_norm": 31.725608825683594,
1474
+ "learning_rate": 3.671875e-08,
1475
+ "logits/chosen": 0.325714111328125,
1476
+ "logits/rejected": 0.30804443359375,
1477
+ "logps/chosen": -166.0625,
1478
+ "logps/rejected": -142.1875,
1479
+ "loss": 2.7952,
1480
+ "nll_loss": 2.1650390625,
1481
+ "rewards/accuracies": 0.6484375,
1482
+ "rewards/chosen": 0.95263671875,
1483
+ "rewards/margins": 0.145782470703125,
1484
+ "rewards/rejected": 0.806884765625,
1485
+ "step": 89
1486
+ },
1487
+ {
1488
+ "epoch": 3.3341067285382833,
1489
+ "grad_norm": 24.72932243347168,
1490
+ "learning_rate": 3.59375e-08,
1491
+ "logits/chosen": 0.344940185546875,
1492
+ "logits/rejected": 0.322998046875,
1493
+ "logps/chosen": -153.3125,
1494
+ "logps/rejected": -149.625,
1495
+ "loss": 2.7468,
1496
+ "nll_loss": 2.07080078125,
1497
+ "rewards/accuracies": 0.4609375,
1498
+ "rewards/chosen": 0.89794921875,
1499
+ "rewards/margins": 0.045196533203125,
1500
+ "rewards/rejected": 0.852783203125,
1501
+ "step": 90
1502
+ },
1503
+ {
1504
+ "epoch": 3.37122969837587,
1505
+ "grad_norm": 32.73944854736328,
1506
+ "learning_rate": 3.515625e-08,
1507
+ "logits/chosen": 0.354888916015625,
1508
+ "logits/rejected": 0.333740234375,
1509
+ "logps/chosen": -162.0625,
1510
+ "logps/rejected": -136.5625,
1511
+ "loss": 2.7894,
1512
+ "nll_loss": 2.162109375,
1513
+ "rewards/accuracies": 0.6171875,
1514
+ "rewards/chosen": 0.947998046875,
1515
+ "rewards/margins": 0.153228759765625,
1516
+ "rewards/rejected": 0.794921875,
1517
+ "step": 91
1518
+ },
1519
+ {
1520
+ "epoch": 3.408352668213457,
1521
+ "grad_norm": 28.821645736694336,
1522
+ "learning_rate": 3.4375e-08,
1523
+ "logits/chosen": 0.386474609375,
1524
+ "logits/rejected": 0.38519287109375,
1525
+ "logps/chosen": -160.5,
1526
+ "logps/rejected": -141.28125,
1527
+ "loss": 2.778,
1528
+ "nll_loss": 2.13037109375,
1529
+ "rewards/accuracies": 0.5703125,
1530
+ "rewards/chosen": 0.95947265625,
1531
+ "rewards/margins": 0.1248779296875,
1532
+ "rewards/rejected": 0.8349609375,
1533
+ "step": 92
1534
+ },
1535
+ {
1536
+ "epoch": 3.445475638051044,
1537
+ "grad_norm": 28.84639549255371,
1538
+ "learning_rate": 3.3593749999999996e-08,
1539
+ "logits/chosen": 0.35491943359375,
1540
+ "logits/rejected": 0.34991455078125,
1541
+ "logps/chosen": -162.3125,
1542
+ "logps/rejected": -148.1875,
1543
+ "loss": 2.7994,
1544
+ "nll_loss": 2.14794921875,
1545
+ "rewards/accuracies": 0.5390625,
1546
+ "rewards/chosen": 0.9541015625,
1547
+ "rewards/margins": 0.0999755859375,
1548
+ "rewards/rejected": 0.853759765625,
1549
+ "step": 93
1550
+ },
1551
+ {
1552
+ "epoch": 3.4825986078886313,
1553
+ "grad_norm": 29.14773178100586,
1554
+ "learning_rate": 3.28125e-08,
1555
+ "logits/chosen": 0.34576416015625,
1556
+ "logits/rejected": 0.294586181640625,
1557
+ "logps/chosen": -163.375,
1558
+ "logps/rejected": -143.78125,
1559
+ "loss": 2.8074,
1560
+ "nll_loss": 2.15966796875,
1561
+ "rewards/accuracies": 0.5546875,
1562
+ "rewards/chosen": 0.953857421875,
1563
+ "rewards/margins": 0.11199951171875,
1564
+ "rewards/rejected": 0.841796875,
1565
+ "step": 94
1566
+ },
1567
+ {
1568
+ "epoch": 3.519721577726218,
1569
+ "grad_norm": 28.385051727294922,
1570
+ "learning_rate": 3.203125e-08,
1571
+ "logits/chosen": 0.361328125,
1572
+ "logits/rejected": 0.368896484375,
1573
+ "logps/chosen": -163.875,
1574
+ "logps/rejected": -144.5625,
1575
+ "loss": 2.8059,
1576
+ "nll_loss": 2.14208984375,
1577
+ "rewards/accuracies": 0.484375,
1578
+ "rewards/chosen": 0.9453125,
1579
+ "rewards/margins": 0.0822296142578125,
1580
+ "rewards/rejected": 0.86279296875,
1581
+ "step": 95
1582
+ },
1583
+ {
1584
+ "epoch": 3.556844547563805,
1585
+ "grad_norm": 30.213552474975586,
1586
+ "learning_rate": 3.125e-08,
1587
+ "logits/chosen": 0.39837646484375,
1588
+ "logits/rejected": 0.37939453125,
1589
+ "logps/chosen": -162.9375,
1590
+ "logps/rejected": -140.46875,
1591
+ "loss": 2.7784,
1592
+ "nll_loss": 2.14501953125,
1593
+ "rewards/accuracies": 0.546875,
1594
+ "rewards/chosen": 0.977783203125,
1595
+ "rewards/margins": 0.1444854736328125,
1596
+ "rewards/rejected": 0.833984375,
1597
+ "step": 96
1598
+ },
1599
+ {
1600
+ "epoch": 3.593967517401392,
1601
+ "grad_norm": 32.763328552246094,
1602
+ "learning_rate": 3.046875e-08,
1603
+ "logits/chosen": 0.3245849609375,
1604
+ "logits/rejected": 0.333984375,
1605
+ "logps/chosen": -168.5625,
1606
+ "logps/rejected": -146.40625,
1607
+ "loss": 2.8275,
1608
+ "nll_loss": 2.201171875,
1609
+ "rewards/accuracies": 0.53125,
1610
+ "rewards/chosen": 1.023193359375,
1611
+ "rewards/margins": 0.15716552734375,
1612
+ "rewards/rejected": 0.86669921875,
1613
+ "step": 97
1614
+ },
1615
+ {
1616
+ "epoch": 3.6310904872389793,
1617
+ "grad_norm": 28.763147354125977,
1618
+ "learning_rate": 2.96875e-08,
1619
+ "logits/chosen": 0.4039306640625,
1620
+ "logits/rejected": 0.373779296875,
1621
+ "logps/chosen": -162.625,
1622
+ "logps/rejected": -142.9375,
1623
+ "loss": 2.8257,
1624
+ "nll_loss": 2.17041015625,
1625
+ "rewards/accuracies": 0.5078125,
1626
+ "rewards/chosen": 0.997314453125,
1627
+ "rewards/margins": 0.10540771484375,
1628
+ "rewards/rejected": 0.892333984375,
1629
+ "step": 98
1630
+ },
1631
+ {
1632
+ "epoch": 3.668213457076566,
1633
+ "grad_norm": 30.689043045043945,
1634
+ "learning_rate": 2.8906249999999998e-08,
1635
+ "logits/chosen": 0.359619140625,
1636
+ "logits/rejected": 0.330169677734375,
1637
+ "logps/chosen": -165.625,
1638
+ "logps/rejected": -142.09375,
1639
+ "loss": 2.8357,
1640
+ "nll_loss": 2.1826171875,
1641
+ "rewards/accuracies": 0.546875,
1642
+ "rewards/chosen": 1.013427734375,
1643
+ "rewards/margins": 0.1118927001953125,
1644
+ "rewards/rejected": 0.901611328125,
1645
+ "step": 99
1646
+ },
1647
+ {
1648
+ "epoch": 3.705336426914153,
1649
+ "grad_norm": 26.55453872680664,
1650
+ "learning_rate": 2.8125e-08,
1651
+ "logits/chosen": 0.319427490234375,
1652
+ "logits/rejected": 0.313446044921875,
1653
+ "logps/chosen": -159.625,
1654
+ "logps/rejected": -146.21875,
1655
+ "loss": 2.8075,
1656
+ "nll_loss": 2.14453125,
1657
+ "rewards/accuracies": 0.4921875,
1658
+ "rewards/chosen": 0.961669921875,
1659
+ "rewards/margins": 0.082550048828125,
1660
+ "rewards/rejected": 0.87890625,
1661
+ "step": 100
1662
+ },
1663
+ {
1664
+ "epoch": 3.74245939675174,
1665
+ "grad_norm": 29.098655700683594,
1666
+ "learning_rate": 2.734375e-08,
1667
+ "logits/chosen": 0.3113555908203125,
1668
+ "logits/rejected": 0.312286376953125,
1669
+ "logps/chosen": -169.25,
1670
+ "logps/rejected": -155.03125,
1671
+ "loss": 2.8762,
1672
+ "nll_loss": 2.220703125,
1673
+ "rewards/accuracies": 0.4921875,
1674
+ "rewards/chosen": 1.02880859375,
1675
+ "rewards/margins": 0.09100341796875,
1676
+ "rewards/rejected": 0.9384765625,
1677
+ "step": 101
1678
+ },
1679
+ {
1680
+ "epoch": 3.7795823665893273,
1681
+ "grad_norm": 30.016429901123047,
1682
+ "learning_rate": 2.6562499999999997e-08,
1683
+ "logits/chosen": 0.3350830078125,
1684
+ "logits/rejected": 0.32012939453125,
1685
+ "logps/chosen": -163.75,
1686
+ "logps/rejected": -143.78125,
1687
+ "loss": 2.8272,
1688
+ "nll_loss": 2.1796875,
1689
+ "rewards/accuracies": 0.53125,
1690
+ "rewards/chosen": 1.017578125,
1691
+ "rewards/margins": 0.116119384765625,
1692
+ "rewards/rejected": 0.90087890625,
1693
+ "step": 102
1694
+ },
1695
+ {
1696
+ "epoch": 3.816705336426914,
1697
+ "grad_norm": 28.877973556518555,
1698
+ "learning_rate": 2.5781249999999998e-08,
1699
+ "logits/chosen": 0.34783935546875,
1700
+ "logits/rejected": 0.335540771484375,
1701
+ "logps/chosen": -161.375,
1702
+ "logps/rejected": -144.59375,
1703
+ "loss": 2.8558,
1704
+ "nll_loss": 2.2099609375,
1705
+ "rewards/accuracies": 0.53125,
1706
+ "rewards/chosen": 1.007080078125,
1707
+ "rewards/margins": 0.1160125732421875,
1708
+ "rewards/rejected": 0.890625,
1709
+ "step": 103
1710
+ },
1711
+ {
1712
+ "epoch": 3.853828306264501,
1713
+ "grad_norm": 28.23018455505371,
1714
+ "learning_rate": 2.5e-08,
1715
+ "logits/chosen": 0.37432861328125,
1716
+ "logits/rejected": 0.29962158203125,
1717
+ "logps/chosen": -164.375,
1718
+ "logps/rejected": -148.0,
1719
+ "loss": 2.7842,
1720
+ "nll_loss": 2.13134765625,
1721
+ "rewards/accuracies": 0.53125,
1722
+ "rewards/chosen": 1.01220703125,
1723
+ "rewards/margins": 0.1003265380859375,
1724
+ "rewards/rejected": 0.911376953125,
1725
+ "step": 104
1726
+ },
1727
+ {
1728
+ "epoch": 3.890951276102088,
1729
+ "grad_norm": 30.10898780822754,
1730
+ "learning_rate": 2.421875e-08,
1731
+ "logits/chosen": 0.3602294921875,
1732
+ "logits/rejected": 0.346923828125,
1733
+ "logps/chosen": -163.9375,
1734
+ "logps/rejected": -143.0625,
1735
+ "loss": 2.8019,
1736
+ "nll_loss": 2.1611328125,
1737
+ "rewards/accuracies": 0.6015625,
1738
+ "rewards/chosen": 0.990966796875,
1739
+ "rewards/margins": 0.13226318359375,
1740
+ "rewards/rejected": 0.858154296875,
1741
+ "step": 105
1742
+ },
1743
+ {
1744
+ "epoch": 3.9280742459396754,
1745
+ "grad_norm": 27.816171646118164,
1746
+ "learning_rate": 2.34375e-08,
1747
+ "logits/chosen": 0.35302734375,
1748
+ "logits/rejected": 0.327880859375,
1749
+ "logps/chosen": -163.625,
1750
+ "logps/rejected": -144.46875,
1751
+ "loss": 2.7886,
1752
+ "nll_loss": 2.1376953125,
1753
+ "rewards/accuracies": 0.484375,
1754
+ "rewards/chosen": 0.99365234375,
1755
+ "rewards/margins": 0.1105499267578125,
1756
+ "rewards/rejected": 0.883056640625,
1757
+ "step": 106
1758
+ },
1759
+ {
1760
+ "epoch": 3.965197215777262,
1761
+ "grad_norm": 30.782146453857422,
1762
+ "learning_rate": 2.2656249999999998e-08,
1763
+ "logits/chosen": 0.367279052734375,
1764
+ "logits/rejected": 0.34185791015625,
1765
+ "logps/chosen": -163.5,
1766
+ "logps/rejected": -141.5,
1767
+ "loss": 2.7767,
1768
+ "nll_loss": 2.146484375,
1769
+ "rewards/accuracies": 0.625,
1770
+ "rewards/chosen": 1.034912109375,
1771
+ "rewards/margins": 0.15887451171875,
1772
+ "rewards/rejected": 0.875732421875,
1773
+ "step": 107
1774
+ },
1775
+ {
1776
+ "epoch": 4.0,
1777
+ "grad_norm": 32.44054412841797,
1778
+ "learning_rate": 2.1875e-08,
1779
+ "logits/chosen": 0.37871092557907104,
1780
+ "logits/rejected": 0.34069010615348816,
1781
+ "logps/chosen": -164.93333435058594,
1782
+ "logps/rejected": -138.6666717529297,
1783
+ "loss": 2.8432,
1784
+ "nll_loss": 2.211458444595337,
1785
+ "rewards/accuracies": 0.5833333134651184,
1786
+ "rewards/chosen": 1.03125,
1787
+ "rewards/margins": 0.14685872197151184,
1788
+ "rewards/rejected": 0.8841145634651184,
1789
+ "step": 108
1790
+ },
1791
+ {
1792
+ "epoch": 4.0,
1793
+ "eval_logits/chosen": 0.3561197817325592,
1794
+ "eval_logits/rejected": 0.3212890625,
1795
+ "eval_logps/chosen": -161.8333282470703,
1796
+ "eval_logps/rejected": -141.8333282470703,
1797
+ "eval_loss": 2.7526042461395264,
1798
+ "eval_nll_loss": 2.11328125,
1799
+ "eval_rewards/accuracies": 0.5677083134651184,
1800
+ "eval_rewards/chosen": 1.01171875,
1801
+ "eval_rewards/margins": 0.1364542692899704,
1802
+ "eval_rewards/rejected": 0.8756510615348816,
1803
+ "eval_runtime": 102.8297,
1804
+ "eval_samples_per_second": 3.734,
1805
+ "eval_steps_per_second": 0.058,
1806
+ "step": 108
1807
+ },
1808
+ {
1809
+ "epoch": 4.037122969837587,
1810
+ "grad_norm": 31.294692993164062,
1811
+ "learning_rate": 2.109375e-08,
1812
+ "logits/chosen": 0.33905029296875,
1813
+ "logits/rejected": 0.32745361328125,
1814
+ "logps/chosen": -165.125,
1815
+ "logps/rejected": -144.09375,
1816
+ "loss": 2.8858,
1817
+ "nll_loss": 2.2529296875,
1818
+ "rewards/accuracies": 0.53125,
1819
+ "rewards/chosen": 1.063232421875,
1820
+ "rewards/margins": 0.144775390625,
1821
+ "rewards/rejected": 0.91845703125,
1822
+ "step": 109
1823
+ },
1824
+ {
1825
+ "epoch": 4.0742459396751745,
1826
+ "grad_norm": 29.496854782104492,
1827
+ "learning_rate": 2.0312499999999997e-08,
1828
+ "logits/chosen": 0.4388427734375,
1829
+ "logits/rejected": 0.37957763671875,
1830
+ "logps/chosen": -161.5,
1831
+ "logps/rejected": -143.40625,
1832
+ "loss": 2.8021,
1833
+ "nll_loss": 2.14990234375,
1834
+ "rewards/accuracies": 0.4921875,
1835
+ "rewards/chosen": 0.9970703125,
1836
+ "rewards/margins": 0.10546875,
1837
+ "rewards/rejected": 0.892333984375,
1838
+ "step": 110
1839
+ },
1840
+ {
1841
+ "epoch": 4.111368909512761,
1842
+ "grad_norm": 29.509689331054688,
1843
+ "learning_rate": 1.9531249999999998e-08,
1844
+ "logits/chosen": 0.35552978515625,
1845
+ "logits/rejected": 0.3720703125,
1846
+ "logps/chosen": -168.25,
1847
+ "logps/rejected": -143.53125,
1848
+ "loss": 2.8298,
1849
+ "nll_loss": 2.18017578125,
1850
+ "rewards/accuracies": 0.5234375,
1851
+ "rewards/chosen": 1.0341796875,
1852
+ "rewards/margins": 0.121368408203125,
1853
+ "rewards/rejected": 0.91357421875,
1854
+ "step": 111
1855
+ },
1856
+ {
1857
+ "epoch": 4.148491879350348,
1858
+ "grad_norm": 27.970518112182617,
1859
+ "learning_rate": 1.875e-08,
1860
+ "logits/chosen": 0.35711669921875,
1861
+ "logits/rejected": 0.347930908203125,
1862
+ "logps/chosen": -160.875,
1863
+ "logps/rejected": -148.6875,
1864
+ "loss": 2.8164,
1865
+ "nll_loss": 2.16943359375,
1866
+ "rewards/accuracies": 0.5390625,
1867
+ "rewards/chosen": 1.0263671875,
1868
+ "rewards/margins": 0.12030029296875,
1869
+ "rewards/rejected": 0.90478515625,
1870
+ "step": 112
1871
+ },
1872
+ {
1873
+ "epoch": 4.185614849187935,
1874
+ "grad_norm": 27.921180725097656,
1875
+ "learning_rate": 1.796875e-08,
1876
+ "logits/chosen": 0.3526611328125,
1877
+ "logits/rejected": 0.3419189453125,
1878
+ "logps/chosen": -165.8125,
1879
+ "logps/rejected": -148.375,
1880
+ "loss": 2.8152,
1881
+ "nll_loss": 2.154296875,
1882
+ "rewards/accuracies": 0.4921875,
1883
+ "rewards/chosen": 1.027587890625,
1884
+ "rewards/margins": 0.085540771484375,
1885
+ "rewards/rejected": 0.94287109375,
1886
+ "step": 113
1887
+ },
1888
+ {
1889
+ "epoch": 4.222737819025522,
1890
+ "grad_norm": 28.095144271850586,
1891
+ "learning_rate": 1.71875e-08,
1892
+ "logits/chosen": 0.345458984375,
1893
+ "logits/rejected": 0.31927490234375,
1894
+ "logps/chosen": -158.90625,
1895
+ "logps/rejected": -148.96875,
1896
+ "loss": 2.7949,
1897
+ "nll_loss": 2.140625,
1898
+ "rewards/accuracies": 0.5234375,
1899
+ "rewards/chosen": 1.04345703125,
1900
+ "rewards/margins": 0.0972442626953125,
1901
+ "rewards/rejected": 0.9462890625,
1902
+ "step": 114
1903
+ },
1904
+ {
1905
+ "epoch": 4.259860788863109,
1906
+ "grad_norm": 28.238088607788086,
1907
+ "learning_rate": 1.640625e-08,
1908
+ "logits/chosen": 0.34893798828125,
1909
+ "logits/rejected": 0.307281494140625,
1910
+ "logps/chosen": -157.4375,
1911
+ "logps/rejected": -144.75,
1912
+ "loss": 2.8239,
1913
+ "nll_loss": 2.16357421875,
1914
+ "rewards/accuracies": 0.5,
1915
+ "rewards/chosen": 1.010009765625,
1916
+ "rewards/margins": 0.09295654296875,
1917
+ "rewards/rejected": 0.91650390625,
1918
+ "step": 115
1919
+ },
1920
+ {
1921
+ "epoch": 4.296983758700696,
1922
+ "grad_norm": 27.830520629882812,
1923
+ "learning_rate": 1.5625e-08,
1924
+ "logits/chosen": 0.36151123046875,
1925
+ "logits/rejected": 0.3563232421875,
1926
+ "logps/chosen": -165.1875,
1927
+ "logps/rejected": -145.09375,
1928
+ "loss": 2.8053,
1929
+ "nll_loss": 2.1572265625,
1930
+ "rewards/accuracies": 0.5546875,
1931
+ "rewards/chosen": 1.0478515625,
1932
+ "rewards/margins": 0.11956787109375,
1933
+ "rewards/rejected": 0.927490234375,
1934
+ "step": 116
1935
+ },
1936
+ {
1937
+ "epoch": 4.334106728538283,
1938
+ "grad_norm": 28.707717895507812,
1939
+ "learning_rate": 1.484375e-08,
1940
+ "logits/chosen": 0.39031982421875,
1941
+ "logits/rejected": 0.36114501953125,
1942
+ "logps/chosen": -164.1875,
1943
+ "logps/rejected": -145.84375,
1944
+ "loss": 2.8093,
1945
+ "nll_loss": 2.15478515625,
1946
+ "rewards/accuracies": 0.53125,
1947
+ "rewards/chosen": 1.041259765625,
1948
+ "rewards/margins": 0.098602294921875,
1949
+ "rewards/rejected": 0.943603515625,
1950
+ "step": 117
1951
+ },
1952
+ {
1953
+ "epoch": 4.3712296983758705,
1954
+ "grad_norm": 28.201648712158203,
1955
+ "learning_rate": 1.40625e-08,
1956
+ "logits/chosen": 0.3319091796875,
1957
+ "logits/rejected": 0.301177978515625,
1958
+ "logps/chosen": -166.875,
1959
+ "logps/rejected": -148.9375,
1960
+ "loss": 2.7664,
1961
+ "nll_loss": 2.1240234375,
1962
+ "rewards/accuracies": 0.5390625,
1963
+ "rewards/chosen": 1.051513671875,
1964
+ "rewards/margins": 0.129791259765625,
1965
+ "rewards/rejected": 0.920654296875,
1966
+ "step": 118
1967
+ },
1968
+ {
1969
+ "epoch": 4.408352668213457,
1970
+ "grad_norm": 25.48500633239746,
1971
+ "learning_rate": 1.3281249999999999e-08,
1972
+ "logits/chosen": 0.35797119140625,
1973
+ "logits/rejected": 0.34332275390625,
1974
+ "logps/chosen": -160.9375,
1975
+ "logps/rejected": -151.0625,
1976
+ "loss": 2.7906,
1977
+ "nll_loss": 2.115234375,
1978
+ "rewards/accuracies": 0.4296875,
1979
+ "rewards/chosen": 1.037353515625,
1980
+ "rewards/margins": 0.0632781982421875,
1981
+ "rewards/rejected": 0.97509765625,
1982
+ "step": 119
1983
+ },
1984
+ {
1985
+ "epoch": 4.445475638051044,
1986
+ "grad_norm": 27.510597229003906,
1987
+ "learning_rate": 1.25e-08,
1988
+ "logits/chosen": 0.3927001953125,
1989
+ "logits/rejected": 0.381103515625,
1990
+ "logps/chosen": -160.4375,
1991
+ "logps/rejected": -149.46875,
1992
+ "loss": 2.777,
1993
+ "nll_loss": 2.12109375,
1994
+ "rewards/accuracies": 0.5234375,
1995
+ "rewards/chosen": 1.0439453125,
1996
+ "rewards/margins": 0.103424072265625,
1997
+ "rewards/rejected": 0.941162109375,
1998
+ "step": 120
1999
+ },
2000
+ {
2001
+ "epoch": 4.482598607888631,
2002
+ "grad_norm": 26.467517852783203,
2003
+ "learning_rate": 1.171875e-08,
2004
+ "logits/chosen": 0.3460693359375,
2005
+ "logits/rejected": 0.333404541015625,
2006
+ "logps/chosen": -157.0,
2007
+ "logps/rejected": -148.125,
2008
+ "loss": 2.7761,
2009
+ "nll_loss": 2.1083984375,
2010
+ "rewards/accuracies": 0.5546875,
2011
+ "rewards/chosen": 1.01708984375,
2012
+ "rewards/margins": 0.069854736328125,
2013
+ "rewards/rejected": 0.9462890625,
2014
+ "step": 121
2015
+ },
2016
+ {
2017
+ "epoch": 4.519721577726218,
2018
+ "grad_norm": 30.551589965820312,
2019
+ "learning_rate": 1.09375e-08,
2020
+ "logits/chosen": 0.3670654296875,
2021
+ "logits/rejected": 0.32598876953125,
2022
+ "logps/chosen": -160.25,
2023
+ "logps/rejected": -135.75,
2024
+ "loss": 2.7577,
2025
+ "nll_loss": 2.126953125,
2026
+ "rewards/accuracies": 0.59375,
2027
+ "rewards/chosen": 1.037841796875,
2028
+ "rewards/margins": 0.14959716796875,
2029
+ "rewards/rejected": 0.888916015625,
2030
+ "step": 122
2031
+ },
2032
+ {
2033
+ "epoch": 4.556844547563805,
2034
+ "grad_norm": 30.325624465942383,
2035
+ "learning_rate": 1.0156249999999999e-08,
2036
+ "logits/chosen": 0.35888671875,
2037
+ "logits/rejected": 0.361663818359375,
2038
+ "logps/chosen": -167.6875,
2039
+ "logps/rejected": -140.03125,
2040
+ "loss": 2.8104,
2041
+ "nll_loss": 2.1806640625,
2042
+ "rewards/accuracies": 0.6171875,
2043
+ "rewards/chosen": 1.078125,
2044
+ "rewards/margins": 0.1522216796875,
2045
+ "rewards/rejected": 0.9267578125,
2046
+ "step": 123
2047
+ },
2048
+ {
2049
+ "epoch": 4.593967517401392,
2050
+ "grad_norm": 30.124492645263672,
2051
+ "learning_rate": 9.375e-09,
2052
+ "logits/chosen": 0.302703857421875,
2053
+ "logits/rejected": 0.34002685546875,
2054
+ "logps/chosen": -163.625,
2055
+ "logps/rejected": -140.8125,
2056
+ "loss": 2.8342,
2057
+ "nll_loss": 2.197265625,
2058
+ "rewards/accuracies": 0.5625,
2059
+ "rewards/chosen": 1.07177734375,
2060
+ "rewards/margins": 0.1431884765625,
2061
+ "rewards/rejected": 0.927734375,
2062
+ "step": 124
2063
+ },
2064
+ {
2065
+ "epoch": 4.631090487238979,
2066
+ "grad_norm": 27.595609664916992,
2067
+ "learning_rate": 8.59375e-09,
2068
+ "logits/chosen": 0.38677978515625,
2069
+ "logits/rejected": 0.33587646484375,
2070
+ "logps/chosen": -162.3125,
2071
+ "logps/rejected": -142.875,
2072
+ "loss": 2.7643,
2073
+ "nll_loss": 2.11083984375,
2074
+ "rewards/accuracies": 0.546875,
2075
+ "rewards/chosen": 1.028564453125,
2076
+ "rewards/margins": 0.1085205078125,
2077
+ "rewards/rejected": 0.91943359375,
2078
+ "step": 125
2079
+ },
2080
+ {
2081
+ "epoch": 4.6682134570765665,
2082
+ "grad_norm": 30.859743118286133,
2083
+ "learning_rate": 7.8125e-09,
2084
+ "logits/chosen": 0.295379638671875,
2085
+ "logits/rejected": 0.2750244140625,
2086
+ "logps/chosen": -160.6875,
2087
+ "logps/rejected": -133.21875,
2088
+ "loss": 2.8286,
2089
+ "nll_loss": 2.2001953125,
2090
+ "rewards/accuracies": 0.5859375,
2091
+ "rewards/chosen": 1.045654296875,
2092
+ "rewards/margins": 0.16162109375,
2093
+ "rewards/rejected": 0.88330078125,
2094
+ "step": 126
2095
+ },
2096
+ {
2097
+ "epoch": 4.705336426914153,
2098
+ "grad_norm": 30.610149383544922,
2099
+ "learning_rate": 7.03125e-09,
2100
+ "logits/chosen": 0.35791015625,
2101
+ "logits/rejected": 0.337982177734375,
2102
+ "logps/chosen": -163.125,
2103
+ "logps/rejected": -135.25,
2104
+ "loss": 2.7411,
2105
+ "nll_loss": 2.11767578125,
2106
+ "rewards/accuracies": 0.6015625,
2107
+ "rewards/chosen": 1.070556640625,
2108
+ "rewards/margins": 0.1749114990234375,
2109
+ "rewards/rejected": 0.8955078125,
2110
+ "step": 127
2111
+ },
2112
+ {
2113
+ "epoch": 4.74245939675174,
2114
+ "grad_norm": 26.896799087524414,
2115
+ "learning_rate": 6.25e-09,
2116
+ "logits/chosen": 0.3948974609375,
2117
+ "logits/rejected": 0.37274169921875,
2118
+ "logps/chosen": -163.875,
2119
+ "logps/rejected": -151.1875,
2120
+ "loss": 2.8187,
2121
+ "nll_loss": 2.15869140625,
2122
+ "rewards/accuracies": 0.515625,
2123
+ "rewards/chosen": 1.0595703125,
2124
+ "rewards/margins": 0.098236083984375,
2125
+ "rewards/rejected": 0.961669921875,
2126
+ "step": 128
2127
+ },
2128
+ {
2129
+ "epoch": 4.779582366589327,
2130
+ "grad_norm": 28.43315887451172,
2131
+ "learning_rate": 5.46875e-09,
2132
+ "logits/chosen": 0.34173583984375,
2133
+ "logits/rejected": 0.30682373046875,
2134
+ "logps/chosen": -158.3125,
2135
+ "logps/rejected": -144.125,
2136
+ "loss": 2.7525,
2137
+ "nll_loss": 2.10693359375,
2138
+ "rewards/accuracies": 0.5546875,
2139
+ "rewards/chosen": 1.040283203125,
2140
+ "rewards/margins": 0.1310577392578125,
2141
+ "rewards/rejected": 0.908935546875,
2142
+ "step": 129
2143
+ },
2144
+ {
2145
+ "epoch": 4.816705336426914,
2146
+ "grad_norm": 29.862634658813477,
2147
+ "learning_rate": 4.6875e-09,
2148
+ "logits/chosen": 0.390380859375,
2149
+ "logits/rejected": 0.3612060546875,
2150
+ "logps/chosen": -164.0,
2151
+ "logps/rejected": -146.03125,
2152
+ "loss": 2.7642,
2153
+ "nll_loss": 2.13427734375,
2154
+ "rewards/accuracies": 0.5546875,
2155
+ "rewards/chosen": 1.08056640625,
2156
+ "rewards/margins": 0.1512451171875,
2157
+ "rewards/rejected": 0.93017578125,
2158
+ "step": 130
2159
+ },
2160
+ {
2161
+ "epoch": 4.853828306264501,
2162
+ "grad_norm": 32.517127990722656,
2163
+ "learning_rate": 3.90625e-09,
2164
+ "logits/chosen": 0.3182373046875,
2165
+ "logits/rejected": 0.3302001953125,
2166
+ "logps/chosen": -170.875,
2167
+ "logps/rejected": -147.53125,
2168
+ "loss": 2.8578,
2169
+ "nll_loss": 2.21142578125,
2170
+ "rewards/accuracies": 0.578125,
2171
+ "rewards/chosen": 1.061279296875,
2172
+ "rewards/margins": 0.1163330078125,
2173
+ "rewards/rejected": 0.94384765625,
2174
+ "step": 131
2175
+ },
2176
+ {
2177
+ "epoch": 4.890951276102088,
2178
+ "grad_norm": 27.79583168029785,
2179
+ "learning_rate": 3.125e-09,
2180
+ "logits/chosen": 0.36505126953125,
2181
+ "logits/rejected": 0.35943603515625,
2182
+ "logps/chosen": -158.75,
2183
+ "logps/rejected": -144.125,
2184
+ "loss": 2.7755,
2185
+ "nll_loss": 2.11279296875,
2186
+ "rewards/accuracies": 0.5078125,
2187
+ "rewards/chosen": 1.034423828125,
2188
+ "rewards/margins": 0.0912933349609375,
2189
+ "rewards/rejected": 0.943359375,
2190
+ "step": 132
2191
+ },
2192
+ {
2193
+ "epoch": 4.928074245939675,
2194
+ "grad_norm": 29.39356803894043,
2195
+ "learning_rate": 2.34375e-09,
2196
+ "logits/chosen": 0.33624267578125,
2197
+ "logits/rejected": 0.29620361328125,
2198
+ "logps/chosen": -161.6875,
2199
+ "logps/rejected": -150.09375,
2200
+ "loss": 2.8303,
2201
+ "nll_loss": 2.173828125,
2202
+ "rewards/accuracies": 0.5234375,
2203
+ "rewards/chosen": 1.076904296875,
2204
+ "rewards/margins": 0.098663330078125,
2205
+ "rewards/rejected": 0.978515625,
2206
+ "step": 133
2207
+ },
2208
+ {
2209
+ "epoch": 4.965197215777263,
2210
+ "grad_norm": 27.902801513671875,
2211
+ "learning_rate": 1.5625e-09,
2212
+ "logits/chosen": 0.34344482421875,
2213
+ "logits/rejected": 0.33367919921875,
2214
+ "logps/chosen": -163.125,
2215
+ "logps/rejected": -146.59375,
2216
+ "loss": 2.8206,
2217
+ "nll_loss": 2.16650390625,
2218
+ "rewards/accuracies": 0.5703125,
2219
+ "rewards/chosen": 1.05224609375,
2220
+ "rewards/margins": 0.0961456298828125,
2221
+ "rewards/rejected": 0.95556640625,
2222
+ "step": 134
2223
+ },
2224
+ {
2225
+ "epoch": 5.0,
2226
+ "grad_norm": 30.327899932861328,
2227
+ "learning_rate": 7.8125e-10,
2228
+ "logits/chosen": 0.3444661498069763,
2229
+ "logits/rejected": 0.314453125,
2230
+ "logps/chosen": -171.0,
2231
+ "logps/rejected": -152.13333129882812,
2232
+ "loss": 2.8311,
2233
+ "nll_loss": 2.1864583492279053,
2234
+ "rewards/accuracies": 0.5249999761581421,
2235
+ "rewards/chosen": 1.09375,
2236
+ "rewards/margins": 0.12301432341337204,
2237
+ "rewards/rejected": 0.9703124761581421,
2238
+ "step": 135
2239
+ },
2240
+ {
2241
+ "epoch": 5.0,
2242
+ "eval_logits/chosen": 0.35546875,
2243
+ "eval_logits/rejected": 0.3216145932674408,
2244
+ "eval_logps/chosen": -161.5,
2245
+ "eval_logps/rejected": -141.3333282470703,
2246
+ "eval_loss": 2.751953125,
2247
+ "eval_nll_loss": 2.1106770038604736,
2248
+ "eval_rewards/accuracies": 0.5807291865348816,
2249
+ "eval_rewards/chosen": 1.0455728769302368,
2250
+ "eval_rewards/margins": 0.1366984099149704,
2251
+ "eval_rewards/rejected": 0.908203125,
2252
+ "eval_runtime": 102.921,
2253
+ "eval_samples_per_second": 3.731,
2254
+ "eval_steps_per_second": 0.058,
2255
+ "step": 135
2256
+ }
2257
+ ],
2258
+ "logging_steps": 1,
2259
+ "max_steps": 135,
2260
+ "num_input_tokens_seen": 0,
2261
+ "num_train_epochs": 5,
2262
+ "save_steps": 500,
2263
+ "stateful_callbacks": {
2264
+ "TrainerControl": {
2265
+ "args": {
2266
+ "should_epoch_stop": false,
2267
+ "should_evaluate": false,
2268
+ "should_log": false,
2269
+ "should_save": true,
2270
+ "should_training_stop": true
2271
+ },
2272
+ "attributes": {}
2273
+ }
2274
+ },
2275
+ "total_flos": 0.0,
2276
+ "train_batch_size": 1,
2277
+ "trial_name": null,
2278
+ "trial_params": null
2279
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7d337dc52bcee8c54a321fd740931cfc7e30649a7711d66b13408c4d97155dd
3
+ size 8401
vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
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)