ziansu commited on
Commit
acfb53a
·
verified ·
1 Parent(s): aa8d698

Training in progress, step 1200, checkpoint

Browse files
Files changed (30) hide show
  1. checkpoint-1200/README.md +202 -0
  2. checkpoint-1200/adapter_config.json +34 -0
  3. checkpoint-1200/adapter_model.safetensors +3 -0
  4. checkpoint-1200/added_tokens.json +13 -0
  5. checkpoint-1200/global_step1199/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt +3 -0
  6. checkpoint-1200/global_step1199/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt +3 -0
  7. checkpoint-1200/global_step1199/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt +3 -0
  8. checkpoint-1200/global_step1199/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt +3 -0
  9. checkpoint-1200/global_step1199/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt +3 -0
  10. checkpoint-1200/global_step1199/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt +3 -0
  11. checkpoint-1200/global_step1199/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt +3 -0
  12. checkpoint-1200/global_step1199/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt +3 -0
  13. checkpoint-1200/global_step1199/mp_rank_00_model_states.pt +3 -0
  14. checkpoint-1200/latest +1 -0
  15. checkpoint-1200/rng_state_0.pth +3 -0
  16. checkpoint-1200/rng_state_1.pth +3 -0
  17. checkpoint-1200/rng_state_2.pth +3 -0
  18. checkpoint-1200/rng_state_3.pth +3 -0
  19. checkpoint-1200/rng_state_4.pth +3 -0
  20. checkpoint-1200/rng_state_5.pth +3 -0
  21. checkpoint-1200/rng_state_6.pth +3 -0
  22. checkpoint-1200/rng_state_7.pth +3 -0
  23. checkpoint-1200/scheduler.pt +3 -0
  24. checkpoint-1200/special_tokens_map.json +30 -0
  25. checkpoint-1200/tokenizer.json +0 -0
  26. checkpoint-1200/tokenizer.model +3 -0
  27. checkpoint-1200/tokenizer_config.json +133 -0
  28. checkpoint-1200/trainer_state.json +2217 -0
  29. checkpoint-1200/training_args.bin +3 -0
  30. checkpoint-1200/zero_to_fp32.py +674 -0
checkpoint-1200/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: microsoft/Phi-3-mini-4k-instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.14.0
checkpoint-1200/adapter_config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "microsoft/Phi-3-mini-4k-instruct",
5
+ "bias": "none",
6
+ "eva_config": null,
7
+ "exclude_modules": null,
8
+ "fan_in_fan_out": false,
9
+ "inference_mode": true,
10
+ "init_lora_weights": true,
11
+ "layer_replication": null,
12
+ "layers_pattern": null,
13
+ "layers_to_transform": null,
14
+ "loftq_config": {},
15
+ "lora_alpha": 16,
16
+ "lora_bias": false,
17
+ "lora_dropout": 0.0,
18
+ "megatron_config": null,
19
+ "megatron_core": "megatron.core",
20
+ "modules_to_save": null,
21
+ "peft_type": "LORA",
22
+ "r": 8,
23
+ "rank_pattern": {},
24
+ "revision": null,
25
+ "target_modules": [
26
+ "qkv_proj",
27
+ "gate_up_proj",
28
+ "o_proj",
29
+ "down_proj"
30
+ ],
31
+ "task_type": "CAUSAL_LM",
32
+ "use_dora": false,
33
+ "use_rslora": false
34
+ }
checkpoint-1200/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb1ad4d8146d1c0ffaadf909c904bdea6ec9c72fd04281742417f71e91a03fe4
3
+ size 25200088
checkpoint-1200/added_tokens.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "<|assistant|>": 32001,
3
+ "<|endoftext|>": 32000,
4
+ "<|end|>": 32007,
5
+ "<|placeholder1|>": 32002,
6
+ "<|placeholder2|>": 32003,
7
+ "<|placeholder3|>": 32004,
8
+ "<|placeholder4|>": 32005,
9
+ "<|placeholder5|>": 32008,
10
+ "<|placeholder6|>": 32009,
11
+ "<|system|>": 32006,
12
+ "<|user|>": 32010
13
+ }
checkpoint-1200/global_step1199/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:64bef7b4f1d3c45b691381f9fa4a69648f01bb4becdb2e2498af4ab21f43ebab
3
+ size 18881328
checkpoint-1200/global_step1199/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:273638187de33d64bb5d6bbb24b7b7100c398cf434134391b6cd21cfbe84486f
3
+ size 18881328
checkpoint-1200/global_step1199/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccc0a94a8f3a464cd54bbb59b8bbf1eb366c54b4f9cfab101005e1525f8898de
3
+ size 18881328
checkpoint-1200/global_step1199/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f102bcf842ebbf9bdb54fdd056fc129ea39584f3c52cb9f517d4b33564a57c5
3
+ size 18881392
checkpoint-1200/global_step1199/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:08a10c8f1efcd8c80d38874d979cd24c3b6a0b7b1d2b10ca743a3b23c6169ab3
3
+ size 18881392
checkpoint-1200/global_step1199/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:caf7a5599f544ba8c61ccb8c0d7f15a21a5f71832e4068b931d56658c6f2f751
3
+ size 18881392
checkpoint-1200/global_step1199/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad5882e64b0a65018c2c94c29272d2e1d616f4152a7819633c2572f4c343d41f
3
+ size 18881392
checkpoint-1200/global_step1199/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57160f75acff209f9944e46f1304f610eabc320dc8944e2e5e8fb8a76a8fb097
3
+ size 18881392
checkpoint-1200/global_step1199/mp_rank_00_model_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf3d1fbe07f9f7b4b57efc30e65a154bffc1275d30b91aa5916081abe31b4ebe
3
+ size 25379244
checkpoint-1200/latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step1199
checkpoint-1200/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9dd0b885528e55ec25b01a487faef7810481e858198ac24b76aedb3688770c06
3
+ size 15984
checkpoint-1200/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a451e60f451c0ed06e4b0d619be9f7981c5af29ba16d797996e102e4d1fd7514
3
+ size 15984
checkpoint-1200/rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ff4027794d1d9c71e13291169d32d384e8f6078c931f43db354471cbc57d8639
3
+ size 15984
checkpoint-1200/rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40925f5ac9883b8dfe22197d58f18429503331adeff91ce58e72d56b5094171a
3
+ size 15984
checkpoint-1200/rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0890e080f98bcfb81036d2db959cc45209e8c2f67a0dccde184473488395153
3
+ size 15984
checkpoint-1200/rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d8e4b714cb76d14f84bc59d5d9ba706908caddc95de8f17bfbeb87cbce486cb3
3
+ size 15984
checkpoint-1200/rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f5c095f0a000582673860ef2dcfa50f1ba3d6bf9b31cb0a66349b60d581ecbe3
3
+ size 15984
checkpoint-1200/rng_state_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e625b7623f260d65cb1001beba6e4d0df9ed61b3f496d3e767f280a6b73cde8
3
+ size 15984
checkpoint-1200/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35b4cc7927ca0f1b9b45cd00f72746408c82d953cb952c75e7569243d9fa3f0c
3
+ size 1064
checkpoint-1200/special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "<|end|>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<|endoftext|>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
checkpoint-1200/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-1200/tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
3
+ size 499723
checkpoint-1200/tokenizer_config.json ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_eos_token": false,
4
+ "add_prefix_space": null,
5
+ "added_tokens_decoder": {
6
+ "0": {
7
+ "content": "<unk>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false,
12
+ "special": true
13
+ },
14
+ "1": {
15
+ "content": "<s>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false,
20
+ "special": true
21
+ },
22
+ "2": {
23
+ "content": "</s>",
24
+ "lstrip": false,
25
+ "normalized": false,
26
+ "rstrip": true,
27
+ "single_word": false,
28
+ "special": false
29
+ },
30
+ "32000": {
31
+ "content": "<|endoftext|>",
32
+ "lstrip": false,
33
+ "normalized": false,
34
+ "rstrip": false,
35
+ "single_word": false,
36
+ "special": true
37
+ },
38
+ "32001": {
39
+ "content": "<|assistant|>",
40
+ "lstrip": false,
41
+ "normalized": false,
42
+ "rstrip": true,
43
+ "single_word": false,
44
+ "special": true
45
+ },
46
+ "32002": {
47
+ "content": "<|placeholder1|>",
48
+ "lstrip": false,
49
+ "normalized": false,
50
+ "rstrip": true,
51
+ "single_word": false,
52
+ "special": true
53
+ },
54
+ "32003": {
55
+ "content": "<|placeholder2|>",
56
+ "lstrip": false,
57
+ "normalized": false,
58
+ "rstrip": true,
59
+ "single_word": false,
60
+ "special": true
61
+ },
62
+ "32004": {
63
+ "content": "<|placeholder3|>",
64
+ "lstrip": false,
65
+ "normalized": false,
66
+ "rstrip": true,
67
+ "single_word": false,
68
+ "special": true
69
+ },
70
+ "32005": {
71
+ "content": "<|placeholder4|>",
72
+ "lstrip": false,
73
+ "normalized": false,
74
+ "rstrip": true,
75
+ "single_word": false,
76
+ "special": true
77
+ },
78
+ "32006": {
79
+ "content": "<|system|>",
80
+ "lstrip": false,
81
+ "normalized": false,
82
+ "rstrip": true,
83
+ "single_word": false,
84
+ "special": true
85
+ },
86
+ "32007": {
87
+ "content": "<|end|>",
88
+ "lstrip": false,
89
+ "normalized": false,
90
+ "rstrip": false,
91
+ "single_word": false,
92
+ "special": true
93
+ },
94
+ "32008": {
95
+ "content": "<|placeholder5|>",
96
+ "lstrip": false,
97
+ "normalized": false,
98
+ "rstrip": true,
99
+ "single_word": false,
100
+ "special": true
101
+ },
102
+ "32009": {
103
+ "content": "<|placeholder6|>",
104
+ "lstrip": false,
105
+ "normalized": false,
106
+ "rstrip": true,
107
+ "single_word": false,
108
+ "special": true
109
+ },
110
+ "32010": {
111
+ "content": "<|user|>",
112
+ "lstrip": false,
113
+ "normalized": false,
114
+ "rstrip": true,
115
+ "single_word": false,
116
+ "special": true
117
+ }
118
+ },
119
+ "bos_token": "<s>",
120
+ "chat_template": "{% set system_message = 'You are a helpful AI assistant.' %}{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ '<s>' + '<|system|>\n' + system_message + '<|end|>\n' }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|user|>\n' + content + '<|end|>\n<|assistant|>\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|end|>' + '\n' }}{% endif %}{% endfor %}",
121
+ "clean_up_tokenization_spaces": false,
122
+ "eos_token": "<|end|>",
123
+ "extra_special_tokens": {},
124
+ "legacy": false,
125
+ "model_max_length": 4096,
126
+ "pad_token": "<|endoftext|>",
127
+ "padding_side": "right",
128
+ "sp_model_kwargs": {},
129
+ "split_special_tokens": false,
130
+ "tokenizer_class": "LlamaTokenizer",
131
+ "unk_token": "<unk>",
132
+ "use_default_system_prompt": false
133
+ }
checkpoint-1200/trainer_state.json ADDED
@@ -0,0 +1,2217 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0309411259131929,
5
+ "eval_steps": 50,
6
+ "global_step": 1200,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.008594757198109154,
13
+ "grad_norm": 0.06708361208438873,
14
+ "learning_rate": 4.999451708687114e-06,
15
+ "logits/chosen": 14.524938583374023,
16
+ "logits/rejected": 14.82593822479248,
17
+ "logps/chosen": -0.31433865427970886,
18
+ "logps/rejected": -0.32406437397003174,
19
+ "loss": 0.9442,
20
+ "rewards/accuracies": 0.4124999940395355,
21
+ "rewards/chosen": -0.4715079367160797,
22
+ "rewards/margins": 0.014588532969355583,
23
+ "rewards/rejected": -0.48609647154808044,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.017189514396218308,
28
+ "grad_norm": 0.056814808398485184,
29
+ "learning_rate": 4.997807075247147e-06,
30
+ "logits/chosen": 14.309213638305664,
31
+ "logits/rejected": 14.978128433227539,
32
+ "logps/chosen": -0.31283506751060486,
33
+ "logps/rejected": -0.3911947011947632,
34
+ "loss": 0.928,
35
+ "rewards/accuracies": 0.550000011920929,
36
+ "rewards/chosen": -0.46925264596939087,
37
+ "rewards/margins": 0.1175394207239151,
38
+ "rewards/rejected": -0.5867919921875,
39
+ "step": 20
40
+ },
41
+ {
42
+ "epoch": 0.02578427159432746,
43
+ "grad_norm": 0.061199307441711426,
44
+ "learning_rate": 4.9950668210706795e-06,
45
+ "logits/chosen": 14.68384075164795,
46
+ "logits/rejected": 15.338122367858887,
47
+ "logps/chosen": -0.3007296621799469,
48
+ "logps/rejected": -0.3204456865787506,
49
+ "loss": 0.9439,
50
+ "rewards/accuracies": 0.4375,
51
+ "rewards/chosen": -0.45109447836875916,
52
+ "rewards/margins": 0.029573997482657433,
53
+ "rewards/rejected": -0.48066848516464233,
54
+ "step": 30
55
+ },
56
+ {
57
+ "epoch": 0.034379028792436615,
58
+ "grad_norm": 0.08423774689435959,
59
+ "learning_rate": 4.9912321481237616e-06,
60
+ "logits/chosen": 14.39265251159668,
61
+ "logits/rejected": 15.059102058410645,
62
+ "logps/chosen": -0.28216058015823364,
63
+ "logps/rejected": -0.33495840430259705,
64
+ "loss": 0.9184,
65
+ "rewards/accuracies": 0.4124999940395355,
66
+ "rewards/chosen": -0.42324090003967285,
67
+ "rewards/margins": 0.07919676601886749,
68
+ "rewards/rejected": -0.5024376511573792,
69
+ "step": 40
70
+ },
71
+ {
72
+ "epoch": 0.042973785990545764,
73
+ "grad_norm": 0.06052614375948906,
74
+ "learning_rate": 4.986304738420684e-06,
75
+ "logits/chosen": 14.383735656738281,
76
+ "logits/rejected": 15.029413223266602,
77
+ "logps/chosen": -0.27970507740974426,
78
+ "logps/rejected": -0.33213528990745544,
79
+ "loss": 0.9317,
80
+ "rewards/accuracies": 0.5249999761581421,
81
+ "rewards/chosen": -0.4195576310157776,
82
+ "rewards/margins": 0.07864536345005035,
83
+ "rewards/rejected": -0.49820294976234436,
84
+ "step": 50
85
+ },
86
+ {
87
+ "epoch": 0.042973785990545764,
88
+ "eval_logits/chosen": 14.424538612365723,
89
+ "eval_logits/rejected": 15.006633758544922,
90
+ "eval_logps/chosen": -0.2923925220966339,
91
+ "eval_logps/rejected": -0.3531996011734009,
92
+ "eval_loss": 0.9324354529380798,
93
+ "eval_rewards/accuracies": 0.5052631497383118,
94
+ "eval_rewards/chosen": -0.43858882784843445,
95
+ "eval_rewards/margins": 0.09121060371398926,
96
+ "eval_rewards/rejected": -0.5297994017601013,
97
+ "eval_runtime": 26.3759,
98
+ "eval_samples_per_second": 28.549,
99
+ "eval_steps_per_second": 3.602,
100
+ "step": 50
101
+ },
102
+ {
103
+ "epoch": 0.05156854318865492,
104
+ "grad_norm": 0.06899414211511612,
105
+ "learning_rate": 4.980286753286196e-06,
106
+ "logits/chosen": 14.888933181762695,
107
+ "logits/rejected": 15.33955192565918,
108
+ "logps/chosen": -0.2886829972267151,
109
+ "logps/rejected": -0.34016504883766174,
110
+ "loss": 0.9323,
111
+ "rewards/accuracies": 0.5,
112
+ "rewards/chosen": -0.43302449584007263,
113
+ "rewards/margins": 0.07722309231758118,
114
+ "rewards/rejected": -0.5102475881576538,
115
+ "step": 60
116
+ },
117
+ {
118
+ "epoch": 0.060163300386764075,
119
+ "grad_norm": 0.06679105013608932,
120
+ "learning_rate": 4.973180832407471e-06,
121
+ "logits/chosen": 14.721624374389648,
122
+ "logits/rejected": 15.614666938781738,
123
+ "logps/chosen": -0.29435139894485474,
124
+ "logps/rejected": -0.38699784874916077,
125
+ "loss": 0.9172,
126
+ "rewards/accuracies": 0.48750001192092896,
127
+ "rewards/chosen": -0.4415270686149597,
128
+ "rewards/margins": 0.13896968960762024,
129
+ "rewards/rejected": -0.5804967880249023,
130
+ "step": 70
131
+ },
132
+ {
133
+ "epoch": 0.06875805758487323,
134
+ "grad_norm": 0.07169903814792633,
135
+ "learning_rate": 4.964990092676263e-06,
136
+ "logits/chosen": 13.848808288574219,
137
+ "logits/rejected": 14.609800338745117,
138
+ "logps/chosen": -0.26156893372535706,
139
+ "logps/rejected": -0.33030644059181213,
140
+ "loss": 0.9245,
141
+ "rewards/accuracies": 0.48750001192092896,
142
+ "rewards/chosen": -0.3923533856868744,
143
+ "rewards/margins": 0.10310628265142441,
144
+ "rewards/rejected": -0.495459645986557,
145
+ "step": 80
146
+ },
147
+ {
148
+ "epoch": 0.07735281478298238,
149
+ "grad_norm": 0.06593246012926102,
150
+ "learning_rate": 4.9557181268217225e-06,
151
+ "logits/chosen": 14.603567123413086,
152
+ "logits/rejected": 14.994171142578125,
153
+ "logps/chosen": -0.3191321790218353,
154
+ "logps/rejected": -0.3477073311805725,
155
+ "loss": 0.9359,
156
+ "rewards/accuracies": 0.5,
157
+ "rewards/chosen": -0.4786983132362366,
158
+ "rewards/margins": 0.042862698435783386,
159
+ "rewards/rejected": -0.5215609669685364,
160
+ "step": 90
161
+ },
162
+ {
163
+ "epoch": 0.08594757198109153,
164
+ "grad_norm": 0.0718066617846489,
165
+ "learning_rate": 4.9453690018345144e-06,
166
+ "logits/chosen": 13.928094863891602,
167
+ "logits/rejected": 14.792709350585938,
168
+ "logps/chosen": -0.24115696549415588,
169
+ "logps/rejected": -0.3537539839744568,
170
+ "loss": 0.9066,
171
+ "rewards/accuracies": 0.5874999761581421,
172
+ "rewards/chosen": -0.3617354929447174,
173
+ "rewards/margins": 0.16889554262161255,
174
+ "rewards/rejected": -0.5306310653686523,
175
+ "step": 100
176
+ },
177
+ {
178
+ "epoch": 0.08594757198109153,
179
+ "eval_logits/chosen": 14.40036392211914,
180
+ "eval_logits/rejected": 14.97786808013916,
181
+ "eval_logps/chosen": -0.2777771055698395,
182
+ "eval_logps/rejected": -0.3516874611377716,
183
+ "eval_loss": 0.9236211180686951,
184
+ "eval_rewards/accuracies": 0.5052631497383118,
185
+ "eval_rewards/chosen": -0.4166657328605652,
186
+ "eval_rewards/margins": 0.11086549609899521,
187
+ "eval_rewards/rejected": -0.5275312066078186,
188
+ "eval_runtime": 25.8056,
189
+ "eval_samples_per_second": 29.18,
190
+ "eval_steps_per_second": 3.681,
191
+ "step": 100
192
+ },
193
+ {
194
+ "epoch": 0.09454232917920069,
195
+ "grad_norm": 0.06681054830551147,
196
+ "learning_rate": 4.933947257182901e-06,
197
+ "logits/chosen": 14.76116943359375,
198
+ "logits/rejected": 15.001077651977539,
199
+ "logps/chosen": -0.297056645154953,
200
+ "logps/rejected": -0.3221590518951416,
201
+ "loss": 0.929,
202
+ "rewards/accuracies": 0.44999998807907104,
203
+ "rewards/chosen": -0.4455850124359131,
204
+ "rewards/margins": 0.03765357658267021,
205
+ "rewards/rejected": -0.4832385182380676,
206
+ "step": 110
207
+ },
208
+ {
209
+ "epoch": 0.10313708637730984,
210
+ "grad_norm": 0.10024584829807281,
211
+ "learning_rate": 4.921457902821578e-06,
212
+ "logits/chosen": 14.405306816101074,
213
+ "logits/rejected": 15.084524154663086,
214
+ "logps/chosen": -0.2726767361164093,
215
+ "logps/rejected": -0.3543504774570465,
216
+ "loss": 0.9299,
217
+ "rewards/accuracies": 0.5874999761581421,
218
+ "rewards/chosen": -0.40901508927345276,
219
+ "rewards/margins": 0.12251058965921402,
220
+ "rewards/rejected": -0.531525731086731,
221
+ "step": 120
222
+ },
223
+ {
224
+ "epoch": 0.11173184357541899,
225
+ "grad_norm": 0.08629737794399261,
226
+ "learning_rate": 4.907906416994146e-06,
227
+ "logits/chosen": 14.073992729187012,
228
+ "logits/rejected": 14.882128715515137,
229
+ "logps/chosen": -0.2827032506465912,
230
+ "logps/rejected": -0.369393527507782,
231
+ "loss": 0.9109,
232
+ "rewards/accuracies": 0.5375000238418579,
233
+ "rewards/chosen": -0.42405492067337036,
234
+ "rewards/margins": 0.13003548979759216,
235
+ "rewards/rejected": -0.5540903806686401,
236
+ "step": 130
237
+ },
238
+ {
239
+ "epoch": 0.12032660077352815,
240
+ "grad_norm": 0.07973086833953857,
241
+ "learning_rate": 4.893298743830168e-06,
242
+ "logits/chosen": 13.96656322479248,
243
+ "logits/rejected": 14.639463424682617,
244
+ "logps/chosen": -0.28426361083984375,
245
+ "logps/rejected": -0.3899250030517578,
246
+ "loss": 0.9138,
247
+ "rewards/accuracies": 0.5625,
248
+ "rewards/chosen": -0.4263954162597656,
249
+ "rewards/margins": 0.1584920585155487,
250
+ "rewards/rejected": -0.5848874449729919,
251
+ "step": 140
252
+ },
253
+ {
254
+ "epoch": 0.1289213579716373,
255
+ "grad_norm": 0.08767445385456085,
256
+ "learning_rate": 4.8776412907378845e-06,
257
+ "logits/chosen": 13.705177307128906,
258
+ "logits/rejected": 14.19865608215332,
259
+ "logps/chosen": -0.26735779643058777,
260
+ "logps/rejected": -0.34726911783218384,
261
+ "loss": 0.9157,
262
+ "rewards/accuracies": 0.574999988079071,
263
+ "rewards/chosen": -0.40103667974472046,
264
+ "rewards/margins": 0.1198669821023941,
265
+ "rewards/rejected": -0.5209037065505981,
266
+ "step": 150
267
+ },
268
+ {
269
+ "epoch": 0.1289213579716373,
270
+ "eval_logits/chosen": 13.20260238647461,
271
+ "eval_logits/rejected": 13.959339141845703,
272
+ "eval_logps/chosen": -0.27623170614242554,
273
+ "eval_logps/rejected": -0.3724917769432068,
274
+ "eval_loss": 0.909102737903595,
275
+ "eval_rewards/accuracies": 0.557894766330719,
276
+ "eval_rewards/chosen": -0.4143475592136383,
277
+ "eval_rewards/margins": 0.14439010620117188,
278
+ "eval_rewards/rejected": -0.5587376356124878,
279
+ "eval_runtime": 25.7839,
280
+ "eval_samples_per_second": 29.204,
281
+ "eval_steps_per_second": 3.684,
282
+ "step": 150
283
+ },
284
+ {
285
+ "epoch": 0.13751611516974646,
286
+ "grad_norm": 0.09749539196491241,
287
+ "learning_rate": 4.860940925593703e-06,
288
+ "logits/chosen": 13.301411628723145,
289
+ "logits/rejected": 14.054819107055664,
290
+ "logps/chosen": -0.2808162569999695,
291
+ "logps/rejected": -0.39500662684440613,
292
+ "loss": 0.9,
293
+ "rewards/accuracies": 0.512499988079071,
294
+ "rewards/chosen": -0.4212244153022766,
295
+ "rewards/margins": 0.17128555476665497,
296
+ "rewards/rejected": -0.592509925365448,
297
+ "step": 160
298
+ },
299
+ {
300
+ "epoch": 0.1461108723678556,
301
+ "grad_norm": 0.14965052902698517,
302
+ "learning_rate": 4.84320497372973e-06,
303
+ "logits/chosen": 12.261284828186035,
304
+ "logits/rejected": 13.0617036819458,
305
+ "logps/chosen": -0.29266461730003357,
306
+ "logps/rejected": -0.4265298843383789,
307
+ "loss": 0.896,
308
+ "rewards/accuracies": 0.637499988079071,
309
+ "rewards/chosen": -0.43899694085121155,
310
+ "rewards/margins": 0.20079784095287323,
311
+ "rewards/rejected": -0.6397948265075684,
312
+ "step": 170
313
+ },
314
+ {
315
+ "epoch": 0.15470562956596476,
316
+ "grad_norm": 0.13044072687625885,
317
+ "learning_rate": 4.824441214720629e-06,
318
+ "logits/chosen": 11.509119033813477,
319
+ "logits/rejected": 12.31033706665039,
320
+ "logps/chosen": -0.27384257316589355,
321
+ "logps/rejected": -0.3920982778072357,
322
+ "loss": 0.8911,
323
+ "rewards/accuracies": 0.5625,
324
+ "rewards/chosen": -0.4107638895511627,
325
+ "rewards/margins": 0.17738358676433563,
326
+ "rewards/rejected": -0.5881474018096924,
327
+ "step": 180
328
+ },
329
+ {
330
+ "epoch": 0.1633003867640739,
331
+ "grad_norm": 0.16182811558246613,
332
+ "learning_rate": 4.804657878971252e-06,
333
+ "logits/chosen": 10.68933391571045,
334
+ "logits/rejected": 11.632065773010254,
335
+ "logps/chosen": -0.292975515127182,
336
+ "logps/rejected": -0.42257896065711975,
337
+ "loss": 0.9002,
338
+ "rewards/accuracies": 0.6000000238418579,
339
+ "rewards/chosen": -0.4394632875919342,
340
+ "rewards/margins": 0.19440510869026184,
341
+ "rewards/rejected": -0.633868396282196,
342
+ "step": 190
343
+ },
344
+ {
345
+ "epoch": 0.17189514396218306,
346
+ "grad_norm": 0.181160107254982,
347
+ "learning_rate": 4.783863644106502e-06,
348
+ "logits/chosen": 10.593437194824219,
349
+ "logits/rejected": 11.435877799987793,
350
+ "logps/chosen": -0.32495418190956116,
351
+ "logps/rejected": -0.4480825364589691,
352
+ "loss": 0.8773,
353
+ "rewards/accuracies": 0.574999988079071,
354
+ "rewards/chosen": -0.48743128776550293,
355
+ "rewards/margins": 0.18469250202178955,
356
+ "rewards/rejected": -0.6721237897872925,
357
+ "step": 200
358
+ },
359
+ {
360
+ "epoch": 0.17189514396218306,
361
+ "eval_logits/chosen": 9.299257278442383,
362
+ "eval_logits/rejected": 10.055145263671875,
363
+ "eval_logps/chosen": -0.31059205532073975,
364
+ "eval_logps/rejected": -0.47102925181388855,
365
+ "eval_loss": 0.8721462488174438,
366
+ "eval_rewards/accuracies": 0.6105263233184814,
367
+ "eval_rewards/chosen": -0.4658880829811096,
368
+ "eval_rewards/margins": 0.24065588414669037,
369
+ "eval_rewards/rejected": -0.7065439224243164,
370
+ "eval_runtime": 25.78,
371
+ "eval_samples_per_second": 29.209,
372
+ "eval_steps_per_second": 3.685,
373
+ "step": 200
374
+ },
375
+ {
376
+ "epoch": 0.18048990116029223,
377
+ "grad_norm": 0.24912959337234497,
378
+ "learning_rate": 4.762067631165049e-06,
379
+ "logits/chosen": 8.803088188171387,
380
+ "logits/rejected": 9.326388359069824,
381
+ "logps/chosen": -0.3249451816082001,
382
+ "logps/rejected": -0.44993042945861816,
383
+ "loss": 0.8484,
384
+ "rewards/accuracies": 0.5375000238418579,
385
+ "rewards/chosen": -0.4874177575111389,
386
+ "rewards/margins": 0.18747788667678833,
387
+ "rewards/rejected": -0.6748956441879272,
388
+ "step": 210
389
+ },
390
+ {
391
+ "epoch": 0.18908465835840138,
392
+ "grad_norm": 0.319579541683197,
393
+ "learning_rate": 4.7392794005985324e-06,
394
+ "logits/chosen": 6.257112979888916,
395
+ "logits/rejected": 7.168400764465332,
396
+ "logps/chosen": -0.335318386554718,
397
+ "logps/rejected": -0.5439311265945435,
398
+ "loss": 0.8499,
399
+ "rewards/accuracies": 0.637499988079071,
400
+ "rewards/chosen": -0.5029775500297546,
401
+ "rewards/margins": 0.31291908025741577,
402
+ "rewards/rejected": -0.8158966302871704,
403
+ "step": 220
404
+ },
405
+ {
406
+ "epoch": 0.19767941555651053,
407
+ "grad_norm": 0.31494757533073425,
408
+ "learning_rate": 4.715508948078037e-06,
409
+ "logits/chosen": 5.725883960723877,
410
+ "logits/rejected": 5.9254865646362305,
411
+ "logps/chosen": -0.3735908567905426,
412
+ "logps/rejected": -0.5729750394821167,
413
+ "loss": 0.826,
414
+ "rewards/accuracies": 0.6499999761581421,
415
+ "rewards/chosen": -0.5603862404823303,
416
+ "rewards/margins": 0.2990763187408447,
417
+ "rewards/rejected": -0.859462559223175,
418
+ "step": 230
419
+ },
420
+ {
421
+ "epoch": 0.20627417275461968,
422
+ "grad_norm": 0.46439653635025024,
423
+ "learning_rate": 4.690766700109659e-06,
424
+ "logits/chosen": 5.059751033782959,
425
+ "logits/rejected": 5.128623008728027,
426
+ "logps/chosen": -0.4083784222602844,
427
+ "logps/rejected": -0.6792675852775574,
428
+ "loss": 0.7992,
429
+ "rewards/accuracies": 0.637499988079071,
430
+ "rewards/chosen": -0.612567663192749,
431
+ "rewards/margins": 0.40633392333984375,
432
+ "rewards/rejected": -1.0189014673233032,
433
+ "step": 240
434
+ },
435
+ {
436
+ "epoch": 0.21486892995272883,
437
+ "grad_norm": 0.42406076192855835,
438
+ "learning_rate": 4.665063509461098e-06,
439
+ "logits/chosen": 4.128974437713623,
440
+ "logits/rejected": 4.141166687011719,
441
+ "logps/chosen": -0.4256651997566223,
442
+ "logps/rejected": -0.7279168367385864,
443
+ "loss": 0.7848,
444
+ "rewards/accuracies": 0.625,
445
+ "rewards/chosen": -0.6384977102279663,
446
+ "rewards/margins": 0.4533773958683014,
447
+ "rewards/rejected": -1.0918750762939453,
448
+ "step": 250
449
+ },
450
+ {
451
+ "epoch": 0.21486892995272883,
452
+ "eval_logits/chosen": 3.800307512283325,
453
+ "eval_logits/rejected": 3.1472771167755127,
454
+ "eval_logps/chosen": -0.4563433527946472,
455
+ "eval_logps/rejected": -0.8247694373130798,
456
+ "eval_loss": 0.7728626728057861,
457
+ "eval_rewards/accuracies": 0.6526315808296204,
458
+ "eval_rewards/chosen": -0.6845150589942932,
459
+ "eval_rewards/margins": 0.5526391267776489,
460
+ "eval_rewards/rejected": -1.237154245376587,
461
+ "eval_runtime": 25.7836,
462
+ "eval_samples_per_second": 29.205,
463
+ "eval_steps_per_second": 3.685,
464
+ "step": 250
465
+ },
466
+ {
467
+ "epoch": 0.22346368715083798,
468
+ "grad_norm": 0.4071955680847168,
469
+ "learning_rate": 4.638410650401267e-06,
470
+ "logits/chosen": 3.169527530670166,
471
+ "logits/rejected": 2.603461503982544,
472
+ "logps/chosen": -0.5029922723770142,
473
+ "logps/rejected": -0.9469219446182251,
474
+ "loss": 0.7273,
475
+ "rewards/accuracies": 0.7124999761581421,
476
+ "rewards/chosen": -0.7544883489608765,
477
+ "rewards/margins": 0.6658946871757507,
478
+ "rewards/rejected": -1.4203828573226929,
479
+ "step": 260
480
+ },
481
+ {
482
+ "epoch": 0.23205844434894715,
483
+ "grad_norm": 0.6253886222839355,
484
+ "learning_rate": 4.610819813755038e-06,
485
+ "logits/chosen": 3.8718018531799316,
486
+ "logits/rejected": 2.569753646850586,
487
+ "logps/chosen": -0.4955294132232666,
488
+ "logps/rejected": -0.8811863660812378,
489
+ "loss": 0.7483,
490
+ "rewards/accuracies": 0.6875,
491
+ "rewards/chosen": -0.7432941198348999,
492
+ "rewards/margins": 0.5784854888916016,
493
+ "rewards/rejected": -1.321779489517212,
494
+ "step": 270
495
+ },
496
+ {
497
+ "epoch": 0.2406532015470563,
498
+ "grad_norm": 0.5592113733291626,
499
+ "learning_rate": 4.582303101775249e-06,
500
+ "logits/chosen": 3.4818286895751953,
501
+ "logits/rejected": 2.428328275680542,
502
+ "logps/chosen": -0.5700691342353821,
503
+ "logps/rejected": -1.010145664215088,
504
+ "loss": 0.7165,
505
+ "rewards/accuracies": 0.612500011920929,
506
+ "rewards/chosen": -0.8551036715507507,
507
+ "rewards/margins": 0.6601148843765259,
508
+ "rewards/rejected": -1.5152184963226318,
509
+ "step": 280
510
+ },
511
+ {
512
+ "epoch": 0.24924795874516545,
513
+ "grad_norm": 0.8438608050346375,
514
+ "learning_rate": 4.55287302283426e-06,
515
+ "logits/chosen": 2.5937914848327637,
516
+ "logits/rejected": 1.8570162057876587,
517
+ "logps/chosen": -0.592321515083313,
518
+ "logps/rejected": -1.1775600910186768,
519
+ "loss": 0.6685,
520
+ "rewards/accuracies": 0.6000000238418579,
521
+ "rewards/chosen": -0.8884822130203247,
522
+ "rewards/margins": 0.8778578042984009,
523
+ "rewards/rejected": -1.7663400173187256,
524
+ "step": 290
525
+ },
526
+ {
527
+ "epoch": 0.2578427159432746,
528
+ "grad_norm": 2.9559757709503174,
529
+ "learning_rate": 4.522542485937369e-06,
530
+ "logits/chosen": 3.2419090270996094,
531
+ "logits/rejected": 1.9082870483398438,
532
+ "logps/chosen": -0.6832663416862488,
533
+ "logps/rejected": -1.5631868839263916,
534
+ "loss": 0.6009,
535
+ "rewards/accuracies": 0.625,
536
+ "rewards/chosen": -1.0248994827270508,
537
+ "rewards/margins": 1.3198809623718262,
538
+ "rewards/rejected": -2.344780445098877,
539
+ "step": 300
540
+ },
541
+ {
542
+ "epoch": 0.2578427159432746,
543
+ "eval_logits/chosen": 2.5470504760742188,
544
+ "eval_logits/rejected": 1.492888331413269,
545
+ "eval_logps/chosen": -0.7285813689231873,
546
+ "eval_logps/rejected": -1.8318607807159424,
547
+ "eval_loss": 0.5855891704559326,
548
+ "eval_rewards/accuracies": 0.7052631378173828,
549
+ "eval_rewards/chosen": -1.092872142791748,
550
+ "eval_rewards/margins": 1.6549187898635864,
551
+ "eval_rewards/rejected": -2.747790813446045,
552
+ "eval_runtime": 25.8105,
553
+ "eval_samples_per_second": 29.174,
554
+ "eval_steps_per_second": 3.681,
555
+ "step": 300
556
+ },
557
+ {
558
+ "epoch": 0.2664374731413838,
559
+ "grad_norm": 1.4503060579299927,
560
+ "learning_rate": 4.491324795060491e-06,
561
+ "logits/chosen": 1.6672757863998413,
562
+ "logits/rejected": 0.7888604402542114,
563
+ "logps/chosen": -0.769140899181366,
564
+ "logps/rejected": -2.0822532176971436,
565
+ "loss": 0.512,
566
+ "rewards/accuracies": 0.75,
567
+ "rewards/chosen": -1.1537113189697266,
568
+ "rewards/margins": 1.9696683883666992,
569
+ "rewards/rejected": -3.123379945755005,
570
+ "step": 310
571
+ },
572
+ {
573
+ "epoch": 0.2750322303394929,
574
+ "grad_norm": 0.36741188168525696,
575
+ "learning_rate": 4.4592336433146e-06,
576
+ "logits/chosen": 2.6584715843200684,
577
+ "logits/rejected": 1.835911750793457,
578
+ "logps/chosen": -0.8400143384933472,
579
+ "logps/rejected": -1.9262489080429077,
580
+ "loss": 0.5405,
581
+ "rewards/accuracies": 0.6499999761581421,
582
+ "rewards/chosen": -1.2600215673446655,
583
+ "rewards/margins": 1.6293519735336304,
584
+ "rewards/rejected": -2.889373302459717,
585
+ "step": 320
586
+ },
587
+ {
588
+ "epoch": 0.28362698753760207,
589
+ "grad_norm": 0.6233783960342407,
590
+ "learning_rate": 4.426283106939474e-06,
591
+ "logits/chosen": 3.2203617095947266,
592
+ "logits/rejected": 2.3215420246124268,
593
+ "logps/chosen": -0.7985933423042297,
594
+ "logps/rejected": -2.4170174598693848,
595
+ "loss": 0.5335,
596
+ "rewards/accuracies": 0.637499988079071,
597
+ "rewards/chosen": -1.197890043258667,
598
+ "rewards/margins": 2.427635669708252,
599
+ "rewards/rejected": -3.625525712966919,
600
+ "step": 330
601
+ },
602
+ {
603
+ "epoch": 0.2922217447357112,
604
+ "grad_norm": 1.0881849527359009,
605
+ "learning_rate": 4.3924876391293915e-06,
606
+ "logits/chosen": 2.229017734527588,
607
+ "logits/rejected": 1.2251309156417847,
608
+ "logps/chosen": -0.8058193325996399,
609
+ "logps/rejected": -2.810622215270996,
610
+ "loss": 0.4903,
611
+ "rewards/accuracies": 0.7124999761581421,
612
+ "rewards/chosen": -1.2087291479110718,
613
+ "rewards/margins": 3.007204294204712,
614
+ "rewards/rejected": -4.215933799743652,
615
+ "step": 340
616
+ },
617
+ {
618
+ "epoch": 0.30081650193382037,
619
+ "grad_norm": 4.168415069580078,
620
+ "learning_rate": 4.357862063693486e-06,
621
+ "logits/chosen": 2.4198296070098877,
622
+ "logits/rejected": 1.5391919612884521,
623
+ "logps/chosen": -1.010558843612671,
624
+ "logps/rejected": -2.2362923622131348,
625
+ "loss": 0.5249,
626
+ "rewards/accuracies": 0.5625,
627
+ "rewards/chosen": -1.515838384628296,
628
+ "rewards/margins": 1.8385999202728271,
629
+ "rewards/rejected": -3.3544387817382812,
630
+ "step": 350
631
+ },
632
+ {
633
+ "epoch": 0.30081650193382037,
634
+ "eval_logits/chosen": 2.996535539627075,
635
+ "eval_logits/rejected": 2.064058303833008,
636
+ "eval_logps/chosen": -0.8687878847122192,
637
+ "eval_logps/rejected": -2.9790267944335938,
638
+ "eval_loss": 0.5171241760253906,
639
+ "eval_rewards/accuracies": 0.7263157963752747,
640
+ "eval_rewards/chosen": -1.3031818866729736,
641
+ "eval_rewards/margins": 3.165358781814575,
642
+ "eval_rewards/rejected": -4.468540668487549,
643
+ "eval_runtime": 25.8152,
644
+ "eval_samples_per_second": 29.169,
645
+ "eval_steps_per_second": 3.68,
646
+ "step": 350
647
+ },
648
+ {
649
+ "epoch": 0.3094112591319295,
650
+ "grad_norm": 0.5646592378616333,
651
+ "learning_rate": 4.322421568553529e-06,
652
+ "logits/chosen": 3.050445556640625,
653
+ "logits/rejected": 2.0960793495178223,
654
+ "logps/chosen": -0.7702202796936035,
655
+ "logps/rejected": -2.5967533588409424,
656
+ "loss": 0.5067,
657
+ "rewards/accuracies": 0.675000011920929,
658
+ "rewards/chosen": -1.1553303003311157,
659
+ "rewards/margins": 2.739799737930298,
660
+ "rewards/rejected": -3.895130157470703,
661
+ "step": 360
662
+ },
663
+ {
664
+ "epoch": 0.31800601633003867,
665
+ "grad_norm": 0.5547713041305542,
666
+ "learning_rate": 4.286181699082008e-06,
667
+ "logits/chosen": 2.7148895263671875,
668
+ "logits/rejected": 1.9958852529525757,
669
+ "logps/chosen": -0.9548311233520508,
670
+ "logps/rejected": -3.1348252296447754,
671
+ "loss": 0.4726,
672
+ "rewards/accuracies": 0.800000011920929,
673
+ "rewards/chosen": -1.4322465658187866,
674
+ "rewards/margins": 3.269991397857666,
675
+ "rewards/rejected": -4.702237606048584,
676
+ "step": 370
677
+ },
678
+ {
679
+ "epoch": 0.3266007735281478,
680
+ "grad_norm": 3.4396660327911377,
681
+ "learning_rate": 4.249158351283414e-06,
682
+ "logits/chosen": 2.586766004562378,
683
+ "logits/rejected": 2.070089340209961,
684
+ "logps/chosen": -0.9903923273086548,
685
+ "logps/rejected": -3.0135743618011475,
686
+ "loss": 0.4801,
687
+ "rewards/accuracies": 0.6625000238418579,
688
+ "rewards/chosen": -1.485588550567627,
689
+ "rewards/margins": 3.034773349761963,
690
+ "rewards/rejected": -4.52036190032959,
691
+ "step": 380
692
+ },
693
+ {
694
+ "epoch": 0.33519553072625696,
695
+ "grad_norm": 0.9405317306518555,
696
+ "learning_rate": 4.211367764821722e-06,
697
+ "logits/chosen": 4.370789527893066,
698
+ "logits/rejected": 3.165931224822998,
699
+ "logps/chosen": -0.7785463929176331,
700
+ "logps/rejected": -2.456723928451538,
701
+ "loss": 0.4585,
702
+ "rewards/accuracies": 0.5874999761581421,
703
+ "rewards/chosen": -1.1678194999694824,
704
+ "rewards/margins": 2.5172665119171143,
705
+ "rewards/rejected": -3.6850857734680176,
706
+ "step": 390
707
+ },
708
+ {
709
+ "epoch": 0.3437902879243661,
710
+ "grad_norm": 0.7120731472969055,
711
+ "learning_rate": 4.172826515897146e-06,
712
+ "logits/chosen": 3.3425400257110596,
713
+ "logits/rejected": 2.6448545455932617,
714
+ "logps/chosen": -0.9174768328666687,
715
+ "logps/rejected": -3.047037124633789,
716
+ "loss": 0.4771,
717
+ "rewards/accuracies": 0.675000011920929,
718
+ "rewards/chosen": -1.3762153387069702,
719
+ "rewards/margins": 3.194340229034424,
720
+ "rewards/rejected": -4.570555686950684,
721
+ "step": 400
722
+ },
723
+ {
724
+ "epoch": 0.3437902879243661,
725
+ "eval_logits/chosen": 3.548964262008667,
726
+ "eval_logits/rejected": 2.7726428508758545,
727
+ "eval_logps/chosen": -1.0053316354751587,
728
+ "eval_logps/rejected": -3.487654447555542,
729
+ "eval_loss": 0.47841358184814453,
730
+ "eval_rewards/accuracies": 0.7368420958518982,
731
+ "eval_rewards/chosen": -1.5079973936080933,
732
+ "eval_rewards/margins": 3.723484992980957,
733
+ "eval_rewards/rejected": -5.231482028961182,
734
+ "eval_runtime": 25.8148,
735
+ "eval_samples_per_second": 29.169,
736
+ "eval_steps_per_second": 3.68,
737
+ "step": 400
738
+ },
739
+ {
740
+ "epoch": 0.3523850451224753,
741
+ "grad_norm": 2.403956651687622,
742
+ "learning_rate": 4.133551509975264e-06,
743
+ "logits/chosen": 3.2028489112854004,
744
+ "logits/rejected": 2.2486982345581055,
745
+ "logps/chosen": -0.9957242012023926,
746
+ "logps/rejected": -3.243959426879883,
747
+ "loss": 0.4449,
748
+ "rewards/accuracies": 0.637499988079071,
749
+ "rewards/chosen": -1.4935863018035889,
750
+ "rewards/margins": 3.3723526000976562,
751
+ "rewards/rejected": -4.865939140319824,
752
+ "step": 410
753
+ },
754
+ {
755
+ "epoch": 0.36097980232058446,
756
+ "grad_norm": 0.39530256390571594,
757
+ "learning_rate": 4.093559974371725e-06,
758
+ "logits/chosen": 3.8590214252471924,
759
+ "logits/rejected": 3.1420931816101074,
760
+ "logps/chosen": -0.9541120529174805,
761
+ "logps/rejected": -3.0112829208374023,
762
+ "loss": 0.4598,
763
+ "rewards/accuracies": 0.699999988079071,
764
+ "rewards/chosen": -1.4311680793762207,
765
+ "rewards/margins": 3.085756301879883,
766
+ "rewards/rejected": -4.5169243812561035,
767
+ "step": 420
768
+ },
769
+ {
770
+ "epoch": 0.3695745595186936,
771
+ "grad_norm": 0.29451707005500793,
772
+ "learning_rate": 4.052869450695776e-06,
773
+ "logits/chosen": 4.697268486022949,
774
+ "logits/rejected": 3.7647697925567627,
775
+ "logps/chosen": -1.1037578582763672,
776
+ "logps/rejected": -3.8626160621643066,
777
+ "loss": 0.4275,
778
+ "rewards/accuracies": 0.75,
779
+ "rewards/chosen": -1.6556367874145508,
780
+ "rewards/margins": 4.138287544250488,
781
+ "rewards/rejected": -5.793923854827881,
782
+ "step": 430
783
+ },
784
+ {
785
+ "epoch": 0.37816931671680276,
786
+ "grad_norm": 0.5065125823020935,
787
+ "learning_rate": 4.011497787155938e-06,
788
+ "logits/chosen": 3.5233864784240723,
789
+ "logits/rejected": 2.798567533493042,
790
+ "logps/chosen": -1.1753087043762207,
791
+ "logps/rejected": -4.171238899230957,
792
+ "loss": 0.4132,
793
+ "rewards/accuracies": 0.75,
794
+ "rewards/chosen": -1.7629629373550415,
795
+ "rewards/margins": 4.493895053863525,
796
+ "rewards/rejected": -6.256857872009277,
797
+ "step": 440
798
+ },
799
+ {
800
+ "epoch": 0.3867640739149119,
801
+ "grad_norm": 1.414167881011963,
802
+ "learning_rate": 3.969463130731183e-06,
803
+ "logits/chosen": 4.733740329742432,
804
+ "logits/rejected": 4.114102363586426,
805
+ "logps/chosen": -1.1846634149551392,
806
+ "logps/rejected": -4.04649543762207,
807
+ "loss": 0.4266,
808
+ "rewards/accuracies": 0.7749999761581421,
809
+ "rewards/chosen": -1.776995301246643,
810
+ "rewards/margins": 4.292747974395752,
811
+ "rewards/rejected": -6.0697431564331055,
812
+ "step": 450
813
+ },
814
+ {
815
+ "epoch": 0.3867640739149119,
816
+ "eval_logits/chosen": 4.25229549407959,
817
+ "eval_logits/rejected": 3.900564193725586,
818
+ "eval_logps/chosen": -1.410205602645874,
819
+ "eval_logps/rejected": -4.276910781860352,
820
+ "eval_loss": 0.4397798478603363,
821
+ "eval_rewards/accuracies": 0.800000011920929,
822
+ "eval_rewards/chosen": -2.1153085231781006,
823
+ "eval_rewards/margins": 4.3000569343566895,
824
+ "eval_rewards/rejected": -6.415364742279053,
825
+ "eval_runtime": 25.7968,
826
+ "eval_samples_per_second": 29.19,
827
+ "eval_steps_per_second": 3.683,
828
+ "step": 450
829
+ },
830
+ {
831
+ "epoch": 0.39535883111302106,
832
+ "grad_norm": 1.7992101907730103,
833
+ "learning_rate": 3.92678391921108e-06,
834
+ "logits/chosen": 4.718934059143066,
835
+ "logits/rejected": 4.305315971374512,
836
+ "logps/chosen": -1.423595666885376,
837
+ "logps/rejected": -3.9873733520507812,
838
+ "loss": 0.367,
839
+ "rewards/accuracies": 0.762499988079071,
840
+ "rewards/chosen": -2.1353936195373535,
841
+ "rewards/margins": 3.8456661701202393,
842
+ "rewards/rejected": -5.981060028076172,
843
+ "step": 460
844
+ },
845
+ {
846
+ "epoch": 0.4039535883111302,
847
+ "grad_norm": 1.9630879163742065,
848
+ "learning_rate": 3.88347887310836e-06,
849
+ "logits/chosen": 3.8371150493621826,
850
+ "logits/rejected": 3.5719306468963623,
851
+ "logps/chosen": -2.0386481285095215,
852
+ "logps/rejected": -4.779314994812012,
853
+ "loss": 0.3957,
854
+ "rewards/accuracies": 0.8374999761581421,
855
+ "rewards/chosen": -3.0579724311828613,
856
+ "rewards/margins": 4.1110005378723145,
857
+ "rewards/rejected": -7.168972969055176,
858
+ "step": 470
859
+ },
860
+ {
861
+ "epoch": 0.41254834550923936,
862
+ "grad_norm": 4.952139854431152,
863
+ "learning_rate": 3.839566987447492e-06,
864
+ "logits/chosen": 5.024113655090332,
865
+ "logits/rejected": 4.880651950836182,
866
+ "logps/chosen": -2.3612470626831055,
867
+ "logps/rejected": -4.691690444946289,
868
+ "loss": 0.3808,
869
+ "rewards/accuracies": 0.824999988079071,
870
+ "rewards/chosen": -3.5418708324432373,
871
+ "rewards/margins": 3.495664596557617,
872
+ "rewards/rejected": -7.037535190582275,
873
+ "step": 480
874
+ },
875
+ {
876
+ "epoch": 0.4211431027073485,
877
+ "grad_norm": 2.832200527191162,
878
+ "learning_rate": 3.795067523432826e-06,
879
+ "logits/chosen": 4.689079284667969,
880
+ "logits/rejected": 4.435003280639648,
881
+ "logps/chosen": -2.437671184539795,
882
+ "logps/rejected": -5.003944396972656,
883
+ "loss": 0.3555,
884
+ "rewards/accuracies": 0.875,
885
+ "rewards/chosen": -3.6565067768096924,
886
+ "rewards/margins": 3.849409580230713,
887
+ "rewards/rejected": -7.505916595458984,
888
+ "step": 490
889
+ },
890
+ {
891
+ "epoch": 0.42973785990545765,
892
+ "grad_norm": 3.2997682094573975,
893
+ "learning_rate": 3.7500000000000005e-06,
894
+ "logits/chosen": 4.286547660827637,
895
+ "logits/rejected": 4.181652545928955,
896
+ "logps/chosen": -2.3390612602233887,
897
+ "logps/rejected": -4.888935565948486,
898
+ "loss": 0.3211,
899
+ "rewards/accuracies": 0.824999988079071,
900
+ "rewards/chosen": -3.508591890335083,
901
+ "rewards/margins": 3.8248119354248047,
902
+ "rewards/rejected": -7.333403587341309,
903
+ "step": 500
904
+ },
905
+ {
906
+ "epoch": 0.42973785990545765,
907
+ "eval_logits/chosen": 3.5323922634124756,
908
+ "eval_logits/rejected": 3.5116958618164062,
909
+ "eval_logps/chosen": -2.679356575012207,
910
+ "eval_logps/rejected": -5.927057266235352,
911
+ "eval_loss": 0.37026864290237427,
912
+ "eval_rewards/accuracies": 0.8736842274665833,
913
+ "eval_rewards/chosen": -4.019035339355469,
914
+ "eval_rewards/margins": 4.871551036834717,
915
+ "eval_rewards/rejected": -8.890586853027344,
916
+ "eval_runtime": 25.8768,
917
+ "eval_samples_per_second": 29.099,
918
+ "eval_steps_per_second": 3.671,
919
+ "step": 500
920
+ },
921
+ {
922
+ "epoch": 0.4383326171035668,
923
+ "grad_norm": 2.6194217205047607,
924
+ "learning_rate": 3.7043841852542884e-06,
925
+ "logits/chosen": 4.265946388244629,
926
+ "logits/rejected": 3.7863662242889404,
927
+ "logps/chosen": -2.279764413833618,
928
+ "logps/rejected": -4.825397968292236,
929
+ "loss": 0.3375,
930
+ "rewards/accuracies": 0.887499988079071,
931
+ "rewards/chosen": -3.4196460247039795,
932
+ "rewards/margins": 3.818450927734375,
933
+ "rewards/rejected": -7.238096714019775,
934
+ "step": 510
935
+ },
936
+ {
937
+ "epoch": 0.44692737430167595,
938
+ "grad_norm": 4.434008598327637,
939
+ "learning_rate": 3.658240087799655e-06,
940
+ "logits/chosen": 3.4983534812927246,
941
+ "logits/rejected": 3.3109116554260254,
942
+ "logps/chosen": -2.7063632011413574,
943
+ "logps/rejected": -5.9535369873046875,
944
+ "loss": 0.3388,
945
+ "rewards/accuracies": 0.887499988079071,
946
+ "rewards/chosen": -4.059545040130615,
947
+ "rewards/margins": 4.870760917663574,
948
+ "rewards/rejected": -8.930305480957031,
949
+ "step": 520
950
+ },
951
+ {
952
+ "epoch": 0.45552213149978515,
953
+ "grad_norm": 2.4193809032440186,
954
+ "learning_rate": 3.611587947962319e-06,
955
+ "logits/chosen": 3.4949145317077637,
956
+ "logits/rejected": 3.402980089187622,
957
+ "logps/chosen": -2.5146005153656006,
958
+ "logps/rejected": -5.63289737701416,
959
+ "loss": 0.3281,
960
+ "rewards/accuracies": 0.8500000238418579,
961
+ "rewards/chosen": -3.7719013690948486,
962
+ "rewards/margins": 4.677445411682129,
963
+ "rewards/rejected": -8.449346542358398,
964
+ "step": 530
965
+ },
966
+ {
967
+ "epoch": 0.4641168886978943,
968
+ "grad_norm": 10.788633346557617,
969
+ "learning_rate": 3.564448228912682e-06,
970
+ "logits/chosen": 3.3073112964630127,
971
+ "logits/rejected": 3.163470506668091,
972
+ "logps/chosen": -2.258653163909912,
973
+ "logps/rejected": -5.642867088317871,
974
+ "loss": 0.3565,
975
+ "rewards/accuracies": 0.9125000238418579,
976
+ "rewards/chosen": -3.387979507446289,
977
+ "rewards/margins": 5.076320648193359,
978
+ "rewards/rejected": -8.464300155639648,
979
+ "step": 540
980
+ },
981
+ {
982
+ "epoch": 0.47271164589600345,
983
+ "grad_norm": 1.6846323013305664,
984
+ "learning_rate": 3.516841607689501e-06,
985
+ "logits/chosen": 3.476361036300659,
986
+ "logits/rejected": 3.375828504562378,
987
+ "logps/chosen": -2.5325064659118652,
988
+ "logps/rejected": -5.799270153045654,
989
+ "loss": 0.3103,
990
+ "rewards/accuracies": 0.887499988079071,
991
+ "rewards/chosen": -3.798759937286377,
992
+ "rewards/margins": 4.900145530700684,
993
+ "rewards/rejected": -8.698905944824219,
994
+ "step": 550
995
+ },
996
+ {
997
+ "epoch": 0.47271164589600345,
998
+ "eval_logits/chosen": 3.451749563217163,
999
+ "eval_logits/rejected": 3.3771002292633057,
1000
+ "eval_logps/chosen": -2.9835667610168457,
1001
+ "eval_logps/rejected": -6.5389509201049805,
1002
+ "eval_loss": 0.32732319831848145,
1003
+ "eval_rewards/accuracies": 0.9052631855010986,
1004
+ "eval_rewards/chosen": -4.475350379943848,
1005
+ "eval_rewards/margins": 5.333076000213623,
1006
+ "eval_rewards/rejected": -9.808425903320312,
1007
+ "eval_runtime": 25.8141,
1008
+ "eval_samples_per_second": 29.17,
1009
+ "eval_steps_per_second": 3.68,
1010
+ "step": 550
1011
+ },
1012
+ {
1013
+ "epoch": 0.4813064030941126,
1014
+ "grad_norm": 2.7683331966400146,
1015
+ "learning_rate": 3.4687889661302577e-06,
1016
+ "logits/chosen": 2.468799591064453,
1017
+ "logits/rejected": 2.4257254600524902,
1018
+ "logps/chosen": -2.6801788806915283,
1019
+ "logps/rejected": -6.136897087097168,
1020
+ "loss": 0.3171,
1021
+ "rewards/accuracies": 0.887499988079071,
1022
+ "rewards/chosen": -4.020268440246582,
1023
+ "rewards/margins": 5.185078144073486,
1024
+ "rewards/rejected": -9.205347061157227,
1025
+ "step": 560
1026
+ },
1027
+ {
1028
+ "epoch": 0.48990116029222175,
1029
+ "grad_norm": 11.559685707092285,
1030
+ "learning_rate": 3.4203113817116955e-06,
1031
+ "logits/chosen": 3.6535427570343018,
1032
+ "logits/rejected": 3.583962917327881,
1033
+ "logps/chosen": -2.7984983921051025,
1034
+ "logps/rejected": -6.468808650970459,
1035
+ "loss": 0.331,
1036
+ "rewards/accuracies": 0.949999988079071,
1037
+ "rewards/chosen": -4.197747707366943,
1038
+ "rewards/margins": 5.505465984344482,
1039
+ "rewards/rejected": -9.703214645385742,
1040
+ "step": 570
1041
+ },
1042
+ {
1043
+ "epoch": 0.4984959174903309,
1044
+ "grad_norm": 7.691457271575928,
1045
+ "learning_rate": 3.3714301183045382e-06,
1046
+ "logits/chosen": 2.7467286586761475,
1047
+ "logits/rejected": 2.4911317825317383,
1048
+ "logps/chosen": -2.4207634925842285,
1049
+ "logps/rejected": -6.385074138641357,
1050
+ "loss": 0.2557,
1051
+ "rewards/accuracies": 0.9750000238418579,
1052
+ "rewards/chosen": -3.631145477294922,
1053
+ "rewards/margins": 5.946464538574219,
1054
+ "rewards/rejected": -9.577610969543457,
1055
+ "step": 580
1056
+ },
1057
+ {
1058
+ "epoch": 0.50709067468844,
1059
+ "grad_norm": 5.381045341491699,
1060
+ "learning_rate": 3.3221666168464584e-06,
1061
+ "logits/chosen": 3.239227294921875,
1062
+ "logits/rejected": 3.1982555389404297,
1063
+ "logps/chosen": -3.2162883281707764,
1064
+ "logps/rejected": -6.693168640136719,
1065
+ "loss": 0.2921,
1066
+ "rewards/accuracies": 0.949999988079071,
1067
+ "rewards/chosen": -4.824432373046875,
1068
+ "rewards/margins": 5.215321063995361,
1069
+ "rewards/rejected": -10.039752960205078,
1070
+ "step": 590
1071
+ },
1072
+ {
1073
+ "epoch": 0.5156854318865493,
1074
+ "grad_norm": 2.616710901260376,
1075
+ "learning_rate": 3.272542485937369e-06,
1076
+ "logits/chosen": 2.7326784133911133,
1077
+ "logits/rejected": 2.4248623847961426,
1078
+ "logps/chosen": -3.001952648162842,
1079
+ "logps/rejected": -6.597250461578369,
1080
+ "loss": 0.3225,
1081
+ "rewards/accuracies": 0.9125000238418579,
1082
+ "rewards/chosen": -4.502928733825684,
1083
+ "rewards/margins": 5.392947196960449,
1084
+ "rewards/rejected": -9.89587688446045,
1085
+ "step": 600
1086
+ },
1087
+ {
1088
+ "epoch": 0.5156854318865493,
1089
+ "eval_logits/chosen": 3.1995701789855957,
1090
+ "eval_logits/rejected": 3.2761952877044678,
1091
+ "eval_logps/chosen": -3.166325330734253,
1092
+ "eval_logps/rejected": -7.062953472137451,
1093
+ "eval_loss": 0.29786577820777893,
1094
+ "eval_rewards/accuracies": 0.9263157844543457,
1095
+ "eval_rewards/chosen": -4.74948787689209,
1096
+ "eval_rewards/margins": 5.84494161605835,
1097
+ "eval_rewards/rejected": -10.594429969787598,
1098
+ "eval_runtime": 25.8074,
1099
+ "eval_samples_per_second": 29.178,
1100
+ "eval_steps_per_second": 3.681,
1101
+ "step": 600
1102
+ },
1103
+ {
1104
+ "epoch": 0.5242801890846583,
1105
+ "grad_norm": 3.5550060272216797,
1106
+ "learning_rate": 3.222579492361179e-06,
1107
+ "logits/chosen": 2.8619818687438965,
1108
+ "logits/rejected": 3.014125347137451,
1109
+ "logps/chosen": -2.9217543601989746,
1110
+ "logps/rejected": -6.349586486816406,
1111
+ "loss": 0.2516,
1112
+ "rewards/accuracies": 0.9125000238418579,
1113
+ "rewards/chosen": -4.382631301879883,
1114
+ "rewards/margins": 5.141747951507568,
1115
+ "rewards/rejected": -9.52437973022461,
1116
+ "step": 610
1117
+ },
1118
+ {
1119
+ "epoch": 0.5328749462827675,
1120
+ "grad_norm": 3.6020898818969727,
1121
+ "learning_rate": 3.1722995515381644e-06,
1122
+ "logits/chosen": 2.770552635192871,
1123
+ "logits/rejected": 2.9711716175079346,
1124
+ "logps/chosen": -2.8700039386749268,
1125
+ "logps/rejected": -6.527164459228516,
1126
+ "loss": 0.3062,
1127
+ "rewards/accuracies": 0.925000011920929,
1128
+ "rewards/chosen": -4.305006504058838,
1129
+ "rewards/margins": 5.485739707946777,
1130
+ "rewards/rejected": -9.790745735168457,
1131
+ "step": 620
1132
+ },
1133
+ {
1134
+ "epoch": 0.5414697034808766,
1135
+ "grad_norm": 2.2209339141845703,
1136
+ "learning_rate": 3.121724717912138e-06,
1137
+ "logits/chosen": 3.1150155067443848,
1138
+ "logits/rejected": 3.038687229156494,
1139
+ "logps/chosen": -2.9821603298187256,
1140
+ "logps/rejected": -6.432187080383301,
1141
+ "loss": 0.2726,
1142
+ "rewards/accuracies": 0.9125000238418579,
1143
+ "rewards/chosen": -4.473240852355957,
1144
+ "rewards/margins": 5.175040245056152,
1145
+ "rewards/rejected": -9.648280143737793,
1146
+ "step": 630
1147
+ },
1148
+ {
1149
+ "epoch": 0.5500644606789858,
1150
+ "grad_norm": 2.148709297180176,
1151
+ "learning_rate": 3.0708771752766397e-06,
1152
+ "logits/chosen": 3.256204605102539,
1153
+ "logits/rejected": 2.9765384197235107,
1154
+ "logps/chosen": -2.7979862689971924,
1155
+ "logps/rejected": -6.7102813720703125,
1156
+ "loss": 0.2861,
1157
+ "rewards/accuracies": 0.9125000238418579,
1158
+ "rewards/chosen": -4.196979522705078,
1159
+ "rewards/margins": 5.868442535400391,
1160
+ "rewards/rejected": -10.065422058105469,
1161
+ "step": 640
1162
+ },
1163
+ {
1164
+ "epoch": 0.5586592178770949,
1165
+ "grad_norm": 7.988170623779297,
1166
+ "learning_rate": 3.019779227044398e-06,
1167
+ "logits/chosen": 2.7312607765197754,
1168
+ "logits/rejected": 2.705409526824951,
1169
+ "logps/chosen": -2.8119194507598877,
1170
+ "logps/rejected": -6.627874851226807,
1171
+ "loss": 0.2507,
1172
+ "rewards/accuracies": 0.9125000238418579,
1173
+ "rewards/chosen": -4.217879295349121,
1174
+ "rewards/margins": 5.723933219909668,
1175
+ "rewards/rejected": -9.941811561584473,
1176
+ "step": 650
1177
+ },
1178
+ {
1179
+ "epoch": 0.5586592178770949,
1180
+ "eval_logits/chosen": 2.9934144020080566,
1181
+ "eval_logits/rejected": 3.124950647354126,
1182
+ "eval_logps/chosen": -3.1934540271759033,
1183
+ "eval_logps/rejected": -7.429007053375244,
1184
+ "eval_loss": 0.28581172227859497,
1185
+ "eval_rewards/accuracies": 0.9368420839309692,
1186
+ "eval_rewards/chosen": -4.7901811599731445,
1187
+ "eval_rewards/margins": 6.353330135345459,
1188
+ "eval_rewards/rejected": -11.143510818481445,
1189
+ "eval_runtime": 25.8075,
1190
+ "eval_samples_per_second": 29.178,
1191
+ "eval_steps_per_second": 3.681,
1192
+ "step": 650
1193
+ },
1194
+ {
1195
+ "epoch": 0.5672539750752041,
1196
+ "grad_norm": 3.133023500442505,
1197
+ "learning_rate": 2.9684532864643123e-06,
1198
+ "logits/chosen": 3.3388328552246094,
1199
+ "logits/rejected": 3.3477108478546143,
1200
+ "logps/chosen": -3.3007023334503174,
1201
+ "logps/rejected": -6.496421813964844,
1202
+ "loss": 0.2635,
1203
+ "rewards/accuracies": 0.9125000238418579,
1204
+ "rewards/chosen": -4.951053619384766,
1205
+ "rewards/margins": 4.793579578399658,
1206
+ "rewards/rejected": -9.744632720947266,
1207
+ "step": 660
1208
+ },
1209
+ {
1210
+ "epoch": 0.5758487322733132,
1211
+ "grad_norm": 3.6694588661193848,
1212
+ "learning_rate": 2.9169218667902562e-06,
1213
+ "logits/chosen": 3.1316323280334473,
1214
+ "logits/rejected": 2.8977527618408203,
1215
+ "logps/chosen": -2.8769032955169678,
1216
+ "logps/rejected": -5.834782123565674,
1217
+ "loss": 0.2854,
1218
+ "rewards/accuracies": 0.875,
1219
+ "rewards/chosen": -4.31535530090332,
1220
+ "rewards/margins": 4.4368181228637695,
1221
+ "rewards/rejected": -8.752172470092773,
1222
+ "step": 670
1223
+ },
1224
+ {
1225
+ "epoch": 0.5844434894714224,
1226
+ "grad_norm": 3.409086227416992,
1227
+ "learning_rate": 2.8652075714060296e-06,
1228
+ "logits/chosen": 2.726431369781494,
1229
+ "logits/rejected": 2.9928715229034424,
1230
+ "logps/chosen": -3.0180654525756836,
1231
+ "logps/rejected": -7.160694122314453,
1232
+ "loss": 0.2461,
1233
+ "rewards/accuracies": 0.949999988079071,
1234
+ "rewards/chosen": -4.527098178863525,
1235
+ "rewards/margins": 6.213942050933838,
1236
+ "rewards/rejected": -10.741040229797363,
1237
+ "step": 680
1238
+ },
1239
+ {
1240
+ "epoch": 0.5930382466695315,
1241
+ "grad_norm": 5.1549153327941895,
1242
+ "learning_rate": 2.813333083910761e-06,
1243
+ "logits/chosen": 3.371833086013794,
1244
+ "logits/rejected": 3.317675828933716,
1245
+ "logps/chosen": -3.1700775623321533,
1246
+ "logps/rejected": -6.8295745849609375,
1247
+ "loss": 0.2458,
1248
+ "rewards/accuracies": 0.949999988079071,
1249
+ "rewards/chosen": -4.755115985870361,
1250
+ "rewards/margins": 5.489245891571045,
1251
+ "rewards/rejected": -10.244361877441406,
1252
+ "step": 690
1253
+ },
1254
+ {
1255
+ "epoch": 0.6016330038676407,
1256
+ "grad_norm": 4.8320112228393555,
1257
+ "learning_rate": 2.761321158169134e-06,
1258
+ "logits/chosen": 2.5933046340942383,
1259
+ "logits/rejected": 2.7114923000335693,
1260
+ "logps/chosen": -3.2017345428466797,
1261
+ "logps/rejected": -7.420884609222412,
1262
+ "loss": 0.2464,
1263
+ "rewards/accuracies": 0.925000011920929,
1264
+ "rewards/chosen": -4.8026018142700195,
1265
+ "rewards/margins": 6.328725814819336,
1266
+ "rewards/rejected": -11.131326675415039,
1267
+ "step": 700
1268
+ },
1269
+ {
1270
+ "epoch": 0.6016330038676407,
1271
+ "eval_logits/chosen": 2.7770564556121826,
1272
+ "eval_logits/rejected": 2.9842188358306885,
1273
+ "eval_logps/chosen": -3.5835421085357666,
1274
+ "eval_logps/rejected": -8.199637413024902,
1275
+ "eval_loss": 0.26223084330558777,
1276
+ "eval_rewards/accuracies": 0.9263157844543457,
1277
+ "eval_rewards/chosen": -5.375312805175781,
1278
+ "eval_rewards/margins": 6.924142360687256,
1279
+ "eval_rewards/rejected": -12.299455642700195,
1280
+ "eval_runtime": 25.8482,
1281
+ "eval_samples_per_second": 29.132,
1282
+ "eval_steps_per_second": 3.675,
1283
+ "step": 700
1284
+ },
1285
+ {
1286
+ "epoch": 0.6102277610657499,
1287
+ "grad_norm": 3.6024580001831055,
1288
+ "learning_rate": 2.70919460833079e-06,
1289
+ "logits/chosen": 2.4713566303253174,
1290
+ "logits/rejected": 2.523773193359375,
1291
+ "logps/chosen": -2.962407112121582,
1292
+ "logps/rejected": -7.556809902191162,
1293
+ "loss": 0.2785,
1294
+ "rewards/accuracies": 0.9624999761581421,
1295
+ "rewards/chosen": -4.443611145019531,
1296
+ "rewards/margins": 6.891604423522949,
1297
+ "rewards/rejected": -11.33521556854248,
1298
+ "step": 710
1299
+ },
1300
+ {
1301
+ "epoch": 0.618822518263859,
1302
+ "grad_norm": 3.6402506828308105,
1303
+ "learning_rate": 2.6569762988232838e-06,
1304
+ "logits/chosen": 3.112736940383911,
1305
+ "logits/rejected": 2.901930809020996,
1306
+ "logps/chosen": -3.1669626235961914,
1307
+ "logps/rejected": -7.049059867858887,
1308
+ "loss": 0.2623,
1309
+ "rewards/accuracies": 0.925000011920929,
1310
+ "rewards/chosen": -4.750443935394287,
1311
+ "rewards/margins": 5.823145866394043,
1312
+ "rewards/rejected": -10.573590278625488,
1313
+ "step": 720
1314
+ },
1315
+ {
1316
+ "epoch": 0.6274172754619682,
1317
+ "grad_norm": 9.418655395507812,
1318
+ "learning_rate": 2.604689134322999e-06,
1319
+ "logits/chosen": 2.8671321868896484,
1320
+ "logits/rejected": 2.763396739959717,
1321
+ "logps/chosen": -3.1475024223327637,
1322
+ "logps/rejected": -7.338767051696777,
1323
+ "loss": 0.2702,
1324
+ "rewards/accuracies": 0.9125000238418579,
1325
+ "rewards/chosen": -4.721253871917725,
1326
+ "rewards/margins": 6.286896705627441,
1327
+ "rewards/rejected": -11.008151054382324,
1328
+ "step": 730
1329
+ },
1330
+ {
1331
+ "epoch": 0.6360120326600773,
1332
+ "grad_norm": 6.016907215118408,
1333
+ "learning_rate": 2.5523560497083927e-06,
1334
+ "logits/chosen": 2.8831398487091064,
1335
+ "logits/rejected": 2.9516844749450684,
1336
+ "logps/chosen": -3.055087089538574,
1337
+ "logps/rejected": -7.092196464538574,
1338
+ "loss": 0.2103,
1339
+ "rewards/accuracies": 0.949999988079071,
1340
+ "rewards/chosen": -4.5826311111450195,
1341
+ "rewards/margins": 6.0556640625,
1342
+ "rewards/rejected": -10.638293266296387,
1343
+ "step": 740
1344
+ },
1345
+ {
1346
+ "epoch": 0.6446067898581865,
1347
+ "grad_norm": 2.791388988494873,
1348
+ "learning_rate": 2.5e-06,
1349
+ "logits/chosen": 3.5676627159118652,
1350
+ "logits/rejected": 3.4635086059570312,
1351
+ "logps/chosen": -2.8061039447784424,
1352
+ "logps/rejected": -6.826286315917969,
1353
+ "loss": 0.2545,
1354
+ "rewards/accuracies": 0.949999988079071,
1355
+ "rewards/chosen": -4.209155082702637,
1356
+ "rewards/margins": 6.030273914337158,
1357
+ "rewards/rejected": -10.239428520202637,
1358
+ "step": 750
1359
+ },
1360
+ {
1361
+ "epoch": 0.6446067898581865,
1362
+ "eval_logits/chosen": 2.94565749168396,
1363
+ "eval_logits/rejected": 3.170260190963745,
1364
+ "eval_logps/chosen": -3.455902338027954,
1365
+ "eval_logps/rejected": -8.303979873657227,
1366
+ "eval_loss": 0.2537091076374054,
1367
+ "eval_rewards/accuracies": 0.9368420839309692,
1368
+ "eval_rewards/chosen": -5.1838531494140625,
1369
+ "eval_rewards/margins": 7.272115707397461,
1370
+ "eval_rewards/rejected": -12.455968856811523,
1371
+ "eval_runtime": 25.8128,
1372
+ "eval_samples_per_second": 29.172,
1373
+ "eval_steps_per_second": 3.68,
1374
+ "step": 750
1375
+ },
1376
+ {
1377
+ "epoch": 0.6532015470562956,
1378
+ "grad_norm": 2.978318452835083,
1379
+ "learning_rate": 2.447643950291608e-06,
1380
+ "logits/chosen": 3.1550345420837402,
1381
+ "logits/rejected": 2.9587855339050293,
1382
+ "logps/chosen": -3.403642177581787,
1383
+ "logps/rejected": -7.627197265625,
1384
+ "loss": 0.2261,
1385
+ "rewards/accuracies": 0.9375,
1386
+ "rewards/chosen": -5.10546350479126,
1387
+ "rewards/margins": 6.335333824157715,
1388
+ "rewards/rejected": -11.440796852111816,
1389
+ "step": 760
1390
+ },
1391
+ {
1392
+ "epoch": 0.6617963042544048,
1393
+ "grad_norm": 4.7188801765441895,
1394
+ "learning_rate": 2.3953108656770018e-06,
1395
+ "logits/chosen": 3.7232565879821777,
1396
+ "logits/rejected": 3.5991597175598145,
1397
+ "logps/chosen": -3.6658260822296143,
1398
+ "logps/rejected": -7.148935794830322,
1399
+ "loss": 0.2703,
1400
+ "rewards/accuracies": 0.862500011920929,
1401
+ "rewards/chosen": -5.498739242553711,
1402
+ "rewards/margins": 5.22466516494751,
1403
+ "rewards/rejected": -10.723405838012695,
1404
+ "step": 770
1405
+ },
1406
+ {
1407
+ "epoch": 0.6703910614525139,
1408
+ "grad_norm": 4.847439289093018,
1409
+ "learning_rate": 2.3430237011767166e-06,
1410
+ "logits/chosen": 2.366446018218994,
1411
+ "logits/rejected": 2.4268651008605957,
1412
+ "logps/chosen": -3.6872811317443848,
1413
+ "logps/rejected": -8.306299209594727,
1414
+ "loss": 0.2493,
1415
+ "rewards/accuracies": 0.987500011920929,
1416
+ "rewards/chosen": -5.530921459197998,
1417
+ "rewards/margins": 6.928528785705566,
1418
+ "rewards/rejected": -12.459450721740723,
1419
+ "step": 780
1420
+ },
1421
+ {
1422
+ "epoch": 0.6789858186506231,
1423
+ "grad_norm": 3.72248911857605,
1424
+ "learning_rate": 2.290805391669212e-06,
1425
+ "logits/chosen": 2.7826573848724365,
1426
+ "logits/rejected": 2.9013209342956543,
1427
+ "logps/chosen": -3.5557899475097656,
1428
+ "logps/rejected": -7.959009647369385,
1429
+ "loss": 0.2461,
1430
+ "rewards/accuracies": 0.925000011920929,
1431
+ "rewards/chosen": -5.33368444442749,
1432
+ "rewards/margins": 6.604828834533691,
1433
+ "rewards/rejected": -11.938512802124023,
1434
+ "step": 790
1435
+ },
1436
+ {
1437
+ "epoch": 0.6875805758487322,
1438
+ "grad_norm": 4.7869343757629395,
1439
+ "learning_rate": 2.238678841830867e-06,
1440
+ "logits/chosen": 2.5417182445526123,
1441
+ "logits/rejected": 2.938063621520996,
1442
+ "logps/chosen": -3.5474331378936768,
1443
+ "logps/rejected": -7.563382148742676,
1444
+ "loss": 0.2483,
1445
+ "rewards/accuracies": 0.925000011920929,
1446
+ "rewards/chosen": -5.3211493492126465,
1447
+ "rewards/margins": 6.023923873901367,
1448
+ "rewards/rejected": -11.345073699951172,
1449
+ "step": 800
1450
+ },
1451
+ {
1452
+ "epoch": 0.6875805758487322,
1453
+ "eval_logits/chosen": 2.9645943641662598,
1454
+ "eval_logits/rejected": 3.2736430168151855,
1455
+ "eval_logps/chosen": -3.4982783794403076,
1456
+ "eval_logps/rejected": -8.614095687866211,
1457
+ "eval_loss": 0.24035032093524933,
1458
+ "eval_rewards/accuracies": 0.9263157844543457,
1459
+ "eval_rewards/chosen": -5.24741792678833,
1460
+ "eval_rewards/margins": 7.673725128173828,
1461
+ "eval_rewards/rejected": -12.921142578125,
1462
+ "eval_runtime": 25.8098,
1463
+ "eval_samples_per_second": 29.175,
1464
+ "eval_steps_per_second": 3.681,
1465
+ "step": 800
1466
+ },
1467
+ {
1468
+ "epoch": 0.6961753330468414,
1469
+ "grad_norm": 5.635983943939209,
1470
+ "learning_rate": 2.186666916089239e-06,
1471
+ "logits/chosen": 2.7499587535858154,
1472
+ "logits/rejected": 2.6932997703552246,
1473
+ "logps/chosen": -3.2679648399353027,
1474
+ "logps/rejected": -8.038375854492188,
1475
+ "loss": 0.2647,
1476
+ "rewards/accuracies": 0.925000011920929,
1477
+ "rewards/chosen": -4.901947975158691,
1478
+ "rewards/margins": 7.155615329742432,
1479
+ "rewards/rejected": -12.057561874389648,
1480
+ "step": 810
1481
+ },
1482
+ {
1483
+ "epoch": 0.7047700902449506,
1484
+ "grad_norm": 4.1279215812683105,
1485
+ "learning_rate": 2.134792428593971e-06,
1486
+ "logits/chosen": 3.1286826133728027,
1487
+ "logits/rejected": 3.3689827919006348,
1488
+ "logps/chosen": -3.5489754676818848,
1489
+ "logps/rejected": -8.034095764160156,
1490
+ "loss": 0.1888,
1491
+ "rewards/accuracies": 0.8999999761581421,
1492
+ "rewards/chosen": -5.323462963104248,
1493
+ "rewards/margins": 6.727679252624512,
1494
+ "rewards/rejected": -12.051143646240234,
1495
+ "step": 820
1496
+ },
1497
+ {
1498
+ "epoch": 0.7133648474430597,
1499
+ "grad_norm": 4.562918186187744,
1500
+ "learning_rate": 2.0830781332097446e-06,
1501
+ "logits/chosen": 2.635140895843506,
1502
+ "logits/rejected": 2.909487247467041,
1503
+ "logps/chosen": -3.598254442214966,
1504
+ "logps/rejected": -8.546786308288574,
1505
+ "loss": 0.2141,
1506
+ "rewards/accuracies": 0.949999988079071,
1507
+ "rewards/chosen": -5.39738130569458,
1508
+ "rewards/margins": 7.422798156738281,
1509
+ "rewards/rejected": -12.820180892944336,
1510
+ "step": 830
1511
+ },
1512
+ {
1513
+ "epoch": 0.7219596046411689,
1514
+ "grad_norm": 5.229101657867432,
1515
+ "learning_rate": 2.031546713535688e-06,
1516
+ "logits/chosen": 3.0485472679138184,
1517
+ "logits/rejected": 3.2676749229431152,
1518
+ "logps/chosen": -3.5765304565429688,
1519
+ "logps/rejected": -8.731932640075684,
1520
+ "loss": 0.1953,
1521
+ "rewards/accuracies": 0.949999988079071,
1522
+ "rewards/chosen": -5.364795207977295,
1523
+ "rewards/margins": 7.7331037521362305,
1524
+ "rewards/rejected": -13.097898483276367,
1525
+ "step": 840
1526
+ },
1527
+ {
1528
+ "epoch": 0.730554361839278,
1529
+ "grad_norm": 3.0395517349243164,
1530
+ "learning_rate": 1.9802207729556023e-06,
1531
+ "logits/chosen": 3.242750644683838,
1532
+ "logits/rejected": 3.410076141357422,
1533
+ "logps/chosen": -3.5732295513153076,
1534
+ "logps/rejected": -7.774885654449463,
1535
+ "loss": 0.2395,
1536
+ "rewards/accuracies": 0.8999999761581421,
1537
+ "rewards/chosen": -5.359843730926514,
1538
+ "rewards/margins": 6.302483558654785,
1539
+ "rewards/rejected": -11.662328720092773,
1540
+ "step": 850
1541
+ },
1542
+ {
1543
+ "epoch": 0.730554361839278,
1544
+ "eval_logits/chosen": 2.9314279556274414,
1545
+ "eval_logits/rejected": 3.276207208633423,
1546
+ "eval_logps/chosen": -3.8869926929473877,
1547
+ "eval_logps/rejected": -9.050567626953125,
1548
+ "eval_loss": 0.2363433688879013,
1549
+ "eval_rewards/accuracies": 0.9263157844543457,
1550
+ "eval_rewards/chosen": -5.830489158630371,
1551
+ "eval_rewards/margins": 7.745361328125,
1552
+ "eval_rewards/rejected": -13.575852394104004,
1553
+ "eval_runtime": 25.8152,
1554
+ "eval_samples_per_second": 29.169,
1555
+ "eval_steps_per_second": 3.68,
1556
+ "step": 850
1557
+ },
1558
+ {
1559
+ "epoch": 0.7391491190373872,
1560
+ "grad_norm": 3.7891011238098145,
1561
+ "learning_rate": 1.9291228247233607e-06,
1562
+ "logits/chosen": 1.9879547357559204,
1563
+ "logits/rejected": 2.725163698196411,
1564
+ "logps/chosen": -3.3898093700408936,
1565
+ "logps/rejected": -8.569767951965332,
1566
+ "loss": 0.2234,
1567
+ "rewards/accuracies": 0.925000011920929,
1568
+ "rewards/chosen": -5.084713935852051,
1569
+ "rewards/margins": 7.769936561584473,
1570
+ "rewards/rejected": -12.854650497436523,
1571
+ "step": 860
1572
+ },
1573
+ {
1574
+ "epoch": 0.7477438762354963,
1575
+ "grad_norm": 2.793806552886963,
1576
+ "learning_rate": 1.8782752820878636e-06,
1577
+ "logits/chosen": 2.8827013969421387,
1578
+ "logits/rejected": 3.106083631515503,
1579
+ "logps/chosen": -3.376796245574951,
1580
+ "logps/rejected": -8.713462829589844,
1581
+ "loss": 0.2045,
1582
+ "rewards/accuracies": 0.949999988079071,
1583
+ "rewards/chosen": -5.065195083618164,
1584
+ "rewards/margins": 8.004999160766602,
1585
+ "rewards/rejected": -13.07019329071045,
1586
+ "step": 870
1587
+ },
1588
+ {
1589
+ "epoch": 0.7563386334336055,
1590
+ "grad_norm": 3.8785598278045654,
1591
+ "learning_rate": 1.827700448461836e-06,
1592
+ "logits/chosen": 3.18955659866333,
1593
+ "logits/rejected": 3.332995653152466,
1594
+ "logps/chosen": -3.6388778686523438,
1595
+ "logps/rejected": -8.38855266571045,
1596
+ "loss": 0.231,
1597
+ "rewards/accuracies": 0.9125000238418579,
1598
+ "rewards/chosen": -5.458316802978516,
1599
+ "rewards/margins": 7.124513149261475,
1600
+ "rewards/rejected": -12.582829475402832,
1601
+ "step": 880
1602
+ },
1603
+ {
1604
+ "epoch": 0.7649333906317146,
1605
+ "grad_norm": 3.2375245094299316,
1606
+ "learning_rate": 1.7774205076388207e-06,
1607
+ "logits/chosen": 3.7443747520446777,
1608
+ "logits/rejected": 3.607149839401245,
1609
+ "logps/chosen": -3.7576351165771484,
1610
+ "logps/rejected": -8.150545120239258,
1611
+ "loss": 0.2398,
1612
+ "rewards/accuracies": 0.8999999761581421,
1613
+ "rewards/chosen": -5.636452674865723,
1614
+ "rewards/margins": 6.589364528656006,
1615
+ "rewards/rejected": -12.22581672668457,
1616
+ "step": 890
1617
+ },
1618
+ {
1619
+ "epoch": 0.7735281478298238,
1620
+ "grad_norm": 3.1005704402923584,
1621
+ "learning_rate": 1.7274575140626318e-06,
1622
+ "logits/chosen": 2.988206386566162,
1623
+ "logits/rejected": 3.322185516357422,
1624
+ "logps/chosen": -3.125667095184326,
1625
+ "logps/rejected": -8.629180908203125,
1626
+ "loss": 0.2388,
1627
+ "rewards/accuracies": 0.9750000238418579,
1628
+ "rewards/chosen": -4.68850040435791,
1629
+ "rewards/margins": 8.255270957946777,
1630
+ "rewards/rejected": -12.943771362304688,
1631
+ "step": 900
1632
+ },
1633
+ {
1634
+ "epoch": 0.7735281478298238,
1635
+ "eval_logits/chosen": 3.079637289047241,
1636
+ "eval_logits/rejected": 3.3415913581848145,
1637
+ "eval_logps/chosen": -3.6503636837005615,
1638
+ "eval_logps/rejected": -8.939910888671875,
1639
+ "eval_loss": 0.22793449461460114,
1640
+ "eval_rewards/accuracies": 0.9157894849777222,
1641
+ "eval_rewards/chosen": -5.4755449295043945,
1642
+ "eval_rewards/margins": 7.934320449829102,
1643
+ "eval_rewards/rejected": -13.409865379333496,
1644
+ "eval_runtime": 25.8118,
1645
+ "eval_samples_per_second": 29.173,
1646
+ "eval_steps_per_second": 3.68,
1647
+ "step": 900
1648
+ },
1649
+ {
1650
+ "epoch": 0.7821229050279329,
1651
+ "grad_norm": 2.3264434337615967,
1652
+ "learning_rate": 1.677833383153542e-06,
1653
+ "logits/chosen": 2.6580982208251953,
1654
+ "logits/rejected": 2.6074371337890625,
1655
+ "logps/chosen": -3.126781702041626,
1656
+ "logps/rejected": -8.055559158325195,
1657
+ "loss": 0.2267,
1658
+ "rewards/accuracies": 0.949999988079071,
1659
+ "rewards/chosen": -4.69017219543457,
1660
+ "rewards/margins": 7.393167018890381,
1661
+ "rewards/rejected": -12.083338737487793,
1662
+ "step": 910
1663
+ },
1664
+ {
1665
+ "epoch": 0.7907176622260421,
1666
+ "grad_norm": 5.1758880615234375,
1667
+ "learning_rate": 1.6285698816954626e-06,
1668
+ "logits/chosen": 3.202477216720581,
1669
+ "logits/rejected": 3.286799907684326,
1670
+ "logps/chosen": -3.5538082122802734,
1671
+ "logps/rejected": -8.186088562011719,
1672
+ "loss": 0.2051,
1673
+ "rewards/accuracies": 0.925000011920929,
1674
+ "rewards/chosen": -5.330712795257568,
1675
+ "rewards/margins": 6.94842004776001,
1676
+ "rewards/rejected": -12.279133796691895,
1677
+ "step": 920
1678
+ },
1679
+ {
1680
+ "epoch": 0.7993124194241513,
1681
+ "grad_norm": 1.7054041624069214,
1682
+ "learning_rate": 1.5796886182883053e-06,
1683
+ "logits/chosen": 3.129915714263916,
1684
+ "logits/rejected": 3.352510929107666,
1685
+ "logps/chosen": -3.1991374492645264,
1686
+ "logps/rejected": -8.518336296081543,
1687
+ "loss": 0.2353,
1688
+ "rewards/accuracies": 0.949999988079071,
1689
+ "rewards/chosen": -4.798706531524658,
1690
+ "rewards/margins": 7.978797912597656,
1691
+ "rewards/rejected": -12.777504920959473,
1692
+ "step": 930
1693
+ },
1694
+ {
1695
+ "epoch": 0.8079071766222604,
1696
+ "grad_norm": 2.9931082725524902,
1697
+ "learning_rate": 1.5312110338697427e-06,
1698
+ "logits/chosen": 2.5596461296081543,
1699
+ "logits/rejected": 2.5915169715881348,
1700
+ "logps/chosen": -3.3413949012756348,
1701
+ "logps/rejected": -7.961850643157959,
1702
+ "loss": 0.2516,
1703
+ "rewards/accuracies": 0.875,
1704
+ "rewards/chosen": -5.012092113494873,
1705
+ "rewards/margins": 6.930683135986328,
1706
+ "rewards/rejected": -11.942774772644043,
1707
+ "step": 940
1708
+ },
1709
+ {
1710
+ "epoch": 0.8165019338203696,
1711
+ "grad_norm": 4.107710838317871,
1712
+ "learning_rate": 1.4831583923105e-06,
1713
+ "logits/chosen": 2.8222692012786865,
1714
+ "logits/rejected": 2.910057306289673,
1715
+ "logps/chosen": -3.256718397140503,
1716
+ "logps/rejected": -8.105169296264648,
1717
+ "loss": 0.2299,
1718
+ "rewards/accuracies": 0.9624999761581421,
1719
+ "rewards/chosen": -4.885077476501465,
1720
+ "rewards/margins": 7.272677421569824,
1721
+ "rewards/rejected": -12.157754898071289,
1722
+ "step": 950
1723
+ },
1724
+ {
1725
+ "epoch": 0.8165019338203696,
1726
+ "eval_logits/chosen": 3.0568909645080566,
1727
+ "eval_logits/rejected": 3.3856008052825928,
1728
+ "eval_logps/chosen": -3.493980646133423,
1729
+ "eval_logps/rejected": -8.99133014678955,
1730
+ "eval_loss": 0.2293253093957901,
1731
+ "eval_rewards/accuracies": 0.9368420839309692,
1732
+ "eval_rewards/chosen": -5.240970611572266,
1733
+ "eval_rewards/margins": 8.246024131774902,
1734
+ "eval_rewards/rejected": -13.486994743347168,
1735
+ "eval_runtime": 25.8236,
1736
+ "eval_samples_per_second": 29.159,
1737
+ "eval_steps_per_second": 3.679,
1738
+ "step": 950
1739
+ },
1740
+ {
1741
+ "epoch": 0.8250966910184787,
1742
+ "grad_norm": 4.835093021392822,
1743
+ "learning_rate": 1.4355517710873184e-06,
1744
+ "logits/chosen": 3.3382835388183594,
1745
+ "logits/rejected": 3.328683376312256,
1746
+ "logps/chosen": -3.7020835876464844,
1747
+ "logps/rejected": -8.166964530944824,
1748
+ "loss": 0.2091,
1749
+ "rewards/accuracies": 0.9375,
1750
+ "rewards/chosen": -5.553124904632568,
1751
+ "rewards/margins": 6.697320461273193,
1752
+ "rewards/rejected": -12.250445365905762,
1753
+ "step": 960
1754
+ },
1755
+ {
1756
+ "epoch": 0.8336914482165879,
1757
+ "grad_norm": 4.466915130615234,
1758
+ "learning_rate": 1.388412052037682e-06,
1759
+ "logits/chosen": 2.396477222442627,
1760
+ "logits/rejected": 2.636380672454834,
1761
+ "logps/chosen": -3.1254048347473145,
1762
+ "logps/rejected": -9.067309379577637,
1763
+ "loss": 0.2295,
1764
+ "rewards/accuracies": 0.949999988079071,
1765
+ "rewards/chosen": -4.688107013702393,
1766
+ "rewards/margins": 8.912858963012695,
1767
+ "rewards/rejected": -13.600967407226562,
1768
+ "step": 970
1769
+ },
1770
+ {
1771
+ "epoch": 0.842286205414697,
1772
+ "grad_norm": 2.845311403274536,
1773
+ "learning_rate": 1.3417599122003464e-06,
1774
+ "logits/chosen": 2.760969638824463,
1775
+ "logits/rejected": 2.9475345611572266,
1776
+ "logps/chosen": -3.6206259727478027,
1777
+ "logps/rejected": -9.263578414916992,
1778
+ "loss": 0.1765,
1779
+ "rewards/accuracies": 0.9624999761581421,
1780
+ "rewards/chosen": -5.430939674377441,
1781
+ "rewards/margins": 8.464428901672363,
1782
+ "rewards/rejected": -13.895367622375488,
1783
+ "step": 980
1784
+ },
1785
+ {
1786
+ "epoch": 0.8508809626128062,
1787
+ "grad_norm": 4.265077590942383,
1788
+ "learning_rate": 1.2956158147457116e-06,
1789
+ "logits/chosen": 3.060234546661377,
1790
+ "logits/rejected": 3.3040313720703125,
1791
+ "logps/chosen": -3.448155641555786,
1792
+ "logps/rejected": -8.33607292175293,
1793
+ "loss": 0.2154,
1794
+ "rewards/accuracies": 0.9375,
1795
+ "rewards/chosen": -5.172233581542969,
1796
+ "rewards/margins": 7.3318772315979,
1797
+ "rewards/rejected": -12.504110336303711,
1798
+ "step": 990
1799
+ },
1800
+ {
1801
+ "epoch": 0.8594757198109153,
1802
+ "grad_norm": 2.6619937419891357,
1803
+ "learning_rate": 1.2500000000000007e-06,
1804
+ "logits/chosen": 3.205899715423584,
1805
+ "logits/rejected": 3.2403626441955566,
1806
+ "logps/chosen": -2.9734084606170654,
1807
+ "logps/rejected": -7.3107008934021,
1808
+ "loss": 0.1983,
1809
+ "rewards/accuracies": 0.925000011920929,
1810
+ "rewards/chosen": -4.460112571716309,
1811
+ "rewards/margins": 6.5059404373168945,
1812
+ "rewards/rejected": -10.966052055358887,
1813
+ "step": 1000
1814
+ },
1815
+ {
1816
+ "epoch": 0.8594757198109153,
1817
+ "eval_logits/chosen": 3.0807368755340576,
1818
+ "eval_logits/rejected": 3.4346303939819336,
1819
+ "eval_logps/chosen": -3.660179376602173,
1820
+ "eval_logps/rejected": -9.305243492126465,
1821
+ "eval_loss": 0.2214924544095993,
1822
+ "eval_rewards/accuracies": 0.9263157844543457,
1823
+ "eval_rewards/chosen": -5.490269184112549,
1824
+ "eval_rewards/margins": 8.467597007751465,
1825
+ "eval_rewards/rejected": -13.957863807678223,
1826
+ "eval_runtime": 25.8036,
1827
+ "eval_samples_per_second": 29.182,
1828
+ "eval_steps_per_second": 3.682,
1829
+ "step": 1000
1830
+ },
1831
+ {
1832
+ "epoch": 0.8680704770090245,
1833
+ "grad_norm": 3.969745635986328,
1834
+ "learning_rate": 1.204932476567175e-06,
1835
+ "logits/chosen": 3.2750396728515625,
1836
+ "logits/rejected": 3.5089526176452637,
1837
+ "logps/chosen": -3.4845385551452637,
1838
+ "logps/rejected": -7.859654903411865,
1839
+ "loss": 0.2136,
1840
+ "rewards/accuracies": 0.949999988079071,
1841
+ "rewards/chosen": -5.226807594299316,
1842
+ "rewards/margins": 6.562674522399902,
1843
+ "rewards/rejected": -11.789482116699219,
1844
+ "step": 1010
1845
+ },
1846
+ {
1847
+ "epoch": 0.8766652342071336,
1848
+ "grad_norm": 10.868583679199219,
1849
+ "learning_rate": 1.160433012552508e-06,
1850
+ "logits/chosen": 3.5371367931365967,
1851
+ "logits/rejected": 3.6224236488342285,
1852
+ "logps/chosen": -3.4987919330596924,
1853
+ "logps/rejected": -8.714313507080078,
1854
+ "loss": 0.2506,
1855
+ "rewards/accuracies": 0.925000011920929,
1856
+ "rewards/chosen": -5.248188018798828,
1857
+ "rewards/margins": 7.8232831954956055,
1858
+ "rewards/rejected": -13.071470260620117,
1859
+ "step": 1020
1860
+ },
1861
+ {
1862
+ "epoch": 0.8852599914052428,
1863
+ "grad_norm": 13.34189510345459,
1864
+ "learning_rate": 1.11652112689164e-06,
1865
+ "logits/chosen": 3.0136570930480957,
1866
+ "logits/rejected": 3.227113723754883,
1867
+ "logps/chosen": -3.607630968093872,
1868
+ "logps/rejected": -8.083453178405762,
1869
+ "loss": 0.2267,
1870
+ "rewards/accuracies": 0.9125000238418579,
1871
+ "rewards/chosen": -5.411446571350098,
1872
+ "rewards/margins": 6.7137346267700195,
1873
+ "rewards/rejected": -12.125181198120117,
1874
+ "step": 1030
1875
+ },
1876
+ {
1877
+ "epoch": 0.8938547486033519,
1878
+ "grad_norm": 3.0865557193756104,
1879
+ "learning_rate": 1.073216080788921e-06,
1880
+ "logits/chosen": 3.20233154296875,
1881
+ "logits/rejected": 3.1444244384765625,
1882
+ "logps/chosen": -3.677372455596924,
1883
+ "logps/rejected": -8.034679412841797,
1884
+ "loss": 0.2019,
1885
+ "rewards/accuracies": 0.887499988079071,
1886
+ "rewards/chosen": -5.516058921813965,
1887
+ "rewards/margins": 6.535961151123047,
1888
+ "rewards/rejected": -12.052019119262695,
1889
+ "step": 1040
1890
+ },
1891
+ {
1892
+ "epoch": 0.9024495058014611,
1893
+ "grad_norm": 2.896090507507324,
1894
+ "learning_rate": 1.0305368692688175e-06,
1895
+ "logits/chosen": 2.8652000427246094,
1896
+ "logits/rejected": 3.20027494430542,
1897
+ "logps/chosen": -3.3209357261657715,
1898
+ "logps/rejected": -8.070123672485352,
1899
+ "loss": 0.191,
1900
+ "rewards/accuracies": 0.949999988079071,
1901
+ "rewards/chosen": -4.981403350830078,
1902
+ "rewards/margins": 7.123780727386475,
1903
+ "rewards/rejected": -12.105184555053711,
1904
+ "step": 1050
1905
+ },
1906
+ {
1907
+ "epoch": 0.9024495058014611,
1908
+ "eval_logits/chosen": 3.1477556228637695,
1909
+ "eval_logits/rejected": 3.4350297451019287,
1910
+ "eval_logps/chosen": -3.7262520790100098,
1911
+ "eval_logps/rejected": -9.287687301635742,
1912
+ "eval_loss": 0.21767112612724304,
1913
+ "eval_rewards/accuracies": 0.9368420839309692,
1914
+ "eval_rewards/chosen": -5.589378833770752,
1915
+ "eval_rewards/margins": 8.342151641845703,
1916
+ "eval_rewards/rejected": -13.931530952453613,
1917
+ "eval_runtime": 25.816,
1918
+ "eval_samples_per_second": 29.168,
1919
+ "eval_steps_per_second": 3.68,
1920
+ "step": 1050
1921
+ },
1922
+ {
1923
+ "epoch": 0.9110442629995703,
1924
+ "grad_norm": 5.839542865753174,
1925
+ "learning_rate": 9.88502212844063e-07,
1926
+ "logits/chosen": 3.563368320465088,
1927
+ "logits/rejected": 3.3060708045959473,
1928
+ "logps/chosen": -3.2863879203796387,
1929
+ "logps/rejected": -8.566315650939941,
1930
+ "loss": 0.2254,
1931
+ "rewards/accuracies": 0.987500011920929,
1932
+ "rewards/chosen": -4.929581642150879,
1933
+ "rewards/margins": 7.919892311096191,
1934
+ "rewards/rejected": -12.849472045898438,
1935
+ "step": 1060
1936
+ },
1937
+ {
1938
+ "epoch": 0.9196390201976794,
1939
+ "grad_norm": 4.452107906341553,
1940
+ "learning_rate": 9.471305493042243e-07,
1941
+ "logits/chosen": 2.7673869132995605,
1942
+ "logits/rejected": 3.075416088104248,
1943
+ "logps/chosen": -3.4145348072052,
1944
+ "logps/rejected": -8.784387588500977,
1945
+ "loss": 0.1987,
1946
+ "rewards/accuracies": 0.9624999761581421,
1947
+ "rewards/chosen": -5.12180233001709,
1948
+ "rewards/margins": 8.054778099060059,
1949
+ "rewards/rejected": -13.176579475402832,
1950
+ "step": 1070
1951
+ },
1952
+ {
1953
+ "epoch": 0.9282337773957886,
1954
+ "grad_norm": 3.9341001510620117,
1955
+ "learning_rate": 9.064400256282757e-07,
1956
+ "logits/chosen": 2.969014883041382,
1957
+ "logits/rejected": 3.198265790939331,
1958
+ "logps/chosen": -3.6482291221618652,
1959
+ "logps/rejected": -9.581583023071289,
1960
+ "loss": 0.1973,
1961
+ "rewards/accuracies": 0.9624999761581421,
1962
+ "rewards/chosen": -5.472344398498535,
1963
+ "rewards/margins": 8.900031089782715,
1964
+ "rewards/rejected": -14.37237548828125,
1965
+ "step": 1080
1966
+ },
1967
+ {
1968
+ "epoch": 0.9368285345938977,
1969
+ "grad_norm": 3.811671733856201,
1970
+ "learning_rate": 8.664484900247363e-07,
1971
+ "logits/chosen": 3.182769298553467,
1972
+ "logits/rejected": 3.1134133338928223,
1973
+ "logps/chosen": -3.3463242053985596,
1974
+ "logps/rejected": -8.769086837768555,
1975
+ "loss": 0.2126,
1976
+ "rewards/accuracies": 0.949999988079071,
1977
+ "rewards/chosen": -5.019486427307129,
1978
+ "rewards/margins": 8.134143829345703,
1979
+ "rewards/rejected": -13.153630256652832,
1980
+ "step": 1090
1981
+ },
1982
+ {
1983
+ "epoch": 0.9454232917920069,
1984
+ "grad_norm": 3.094046115875244,
1985
+ "learning_rate": 8.271734841028553e-07,
1986
+ "logits/chosen": 3.4017093181610107,
1987
+ "logits/rejected": 3.578439235687256,
1988
+ "logps/chosen": -3.4635825157165527,
1989
+ "logps/rejected": -8.677824020385742,
1990
+ "loss": 0.1904,
1991
+ "rewards/accuracies": 0.9624999761581421,
1992
+ "rewards/chosen": -5.195374011993408,
1993
+ "rewards/margins": 7.821363925933838,
1994
+ "rewards/rejected": -13.01673698425293,
1995
+ "step": 1100
1996
+ },
1997
+ {
1998
+ "epoch": 0.9454232917920069,
1999
+ "eval_logits/chosen": 3.0800373554229736,
2000
+ "eval_logits/rejected": 3.422905921936035,
2001
+ "eval_logps/chosen": -3.708962917327881,
2002
+ "eval_logps/rejected": -9.360494613647461,
2003
+ "eval_loss": 0.21661873161792755,
2004
+ "eval_rewards/accuracies": 0.9368420839309692,
2005
+ "eval_rewards/chosen": -5.563443660736084,
2006
+ "eval_rewards/margins": 8.47729778289795,
2007
+ "eval_rewards/rejected": -14.040741920471191,
2008
+ "eval_runtime": 25.8111,
2009
+ "eval_samples_per_second": 29.174,
2010
+ "eval_steps_per_second": 3.681,
2011
+ "step": 1100
2012
+ },
2013
+ {
2014
+ "epoch": 0.954018048990116,
2015
+ "grad_norm": 4.002285003662109,
2016
+ "learning_rate": 7.886322351782782e-07,
2017
+ "logits/chosen": 2.8665084838867188,
2018
+ "logits/rejected": 3.0208096504211426,
2019
+ "logps/chosen": -3.6410727500915527,
2020
+ "logps/rejected": -8.900314331054688,
2021
+ "loss": 0.1988,
2022
+ "rewards/accuracies": 0.9624999761581421,
2023
+ "rewards/chosen": -5.461609363555908,
2024
+ "rewards/margins": 7.888862609863281,
2025
+ "rewards/rejected": -13.350473403930664,
2026
+ "step": 1110
2027
+ },
2028
+ {
2029
+ "epoch": 0.9626128061882252,
2030
+ "grad_norm": 2.3234140872955322,
2031
+ "learning_rate": 7.508416487165862e-07,
2032
+ "logits/chosen": 2.740455389022827,
2033
+ "logits/rejected": 3.0439486503601074,
2034
+ "logps/chosen": -3.088730573654175,
2035
+ "logps/rejected": -8.612493515014648,
2036
+ "loss": 0.1683,
2037
+ "rewards/accuracies": 0.949999988079071,
2038
+ "rewards/chosen": -4.633096218109131,
2039
+ "rewards/margins": 8.28564453125,
2040
+ "rewards/rejected": -12.918741226196289,
2041
+ "step": 1120
2042
+ },
2043
+ {
2044
+ "epoch": 0.9712075633863343,
2045
+ "grad_norm": 4.9889984130859375,
2046
+ "learning_rate": 7.138183009179922e-07,
2047
+ "logits/chosen": 3.4702610969543457,
2048
+ "logits/rejected": 3.582740068435669,
2049
+ "logps/chosen": -3.6305007934570312,
2050
+ "logps/rejected": -9.135300636291504,
2051
+ "loss": 0.1859,
2052
+ "rewards/accuracies": 0.8999999761581421,
2053
+ "rewards/chosen": -5.445751190185547,
2054
+ "rewards/margins": 8.257198333740234,
2055
+ "rewards/rejected": -13.702949523925781,
2056
+ "step": 1130
2057
+ },
2058
+ {
2059
+ "epoch": 0.9798023205844435,
2060
+ "grad_norm": 4.3878254890441895,
2061
+ "learning_rate": 6.775784314464717e-07,
2062
+ "logits/chosen": 3.0020623207092285,
2063
+ "logits/rejected": 3.030848503112793,
2064
+ "logps/chosen": -3.347019910812378,
2065
+ "logps/rejected": -8.666067123413086,
2066
+ "loss": 0.175,
2067
+ "rewards/accuracies": 0.9375,
2068
+ "rewards/chosen": -5.020529747009277,
2069
+ "rewards/margins": 7.978570461273193,
2070
+ "rewards/rejected": -12.999099731445312,
2071
+ "step": 1140
2072
+ },
2073
+ {
2074
+ "epoch": 0.9883970777825526,
2075
+ "grad_norm": 2.510340929031372,
2076
+ "learning_rate": 6.421379363065142e-07,
2077
+ "logits/chosen": 3.489284038543701,
2078
+ "logits/rejected": 3.555269956588745,
2079
+ "logps/chosen": -3.5305256843566895,
2080
+ "logps/rejected": -8.848383903503418,
2081
+ "loss": 0.1612,
2082
+ "rewards/accuracies": 0.9624999761581421,
2083
+ "rewards/chosen": -5.295788764953613,
2084
+ "rewards/margins": 7.976788520812988,
2085
+ "rewards/rejected": -13.272577285766602,
2086
+ "step": 1150
2087
+ },
2088
+ {
2089
+ "epoch": 0.9883970777825526,
2090
+ "eval_logits/chosen": 3.0948123931884766,
2091
+ "eval_logits/rejected": 3.442106008529663,
2092
+ "eval_logps/chosen": -3.807739496231079,
2093
+ "eval_logps/rejected": -9.483173370361328,
2094
+ "eval_loss": 0.2131287306547165,
2095
+ "eval_rewards/accuracies": 0.9473684430122375,
2096
+ "eval_rewards/chosen": -5.711609363555908,
2097
+ "eval_rewards/margins": 8.513151168823242,
2098
+ "eval_rewards/rejected": -14.224760055541992,
2099
+ "eval_runtime": 25.8752,
2100
+ "eval_samples_per_second": 29.101,
2101
+ "eval_steps_per_second": 3.671,
2102
+ "step": 1150
2103
+ },
2104
+ {
2105
+ "epoch": 0.9969918349806618,
2106
+ "grad_norm": 4.808183193206787,
2107
+ "learning_rate": 6.075123608706093e-07,
2108
+ "logits/chosen": 3.347472667694092,
2109
+ "logits/rejected": 3.43141508102417,
2110
+ "logps/chosen": -3.588602066040039,
2111
+ "logps/rejected": -8.802007675170898,
2112
+ "loss": 0.22,
2113
+ "rewards/accuracies": 0.9375,
2114
+ "rewards/chosen": -5.382903099060059,
2115
+ "rewards/margins": 7.820107460021973,
2116
+ "rewards/rejected": -13.203012466430664,
2117
+ "step": 1160
2118
+ },
2119
+ {
2120
+ "epoch": 1.0051568543188656,
2121
+ "grad_norm": 4.898995399475098,
2122
+ "learning_rate": 5.737168930605272e-07,
2123
+ "logits/chosen": 3.6856889724731445,
2124
+ "logits/rejected": 3.534011125564575,
2125
+ "logps/chosen": -3.324617862701416,
2126
+ "logps/rejected": -8.283414840698242,
2127
+ "loss": 0.1738,
2128
+ "rewards/accuracies": 0.9605262875556946,
2129
+ "rewards/chosen": -4.986927032470703,
2130
+ "rewards/margins": 7.438195705413818,
2131
+ "rewards/rejected": -12.425122261047363,
2132
+ "step": 1170
2133
+ },
2134
+ {
2135
+ "epoch": 1.0137516115169747,
2136
+ "grad_norm": 4.611911773681641,
2137
+ "learning_rate": 5.407663566854008e-07,
2138
+ "logits/chosen": 2.828855276107788,
2139
+ "logits/rejected": 3.037381172180176,
2140
+ "logps/chosen": -3.393272876739502,
2141
+ "logps/rejected": -8.736349105834961,
2142
+ "loss": 0.2176,
2143
+ "rewards/accuracies": 0.9624999761581421,
2144
+ "rewards/chosen": -5.089909553527832,
2145
+ "rewards/margins": 8.014615058898926,
2146
+ "rewards/rejected": -13.104524612426758,
2147
+ "step": 1180
2148
+ },
2149
+ {
2150
+ "epoch": 1.0223463687150838,
2151
+ "grad_norm": 1.748143196105957,
2152
+ "learning_rate": 5.086752049395094e-07,
2153
+ "logits/chosen": 3.6575450897216797,
2154
+ "logits/rejected": 3.499067783355713,
2155
+ "logps/chosen": -3.4004104137420654,
2156
+ "logps/rejected": -8.438041687011719,
2157
+ "loss": 0.1689,
2158
+ "rewards/accuracies": 0.9624999761581421,
2159
+ "rewards/chosen": -5.100615501403809,
2160
+ "rewards/margins": 7.5564470291137695,
2161
+ "rewards/rejected": -12.657061576843262,
2162
+ "step": 1190
2163
+ },
2164
+ {
2165
+ "epoch": 1.0309411259131929,
2166
+ "grad_norm": 2.34837007522583,
2167
+ "learning_rate": 4.774575140626317e-07,
2168
+ "logits/chosen": 3.2288806438446045,
2169
+ "logits/rejected": 3.2293314933776855,
2170
+ "logps/chosen": -3.5406124591827393,
2171
+ "logps/rejected": -7.73020076751709,
2172
+ "loss": 0.2052,
2173
+ "rewards/accuracies": 0.9125000238418579,
2174
+ "rewards/chosen": -5.310919284820557,
2175
+ "rewards/margins": 6.284381866455078,
2176
+ "rewards/rejected": -11.595300674438477,
2177
+ "step": 1200
2178
+ },
2179
+ {
2180
+ "epoch": 1.0309411259131929,
2181
+ "eval_logits/chosen": 3.0892465114593506,
2182
+ "eval_logits/rejected": 3.4358115196228027,
2183
+ "eval_logps/chosen": -3.6969687938690186,
2184
+ "eval_logps/rejected": -9.432202339172363,
2185
+ "eval_loss": 0.2119421511888504,
2186
+ "eval_rewards/accuracies": 0.9473684430122375,
2187
+ "eval_rewards/chosen": -5.54545259475708,
2188
+ "eval_rewards/margins": 8.602849960327148,
2189
+ "eval_rewards/rejected": -14.148303031921387,
2190
+ "eval_runtime": 25.8416,
2191
+ "eval_samples_per_second": 29.139,
2192
+ "eval_steps_per_second": 3.676,
2193
+ "step": 1200
2194
+ }
2195
+ ],
2196
+ "logging_steps": 10,
2197
+ "max_steps": 1500,
2198
+ "num_input_tokens_seen": 0,
2199
+ "num_train_epochs": 2,
2200
+ "save_steps": 50,
2201
+ "stateful_callbacks": {
2202
+ "TrainerControl": {
2203
+ "args": {
2204
+ "should_epoch_stop": false,
2205
+ "should_evaluate": false,
2206
+ "should_log": false,
2207
+ "should_save": true,
2208
+ "should_training_stop": false
2209
+ },
2210
+ "attributes": {}
2211
+ }
2212
+ },
2213
+ "total_flos": 2.732990299360985e+18,
2214
+ "train_batch_size": 1,
2215
+ "trial_name": null,
2216
+ "trial_params": null
2217
+ }
checkpoint-1200/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2cc4cd97bd7e66b9f8e8e61852db449abcbbbd1c049a5ad37ec259f9f39597e8
3
+ size 7224
checkpoint-1200/zero_to_fp32.py ADDED
@@ -0,0 +1,674 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 json
25
+ from tqdm import tqdm
26
+ from collections import OrderedDict
27
+ from dataclasses import dataclass
28
+
29
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
30
+ # DeepSpeed data structures it has to be available in the current python environment.
31
+ from deepspeed.utils import logger
32
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
33
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
34
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
35
+
36
+
37
+ @dataclass
38
+ class zero_model_state:
39
+ buffers: dict()
40
+ param_shapes: dict()
41
+ shared_params: list
42
+ ds_version: int
43
+ frozen_param_shapes: dict()
44
+ frozen_param_fragments: dict()
45
+
46
+
47
+ debug = 0
48
+
49
+ # load to cpu
50
+ device = torch.device('cpu')
51
+
52
+
53
+ def atoi(text):
54
+ return int(text) if text.isdigit() else text
55
+
56
+
57
+ def natural_keys(text):
58
+ '''
59
+ alist.sort(key=natural_keys) sorts in human order
60
+ http://nedbatchelder.com/blog/200712/human_sorting.html
61
+ (See Toothy's implementation in the comments)
62
+ '''
63
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
64
+
65
+
66
+ def get_model_state_file(checkpoint_dir, zero_stage):
67
+ if not os.path.isdir(checkpoint_dir):
68
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
69
+
70
+ # there should be only one file
71
+ if zero_stage <= 2:
72
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
73
+ elif zero_stage == 3:
74
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
75
+
76
+ if not os.path.exists(file):
77
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
78
+
79
+ return file
80
+
81
+
82
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
83
+ # XXX: need to test that this simple glob rule works for multi-node setup too
84
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
85
+
86
+ if len(ckpt_files) == 0:
87
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
88
+
89
+ return ckpt_files
90
+
91
+
92
+ def get_optim_files(checkpoint_dir):
93
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
94
+
95
+
96
+ def get_model_state_files(checkpoint_dir):
97
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
98
+
99
+
100
+ def parse_model_states(files):
101
+ zero_model_states = []
102
+ for file in files:
103
+ state_dict = torch.load(file, map_location=device)
104
+
105
+ if BUFFER_NAMES not in state_dict:
106
+ raise ValueError(f"{file} is not a model state checkpoint")
107
+ buffer_names = state_dict[BUFFER_NAMES]
108
+ if debug:
109
+ print("Found buffers:", buffer_names)
110
+
111
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
112
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
113
+ param_shapes = state_dict[PARAM_SHAPES]
114
+
115
+ # collect parameters that are included in param_shapes
116
+ param_names = []
117
+ for s in param_shapes:
118
+ for name in s.keys():
119
+ param_names.append(name)
120
+
121
+ # update with frozen parameters
122
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
123
+ if frozen_param_shapes is not None:
124
+ if debug:
125
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
126
+ param_names += list(frozen_param_shapes.keys())
127
+
128
+ # handle shared params
129
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
130
+
131
+ ds_version = state_dict.get(DS_VERSION, None)
132
+
133
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
134
+
135
+ z_model_state = zero_model_state(buffers=buffers,
136
+ param_shapes=param_shapes,
137
+ shared_params=shared_params,
138
+ ds_version=ds_version,
139
+ frozen_param_shapes=frozen_param_shapes,
140
+ frozen_param_fragments=frozen_param_fragments)
141
+ zero_model_states.append(z_model_state)
142
+
143
+ return zero_model_states
144
+
145
+
146
+ def parse_optim_states(files, ds_checkpoint_dir):
147
+ total_files = len(files)
148
+ state_dicts = []
149
+ for f in files:
150
+ state_dict = torch.load(f, map_location=device)
151
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
152
+ # and also handle the case where it was already removed by another helper script
153
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
154
+ state_dicts.append(state_dict)
155
+
156
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
157
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
158
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
159
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
160
+
161
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
162
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
163
+ # use the max of the partition_count to get the dp world_size.
164
+
165
+ if type(world_size) is list:
166
+ world_size = max(world_size)
167
+
168
+ if world_size != total_files:
169
+ raise ValueError(
170
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
171
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
172
+ )
173
+
174
+ # the groups are named differently in each stage
175
+ if zero_stage <= 2:
176
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
177
+ elif zero_stage == 3:
178
+ fp32_groups_key = FP32_FLAT_GROUPS
179
+ else:
180
+ raise ValueError(f"unknown zero stage {zero_stage}")
181
+
182
+ if zero_stage <= 2:
183
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
184
+ elif zero_stage == 3:
185
+ # if there is more than one param group, there will be multiple flattened tensors - one
186
+ # flattened tensor per group - for simplicity merge them into a single tensor
187
+ #
188
+ # XXX: could make the script more memory efficient for when there are multiple groups - it
189
+ # will require matching the sub-lists of param_shapes for each param group flattened tensor
190
+
191
+ fp32_flat_groups = [
192
+ torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
193
+ ]
194
+
195
+ return zero_stage, world_size, fp32_flat_groups
196
+
197
+
198
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
199
+ """
200
+ Returns fp32 state_dict reconstructed from ds checkpoint
201
+
202
+ Args:
203
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
204
+
205
+ """
206
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
207
+
208
+ optim_files = get_optim_files(ds_checkpoint_dir)
209
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
210
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
211
+
212
+ model_files = get_model_state_files(ds_checkpoint_dir)
213
+
214
+ zero_model_states = parse_model_states(model_files)
215
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
216
+
217
+ if zero_stage <= 2:
218
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
219
+ exclude_frozen_parameters)
220
+ elif zero_stage == 3:
221
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
222
+ exclude_frozen_parameters)
223
+
224
+
225
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
226
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
227
+ return
228
+
229
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
230
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
231
+
232
+ if debug:
233
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
234
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
235
+
236
+ wanted_params = len(frozen_param_shapes)
237
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
238
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
239
+ print(f'Frozen params: Have {avail_numel} numels to process.')
240
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
241
+
242
+ total_params = 0
243
+ total_numel = 0
244
+ for name, shape in frozen_param_shapes.items():
245
+ total_params += 1
246
+ unpartitioned_numel = shape.numel()
247
+ total_numel += unpartitioned_numel
248
+
249
+ state_dict[name] = frozen_param_fragments[name]
250
+
251
+ if debug:
252
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
253
+
254
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
255
+
256
+
257
+ def _has_callable(obj, fn):
258
+ attr = getattr(obj, fn, None)
259
+ return callable(attr)
260
+
261
+
262
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
263
+ param_shapes = zero_model_states[0].param_shapes
264
+
265
+ # Reconstruction protocol:
266
+ #
267
+ # XXX: document this
268
+
269
+ if debug:
270
+ for i in range(world_size):
271
+ for j in range(len(fp32_flat_groups[0])):
272
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
273
+
274
+ # XXX: memory usage doubles here (zero2)
275
+ num_param_groups = len(fp32_flat_groups[0])
276
+ merged_single_partition_of_fp32_groups = []
277
+ for i in range(num_param_groups):
278
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
279
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
280
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
281
+ avail_numel = sum(
282
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
283
+
284
+ if debug:
285
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
286
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
287
+ # not asserting if there is a mismatch due to possible padding
288
+ print(f"Have {avail_numel} numels to process.")
289
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
290
+
291
+ # params
292
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
293
+ # out-of-core computing solution
294
+ total_numel = 0
295
+ total_params = 0
296
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
297
+ offset = 0
298
+ avail_numel = full_single_fp32_vector.numel()
299
+ for name, shape in shapes.items():
300
+
301
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
302
+ total_numel += unpartitioned_numel
303
+ total_params += 1
304
+
305
+ if debug:
306
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
307
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
308
+ offset += unpartitioned_numel
309
+
310
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
311
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
312
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
313
+ # live optimizer object, so we are checking that the numbers are within the right range
314
+ align_to = 2 * world_size
315
+
316
+ def zero2_align(x):
317
+ return align_to * math.ceil(x / align_to)
318
+
319
+ if debug:
320
+ print(f"original offset={offset}, avail_numel={avail_numel}")
321
+
322
+ offset = zero2_align(offset)
323
+ avail_numel = zero2_align(avail_numel)
324
+
325
+ if debug:
326
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
327
+
328
+ # Sanity check
329
+ if offset != avail_numel:
330
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
331
+
332
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
333
+
334
+
335
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
336
+ exclude_frozen_parameters):
337
+ state_dict = OrderedDict()
338
+
339
+ # buffers
340
+ buffers = zero_model_states[0].buffers
341
+ state_dict.update(buffers)
342
+ if debug:
343
+ print(f"added {len(buffers)} buffers")
344
+
345
+ if not exclude_frozen_parameters:
346
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
347
+
348
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
349
+
350
+ # recover shared parameters
351
+ for pair in zero_model_states[0].shared_params:
352
+ if pair[1] in state_dict:
353
+ state_dict[pair[0]] = state_dict[pair[1]]
354
+
355
+ return state_dict
356
+
357
+
358
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
359
+ remainder = unpartitioned_numel % world_size
360
+ padding_numel = (world_size - remainder) if remainder else 0
361
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
362
+ return partitioned_numel, padding_numel
363
+
364
+
365
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
366
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
367
+ return
368
+
369
+ if debug:
370
+ for i in range(world_size):
371
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
372
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
373
+
374
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
375
+ wanted_params = len(frozen_param_shapes)
376
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
377
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
378
+ print(f'Frozen params: Have {avail_numel} numels to process.')
379
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
380
+
381
+ total_params = 0
382
+ total_numel = 0
383
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
384
+ total_params += 1
385
+ unpartitioned_numel = shape.numel()
386
+ total_numel += unpartitioned_numel
387
+
388
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
389
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
390
+
391
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
392
+
393
+ if debug:
394
+ print(
395
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
396
+ )
397
+
398
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
399
+
400
+
401
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
402
+ param_shapes = zero_model_states[0].param_shapes
403
+ avail_numel = fp32_flat_groups[0].numel() * world_size
404
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
405
+ # param, re-consolidating each param, while dealing with padding if any
406
+
407
+ # merge list of dicts, preserving order
408
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
409
+
410
+ if debug:
411
+ for i in range(world_size):
412
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
413
+
414
+ wanted_params = len(param_shapes)
415
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
416
+ # not asserting if there is a mismatch due to possible padding
417
+ avail_numel = fp32_flat_groups[0].numel() * world_size
418
+ print(f"Trainable params: Have {avail_numel} numels to process.")
419
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
420
+
421
+ # params
422
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
423
+ # out-of-core computing solution
424
+ offset = 0
425
+ total_numel = 0
426
+ total_params = 0
427
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
428
+ unpartitioned_numel = shape.numel()
429
+ total_numel += unpartitioned_numel
430
+ total_params += 1
431
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
432
+
433
+ if debug:
434
+ print(
435
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
436
+ )
437
+
438
+ # XXX: memory usage doubles here
439
+ state_dict[name] = torch.cat(
440
+ tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
441
+ 0).narrow(0, 0, unpartitioned_numel).view(shape)
442
+ offset += partitioned_numel
443
+
444
+ offset *= world_size
445
+
446
+ # Sanity check
447
+ if offset != avail_numel:
448
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
449
+
450
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
451
+
452
+
453
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
454
+ exclude_frozen_parameters):
455
+ state_dict = OrderedDict()
456
+
457
+ # buffers
458
+ buffers = zero_model_states[0].buffers
459
+ state_dict.update(buffers)
460
+ if debug:
461
+ print(f"added {len(buffers)} buffers")
462
+
463
+ if not exclude_frozen_parameters:
464
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
465
+
466
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
467
+
468
+ # recover shared parameters
469
+ for pair in zero_model_states[0].shared_params:
470
+ if pair[1] in state_dict:
471
+ state_dict[pair[0]] = state_dict[pair[1]]
472
+
473
+ return state_dict
474
+
475
+
476
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
477
+ """
478
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
479
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
480
+ via a model hub.
481
+
482
+ Args:
483
+ - ``checkpoint_dir``: path to the desired checkpoint folder
484
+ - ``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``
485
+ - ``exclude_frozen_parameters``: exclude frozen parameters
486
+
487
+ Returns:
488
+ - pytorch ``state_dict``
489
+
490
+ Note: this approach may not work if your application doesn't have sufficient free CPU memory and
491
+ you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
492
+ the checkpoint.
493
+
494
+ A typical usage might be ::
495
+
496
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
497
+ # do the training and checkpoint saving
498
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
499
+ model = model.cpu() # move to cpu
500
+ model.load_state_dict(state_dict)
501
+ # submit to model hub or save the model to share with others
502
+
503
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
504
+ application. i.e. you will need to re-initialize the deepspeed engine, since
505
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
506
+
507
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
508
+
509
+ """
510
+ if tag is None:
511
+ latest_path = os.path.join(checkpoint_dir, 'latest')
512
+ if os.path.isfile(latest_path):
513
+ with open(latest_path, 'r') as fd:
514
+ tag = fd.read().strip()
515
+ else:
516
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
517
+
518
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
519
+
520
+ if not os.path.isdir(ds_checkpoint_dir):
521
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
522
+
523
+ return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
524
+
525
+
526
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
527
+ output_dir,
528
+ max_shard_size="5GB",
529
+ safe_serialization=False,
530
+ tag=None,
531
+ exclude_frozen_parameters=False):
532
+ """
533
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
534
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
535
+
536
+ Args:
537
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
538
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
539
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
540
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
541
+ - ``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``
542
+ - ``exclude_frozen_parameters``: exclude frozen parameters
543
+ """
544
+ # Dependency pre-check
545
+ if safe_serialization:
546
+ try:
547
+ from safetensors.torch import save_file
548
+ except ImportError:
549
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
550
+ raise
551
+ if max_shard_size is not None:
552
+ try:
553
+ from huggingface_hub import split_torch_state_dict_into_shards
554
+ except ImportError:
555
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
556
+ raise
557
+
558
+ # Convert zero checkpoint to state_dict
559
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
560
+
561
+ # Shard the model if it is too big.
562
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
563
+ if max_shard_size is not None:
564
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
565
+ state_dict_split = split_torch_state_dict_into_shards(state_dict,
566
+ filename_pattern=filename_pattern,
567
+ max_shard_size=max_shard_size)
568
+ else:
569
+ from collections import namedtuple
570
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
571
+ state_dict_split = StateDictSplit(is_sharded=False,
572
+ filename_to_tensors={weights_name: list(state_dict.keys())})
573
+
574
+ # Save the model
575
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
576
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
577
+ shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
578
+ output_path = os.path.join(output_dir, shard_file)
579
+ if safe_serialization:
580
+ save_file(shard, output_path, metadata={"format": "pt"})
581
+ else:
582
+ torch.save(shard, output_path)
583
+
584
+ # Save index if sharded
585
+ if state_dict_split.is_sharded:
586
+ index = {
587
+ "metadata": state_dict_split.metadata,
588
+ "weight_map": state_dict_split.tensor_to_filename,
589
+ }
590
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
591
+ save_index_file = os.path.join(output_dir, save_index_file)
592
+ with open(save_index_file, "w", encoding="utf-8") as f:
593
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
594
+ f.write(content)
595
+
596
+
597
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
598
+ """
599
+ 1. Put the provided model to cpu
600
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
601
+ 3. Load it into the provided model
602
+
603
+ Args:
604
+ - ``model``: the model object to update
605
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
606
+ - ``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``
607
+
608
+ Returns:
609
+ - ``model`: modified model
610
+
611
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
612
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
613
+ conveniently placed for you in the checkpoint folder.
614
+
615
+ A typical usage might be ::
616
+
617
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
618
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
619
+ # submit to model hub or save the model to share with others
620
+
621
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
622
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
623
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
624
+
625
+ """
626
+ logger.info(f"Extracting fp32 weights")
627
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
628
+
629
+ logger.info(f"Overwriting model with fp32 weights")
630
+ model = model.cpu()
631
+ model.load_state_dict(state_dict, strict=False)
632
+
633
+ return model
634
+
635
+
636
+ if __name__ == "__main__":
637
+ parser = argparse.ArgumentParser()
638
+ parser.add_argument("checkpoint_dir",
639
+ type=str,
640
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
641
+ parser.add_argument("output_dir",
642
+ type=str,
643
+ help="directory to the pytorch fp32 state_dict output files"
644
+ "(e.g. path/checkpoint-12-output/)")
645
+ parser.add_argument(
646
+ "--max_shard_size",
647
+ type=str,
648
+ default="5GB",
649
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
650
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
651
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
652
+ "without CPU OOM issues.")
653
+ parser.add_argument(
654
+ "--safe_serialization",
655
+ default=False,
656
+ action='store_true',
657
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
658
+ parser.add_argument("-t",
659
+ "--tag",
660
+ type=str,
661
+ default=None,
662
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
663
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
664
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
665
+ args = parser.parse_args()
666
+
667
+ debug = args.debug
668
+
669
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
670
+ args.output_dir,
671
+ max_shard_size=args.max_shard_size,
672
+ safe_serialization=args.safe_serialization,
673
+ tag=args.tag,
674
+ exclude_frozen_parameters=args.exclude_frozen_parameters)