Hugging Face's logo

minhaeoh
/
diffu-cd

Safetensors
Model card Files
xet
 Community
minhaeoh commited on
Commit
aab66ab
·
verified ·
1 Parent(s): edaca50

Upload checkpoint from unmask_tags_math_self_distill_INP_PAR_u0.001-1.0_gold1_target1_ce0.0

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +1 -0
  2. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/README.md +202 -0
  3. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/adapter_config.json +39 -0
  4. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/adapter_model.safetensors +3 -0
  5. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/optimizer.pt +3 -0
  6. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/rng_state_0.pth +3 -0
  7. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/rng_state_1.pth +3 -0
  8. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/scheduler.pt +3 -0
  9. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/trainer_state.json +273 -0
  10. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/training_args.bin +3 -0
  11. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/README.md +202 -0
  12. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/adapter_config.json +39 -0
  13. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/adapter_model.safetensors +3 -0
  14. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/optimizer.pt +3 -0
  15. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/rng_state_0.pth +3 -0
  16. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/rng_state_1.pth +3 -0
  17. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/scheduler.pt +3 -0
  18. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/trainer_state.json +2433 -0
  19. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/training_args.bin +3 -0
  20. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/README.md +202 -0
  21. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/adapter_config.json +39 -0
  22. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/adapter_model.safetensors +3 -0
  23. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/optimizer.pt +3 -0
  24. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/rng_state_0.pth +3 -0
  25. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/rng_state_1.pth +3 -0
  26. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/scheduler.pt +3 -0
  27. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/trainer_state.json +2673 -0
  28. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/training_args.bin +3 -0
  29. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/README.md +202 -0
  30. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/adapter_config.json +39 -0
  31. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/adapter_model.safetensors +3 -0
  32. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/optimizer.pt +3 -0
  33. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/rng_state_0.pth +3 -0
  34. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/rng_state_1.pth +3 -0
  35. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/scheduler.pt +3 -0
  36. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/trainer_state.json +2913 -0
  37. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/training_args.bin +3 -0
  38. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/README.md +202 -0
  39. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/adapter_config.json +39 -0
  40. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/adapter_model.safetensors +3 -0
  41. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/optimizer.pt +3 -0
  42. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/rng_state_0.pth +3 -0
  43. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/rng_state_1.pth +3 -0
  44. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/scheduler.pt +3 -0
  45. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/trainer_state.json +513 -0
  46. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/training_args.bin +3 -0
  47. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/README.md +202 -0
  48. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/adapter_config.json +39 -0
  49. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/adapter_model.safetensors +3 -0
  50. math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/optimizer.pt +3 -0
.gitattributes CHANGED
@@ -40,3 +40,4 @@ math/inp/SD-INP/math_self_distill_INP_u0.8-1.0_gold1_target1_ce0.5/debug_trainin
40
  math/SFT/inp-onehot_gold1_target1_ce0.5/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
41
  math/INP-PAR-REVERSE/unmask_tags_gold1_target1_ce0.5/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
42
  math/INP-PAR-REVERSE/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
40
  math/SFT/inp-onehot_gold1_target1_ce0.5/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
41
  math/INP-PAR-REVERSE/unmask_tags_gold1_target1_ce0.5/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
42
  math/INP-PAR-REVERSE/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
43
+ math/INP-PAR/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: GSAI-ML/LLaDA-8B-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 0.15.1
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/adapter_config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "GSAI-ML/LLaDA-8B-Instruct",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 64,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "r": 128,
24
+ "rank_pattern": {},
25
+ "revision": null,
26
+ "target_modules": [
27
+ "down_proj",
28
+ "up_proj",
29
+ "gate_proj",
30
+ "v_proj",
31
+ "k_proj",
32
+ "o_proj",
33
+ "q_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e0c745ab29c29d058af475735923aebe2c8f1ad9261e13068a1445d99180deb
3
+ size 2406624648
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:431b247c6736b42b33ac8dea459f3c87476c22b983a06f88c6df8f6fe05384ec
3
+ size 671304442
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:64fc209e4f8339f7f64f09116abc3335d3843ca29d9410394ed6bfd3431afc50
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27f1021fb57112918a3a6091b09b0ccd50cb071a2324c12ae9afcc9851ee8bd3
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d8cdfceac9f7917b978dca661a3b8e04187faea5d5f6bd7b462d61d8234d57f
3
+ size 1064
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/trainer_state.json ADDED
@@ -0,0 +1,273 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.21333333333333335,
5
+ "eval_steps": 500,
6
+ "global_step": 100,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "avg_mask_ratio": 0.4933756332669873,
13
+ "avg_response_length": 252.0625,
14
+ "avg_student_mask_ratio": 0.4933756332669873,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 0.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 1.0,
19
+ "batch_inp_par_reverse_frac": 0.0,
20
+ "batch_rl_frac": 0.0,
21
+ "batch_sft_frac": 0.0,
22
+ "batch_soft_sft_frac": 0.0,
23
+ "batch_tf_frac": 0.0,
24
+ "ce_loss": 0.0,
25
+ "epoch": 0.021333333333333333,
26
+ "grad_norm": 0.1650390625,
27
+ "kd_loss": 0.24795629351958723,
28
+ "learning_rate": 3e-06,
29
+ "loss": 0.2758,
30
+ "masked_tokens": 120.975,
31
+ "mean_t": 0.5145528071501758,
32
+ "step": 10,
33
+ "student_masked_tokens": 120.975
34
+ },
35
+ {
36
+ "avg_mask_ratio": 0.41923869140446185,
37
+ "avg_response_length": 221.7125,
38
+ "avg_student_mask_ratio": 0.41923869140446185,
39
+ "batch_ainp_frac": 0.0,
40
+ "batch_inp_frac": 0.0,
41
+ "batch_inp_oh_frac": 0.0,
42
+ "batch_inp_par_frac": 1.0,
43
+ "batch_inp_par_reverse_frac": 0.0,
44
+ "batch_rl_frac": 0.0,
45
+ "batch_sft_frac": 0.0,
46
+ "batch_soft_sft_frac": 0.0,
47
+ "batch_tf_frac": 0.0,
48
+ "ce_loss": 0.0,
49
+ "epoch": 0.042666666666666665,
50
+ "grad_norm": 0.388671875,
51
+ "kd_loss": 0.21509853232191353,
52
+ "learning_rate": 3e-06,
53
+ "loss": 0.2046,
54
+ "masked_tokens": 86.1875,
55
+ "mean_t": 0.43874448732240123,
56
+ "step": 20,
57
+ "student_masked_tokens": 86.1875
58
+ },
59
+ {
60
+ "avg_mask_ratio": 0.4542569225654006,
61
+ "avg_response_length": 231.45,
62
+ "avg_student_mask_ratio": 0.4542569225654006,
63
+ "batch_ainp_frac": 0.0,
64
+ "batch_inp_frac": 0.0,
65
+ "batch_inp_oh_frac": 0.0,
66
+ "batch_inp_par_frac": 1.0,
67
+ "batch_inp_par_reverse_frac": 0.0,
68
+ "batch_rl_frac": 0.0,
69
+ "batch_sft_frac": 0.0,
70
+ "batch_soft_sft_frac": 0.0,
71
+ "batch_tf_frac": 0.0,
72
+ "ce_loss": 0.0,
73
+ "epoch": 0.064,
74
+ "grad_norm": 0.52734375,
75
+ "kd_loss": 0.19756361616970963,
76
+ "learning_rate": 3e-06,
77
+ "loss": 0.1976,
78
+ "masked_tokens": 116.2875,
79
+ "mean_t": 0.4803953981841914,
80
+ "step": 30,
81
+ "student_masked_tokens": 116.2875
82
+ },
83
+ {
84
+ "avg_mask_ratio": 0.41855402445653456,
85
+ "avg_response_length": 214.125,
86
+ "avg_student_mask_ratio": 0.41855402445653456,
87
+ "batch_ainp_frac": 0.0,
88
+ "batch_inp_frac": 0.0,
89
+ "batch_inp_oh_frac": 0.0,
90
+ "batch_inp_par_frac": 1.0,
91
+ "batch_inp_par_reverse_frac": 0.0,
92
+ "batch_rl_frac": 0.0,
93
+ "batch_sft_frac": 0.0,
94
+ "batch_soft_sft_frac": 0.0,
95
+ "batch_tf_frac": 0.0,
96
+ "ce_loss": 0.0,
97
+ "epoch": 0.08533333333333333,
98
+ "grad_norm": 0.263671875,
99
+ "kd_loss": 0.1576978968325534,
100
+ "learning_rate": 3e-06,
101
+ "loss": 0.1551,
102
+ "masked_tokens": 94.225,
103
+ "mean_t": 0.4569831106782658,
104
+ "step": 40,
105
+ "student_masked_tokens": 94.225
106
+ },
107
+ {
108
+ "avg_mask_ratio": 0.4331560767372139,
109
+ "avg_response_length": 222.225,
110
+ "avg_student_mask_ratio": 0.4331560767372139,
111
+ "batch_ainp_frac": 0.0,
112
+ "batch_inp_frac": 0.0,
113
+ "batch_inp_oh_frac": 0.0,
114
+ "batch_inp_par_frac": 1.0,
115
+ "batch_inp_par_reverse_frac": 0.0,
116
+ "batch_rl_frac": 0.0,
117
+ "batch_sft_frac": 0.0,
118
+ "batch_soft_sft_frac": 0.0,
119
+ "batch_tf_frac": 0.0,
120
+ "ce_loss": 0.0,
121
+ "epoch": 0.10666666666666667,
122
+ "grad_norm": 0.125,
123
+ "kd_loss": 0.17712681048956255,
124
+ "learning_rate": 3e-06,
125
+ "loss": 0.1648,
126
+ "masked_tokens": 97.825,
127
+ "mean_t": 0.4612453707959503,
128
+ "step": 50,
129
+ "student_masked_tokens": 97.825
130
+ },
131
+ {
132
+ "avg_mask_ratio": 0.4547638618387282,
133
+ "avg_response_length": 242.9,
134
+ "avg_student_mask_ratio": 0.4547638618387282,
135
+ "batch_ainp_frac": 0.0,
136
+ "batch_inp_frac": 0.0,
137
+ "batch_inp_oh_frac": 0.0,
138
+ "batch_inp_par_frac": 1.0,
139
+ "batch_inp_par_reverse_frac": 0.0,
140
+ "batch_rl_frac": 0.0,
141
+ "batch_sft_frac": 0.0,
142
+ "batch_soft_sft_frac": 0.0,
143
+ "batch_tf_frac": 0.0,
144
+ "ce_loss": 0.0,
145
+ "epoch": 0.128,
146
+ "grad_norm": 0.130859375,
147
+ "kd_loss": 0.1745696667137963,
148
+ "learning_rate": 3e-06,
149
+ "loss": 0.1745,
150
+ "masked_tokens": 119.125,
151
+ "mean_t": 0.48597636765334756,
152
+ "step": 60,
153
+ "student_masked_tokens": 119.125
154
+ },
155
+ {
156
+ "avg_mask_ratio": 0.5151988173020072,
157
+ "avg_response_length": 214.3375,
158
+ "avg_student_mask_ratio": 0.5151988173020072,
159
+ "batch_ainp_frac": 0.0,
160
+ "batch_inp_frac": 0.0,
161
+ "batch_inp_oh_frac": 0.0,
162
+ "batch_inp_par_frac": 1.0,
163
+ "batch_inp_par_reverse_frac": 0.0,
164
+ "batch_rl_frac": 0.0,
165
+ "batch_sft_frac": 0.0,
166
+ "batch_soft_sft_frac": 0.0,
167
+ "batch_tf_frac": 0.0,
168
+ "ce_loss": 0.0,
169
+ "epoch": 0.14933333333333335,
170
+ "grad_norm": 0.17578125,
171
+ "kd_loss": 0.1657758254527316,
172
+ "learning_rate": 3e-06,
173
+ "loss": 0.1972,
174
+ "masked_tokens": 111.7875,
175
+ "mean_t": 0.5459650319069624,
176
+ "step": 70,
177
+ "student_masked_tokens": 111.7875
178
+ },
179
+ {
180
+ "avg_mask_ratio": 0.37937068473547697,
181
+ "avg_response_length": 250.2,
182
+ "avg_student_mask_ratio": 0.37937068473547697,
183
+ "batch_ainp_frac": 0.0,
184
+ "batch_inp_frac": 0.0,
185
+ "batch_inp_oh_frac": 0.0,
186
+ "batch_inp_par_frac": 1.0,
187
+ "batch_inp_par_reverse_frac": 0.0,
188
+ "batch_rl_frac": 0.0,
189
+ "batch_sft_frac": 0.0,
190
+ "batch_soft_sft_frac": 0.0,
191
+ "batch_tf_frac": 0.0,
192
+ "ce_loss": 0.0,
193
+ "epoch": 0.17066666666666666,
194
+ "grad_norm": 0.051513671875,
195
+ "kd_loss": 0.13968983994418097,
196
+ "learning_rate": 3e-06,
197
+ "loss": 0.1369,
198
+ "masked_tokens": 95.1875,
199
+ "mean_t": 0.40758824030635876,
200
+ "step": 80,
201
+ "student_masked_tokens": 95.1875
202
+ },
203
+ {
204
+ "avg_mask_ratio": 0.5006106478627771,
205
+ "avg_response_length": 242.1125,
206
+ "avg_student_mask_ratio": 0.5006106478627771,
207
+ "batch_ainp_frac": 0.0,
208
+ "batch_inp_frac": 0.0,
209
+ "batch_inp_oh_frac": 0.0,
210
+ "batch_inp_par_frac": 1.0,
211
+ "batch_inp_par_reverse_frac": 0.0,
212
+ "batch_rl_frac": 0.0,
213
+ "batch_sft_frac": 0.0,
214
+ "batch_soft_sft_frac": 0.0,
215
+ "batch_tf_frac": 0.0,
216
+ "ce_loss": 0.0,
217
+ "epoch": 0.192,
218
+ "grad_norm": 0.22265625,
219
+ "kd_loss": 0.20869405062871707,
220
+ "learning_rate": 3e-06,
221
+ "loss": 0.2007,
222
+ "masked_tokens": 116.3875,
223
+ "mean_t": 0.5297661645396147,
224
+ "step": 90,
225
+ "student_masked_tokens": 116.3875
226
+ },
227
+ {
228
+ "avg_mask_ratio": 0.4596128500183113,
229
+ "avg_response_length": 229.0625,
230
+ "avg_student_mask_ratio": 0.4596128500183113,
231
+ "batch_ainp_frac": 0.0,
232
+ "batch_inp_frac": 0.0,
233
+ "batch_inp_oh_frac": 0.0,
234
+ "batch_inp_par_frac": 1.0,
235
+ "batch_inp_par_reverse_frac": 0.0,
236
+ "batch_rl_frac": 0.0,
237
+ "batch_sft_frac": 0.0,
238
+ "batch_soft_sft_frac": 0.0,
239
+ "batch_tf_frac": 0.0,
240
+ "ce_loss": 0.0,
241
+ "epoch": 0.21333333333333335,
242
+ "grad_norm": 0.1865234375,
243
+ "kd_loss": 0.17640120884607313,
244
+ "learning_rate": 3e-06,
245
+ "loss": 0.1864,
246
+ "masked_tokens": 109.7125,
247
+ "mean_t": 0.4886587227345444,
248
+ "step": 100,
249
+ "student_masked_tokens": 109.7125
250
+ }
251
+ ],
252
+ "logging_steps": 10,
253
+ "max_steps": 1404,
254
+ "num_input_tokens_seen": 0,
255
+ "num_train_epochs": 3,
256
+ "save_steps": 100,
257
+ "stateful_callbacks": {
258
+ "TrainerControl": {
259
+ "args": {
260
+ "should_epoch_stop": false,
261
+ "should_evaluate": false,
262
+ "should_log": false,
263
+ "should_save": true,
264
+ "should_training_stop": false
265
+ },
266
+ "attributes": {}
267
+ }
268
+ },
269
+ "total_flos": 0.0,
270
+ "train_batch_size": 1,
271
+ "trial_name": null,
272
+ "trial_params": null
273
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-100/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c347df37da6e348160afc3fbb65d12595e5064bd8e5bfd591004e5e86a703f42
3
+ size 7992
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: GSAI-ML/LLaDA-8B-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 0.15.1
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/adapter_config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "GSAI-ML/LLaDA-8B-Instruct",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 64,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "r": 128,
24
+ "rank_pattern": {},
25
+ "revision": null,
26
+ "target_modules": [
27
+ "down_proj",
28
+ "up_proj",
29
+ "gate_proj",
30
+ "v_proj",
31
+ "k_proj",
32
+ "o_proj",
33
+ "q_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5dadb69409043d110f0f8c1c799b3827bf022897cf007c0fbc08d6f74f5022d9
3
+ size 2406624648
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:feef74a9920629ff20ab788a1fadcf97aad0450770804f32f725ceb5a5c3731f
3
+ size 671304442
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6cbbccc533dc6035b9eb3e81ab0c37a3544ee2638528b1cb900a84d35f5b76b2
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:112d5e4d41a547b60596d50c5c50332e7991e1c7fbc0e60870f3b01c8a6d0f47
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29d9aa99505fc60c0db1b9cdacaa08b06e8a85c8aaaab4e389667a719fafb9bf
3
+ size 1064
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/trainer_state.json ADDED
@@ -0,0 +1,2433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.1365333333333334,
5
+ "eval_steps": 500,
6
+ "global_step": 1000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "avg_mask_ratio": 0.4933756332669873,
13
+ "avg_response_length": 252.0625,
14
+ "avg_student_mask_ratio": 0.4933756332669873,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 0.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 1.0,
19
+ "batch_inp_par_reverse_frac": 0.0,
20
+ "batch_rl_frac": 0.0,
21
+ "batch_sft_frac": 0.0,
22
+ "batch_soft_sft_frac": 0.0,
23
+ "batch_tf_frac": 0.0,
24
+ "ce_loss": 0.0,
25
+ "epoch": 0.021333333333333333,
26
+ "grad_norm": 0.1650390625,
27
+ "kd_loss": 0.24795629351958723,
28
+ "learning_rate": 3e-06,
29
+ "loss": 0.2758,
30
+ "masked_tokens": 120.975,
31
+ "mean_t": 0.5145528071501758,
32
+ "step": 10,
33
+ "student_masked_tokens": 120.975
34
+ },
35
+ {
36
+ "avg_mask_ratio": 0.41923869140446185,
37
+ "avg_response_length": 221.7125,
38
+ "avg_student_mask_ratio": 0.41923869140446185,
39
+ "batch_ainp_frac": 0.0,
40
+ "batch_inp_frac": 0.0,
41
+ "batch_inp_oh_frac": 0.0,
42
+ "batch_inp_par_frac": 1.0,
43
+ "batch_inp_par_reverse_frac": 0.0,
44
+ "batch_rl_frac": 0.0,
45
+ "batch_sft_frac": 0.0,
46
+ "batch_soft_sft_frac": 0.0,
47
+ "batch_tf_frac": 0.0,
48
+ "ce_loss": 0.0,
49
+ "epoch": 0.042666666666666665,
50
+ "grad_norm": 0.388671875,
51
+ "kd_loss": 0.21509853232191353,
52
+ "learning_rate": 3e-06,
53
+ "loss": 0.2046,
54
+ "masked_tokens": 86.1875,
55
+ "mean_t": 0.43874448732240123,
56
+ "step": 20,
57
+ "student_masked_tokens": 86.1875
58
+ },
59
+ {
60
+ "avg_mask_ratio": 0.4542569225654006,
61
+ "avg_response_length": 231.45,
62
+ "avg_student_mask_ratio": 0.4542569225654006,
63
+ "batch_ainp_frac": 0.0,
64
+ "batch_inp_frac": 0.0,
65
+ "batch_inp_oh_frac": 0.0,
66
+ "batch_inp_par_frac": 1.0,
67
+ "batch_inp_par_reverse_frac": 0.0,
68
+ "batch_rl_frac": 0.0,
69
+ "batch_sft_frac": 0.0,
70
+ "batch_soft_sft_frac": 0.0,
71
+ "batch_tf_frac": 0.0,
72
+ "ce_loss": 0.0,
73
+ "epoch": 0.064,
74
+ "grad_norm": 0.52734375,
75
+ "kd_loss": 0.19756361616970963,
76
+ "learning_rate": 3e-06,
77
+ "loss": 0.1976,
78
+ "masked_tokens": 116.2875,
79
+ "mean_t": 0.4803953981841914,
80
+ "step": 30,
81
+ "student_masked_tokens": 116.2875
82
+ },
83
+ {
84
+ "avg_mask_ratio": 0.41855402445653456,
85
+ "avg_response_length": 214.125,
86
+ "avg_student_mask_ratio": 0.41855402445653456,
87
+ "batch_ainp_frac": 0.0,
88
+ "batch_inp_frac": 0.0,
89
+ "batch_inp_oh_frac": 0.0,
90
+ "batch_inp_par_frac": 1.0,
91
+ "batch_inp_par_reverse_frac": 0.0,
92
+ "batch_rl_frac": 0.0,
93
+ "batch_sft_frac": 0.0,
94
+ "batch_soft_sft_frac": 0.0,
95
+ "batch_tf_frac": 0.0,
96
+ "ce_loss": 0.0,
97
+ "epoch": 0.08533333333333333,
98
+ "grad_norm": 0.263671875,
99
+ "kd_loss": 0.1576978968325534,
100
+ "learning_rate": 3e-06,
101
+ "loss": 0.1551,
102
+ "masked_tokens": 94.225,
103
+ "mean_t": 0.4569831106782658,
104
+ "step": 40,
105
+ "student_masked_tokens": 94.225
106
+ },
107
+ {
108
+ "avg_mask_ratio": 0.4331560767372139,
109
+ "avg_response_length": 222.225,
110
+ "avg_student_mask_ratio": 0.4331560767372139,
111
+ "batch_ainp_frac": 0.0,
112
+ "batch_inp_frac": 0.0,
113
+ "batch_inp_oh_frac": 0.0,
114
+ "batch_inp_par_frac": 1.0,
115
+ "batch_inp_par_reverse_frac": 0.0,
116
+ "batch_rl_frac": 0.0,
117
+ "batch_sft_frac": 0.0,
118
+ "batch_soft_sft_frac": 0.0,
119
+ "batch_tf_frac": 0.0,
120
+ "ce_loss": 0.0,
121
+ "epoch": 0.10666666666666667,
122
+ "grad_norm": 0.125,
123
+ "kd_loss": 0.17712681048956255,
124
+ "learning_rate": 3e-06,
125
+ "loss": 0.1648,
126
+ "masked_tokens": 97.825,
127
+ "mean_t": 0.4612453707959503,
128
+ "step": 50,
129
+ "student_masked_tokens": 97.825
130
+ },
131
+ {
132
+ "avg_mask_ratio": 0.4547638618387282,
133
+ "avg_response_length": 242.9,
134
+ "avg_student_mask_ratio": 0.4547638618387282,
135
+ "batch_ainp_frac": 0.0,
136
+ "batch_inp_frac": 0.0,
137
+ "batch_inp_oh_frac": 0.0,
138
+ "batch_inp_par_frac": 1.0,
139
+ "batch_inp_par_reverse_frac": 0.0,
140
+ "batch_rl_frac": 0.0,
141
+ "batch_sft_frac": 0.0,
142
+ "batch_soft_sft_frac": 0.0,
143
+ "batch_tf_frac": 0.0,
144
+ "ce_loss": 0.0,
145
+ "epoch": 0.128,
146
+ "grad_norm": 0.130859375,
147
+ "kd_loss": 0.1745696667137963,
148
+ "learning_rate": 3e-06,
149
+ "loss": 0.1745,
150
+ "masked_tokens": 119.125,
151
+ "mean_t": 0.48597636765334756,
152
+ "step": 60,
153
+ "student_masked_tokens": 119.125
154
+ },
155
+ {
156
+ "avg_mask_ratio": 0.5151988173020072,
157
+ "avg_response_length": 214.3375,
158
+ "avg_student_mask_ratio": 0.5151988173020072,
159
+ "batch_ainp_frac": 0.0,
160
+ "batch_inp_frac": 0.0,
161
+ "batch_inp_oh_frac": 0.0,
162
+ "batch_inp_par_frac": 1.0,
163
+ "batch_inp_par_reverse_frac": 0.0,
164
+ "batch_rl_frac": 0.0,
165
+ "batch_sft_frac": 0.0,
166
+ "batch_soft_sft_frac": 0.0,
167
+ "batch_tf_frac": 0.0,
168
+ "ce_loss": 0.0,
169
+ "epoch": 0.14933333333333335,
170
+ "grad_norm": 0.17578125,
171
+ "kd_loss": 0.1657758254527316,
172
+ "learning_rate": 3e-06,
173
+ "loss": 0.1972,
174
+ "masked_tokens": 111.7875,
175
+ "mean_t": 0.5459650319069624,
176
+ "step": 70,
177
+ "student_masked_tokens": 111.7875
178
+ },
179
+ {
180
+ "avg_mask_ratio": 0.37937068473547697,
181
+ "avg_response_length": 250.2,
182
+ "avg_student_mask_ratio": 0.37937068473547697,
183
+ "batch_ainp_frac": 0.0,
184
+ "batch_inp_frac": 0.0,
185
+ "batch_inp_oh_frac": 0.0,
186
+ "batch_inp_par_frac": 1.0,
187
+ "batch_inp_par_reverse_frac": 0.0,
188
+ "batch_rl_frac": 0.0,
189
+ "batch_sft_frac": 0.0,
190
+ "batch_soft_sft_frac": 0.0,
191
+ "batch_tf_frac": 0.0,
192
+ "ce_loss": 0.0,
193
+ "epoch": 0.17066666666666666,
194
+ "grad_norm": 0.051513671875,
195
+ "kd_loss": 0.13968983994418097,
196
+ "learning_rate": 3e-06,
197
+ "loss": 0.1369,
198
+ "masked_tokens": 95.1875,
199
+ "mean_t": 0.40758824030635876,
200
+ "step": 80,
201
+ "student_masked_tokens": 95.1875
202
+ },
203
+ {
204
+ "avg_mask_ratio": 0.5006106478627771,
205
+ "avg_response_length": 242.1125,
206
+ "avg_student_mask_ratio": 0.5006106478627771,
207
+ "batch_ainp_frac": 0.0,
208
+ "batch_inp_frac": 0.0,
209
+ "batch_inp_oh_frac": 0.0,
210
+ "batch_inp_par_frac": 1.0,
211
+ "batch_inp_par_reverse_frac": 0.0,
212
+ "batch_rl_frac": 0.0,
213
+ "batch_sft_frac": 0.0,
214
+ "batch_soft_sft_frac": 0.0,
215
+ "batch_tf_frac": 0.0,
216
+ "ce_loss": 0.0,
217
+ "epoch": 0.192,
218
+ "grad_norm": 0.22265625,
219
+ "kd_loss": 0.20869405062871707,
220
+ "learning_rate": 3e-06,
221
+ "loss": 0.2007,
222
+ "masked_tokens": 116.3875,
223
+ "mean_t": 0.5297661645396147,
224
+ "step": 90,
225
+ "student_masked_tokens": 116.3875
226
+ },
227
+ {
228
+ "avg_mask_ratio": 0.4596128500183113,
229
+ "avg_response_length": 229.0625,
230
+ "avg_student_mask_ratio": 0.4596128500183113,
231
+ "batch_ainp_frac": 0.0,
232
+ "batch_inp_frac": 0.0,
233
+ "batch_inp_oh_frac": 0.0,
234
+ "batch_inp_par_frac": 1.0,
235
+ "batch_inp_par_reverse_frac": 0.0,
236
+ "batch_rl_frac": 0.0,
237
+ "batch_sft_frac": 0.0,
238
+ "batch_soft_sft_frac": 0.0,
239
+ "batch_tf_frac": 0.0,
240
+ "ce_loss": 0.0,
241
+ "epoch": 0.21333333333333335,
242
+ "grad_norm": 0.1865234375,
243
+ "kd_loss": 0.17640120884607313,
244
+ "learning_rate": 3e-06,
245
+ "loss": 0.1864,
246
+ "masked_tokens": 109.7125,
247
+ "mean_t": 0.4886587227345444,
248
+ "step": 100,
249
+ "student_masked_tokens": 109.7125
250
+ },
251
+ {
252
+ "avg_mask_ratio": 0.4920400592498481,
253
+ "avg_response_length": 229.9875,
254
+ "avg_student_mask_ratio": 0.4920400592498481,
255
+ "batch_ainp_frac": 0.0,
256
+ "batch_inp_frac": 0.0,
257
+ "batch_inp_oh_frac": 0.0,
258
+ "batch_inp_par_frac": 1.0,
259
+ "batch_inp_par_reverse_frac": 0.0,
260
+ "batch_rl_frac": 0.0,
261
+ "batch_sft_frac": 0.0,
262
+ "batch_soft_sft_frac": 0.0,
263
+ "batch_tf_frac": 0.0,
264
+ "ce_loss": 0.0,
265
+ "epoch": 0.23466666666666666,
266
+ "grad_norm": 0.326171875,
267
+ "kd_loss": 0.2011610215539008,
268
+ "learning_rate": 3e-06,
269
+ "loss": 0.2334,
270
+ "masked_tokens": 109.4,
271
+ "mean_t": 0.5246987929102034,
272
+ "step": 110,
273
+ "student_masked_tokens": 109.4
274
+ },
275
+ {
276
+ "avg_mask_ratio": 0.450224511185661,
277
+ "avg_response_length": 229.225,
278
+ "avg_student_mask_ratio": 0.450224511185661,
279
+ "batch_ainp_frac": 0.0,
280
+ "batch_inp_frac": 0.0,
281
+ "batch_inp_oh_frac": 0.0,
282
+ "batch_inp_par_frac": 1.0,
283
+ "batch_inp_par_reverse_frac": 0.0,
284
+ "batch_rl_frac": 0.0,
285
+ "batch_sft_frac": 0.0,
286
+ "batch_soft_sft_frac": 0.0,
287
+ "batch_tf_frac": 0.0,
288
+ "ce_loss": 0.0,
289
+ "epoch": 0.256,
290
+ "grad_norm": 0.171875,
291
+ "kd_loss": 0.21365654302464918,
292
+ "learning_rate": 3e-06,
293
+ "loss": 0.2021,
294
+ "masked_tokens": 102.6375,
295
+ "mean_t": 0.4805434140143916,
296
+ "step": 120,
297
+ "student_masked_tokens": 102.6375
298
+ },
299
+ {
300
+ "avg_mask_ratio": 0.4397759020910598,
301
+ "avg_response_length": 214.15,
302
+ "avg_student_mask_ratio": 0.4397759020910598,
303
+ "batch_ainp_frac": 0.0,
304
+ "batch_inp_frac": 0.0,
305
+ "batch_inp_oh_frac": 0.0,
306
+ "batch_inp_par_frac": 1.0,
307
+ "batch_inp_par_reverse_frac": 0.0,
308
+ "batch_rl_frac": 0.0,
309
+ "batch_sft_frac": 0.0,
310
+ "batch_soft_sft_frac": 0.0,
311
+ "batch_tf_frac": 0.0,
312
+ "ce_loss": 0.0,
313
+ "epoch": 0.2773333333333333,
314
+ "grad_norm": 0.173828125,
315
+ "kd_loss": 0.13860440934267615,
316
+ "learning_rate": 3e-06,
317
+ "loss": 0.1362,
318
+ "masked_tokens": 98.4,
319
+ "mean_t": 0.47522516988683494,
320
+ "step": 130,
321
+ "student_masked_tokens": 98.4
322
+ },
323
+ {
324
+ "avg_mask_ratio": 0.4890626976499334,
325
+ "avg_response_length": 242.1625,
326
+ "avg_student_mask_ratio": 0.4890626976499334,
327
+ "batch_ainp_frac": 0.0,
328
+ "batch_inp_frac": 0.0,
329
+ "batch_inp_oh_frac": 0.0,
330
+ "batch_inp_par_frac": 1.0,
331
+ "batch_inp_par_reverse_frac": 0.0,
332
+ "batch_rl_frac": 0.0,
333
+ "batch_sft_frac": 0.0,
334
+ "batch_soft_sft_frac": 0.0,
335
+ "batch_tf_frac": 0.0,
336
+ "ce_loss": 0.0,
337
+ "epoch": 0.2986666666666667,
338
+ "grad_norm": 0.27734375,
339
+ "kd_loss": 0.2106460814328841,
340
+ "learning_rate": 3e-06,
341
+ "loss": 0.2054,
342
+ "masked_tokens": 129.725,
343
+ "mean_t": 0.5204090005659964,
344
+ "step": 140,
345
+ "student_masked_tokens": 129.725
346
+ },
347
+ {
348
+ "avg_mask_ratio": 0.47298577734036373,
349
+ "avg_response_length": 262.9875,
350
+ "avg_student_mask_ratio": 0.47298577734036373,
351
+ "batch_ainp_frac": 0.0,
352
+ "batch_inp_frac": 0.0,
353
+ "batch_inp_oh_frac": 0.0,
354
+ "batch_inp_par_frac": 1.0,
355
+ "batch_inp_par_reverse_frac": 0.0,
356
+ "batch_rl_frac": 0.0,
357
+ "batch_sft_frac": 0.0,
358
+ "batch_soft_sft_frac": 0.0,
359
+ "batch_tf_frac": 0.0,
360
+ "ce_loss": 0.0,
361
+ "epoch": 0.32,
362
+ "grad_norm": 0.0673828125,
363
+ "kd_loss": 0.2408599746217078,
364
+ "learning_rate": 3e-06,
365
+ "loss": 0.2276,
366
+ "masked_tokens": 128.375,
367
+ "mean_t": 0.5012552456930279,
368
+ "step": 150,
369
+ "student_masked_tokens": 128.375
370
+ },
371
+ {
372
+ "avg_mask_ratio": 0.5043223856599071,
373
+ "avg_response_length": 217.5,
374
+ "avg_student_mask_ratio": 0.5043223856599071,
375
+ "batch_ainp_frac": 0.0,
376
+ "batch_inp_frac": 0.0,
377
+ "batch_inp_oh_frac": 0.0,
378
+ "batch_inp_par_frac": 1.0,
379
+ "batch_inp_par_reverse_frac": 0.0,
380
+ "batch_rl_frac": 0.0,
381
+ "batch_sft_frac": 0.0,
382
+ "batch_soft_sft_frac": 0.0,
383
+ "batch_tf_frac": 0.0,
384
+ "ce_loss": 0.0,
385
+ "epoch": 0.3413333333333333,
386
+ "grad_norm": 0.2236328125,
387
+ "kd_loss": 0.20484722793833043,
388
+ "learning_rate": 3e-06,
389
+ "loss": 0.2154,
390
+ "masked_tokens": 106.025,
391
+ "mean_t": 0.5408745193795766,
392
+ "step": 160,
393
+ "student_masked_tokens": 106.025
394
+ },
395
+ {
396
+ "avg_mask_ratio": 0.48419030708028005,
397
+ "avg_response_length": 196.2625,
398
+ "avg_student_mask_ratio": 0.48419030708028005,
399
+ "batch_ainp_frac": 0.0,
400
+ "batch_inp_frac": 0.0,
401
+ "batch_inp_oh_frac": 0.0,
402
+ "batch_inp_par_frac": 1.0,
403
+ "batch_inp_par_reverse_frac": 0.0,
404
+ "batch_rl_frac": 0.0,
405
+ "batch_sft_frac": 0.0,
406
+ "batch_soft_sft_frac": 0.0,
407
+ "batch_tf_frac": 0.0,
408
+ "ce_loss": 0.0,
409
+ "epoch": 0.3626666666666667,
410
+ "grad_norm": 0.1611328125,
411
+ "kd_loss": 0.20407032655223248,
412
+ "learning_rate": 3e-06,
413
+ "loss": 0.2007,
414
+ "masked_tokens": 96.2625,
415
+ "mean_t": 0.5134547733236104,
416
+ "step": 170,
417
+ "student_masked_tokens": 96.2625
418
+ },
419
+ {
420
+ "avg_mask_ratio": 0.4485494759515859,
421
+ "avg_response_length": 228.8625,
422
+ "avg_student_mask_ratio": 0.4485494759515859,
423
+ "batch_ainp_frac": 0.0,
424
+ "batch_inp_frac": 0.0,
425
+ "batch_inp_oh_frac": 0.0,
426
+ "batch_inp_par_frac": 1.0,
427
+ "batch_inp_par_reverse_frac": 0.0,
428
+ "batch_rl_frac": 0.0,
429
+ "batch_sft_frac": 0.0,
430
+ "batch_soft_sft_frac": 0.0,
431
+ "batch_tf_frac": 0.0,
432
+ "ce_loss": 0.0,
433
+ "epoch": 0.384,
434
+ "grad_norm": 0.14453125,
435
+ "kd_loss": 0.16957379382825621,
436
+ "learning_rate": 3e-06,
437
+ "loss": 0.1796,
438
+ "masked_tokens": 101.275,
439
+ "mean_t": 0.48226988823735156,
440
+ "step": 180,
441
+ "student_masked_tokens": 101.275
442
+ },
443
+ {
444
+ "avg_mask_ratio": 0.48665878190658984,
445
+ "avg_response_length": 230.5,
446
+ "avg_student_mask_ratio": 0.48665878190658984,
447
+ "batch_ainp_frac": 0.0,
448
+ "batch_inp_frac": 0.0,
449
+ "batch_inp_oh_frac": 0.0,
450
+ "batch_inp_par_frac": 1.0,
451
+ "batch_inp_par_reverse_frac": 0.0,
452
+ "batch_rl_frac": 0.0,
453
+ "batch_sft_frac": 0.0,
454
+ "batch_soft_sft_frac": 0.0,
455
+ "batch_tf_frac": 0.0,
456
+ "ce_loss": 0.0,
457
+ "epoch": 0.4053333333333333,
458
+ "grad_norm": 0.3515625,
459
+ "kd_loss": 0.22313492714965832,
460
+ "learning_rate": 3e-06,
461
+ "loss": 0.2112,
462
+ "masked_tokens": 107.975,
463
+ "mean_t": 0.5165087037021294,
464
+ "step": 190,
465
+ "student_masked_tokens": 107.975
466
+ },
467
+ {
468
+ "avg_mask_ratio": 0.4670982737792656,
469
+ "avg_response_length": 210.9125,
470
+ "avg_student_mask_ratio": 0.4670982737792656,
471
+ "batch_ainp_frac": 0.0,
472
+ "batch_inp_frac": 0.0,
473
+ "batch_inp_oh_frac": 0.0,
474
+ "batch_inp_par_frac": 1.0,
475
+ "batch_inp_par_reverse_frac": 0.0,
476
+ "batch_rl_frac": 0.0,
477
+ "batch_sft_frac": 0.0,
478
+ "batch_soft_sft_frac": 0.0,
479
+ "batch_tf_frac": 0.0,
480
+ "ce_loss": 0.0,
481
+ "epoch": 0.4266666666666667,
482
+ "grad_norm": 0.142578125,
483
+ "kd_loss": 0.15220829088375468,
484
+ "learning_rate": 3e-06,
485
+ "loss": 0.1609,
486
+ "masked_tokens": 98.2125,
487
+ "mean_t": 0.5073627714533359,
488
+ "step": 200,
489
+ "student_masked_tokens": 98.2125
490
+ },
491
+ {
492
+ "avg_mask_ratio": 0.4568137794849463,
493
+ "avg_response_length": 218.7375,
494
+ "avg_student_mask_ratio": 0.4568137794849463,
495
+ "batch_ainp_frac": 0.0,
496
+ "batch_inp_frac": 0.0,
497
+ "batch_inp_oh_frac": 0.0,
498
+ "batch_inp_par_frac": 1.0,
499
+ "batch_inp_par_reverse_frac": 0.0,
500
+ "batch_rl_frac": 0.0,
501
+ "batch_sft_frac": 0.0,
502
+ "batch_soft_sft_frac": 0.0,
503
+ "batch_tf_frac": 0.0,
504
+ "ce_loss": 0.0,
505
+ "epoch": 0.448,
506
+ "grad_norm": 0.1826171875,
507
+ "kd_loss": 0.1580252643583208,
508
+ "learning_rate": 3e-06,
509
+ "loss": 0.1798,
510
+ "masked_tokens": 99.15,
511
+ "mean_t": 0.48430291628465055,
512
+ "step": 210,
513
+ "student_masked_tokens": 99.15
514
+ },
515
+ {
516
+ "avg_mask_ratio": 0.5379857187625021,
517
+ "avg_response_length": 252.1375,
518
+ "avg_student_mask_ratio": 0.5379857187625021,
519
+ "batch_ainp_frac": 0.0,
520
+ "batch_inp_frac": 0.0,
521
+ "batch_inp_oh_frac": 0.0,
522
+ "batch_inp_par_frac": 1.0,
523
+ "batch_inp_par_reverse_frac": 0.0,
524
+ "batch_rl_frac": 0.0,
525
+ "batch_sft_frac": 0.0,
526
+ "batch_soft_sft_frac": 0.0,
527
+ "batch_tf_frac": 0.0,
528
+ "ce_loss": 0.0,
529
+ "epoch": 0.4693333333333333,
530
+ "grad_norm": 0.3203125,
531
+ "kd_loss": 0.2511090909683844,
532
+ "learning_rate": 3e-06,
533
+ "loss": 0.2583,
534
+ "masked_tokens": 135.4,
535
+ "mean_t": 0.570199209311977,
536
+ "step": 220,
537
+ "student_masked_tokens": 135.4
538
+ },
539
+ {
540
+ "avg_mask_ratio": 0.43395056116278286,
541
+ "avg_response_length": 245.2625,
542
+ "avg_student_mask_ratio": 0.43395056116278286,
543
+ "batch_ainp_frac": 0.0,
544
+ "batch_inp_frac": 0.0,
545
+ "batch_inp_oh_frac": 0.0,
546
+ "batch_inp_par_frac": 1.0,
547
+ "batch_inp_par_reverse_frac": 0.0,
548
+ "batch_rl_frac": 0.0,
549
+ "batch_sft_frac": 0.0,
550
+ "batch_soft_sft_frac": 0.0,
551
+ "batch_tf_frac": 0.0,
552
+ "ce_loss": 0.0,
553
+ "epoch": 0.49066666666666664,
554
+ "grad_norm": 0.1767578125,
555
+ "kd_loss": 0.14414861655371852,
556
+ "learning_rate": 3e-06,
557
+ "loss": 0.1964,
558
+ "masked_tokens": 102.5125,
559
+ "mean_t": 0.46891279935371133,
560
+ "step": 230,
561
+ "student_masked_tokens": 102.5125
562
+ },
563
+ {
564
+ "avg_mask_ratio": 0.46948411157354714,
565
+ "avg_response_length": 202.975,
566
+ "avg_student_mask_ratio": 0.46948411157354714,
567
+ "batch_ainp_frac": 0.0,
568
+ "batch_inp_frac": 0.0,
569
+ "batch_inp_oh_frac": 0.0,
570
+ "batch_inp_par_frac": 1.0,
571
+ "batch_inp_par_reverse_frac": 0.0,
572
+ "batch_rl_frac": 0.0,
573
+ "batch_sft_frac": 0.0,
574
+ "batch_soft_sft_frac": 0.0,
575
+ "batch_tf_frac": 0.0,
576
+ "ce_loss": 0.0,
577
+ "epoch": 0.512,
578
+ "grad_norm": 0.1181640625,
579
+ "kd_loss": 0.2197965504183493,
580
+ "learning_rate": 3e-06,
581
+ "loss": 0.243,
582
+ "masked_tokens": 97.0625,
583
+ "mean_t": 0.502228345896583,
584
+ "step": 240,
585
+ "student_masked_tokens": 97.0625
586
+ },
587
+ {
588
+ "avg_mask_ratio": 0.44631263689370826,
589
+ "avg_response_length": 243.425,
590
+ "avg_student_mask_ratio": 0.44631263689370826,
591
+ "batch_ainp_frac": 0.0,
592
+ "batch_inp_frac": 0.0,
593
+ "batch_inp_oh_frac": 0.0,
594
+ "batch_inp_par_frac": 1.0,
595
+ "batch_inp_par_reverse_frac": 0.0,
596
+ "batch_rl_frac": 0.0,
597
+ "batch_sft_frac": 0.0,
598
+ "batch_soft_sft_frac": 0.0,
599
+ "batch_tf_frac": 0.0,
600
+ "ce_loss": 0.0,
601
+ "epoch": 0.5333333333333333,
602
+ "grad_norm": 0.1064453125,
603
+ "kd_loss": 0.2151024747882957,
604
+ "learning_rate": 3e-06,
605
+ "loss": 0.1892,
606
+ "masked_tokens": 107.4625,
607
+ "mean_t": 0.4733429416548461,
608
+ "step": 250,
609
+ "student_masked_tokens": 107.4625
610
+ },
611
+ {
612
+ "avg_mask_ratio": 0.4607314572727773,
613
+ "avg_response_length": 243.375,
614
+ "avg_student_mask_ratio": 0.4607314572727773,
615
+ "batch_ainp_frac": 0.0,
616
+ "batch_inp_frac": 0.0,
617
+ "batch_inp_oh_frac": 0.0,
618
+ "batch_inp_par_frac": 1.0,
619
+ "batch_inp_par_reverse_frac": 0.0,
620
+ "batch_rl_frac": 0.0,
621
+ "batch_sft_frac": 0.0,
622
+ "batch_soft_sft_frac": 0.0,
623
+ "batch_tf_frac": 0.0,
624
+ "ce_loss": 0.0,
625
+ "epoch": 0.5546666666666666,
626
+ "grad_norm": 0.203125,
627
+ "kd_loss": 0.176242933875335,
628
+ "learning_rate": 3e-06,
629
+ "loss": 0.1885,
630
+ "masked_tokens": 110.8875,
631
+ "mean_t": 0.4843149524240289,
632
+ "step": 260,
633
+ "student_masked_tokens": 110.8875
634
+ },
635
+ {
636
+ "avg_mask_ratio": 0.5309946444118395,
637
+ "avg_response_length": 231.6875,
638
+ "avg_student_mask_ratio": 0.5309946444118395,
639
+ "batch_ainp_frac": 0.0,
640
+ "batch_inp_frac": 0.0,
641
+ "batch_inp_oh_frac": 0.0,
642
+ "batch_inp_par_frac": 1.0,
643
+ "batch_inp_par_reverse_frac": 0.0,
644
+ "batch_rl_frac": 0.0,
645
+ "batch_sft_frac": 0.0,
646
+ "batch_soft_sft_frac": 0.0,
647
+ "batch_tf_frac": 0.0,
648
+ "ce_loss": 0.0,
649
+ "epoch": 0.576,
650
+ "grad_norm": 0.208984375,
651
+ "kd_loss": 0.26651088480309115,
652
+ "learning_rate": 3e-06,
653
+ "loss": 0.2828,
654
+ "masked_tokens": 123.2875,
655
+ "mean_t": 0.5643589949700981,
656
+ "step": 270,
657
+ "student_masked_tokens": 123.2875
658
+ },
659
+ {
660
+ "avg_mask_ratio": 0.45879559536697345,
661
+ "avg_response_length": 251.8,
662
+ "avg_student_mask_ratio": 0.45879559536697345,
663
+ "batch_ainp_frac": 0.0,
664
+ "batch_inp_frac": 0.0,
665
+ "batch_inp_oh_frac": 0.0,
666
+ "batch_inp_par_frac": 1.0,
667
+ "batch_inp_par_reverse_frac": 0.0,
668
+ "batch_rl_frac": 0.0,
669
+ "batch_sft_frac": 0.0,
670
+ "batch_soft_sft_frac": 0.0,
671
+ "batch_tf_frac": 0.0,
672
+ "ce_loss": 0.0,
673
+ "epoch": 0.5973333333333334,
674
+ "grad_norm": 0.11474609375,
675
+ "kd_loss": 0.145786481295454,
676
+ "learning_rate": 3e-06,
677
+ "loss": 0.1439,
678
+ "masked_tokens": 125.425,
679
+ "mean_t": 0.47818811538163575,
680
+ "step": 280,
681
+ "student_masked_tokens": 125.425
682
+ },
683
+ {
684
+ "avg_mask_ratio": 0.3955249205115251,
685
+ "avg_response_length": 238.675,
686
+ "avg_student_mask_ratio": 0.3955249205115251,
687
+ "batch_ainp_frac": 0.0,
688
+ "batch_inp_frac": 0.0,
689
+ "batch_inp_oh_frac": 0.0,
690
+ "batch_inp_par_frac": 1.0,
691
+ "batch_inp_par_reverse_frac": 0.0,
692
+ "batch_rl_frac": 0.0,
693
+ "batch_sft_frac": 0.0,
694
+ "batch_soft_sft_frac": 0.0,
695
+ "batch_tf_frac": 0.0,
696
+ "ce_loss": 0.0,
697
+ "epoch": 0.6186666666666667,
698
+ "grad_norm": 0.057373046875,
699
+ "kd_loss": 0.15104623195453543,
700
+ "learning_rate": 3e-06,
701
+ "loss": 0.1578,
702
+ "masked_tokens": 95.2125,
703
+ "mean_t": 0.4088635521940887,
704
+ "step": 290,
705
+ "student_masked_tokens": 95.2125
706
+ },
707
+ {
708
+ "avg_mask_ratio": 0.4504710016073659,
709
+ "avg_response_length": 202.575,
710
+ "avg_student_mask_ratio": 0.4504710016073659,
711
+ "batch_ainp_frac": 0.0,
712
+ "batch_inp_frac": 0.0,
713
+ "batch_inp_oh_frac": 0.0,
714
+ "batch_inp_par_frac": 1.0,
715
+ "batch_inp_par_reverse_frac": 0.0,
716
+ "batch_rl_frac": 0.0,
717
+ "batch_sft_frac": 0.0,
718
+ "batch_soft_sft_frac": 0.0,
719
+ "batch_tf_frac": 0.0,
720
+ "ce_loss": 0.0,
721
+ "epoch": 0.64,
722
+ "grad_norm": 0.150390625,
723
+ "kd_loss": 0.16011972622800386,
724
+ "learning_rate": 3e-06,
725
+ "loss": 0.179,
726
+ "masked_tokens": 90.3375,
727
+ "mean_t": 0.47984200695063917,
728
+ "step": 300,
729
+ "student_masked_tokens": 90.3375
730
+ },
731
+ {
732
+ "avg_mask_ratio": 0.4822002159198746,
733
+ "avg_response_length": 189.9875,
734
+ "avg_student_mask_ratio": 0.4822002159198746,
735
+ "batch_ainp_frac": 0.0,
736
+ "batch_inp_frac": 0.0,
737
+ "batch_inp_oh_frac": 0.0,
738
+ "batch_inp_par_frac": 1.0,
739
+ "batch_inp_par_reverse_frac": 0.0,
740
+ "batch_rl_frac": 0.0,
741
+ "batch_sft_frac": 0.0,
742
+ "batch_soft_sft_frac": 0.0,
743
+ "batch_tf_frac": 0.0,
744
+ "ce_loss": 0.0,
745
+ "epoch": 0.6613333333333333,
746
+ "grad_norm": 0.1630859375,
747
+ "kd_loss": 0.21744939284749734,
748
+ "learning_rate": 3e-06,
749
+ "loss": 0.201,
750
+ "masked_tokens": 92.7,
751
+ "mean_t": 0.5026606284547597,
752
+ "step": 310,
753
+ "student_masked_tokens": 92.7
754
+ },
755
+ {
756
+ "avg_mask_ratio": 0.5024422638700343,
757
+ "avg_response_length": 237.2625,
758
+ "avg_student_mask_ratio": 0.5024422638700343,
759
+ "batch_ainp_frac": 0.0,
760
+ "batch_inp_frac": 0.0,
761
+ "batch_inp_oh_frac": 0.0,
762
+ "batch_inp_par_frac": 1.0,
763
+ "batch_inp_par_reverse_frac": 0.0,
764
+ "batch_rl_frac": 0.0,
765
+ "batch_sft_frac": 0.0,
766
+ "batch_soft_sft_frac": 0.0,
767
+ "batch_tf_frac": 0.0,
768
+ "ce_loss": 0.0,
769
+ "epoch": 0.6826666666666666,
770
+ "grad_norm": 0.1259765625,
771
+ "kd_loss": 0.21489343987664142,
772
+ "learning_rate": 3e-06,
773
+ "loss": 0.2268,
774
+ "masked_tokens": 123.8125,
775
+ "mean_t": 0.5293499688967132,
776
+ "step": 320,
777
+ "student_masked_tokens": 123.8125
778
+ },
779
+ {
780
+ "avg_mask_ratio": 0.510216062690597,
781
+ "avg_response_length": 257.525,
782
+ "avg_student_mask_ratio": 0.510216062690597,
783
+ "batch_ainp_frac": 0.0,
784
+ "batch_inp_frac": 0.0,
785
+ "batch_inp_oh_frac": 0.0,
786
+ "batch_inp_par_frac": 1.0,
787
+ "batch_inp_par_reverse_frac": 0.0,
788
+ "batch_rl_frac": 0.0,
789
+ "batch_sft_frac": 0.0,
790
+ "batch_soft_sft_frac": 0.0,
791
+ "batch_tf_frac": 0.0,
792
+ "ce_loss": 0.0,
793
+ "epoch": 0.704,
794
+ "grad_norm": 0.1337890625,
795
+ "kd_loss": 0.17950079924535203,
796
+ "learning_rate": 3e-06,
797
+ "loss": 0.2007,
798
+ "masked_tokens": 142.8125,
799
+ "mean_t": 0.5426030711154454,
800
+ "step": 330,
801
+ "student_masked_tokens": 142.8125
802
+ },
803
+ {
804
+ "avg_mask_ratio": 0.45074162067612633,
805
+ "avg_response_length": 245.525,
806
+ "avg_student_mask_ratio": 0.45074162067612633,
807
+ "batch_ainp_frac": 0.0,
808
+ "batch_inp_frac": 0.0,
809
+ "batch_inp_oh_frac": 0.0,
810
+ "batch_inp_par_frac": 1.0,
811
+ "batch_inp_par_reverse_frac": 0.0,
812
+ "batch_rl_frac": 0.0,
813
+ "batch_sft_frac": 0.0,
814
+ "batch_soft_sft_frac": 0.0,
815
+ "batch_tf_frac": 0.0,
816
+ "ce_loss": 0.0,
817
+ "epoch": 0.7253333333333334,
818
+ "grad_norm": 0.310546875,
819
+ "kd_loss": 0.14043198096701417,
820
+ "learning_rate": 3e-06,
821
+ "loss": 0.1669,
822
+ "masked_tokens": 116.0875,
823
+ "mean_t": 0.47635243807453664,
824
+ "step": 340,
825
+ "student_masked_tokens": 116.0875
826
+ },
827
+ {
828
+ "avg_mask_ratio": 0.4926802407717332,
829
+ "avg_response_length": 238.2,
830
+ "avg_student_mask_ratio": 0.4926802407717332,
831
+ "batch_ainp_frac": 0.0,
832
+ "batch_inp_frac": 0.0,
833
+ "batch_inp_oh_frac": 0.0,
834
+ "batch_inp_par_frac": 1.0,
835
+ "batch_inp_par_reverse_frac": 0.0,
836
+ "batch_rl_frac": 0.0,
837
+ "batch_sft_frac": 0.0,
838
+ "batch_soft_sft_frac": 0.0,
839
+ "batch_tf_frac": 0.0,
840
+ "ce_loss": 0.0,
841
+ "epoch": 0.7466666666666667,
842
+ "grad_norm": 0.330078125,
843
+ "kd_loss": 0.21977804936059328,
844
+ "learning_rate": 3e-06,
845
+ "loss": 0.2497,
846
+ "masked_tokens": 127.4875,
847
+ "mean_t": 0.5268881446914747,
848
+ "step": 350,
849
+ "student_masked_tokens": 127.4875
850
+ },
851
+ {
852
+ "avg_mask_ratio": 0.4706261330051348,
853
+ "avg_response_length": 251.7375,
854
+ "avg_student_mask_ratio": 0.4706261330051348,
855
+ "batch_ainp_frac": 0.0,
856
+ "batch_inp_frac": 0.0,
857
+ "batch_inp_oh_frac": 0.0,
858
+ "batch_inp_par_frac": 1.0,
859
+ "batch_inp_par_reverse_frac": 0.0,
860
+ "batch_rl_frac": 0.0,
861
+ "batch_sft_frac": 0.0,
862
+ "batch_soft_sft_frac": 0.0,
863
+ "batch_tf_frac": 0.0,
864
+ "ce_loss": 0.0,
865
+ "epoch": 0.768,
866
+ "grad_norm": 0.056640625,
867
+ "kd_loss": 0.29630907970476983,
868
+ "learning_rate": 3e-06,
869
+ "loss": 0.2329,
870
+ "masked_tokens": 116.8625,
871
+ "mean_t": 0.5040419134311378,
872
+ "step": 360,
873
+ "student_masked_tokens": 116.8625
874
+ },
875
+ {
876
+ "avg_mask_ratio": 0.48096118308603764,
877
+ "avg_response_length": 262.25,
878
+ "avg_student_mask_ratio": 0.48096118308603764,
879
+ "batch_ainp_frac": 0.0,
880
+ "batch_inp_frac": 0.0,
881
+ "batch_inp_oh_frac": 0.0,
882
+ "batch_inp_par_frac": 1.0,
883
+ "batch_inp_par_reverse_frac": 0.0,
884
+ "batch_rl_frac": 0.0,
885
+ "batch_sft_frac": 0.0,
886
+ "batch_soft_sft_frac": 0.0,
887
+ "batch_tf_frac": 0.0,
888
+ "ce_loss": 0.0,
889
+ "epoch": 0.7893333333333333,
890
+ "grad_norm": 0.12255859375,
891
+ "kd_loss": 0.20822112379132704,
892
+ "learning_rate": 3e-06,
893
+ "loss": 0.186,
894
+ "masked_tokens": 132.2,
895
+ "mean_t": 0.5127181728370488,
896
+ "step": 370,
897
+ "student_masked_tokens": 132.2
898
+ },
899
+ {
900
+ "avg_mask_ratio": 0.4433969090110622,
901
+ "avg_response_length": 209.7125,
902
+ "avg_student_mask_ratio": 0.4433969090110622,
903
+ "batch_ainp_frac": 0.0,
904
+ "batch_inp_frac": 0.0,
905
+ "batch_inp_oh_frac": 0.0,
906
+ "batch_inp_par_frac": 1.0,
907
+ "batch_inp_par_reverse_frac": 0.0,
908
+ "batch_rl_frac": 0.0,
909
+ "batch_sft_frac": 0.0,
910
+ "batch_soft_sft_frac": 0.0,
911
+ "batch_tf_frac": 0.0,
912
+ "ce_loss": 0.0,
913
+ "epoch": 0.8106666666666666,
914
+ "grad_norm": 0.123046875,
915
+ "kd_loss": 0.14427866181035826,
916
+ "learning_rate": 3e-06,
917
+ "loss": 0.159,
918
+ "masked_tokens": 95.8625,
919
+ "mean_t": 0.4825185665744357,
920
+ "step": 380,
921
+ "student_masked_tokens": 95.8625
922
+ },
923
+ {
924
+ "avg_mask_ratio": 0.4764250977139454,
925
+ "avg_response_length": 226.3875,
926
+ "avg_student_mask_ratio": 0.4764250977139454,
927
+ "batch_ainp_frac": 0.0,
928
+ "batch_inp_frac": 0.0,
929
+ "batch_inp_oh_frac": 0.0,
930
+ "batch_inp_par_frac": 1.0,
931
+ "batch_inp_par_reverse_frac": 0.0,
932
+ "batch_rl_frac": 0.0,
933
+ "batch_sft_frac": 0.0,
934
+ "batch_soft_sft_frac": 0.0,
935
+ "batch_tf_frac": 0.0,
936
+ "ce_loss": 0.0,
937
+ "epoch": 0.832,
938
+ "grad_norm": 0.185546875,
939
+ "kd_loss": 0.18584371552193701,
940
+ "learning_rate": 3e-06,
941
+ "loss": 0.1823,
942
+ "masked_tokens": 113.95,
943
+ "mean_t": 0.5053101469413377,
944
+ "step": 390,
945
+ "student_masked_tokens": 113.95
946
+ },
947
+ {
948
+ "avg_mask_ratio": 0.47088071387261154,
949
+ "avg_response_length": 233.2125,
950
+ "avg_student_mask_ratio": 0.47088071387261154,
951
+ "batch_ainp_frac": 0.0,
952
+ "batch_inp_frac": 0.0,
953
+ "batch_inp_oh_frac": 0.0,
954
+ "batch_inp_par_frac": 1.0,
955
+ "batch_inp_par_reverse_frac": 0.0,
956
+ "batch_rl_frac": 0.0,
957
+ "batch_sft_frac": 0.0,
958
+ "batch_soft_sft_frac": 0.0,
959
+ "batch_tf_frac": 0.0,
960
+ "ce_loss": 0.0,
961
+ "epoch": 0.8533333333333334,
962
+ "grad_norm": 0.2734375,
963
+ "kd_loss": 0.22232839192147366,
964
+ "learning_rate": 3e-06,
965
+ "loss": 0.1961,
966
+ "masked_tokens": 116.675,
967
+ "mean_t": 0.5052250675857067,
968
+ "step": 400,
969
+ "student_masked_tokens": 116.675
970
+ },
971
+ {
972
+ "avg_mask_ratio": 0.4870647343515884,
973
+ "avg_response_length": 228.3875,
974
+ "avg_student_mask_ratio": 0.4870647343515884,
975
+ "batch_ainp_frac": 0.0,
976
+ "batch_inp_frac": 0.0,
977
+ "batch_inp_oh_frac": 0.0,
978
+ "batch_inp_par_frac": 1.0,
979
+ "batch_inp_par_reverse_frac": 0.0,
980
+ "batch_rl_frac": 0.0,
981
+ "batch_sft_frac": 0.0,
982
+ "batch_soft_sft_frac": 0.0,
983
+ "batch_tf_frac": 0.0,
984
+ "ce_loss": 0.0,
985
+ "epoch": 0.8746666666666667,
986
+ "grad_norm": 0.25390625,
987
+ "kd_loss": 0.25729746209006665,
988
+ "learning_rate": 3e-06,
989
+ "loss": 0.2306,
990
+ "masked_tokens": 114.3625,
991
+ "mean_t": 0.5050956419203431,
992
+ "step": 410,
993
+ "student_masked_tokens": 114.3625
994
+ },
995
+ {
996
+ "avg_mask_ratio": 0.440834702400025,
997
+ "avg_response_length": 209.85,
998
+ "avg_student_mask_ratio": 0.440834702400025,
999
+ "batch_ainp_frac": 0.0,
1000
+ "batch_inp_frac": 0.0,
1001
+ "batch_inp_oh_frac": 0.0,
1002
+ "batch_inp_par_frac": 1.0,
1003
+ "batch_inp_par_reverse_frac": 0.0,
1004
+ "batch_rl_frac": 0.0,
1005
+ "batch_sft_frac": 0.0,
1006
+ "batch_soft_sft_frac": 0.0,
1007
+ "batch_tf_frac": 0.0,
1008
+ "ce_loss": 0.0,
1009
+ "epoch": 0.896,
1010
+ "grad_norm": 0.2275390625,
1011
+ "kd_loss": 0.15747290870124503,
1012
+ "learning_rate": 3e-06,
1013
+ "loss": 0.1583,
1014
+ "masked_tokens": 87.575,
1015
+ "mean_t": 0.4769687672611326,
1016
+ "step": 420,
1017
+ "student_masked_tokens": 87.575
1018
+ },
1019
+ {
1020
+ "avg_mask_ratio": 0.4660509963519871,
1021
+ "avg_response_length": 250.9125,
1022
+ "avg_student_mask_ratio": 0.4660509963519871,
1023
+ "batch_ainp_frac": 0.0,
1024
+ "batch_inp_frac": 0.0,
1025
+ "batch_inp_oh_frac": 0.0,
1026
+ "batch_inp_par_frac": 1.0,
1027
+ "batch_inp_par_reverse_frac": 0.0,
1028
+ "batch_rl_frac": 0.0,
1029
+ "batch_sft_frac": 0.0,
1030
+ "batch_soft_sft_frac": 0.0,
1031
+ "batch_tf_frac": 0.0,
1032
+ "ce_loss": 0.0,
1033
+ "epoch": 0.9173333333333333,
1034
+ "grad_norm": 0.392578125,
1035
+ "kd_loss": 0.17299866449352522,
1036
+ "learning_rate": 3e-06,
1037
+ "loss": 0.178,
1038
+ "masked_tokens": 109.675,
1039
+ "mean_t": 0.49262027950026094,
1040
+ "step": 430,
1041
+ "student_masked_tokens": 109.675
1042
+ },
1043
+ {
1044
+ "avg_mask_ratio": 0.4507861359976232,
1045
+ "avg_response_length": 235.875,
1046
+ "avg_student_mask_ratio": 0.4507861359976232,
1047
+ "batch_ainp_frac": 0.0,
1048
+ "batch_inp_frac": 0.0,
1049
+ "batch_inp_oh_frac": 0.0,
1050
+ "batch_inp_par_frac": 1.0,
1051
+ "batch_inp_par_reverse_frac": 0.0,
1052
+ "batch_rl_frac": 0.0,
1053
+ "batch_sft_frac": 0.0,
1054
+ "batch_soft_sft_frac": 0.0,
1055
+ "batch_tf_frac": 0.0,
1056
+ "ce_loss": 0.0,
1057
+ "epoch": 0.9386666666666666,
1058
+ "grad_norm": 0.15234375,
1059
+ "kd_loss": 0.2113636662043291,
1060
+ "learning_rate": 3e-06,
1061
+ "loss": 0.1795,
1062
+ "masked_tokens": 106.95,
1063
+ "mean_t": 0.47046207524836064,
1064
+ "step": 440,
1065
+ "student_masked_tokens": 106.95
1066
+ },
1067
+ {
1068
+ "avg_mask_ratio": 0.4283985076006502,
1069
+ "avg_response_length": 230.95,
1070
+ "avg_student_mask_ratio": 0.4283985076006502,
1071
+ "batch_ainp_frac": 0.0,
1072
+ "batch_inp_frac": 0.0,
1073
+ "batch_inp_oh_frac": 0.0,
1074
+ "batch_inp_par_frac": 1.0,
1075
+ "batch_inp_par_reverse_frac": 0.0,
1076
+ "batch_rl_frac": 0.0,
1077
+ "batch_sft_frac": 0.0,
1078
+ "batch_soft_sft_frac": 0.0,
1079
+ "batch_tf_frac": 0.0,
1080
+ "ce_loss": 0.0,
1081
+ "epoch": 0.96,
1082
+ "grad_norm": 0.16015625,
1083
+ "kd_loss": 0.15304818171161969,
1084
+ "learning_rate": 3e-06,
1085
+ "loss": 0.1724,
1086
+ "masked_tokens": 101.15,
1087
+ "mean_t": 0.45781184462830427,
1088
+ "step": 450,
1089
+ "student_masked_tokens": 101.15
1090
+ },
1091
+ {
1092
+ "avg_mask_ratio": 0.47474822774529457,
1093
+ "avg_response_length": 233.1,
1094
+ "avg_student_mask_ratio": 0.47474822774529457,
1095
+ "batch_ainp_frac": 0.0,
1096
+ "batch_inp_frac": 0.0,
1097
+ "batch_inp_oh_frac": 0.0,
1098
+ "batch_inp_par_frac": 1.0,
1099
+ "batch_inp_par_reverse_frac": 0.0,
1100
+ "batch_rl_frac": 0.0,
1101
+ "batch_sft_frac": 0.0,
1102
+ "batch_soft_sft_frac": 0.0,
1103
+ "batch_tf_frac": 0.0,
1104
+ "ce_loss": 0.0,
1105
+ "epoch": 0.9813333333333333,
1106
+ "grad_norm": 0.06689453125,
1107
+ "kd_loss": 0.1363761811357108,
1108
+ "learning_rate": 3e-06,
1109
+ "loss": 0.171,
1110
+ "masked_tokens": 112.725,
1111
+ "mean_t": 0.5075790266972036,
1112
+ "step": 460,
1113
+ "student_masked_tokens": 112.725
1114
+ },
1115
+ {
1116
+ "avg_mask_ratio": 0.4808142688901474,
1117
+ "avg_response_length": 238.54761904761904,
1118
+ "avg_student_mask_ratio": 0.4808142688901474,
1119
+ "batch_ainp_frac": 0.0,
1120
+ "batch_inp_frac": 0.0,
1121
+ "batch_inp_oh_frac": 0.0,
1122
+ "batch_inp_par_frac": 1.0,
1123
+ "batch_inp_par_reverse_frac": 0.0,
1124
+ "batch_rl_frac": 0.0,
1125
+ "batch_sft_frac": 0.0,
1126
+ "batch_soft_sft_frac": 0.0,
1127
+ "batch_tf_frac": 0.0,
1128
+ "ce_loss": 0.0,
1129
+ "epoch": 1.0042666666666666,
1130
+ "grad_norm": 0.201171875,
1131
+ "kd_loss": 0.22379483340171732,
1132
+ "learning_rate": 3e-06,
1133
+ "loss": 0.2466,
1134
+ "masked_tokens": 113.67857142857143,
1135
+ "mean_t": 0.5031429776822084,
1136
+ "step": 470,
1137
+ "student_masked_tokens": 113.67857142857143
1138
+ },
1139
+ {
1140
+ "avg_mask_ratio": 0.47175657459301873,
1141
+ "avg_response_length": 249.9625,
1142
+ "avg_student_mask_ratio": 0.47175657459301873,
1143
+ "batch_ainp_frac": 0.0,
1144
+ "batch_inp_frac": 0.0,
1145
+ "batch_inp_oh_frac": 0.0,
1146
+ "batch_inp_par_frac": 1.0,
1147
+ "batch_inp_par_reverse_frac": 0.0,
1148
+ "batch_rl_frac": 0.0,
1149
+ "batch_sft_frac": 0.0,
1150
+ "batch_soft_sft_frac": 0.0,
1151
+ "batch_tf_frac": 0.0,
1152
+ "ce_loss": 0.0,
1153
+ "epoch": 1.0256,
1154
+ "grad_norm": 0.166015625,
1155
+ "kd_loss": 0.17491777570117745,
1156
+ "learning_rate": 3e-06,
1157
+ "loss": 0.2029,
1158
+ "masked_tokens": 119.4625,
1159
+ "mean_t": 0.504472183593316,
1160
+ "step": 480,
1161
+ "student_masked_tokens": 119.4625
1162
+ },
1163
+ {
1164
+ "avg_mask_ratio": 0.4564988439786248,
1165
+ "avg_response_length": 238.8875,
1166
+ "avg_student_mask_ratio": 0.4564988439786248,
1167
+ "batch_ainp_frac": 0.0,
1168
+ "batch_inp_frac": 0.0,
1169
+ "batch_inp_oh_frac": 0.0,
1170
+ "batch_inp_par_frac": 1.0,
1171
+ "batch_inp_par_reverse_frac": 0.0,
1172
+ "batch_rl_frac": 0.0,
1173
+ "batch_sft_frac": 0.0,
1174
+ "batch_soft_sft_frac": 0.0,
1175
+ "batch_tf_frac": 0.0,
1176
+ "ce_loss": 0.0,
1177
+ "epoch": 1.0469333333333333,
1178
+ "grad_norm": 0.1279296875,
1179
+ "kd_loss": 0.12884608846077866,
1180
+ "learning_rate": 3e-06,
1181
+ "loss": 0.1536,
1182
+ "masked_tokens": 104.0,
1183
+ "mean_t": 0.4888980514719151,
1184
+ "step": 490,
1185
+ "student_masked_tokens": 104.0
1186
+ },
1187
+ {
1188
+ "avg_mask_ratio": 0.5083174118888565,
1189
+ "avg_response_length": 258.1375,
1190
+ "avg_student_mask_ratio": 0.5083174118888565,
1191
+ "batch_ainp_frac": 0.0,
1192
+ "batch_inp_frac": 0.0,
1193
+ "batch_inp_oh_frac": 0.0,
1194
+ "batch_inp_par_frac": 1.0,
1195
+ "batch_inp_par_reverse_frac": 0.0,
1196
+ "batch_rl_frac": 0.0,
1197
+ "batch_sft_frac": 0.0,
1198
+ "batch_soft_sft_frac": 0.0,
1199
+ "batch_tf_frac": 0.0,
1200
+ "ce_loss": 0.0,
1201
+ "epoch": 1.0682666666666667,
1202
+ "grad_norm": 0.1357421875,
1203
+ "kd_loss": 0.18128383785248586,
1204
+ "learning_rate": 3e-06,
1205
+ "loss": 0.1811,
1206
+ "masked_tokens": 133.7125,
1207
+ "mean_t": 0.5469163245841628,
1208
+ "step": 500,
1209
+ "student_masked_tokens": 133.7125
1210
+ },
1211
+ {
1212
+ "avg_mask_ratio": 0.5130103683215566,
1213
+ "avg_response_length": 246.4875,
1214
+ "avg_student_mask_ratio": 0.5130103683215566,
1215
+ "batch_ainp_frac": 0.0,
1216
+ "batch_inp_frac": 0.0,
1217
+ "batch_inp_oh_frac": 0.0,
1218
+ "batch_inp_par_frac": 1.0,
1219
+ "batch_inp_par_reverse_frac": 0.0,
1220
+ "batch_rl_frac": 0.0,
1221
+ "batch_sft_frac": 0.0,
1222
+ "batch_soft_sft_frac": 0.0,
1223
+ "batch_tf_frac": 0.0,
1224
+ "ce_loss": 0.0,
1225
+ "epoch": 1.0896,
1226
+ "grad_norm": 0.11181640625,
1227
+ "kd_loss": 0.23847924997493805,
1228
+ "learning_rate": 3e-06,
1229
+ "loss": 0.2289,
1230
+ "masked_tokens": 132.5625,
1231
+ "mean_t": 0.5275314710394013,
1232
+ "step": 510,
1233
+ "student_masked_tokens": 132.5625
1234
+ },
1235
+ {
1236
+ "avg_mask_ratio": 0.4543681625276804,
1237
+ "avg_response_length": 199.65,
1238
+ "avg_student_mask_ratio": 0.4543681625276804,
1239
+ "batch_ainp_frac": 0.0,
1240
+ "batch_inp_frac": 0.0,
1241
+ "batch_inp_oh_frac": 0.0,
1242
+ "batch_inp_par_frac": 1.0,
1243
+ "batch_inp_par_reverse_frac": 0.0,
1244
+ "batch_rl_frac": 0.0,
1245
+ "batch_sft_frac": 0.0,
1246
+ "batch_soft_sft_frac": 0.0,
1247
+ "batch_tf_frac": 0.0,
1248
+ "ce_loss": 0.0,
1249
+ "epoch": 1.1109333333333333,
1250
+ "grad_norm": 0.1533203125,
1251
+ "kd_loss": 0.1353249137787543,
1252
+ "learning_rate": 3e-06,
1253
+ "loss": 0.1403,
1254
+ "masked_tokens": 87.9875,
1255
+ "mean_t": 0.49193521235138177,
1256
+ "step": 520,
1257
+ "student_masked_tokens": 87.9875
1258
+ },
1259
+ {
1260
+ "avg_mask_ratio": 0.46718079667771234,
1261
+ "avg_response_length": 218.0875,
1262
+ "avg_student_mask_ratio": 0.46718079667771234,
1263
+ "batch_ainp_frac": 0.0,
1264
+ "batch_inp_frac": 0.0,
1265
+ "batch_inp_oh_frac": 0.0,
1266
+ "batch_inp_par_frac": 1.0,
1267
+ "batch_inp_par_reverse_frac": 0.0,
1268
+ "batch_rl_frac": 0.0,
1269
+ "batch_sft_frac": 0.0,
1270
+ "batch_soft_sft_frac": 0.0,
1271
+ "batch_tf_frac": 0.0,
1272
+ "ce_loss": 0.0,
1273
+ "epoch": 1.1322666666666668,
1274
+ "grad_norm": 0.2109375,
1275
+ "kd_loss": 0.15268151032492625,
1276
+ "learning_rate": 3e-06,
1277
+ "loss": 0.1789,
1278
+ "masked_tokens": 101.3875,
1279
+ "mean_t": 0.49621942077938,
1280
+ "step": 530,
1281
+ "student_masked_tokens": 101.3875
1282
+ },
1283
+ {
1284
+ "avg_mask_ratio": 0.4890203754650429,
1285
+ "avg_response_length": 244.3875,
1286
+ "avg_student_mask_ratio": 0.4890203754650429,
1287
+ "batch_ainp_frac": 0.0,
1288
+ "batch_inp_frac": 0.0,
1289
+ "batch_inp_oh_frac": 0.0,
1290
+ "batch_inp_par_frac": 1.0,
1291
+ "batch_inp_par_reverse_frac": 0.0,
1292
+ "batch_rl_frac": 0.0,
1293
+ "batch_sft_frac": 0.0,
1294
+ "batch_soft_sft_frac": 0.0,
1295
+ "batch_tf_frac": 0.0,
1296
+ "ce_loss": 0.0,
1297
+ "epoch": 1.1536,
1298
+ "grad_norm": 0.27734375,
1299
+ "kd_loss": 0.17835129436630268,
1300
+ "learning_rate": 3e-06,
1301
+ "loss": 0.2173,
1302
+ "masked_tokens": 116.175,
1303
+ "mean_t": 0.5208023569080978,
1304
+ "step": 540,
1305
+ "student_masked_tokens": 116.175
1306
+ },
1307
+ {
1308
+ "avg_mask_ratio": 0.45064474650425834,
1309
+ "avg_response_length": 217.8375,
1310
+ "avg_student_mask_ratio": 0.45064474650425834,
1311
+ "batch_ainp_frac": 0.0,
1312
+ "batch_inp_frac": 0.0,
1313
+ "batch_inp_oh_frac": 0.0,
1314
+ "batch_inp_par_frac": 1.0,
1315
+ "batch_inp_par_reverse_frac": 0.0,
1316
+ "batch_rl_frac": 0.0,
1317
+ "batch_sft_frac": 0.0,
1318
+ "batch_soft_sft_frac": 0.0,
1319
+ "batch_tf_frac": 0.0,
1320
+ "ce_loss": 0.0,
1321
+ "epoch": 1.1749333333333334,
1322
+ "grad_norm": 0.17578125,
1323
+ "kd_loss": 0.16049880692362706,
1324
+ "learning_rate": 3e-06,
1325
+ "loss": 0.1855,
1326
+ "masked_tokens": 98.6375,
1327
+ "mean_t": 0.4767197913257405,
1328
+ "step": 550,
1329
+ "student_masked_tokens": 98.6375
1330
+ },
1331
+ {
1332
+ "avg_mask_ratio": 0.3844255942822201,
1333
+ "avg_response_length": 240.825,
1334
+ "avg_student_mask_ratio": 0.3844255942822201,
1335
+ "batch_ainp_frac": 0.0,
1336
+ "batch_inp_frac": 0.0,
1337
+ "batch_inp_oh_frac": 0.0,
1338
+ "batch_inp_par_frac": 1.0,
1339
+ "batch_inp_par_reverse_frac": 0.0,
1340
+ "batch_rl_frac": 0.0,
1341
+ "batch_sft_frac": 0.0,
1342
+ "batch_soft_sft_frac": 0.0,
1343
+ "batch_tf_frac": 0.0,
1344
+ "ce_loss": 0.0,
1345
+ "epoch": 1.1962666666666666,
1346
+ "grad_norm": 0.5859375,
1347
+ "kd_loss": 0.17605857607457268,
1348
+ "learning_rate": 3e-06,
1349
+ "loss": 0.1885,
1350
+ "masked_tokens": 96.85,
1351
+ "mean_t": 0.4111072298779618,
1352
+ "step": 560,
1353
+ "student_masked_tokens": 96.85
1354
+ },
1355
+ {
1356
+ "avg_mask_ratio": 0.45103558609262107,
1357
+ "avg_response_length": 231.025,
1358
+ "avg_student_mask_ratio": 0.45103558609262107,
1359
+ "batch_ainp_frac": 0.0,
1360
+ "batch_inp_frac": 0.0,
1361
+ "batch_inp_oh_frac": 0.0,
1362
+ "batch_inp_par_frac": 1.0,
1363
+ "batch_inp_par_reverse_frac": 0.0,
1364
+ "batch_rl_frac": 0.0,
1365
+ "batch_sft_frac": 0.0,
1366
+ "batch_soft_sft_frac": 0.0,
1367
+ "batch_tf_frac": 0.0,
1368
+ "ce_loss": 0.0,
1369
+ "epoch": 1.2176,
1370
+ "grad_norm": 0.234375,
1371
+ "kd_loss": 0.1386162672638477,
1372
+ "learning_rate": 3e-06,
1373
+ "loss": 0.1681,
1374
+ "masked_tokens": 105.6875,
1375
+ "mean_t": 0.48533305872697385,
1376
+ "step": 570,
1377
+ "student_masked_tokens": 105.6875
1378
+ },
1379
+ {
1380
+ "avg_mask_ratio": 0.47033366551622746,
1381
+ "avg_response_length": 248.3875,
1382
+ "avg_student_mask_ratio": 0.47033366551622746,
1383
+ "batch_ainp_frac": 0.0,
1384
+ "batch_inp_frac": 0.0,
1385
+ "batch_inp_oh_frac": 0.0,
1386
+ "batch_inp_par_frac": 1.0,
1387
+ "batch_inp_par_reverse_frac": 0.0,
1388
+ "batch_rl_frac": 0.0,
1389
+ "batch_sft_frac": 0.0,
1390
+ "batch_soft_sft_frac": 0.0,
1391
+ "batch_tf_frac": 0.0,
1392
+ "ce_loss": 0.0,
1393
+ "epoch": 1.2389333333333332,
1394
+ "grad_norm": 0.26953125,
1395
+ "kd_loss": 0.17702910760917803,
1396
+ "learning_rate": 3e-06,
1397
+ "loss": 0.1899,
1398
+ "masked_tokens": 125.05,
1399
+ "mean_t": 0.49976949762785805,
1400
+ "step": 580,
1401
+ "student_masked_tokens": 125.05
1402
+ },
1403
+ {
1404
+ "avg_mask_ratio": 0.5230229062028229,
1405
+ "avg_response_length": 241.8125,
1406
+ "avg_student_mask_ratio": 0.5230229062028229,
1407
+ "batch_ainp_frac": 0.0,
1408
+ "batch_inp_frac": 0.0,
1409
+ "batch_inp_oh_frac": 0.0,
1410
+ "batch_inp_par_frac": 1.0,
1411
+ "batch_inp_par_reverse_frac": 0.0,
1412
+ "batch_rl_frac": 0.0,
1413
+ "batch_sft_frac": 0.0,
1414
+ "batch_soft_sft_frac": 0.0,
1415
+ "batch_tf_frac": 0.0,
1416
+ "ce_loss": 0.0,
1417
+ "epoch": 1.2602666666666666,
1418
+ "grad_norm": 0.20703125,
1419
+ "kd_loss": 0.22054996666956866,
1420
+ "learning_rate": 3e-06,
1421
+ "loss": 0.2233,
1422
+ "masked_tokens": 129.0,
1423
+ "mean_t": 0.5521843038732186,
1424
+ "step": 590,
1425
+ "student_masked_tokens": 129.0
1426
+ },
1427
+ {
1428
+ "avg_mask_ratio": 0.44929012526990847,
1429
+ "avg_response_length": 246.7375,
1430
+ "avg_student_mask_ratio": 0.44929012526990847,
1431
+ "batch_ainp_frac": 0.0,
1432
+ "batch_inp_frac": 0.0,
1433
+ "batch_inp_oh_frac": 0.0,
1434
+ "batch_inp_par_frac": 1.0,
1435
+ "batch_inp_par_reverse_frac": 0.0,
1436
+ "batch_rl_frac": 0.0,
1437
+ "batch_sft_frac": 0.0,
1438
+ "batch_soft_sft_frac": 0.0,
1439
+ "batch_tf_frac": 0.0,
1440
+ "ce_loss": 0.0,
1441
+ "epoch": 1.2816,
1442
+ "grad_norm": 0.212890625,
1443
+ "kd_loss": 0.15257543138572202,
1444
+ "learning_rate": 3e-06,
1445
+ "loss": 0.1581,
1446
+ "masked_tokens": 113.9375,
1447
+ "mean_t": 0.4788527532829903,
1448
+ "step": 600,
1449
+ "student_masked_tokens": 113.9375
1450
+ },
1451
+ {
1452
+ "avg_mask_ratio": 0.5061312943696976,
1453
+ "avg_response_length": 237.175,
1454
+ "avg_student_mask_ratio": 0.5061312943696976,
1455
+ "batch_ainp_frac": 0.0,
1456
+ "batch_inp_frac": 0.0,
1457
+ "batch_inp_oh_frac": 0.0,
1458
+ "batch_inp_par_frac": 1.0,
1459
+ "batch_inp_par_reverse_frac": 0.0,
1460
+ "batch_rl_frac": 0.0,
1461
+ "batch_sft_frac": 0.0,
1462
+ "batch_soft_sft_frac": 0.0,
1463
+ "batch_tf_frac": 0.0,
1464
+ "ce_loss": 0.0,
1465
+ "epoch": 1.3029333333333333,
1466
+ "grad_norm": 0.06884765625,
1467
+ "kd_loss": 0.21158275993875578,
1468
+ "learning_rate": 3e-06,
1469
+ "loss": 0.1942,
1470
+ "masked_tokens": 116.675,
1471
+ "mean_t": 0.5445419924799353,
1472
+ "step": 610,
1473
+ "student_masked_tokens": 116.675
1474
+ },
1475
+ {
1476
+ "avg_mask_ratio": 0.4925117701757699,
1477
+ "avg_response_length": 233.2625,
1478
+ "avg_student_mask_ratio": 0.4925117701757699,
1479
+ "batch_ainp_frac": 0.0,
1480
+ "batch_inp_frac": 0.0,
1481
+ "batch_inp_oh_frac": 0.0,
1482
+ "batch_inp_par_frac": 1.0,
1483
+ "batch_inp_par_reverse_frac": 0.0,
1484
+ "batch_rl_frac": 0.0,
1485
+ "batch_sft_frac": 0.0,
1486
+ "batch_soft_sft_frac": 0.0,
1487
+ "batch_tf_frac": 0.0,
1488
+ "ce_loss": 0.0,
1489
+ "epoch": 1.3242666666666667,
1490
+ "grad_norm": 0.1787109375,
1491
+ "kd_loss": 0.1805886138310143,
1492
+ "learning_rate": 3e-06,
1493
+ "loss": 0.1744,
1494
+ "masked_tokens": 111.6875,
1495
+ "mean_t": 0.52490478400141,
1496
+ "step": 620,
1497
+ "student_masked_tokens": 111.6875
1498
+ },
1499
+ {
1500
+ "avg_mask_ratio": 0.5307835865532979,
1501
+ "avg_response_length": 249.65,
1502
+ "avg_student_mask_ratio": 0.5307835865532979,
1503
+ "batch_ainp_frac": 0.0,
1504
+ "batch_inp_frac": 0.0,
1505
+ "batch_inp_oh_frac": 0.0,
1506
+ "batch_inp_par_frac": 1.0,
1507
+ "batch_inp_par_reverse_frac": 0.0,
1508
+ "batch_rl_frac": 0.0,
1509
+ "batch_sft_frac": 0.0,
1510
+ "batch_soft_sft_frac": 0.0,
1511
+ "batch_tf_frac": 0.0,
1512
+ "ce_loss": 0.0,
1513
+ "epoch": 1.3456000000000001,
1514
+ "grad_norm": 0.28125,
1515
+ "kd_loss": 0.25403604302136956,
1516
+ "learning_rate": 3e-06,
1517
+ "loss": 0.2627,
1518
+ "masked_tokens": 126.6875,
1519
+ "mean_t": 0.5564947265549562,
1520
+ "step": 630,
1521
+ "student_masked_tokens": 126.6875
1522
+ },
1523
+ {
1524
+ "avg_mask_ratio": 0.48631439574528484,
1525
+ "avg_response_length": 241.475,
1526
+ "avg_student_mask_ratio": 0.48631439574528484,
1527
+ "batch_ainp_frac": 0.0,
1528
+ "batch_inp_frac": 0.0,
1529
+ "batch_inp_oh_frac": 0.0,
1530
+ "batch_inp_par_frac": 1.0,
1531
+ "batch_inp_par_reverse_frac": 0.0,
1532
+ "batch_rl_frac": 0.0,
1533
+ "batch_sft_frac": 0.0,
1534
+ "batch_soft_sft_frac": 0.0,
1535
+ "batch_tf_frac": 0.0,
1536
+ "ce_loss": 0.0,
1537
+ "epoch": 1.3669333333333333,
1538
+ "grad_norm": 0.2412109375,
1539
+ "kd_loss": 0.1642333460577902,
1540
+ "learning_rate": 3e-06,
1541
+ "loss": 0.1731,
1542
+ "masked_tokens": 122.1125,
1543
+ "mean_t": 0.5093393943971023,
1544
+ "step": 640,
1545
+ "student_masked_tokens": 122.1125
1546
+ },
1547
+ {
1548
+ "avg_mask_ratio": 0.5248487972887232,
1549
+ "avg_response_length": 231.5375,
1550
+ "avg_student_mask_ratio": 0.5248487972887232,
1551
+ "batch_ainp_frac": 0.0,
1552
+ "batch_inp_frac": 0.0,
1553
+ "batch_inp_oh_frac": 0.0,
1554
+ "batch_inp_par_frac": 1.0,
1555
+ "batch_inp_par_reverse_frac": 0.0,
1556
+ "batch_rl_frac": 0.0,
1557
+ "batch_sft_frac": 0.0,
1558
+ "batch_soft_sft_frac": 0.0,
1559
+ "batch_tf_frac": 0.0,
1560
+ "ce_loss": 0.0,
1561
+ "epoch": 1.3882666666666665,
1562
+ "grad_norm": 0.193359375,
1563
+ "kd_loss": 0.2508082524812494,
1564
+ "learning_rate": 3e-06,
1565
+ "loss": 0.2054,
1566
+ "masked_tokens": 127.175,
1567
+ "mean_t": 0.5558586571365595,
1568
+ "step": 650,
1569
+ "student_masked_tokens": 127.175
1570
+ },
1571
+ {
1572
+ "avg_mask_ratio": 0.48489007767057046,
1573
+ "avg_response_length": 246.0125,
1574
+ "avg_student_mask_ratio": 0.48489007767057046,
1575
+ "batch_ainp_frac": 0.0,
1576
+ "batch_inp_frac": 0.0,
1577
+ "batch_inp_oh_frac": 0.0,
1578
+ "batch_inp_par_frac": 1.0,
1579
+ "batch_inp_par_reverse_frac": 0.0,
1580
+ "batch_rl_frac": 0.0,
1581
+ "batch_sft_frac": 0.0,
1582
+ "batch_soft_sft_frac": 0.0,
1583
+ "batch_tf_frac": 0.0,
1584
+ "ce_loss": 0.0,
1585
+ "epoch": 1.4096,
1586
+ "grad_norm": 0.28125,
1587
+ "kd_loss": 0.18781521328146483,
1588
+ "learning_rate": 3e-06,
1589
+ "loss": 0.2045,
1590
+ "masked_tokens": 125.225,
1591
+ "mean_t": 0.51307404555846,
1592
+ "step": 660,
1593
+ "student_masked_tokens": 125.225
1594
+ },
1595
+ {
1596
+ "avg_mask_ratio": 0.4605832444969565,
1597
+ "avg_response_length": 244.5,
1598
+ "avg_student_mask_ratio": 0.4605832444969565,
1599
+ "batch_ainp_frac": 0.0,
1600
+ "batch_inp_frac": 0.0,
1601
+ "batch_inp_oh_frac": 0.0,
1602
+ "batch_inp_par_frac": 1.0,
1603
+ "batch_inp_par_reverse_frac": 0.0,
1604
+ "batch_rl_frac": 0.0,
1605
+ "batch_sft_frac": 0.0,
1606
+ "batch_soft_sft_frac": 0.0,
1607
+ "batch_tf_frac": 0.0,
1608
+ "ce_loss": 0.0,
1609
+ "epoch": 1.4309333333333334,
1610
+ "grad_norm": 0.19140625,
1611
+ "kd_loss": 0.1806626110754223,
1612
+ "learning_rate": 3e-06,
1613
+ "loss": 0.1702,
1614
+ "masked_tokens": 120.825,
1615
+ "mean_t": 0.4907285622088239,
1616
+ "step": 670,
1617
+ "student_masked_tokens": 120.825
1618
+ },
1619
+ {
1620
+ "avg_mask_ratio": 0.4662455078505445,
1621
+ "avg_response_length": 244.6125,
1622
+ "avg_student_mask_ratio": 0.4662455078505445,
1623
+ "batch_ainp_frac": 0.0,
1624
+ "batch_inp_frac": 0.0,
1625
+ "batch_inp_oh_frac": 0.0,
1626
+ "batch_inp_par_frac": 1.0,
1627
+ "batch_inp_par_reverse_frac": 0.0,
1628
+ "batch_rl_frac": 0.0,
1629
+ "batch_sft_frac": 0.0,
1630
+ "batch_soft_sft_frac": 0.0,
1631
+ "batch_tf_frac": 0.0,
1632
+ "ce_loss": 0.0,
1633
+ "epoch": 1.4522666666666666,
1634
+ "grad_norm": 0.16796875,
1635
+ "kd_loss": 0.20038694294766798,
1636
+ "learning_rate": 3e-06,
1637
+ "loss": 0.1824,
1638
+ "masked_tokens": 114.2,
1639
+ "mean_t": 0.4913603452499956,
1640
+ "step": 680,
1641
+ "student_masked_tokens": 114.2
1642
+ },
1643
+ {
1644
+ "avg_mask_ratio": 0.4820630593923852,
1645
+ "avg_response_length": 217.3,
1646
+ "avg_student_mask_ratio": 0.4820630593923852,
1647
+ "batch_ainp_frac": 0.0,
1648
+ "batch_inp_frac": 0.0,
1649
+ "batch_inp_oh_frac": 0.0,
1650
+ "batch_inp_par_frac": 1.0,
1651
+ "batch_inp_par_reverse_frac": 0.0,
1652
+ "batch_rl_frac": 0.0,
1653
+ "batch_sft_frac": 0.0,
1654
+ "batch_soft_sft_frac": 0.0,
1655
+ "batch_tf_frac": 0.0,
1656
+ "ce_loss": 0.0,
1657
+ "epoch": 1.4736,
1658
+ "grad_norm": 0.11279296875,
1659
+ "kd_loss": 0.16563009086588637,
1660
+ "learning_rate": 3e-06,
1661
+ "loss": 0.17,
1662
+ "masked_tokens": 118.4875,
1663
+ "mean_t": 0.5133644798654131,
1664
+ "step": 690,
1665
+ "student_masked_tokens": 118.4875
1666
+ },
1667
+ {
1668
+ "avg_mask_ratio": 0.5206489040749147,
1669
+ "avg_response_length": 216.45,
1670
+ "avg_student_mask_ratio": 0.5206489040749147,
1671
+ "batch_ainp_frac": 0.0,
1672
+ "batch_inp_frac": 0.0,
1673
+ "batch_inp_oh_frac": 0.0,
1674
+ "batch_inp_par_frac": 1.0,
1675
+ "batch_inp_par_reverse_frac": 0.0,
1676
+ "batch_rl_frac": 0.0,
1677
+ "batch_sft_frac": 0.0,
1678
+ "batch_soft_sft_frac": 0.0,
1679
+ "batch_tf_frac": 0.0,
1680
+ "ce_loss": 0.0,
1681
+ "epoch": 1.4949333333333334,
1682
+ "grad_norm": 0.236328125,
1683
+ "kd_loss": 0.23649522811845144,
1684
+ "learning_rate": 3e-06,
1685
+ "loss": 0.2241,
1686
+ "masked_tokens": 118.4375,
1687
+ "mean_t": 0.5625698395539075,
1688
+ "step": 700,
1689
+ "student_masked_tokens": 118.4375
1690
+ },
1691
+ {
1692
+ "avg_mask_ratio": 0.4955552547937259,
1693
+ "avg_response_length": 211.175,
1694
+ "avg_student_mask_ratio": 0.4955552547937259,
1695
+ "batch_ainp_frac": 0.0,
1696
+ "batch_inp_frac": 0.0,
1697
+ "batch_inp_oh_frac": 0.0,
1698
+ "batch_inp_par_frac": 1.0,
1699
+ "batch_inp_par_reverse_frac": 0.0,
1700
+ "batch_rl_frac": 0.0,
1701
+ "batch_sft_frac": 0.0,
1702
+ "batch_soft_sft_frac": 0.0,
1703
+ "batch_tf_frac": 0.0,
1704
+ "ce_loss": 0.0,
1705
+ "epoch": 1.5162666666666667,
1706
+ "grad_norm": 0.275390625,
1707
+ "kd_loss": 0.23970817765721222,
1708
+ "learning_rate": 3e-06,
1709
+ "loss": 0.2184,
1710
+ "masked_tokens": 112.3375,
1711
+ "mean_t": 0.5135623761918395,
1712
+ "step": 710,
1713
+ "student_masked_tokens": 112.3375
1714
+ },
1715
+ {
1716
+ "avg_mask_ratio": 0.4820543818641454,
1717
+ "avg_response_length": 229.1375,
1718
+ "avg_student_mask_ratio": 0.4820543818641454,
1719
+ "batch_ainp_frac": 0.0,
1720
+ "batch_inp_frac": 0.0,
1721
+ "batch_inp_oh_frac": 0.0,
1722
+ "batch_inp_par_frac": 1.0,
1723
+ "batch_inp_par_reverse_frac": 0.0,
1724
+ "batch_rl_frac": 0.0,
1725
+ "batch_sft_frac": 0.0,
1726
+ "batch_soft_sft_frac": 0.0,
1727
+ "batch_tf_frac": 0.0,
1728
+ "ce_loss": 0.0,
1729
+ "epoch": 1.5375999999999999,
1730
+ "grad_norm": 0.435546875,
1731
+ "kd_loss": 0.18955910701470202,
1732
+ "learning_rate": 3e-06,
1733
+ "loss": 0.1978,
1734
+ "masked_tokens": 114.55,
1735
+ "mean_t": 0.5117021896177902,
1736
+ "step": 720,
1737
+ "student_masked_tokens": 114.55
1738
+ },
1739
+ {
1740
+ "avg_mask_ratio": 0.4605119539948646,
1741
+ "avg_response_length": 245.5625,
1742
+ "avg_student_mask_ratio": 0.4605119539948646,
1743
+ "batch_ainp_frac": 0.0,
1744
+ "batch_inp_frac": 0.0,
1745
+ "batch_inp_oh_frac": 0.0,
1746
+ "batch_inp_par_frac": 1.0,
1747
+ "batch_inp_par_reverse_frac": 0.0,
1748
+ "batch_rl_frac": 0.0,
1749
+ "batch_sft_frac": 0.0,
1750
+ "batch_soft_sft_frac": 0.0,
1751
+ "batch_tf_frac": 0.0,
1752
+ "ce_loss": 0.0,
1753
+ "epoch": 1.5589333333333333,
1754
+ "grad_norm": 0.1103515625,
1755
+ "kd_loss": 0.16859328244926958,
1756
+ "learning_rate": 3e-06,
1757
+ "loss": 0.1779,
1758
+ "masked_tokens": 113.85,
1759
+ "mean_t": 0.49050743713742123,
1760
+ "step": 730,
1761
+ "student_masked_tokens": 113.85
1762
+ },
1763
+ {
1764
+ "avg_mask_ratio": 0.5134038798511028,
1765
+ "avg_response_length": 194.125,
1766
+ "avg_student_mask_ratio": 0.5134038798511028,
1767
+ "batch_ainp_frac": 0.0,
1768
+ "batch_inp_frac": 0.0,
1769
+ "batch_inp_oh_frac": 0.0,
1770
+ "batch_inp_par_frac": 1.0,
1771
+ "batch_inp_par_reverse_frac": 0.0,
1772
+ "batch_rl_frac": 0.0,
1773
+ "batch_sft_frac": 0.0,
1774
+ "batch_soft_sft_frac": 0.0,
1775
+ "batch_tf_frac": 0.0,
1776
+ "ce_loss": 0.0,
1777
+ "epoch": 1.5802666666666667,
1778
+ "grad_norm": 0.0830078125,
1779
+ "kd_loss": 0.17122714665274544,
1780
+ "learning_rate": 3e-06,
1781
+ "loss": 0.2018,
1782
+ "masked_tokens": 102.9375,
1783
+ "mean_t": 0.5505168779753149,
1784
+ "step": 740,
1785
+ "student_masked_tokens": 102.9375
1786
+ },
1787
+ {
1788
+ "avg_mask_ratio": 0.4201362137740944,
1789
+ "avg_response_length": 223.55,
1790
+ "avg_student_mask_ratio": 0.4201362137740944,
1791
+ "batch_ainp_frac": 0.0,
1792
+ "batch_inp_frac": 0.0,
1793
+ "batch_inp_oh_frac": 0.0,
1794
+ "batch_inp_par_frac": 1.0,
1795
+ "batch_inp_par_reverse_frac": 0.0,
1796
+ "batch_rl_frac": 0.0,
1797
+ "batch_sft_frac": 0.0,
1798
+ "batch_soft_sft_frac": 0.0,
1799
+ "batch_tf_frac": 0.0,
1800
+ "ce_loss": 0.0,
1801
+ "epoch": 1.6016,
1802
+ "grad_norm": 0.16796875,
1803
+ "kd_loss": 0.19197621339357057,
1804
+ "learning_rate": 3e-06,
1805
+ "loss": 0.1792,
1806
+ "masked_tokens": 94.7125,
1807
+ "mean_t": 0.4437690361432033,
1808
+ "step": 750,
1809
+ "student_masked_tokens": 94.7125
1810
+ },
1811
+ {
1812
+ "avg_mask_ratio": 0.46924527404480615,
1813
+ "avg_response_length": 227.35,
1814
+ "avg_student_mask_ratio": 0.46924527404480615,
1815
+ "batch_ainp_frac": 0.0,
1816
+ "batch_inp_frac": 0.0,
1817
+ "batch_inp_oh_frac": 0.0,
1818
+ "batch_inp_par_frac": 1.0,
1819
+ "batch_inp_par_reverse_frac": 0.0,
1820
+ "batch_rl_frac": 0.0,
1821
+ "batch_sft_frac": 0.0,
1822
+ "batch_soft_sft_frac": 0.0,
1823
+ "batch_tf_frac": 0.0,
1824
+ "ce_loss": 0.0,
1825
+ "epoch": 1.6229333333333333,
1826
+ "grad_norm": 0.8125,
1827
+ "kd_loss": 0.24938117066874382,
1828
+ "learning_rate": 3e-06,
1829
+ "loss": 0.2591,
1830
+ "masked_tokens": 107.5375,
1831
+ "mean_t": 0.49132869170280175,
1832
+ "step": 760,
1833
+ "student_masked_tokens": 107.5375
1834
+ },
1835
+ {
1836
+ "avg_mask_ratio": 0.445551612455165,
1837
+ "avg_response_length": 268.6625,
1838
+ "avg_student_mask_ratio": 0.445551612455165,
1839
+ "batch_ainp_frac": 0.0,
1840
+ "batch_inp_frac": 0.0,
1841
+ "batch_inp_oh_frac": 0.0,
1842
+ "batch_inp_par_frac": 1.0,
1843
+ "batch_inp_par_reverse_frac": 0.0,
1844
+ "batch_rl_frac": 0.0,
1845
+ "batch_sft_frac": 0.0,
1846
+ "batch_soft_sft_frac": 0.0,
1847
+ "batch_tf_frac": 0.0,
1848
+ "ce_loss": 0.0,
1849
+ "epoch": 1.6442666666666668,
1850
+ "grad_norm": 0.1435546875,
1851
+ "kd_loss": 0.1928954417056957,
1852
+ "learning_rate": 3e-06,
1853
+ "loss": 0.1563,
1854
+ "masked_tokens": 124.0875,
1855
+ "mean_t": 0.4667695587326307,
1856
+ "step": 770,
1857
+ "student_masked_tokens": 124.0875
1858
+ },
1859
+ {
1860
+ "avg_mask_ratio": 0.4321410794305848,
1861
+ "avg_response_length": 256.1625,
1862
+ "avg_student_mask_ratio": 0.4321410794305848,
1863
+ "batch_ainp_frac": 0.0,
1864
+ "batch_inp_frac": 0.0,
1865
+ "batch_inp_oh_frac": 0.0,
1866
+ "batch_inp_par_frac": 1.0,
1867
+ "batch_inp_par_reverse_frac": 0.0,
1868
+ "batch_rl_frac": 0.0,
1869
+ "batch_sft_frac": 0.0,
1870
+ "batch_soft_sft_frac": 0.0,
1871
+ "batch_tf_frac": 0.0,
1872
+ "ce_loss": 0.0,
1873
+ "epoch": 1.6656,
1874
+ "grad_norm": 0.2216796875,
1875
+ "kd_loss": 0.15059620087446887,
1876
+ "learning_rate": 3e-06,
1877
+ "loss": 0.1534,
1878
+ "masked_tokens": 117.7,
1879
+ "mean_t": 0.44942845597106496,
1880
+ "step": 780,
1881
+ "student_masked_tokens": 117.7
1882
+ },
1883
+ {
1884
+ "avg_mask_ratio": 0.4697655299096368,
1885
+ "avg_response_length": 240.6125,
1886
+ "avg_student_mask_ratio": 0.4697655299096368,
1887
+ "batch_ainp_frac": 0.0,
1888
+ "batch_inp_frac": 0.0,
1889
+ "batch_inp_oh_frac": 0.0,
1890
+ "batch_inp_par_frac": 1.0,
1891
+ "batch_inp_par_reverse_frac": 0.0,
1892
+ "batch_rl_frac": 0.0,
1893
+ "batch_sft_frac": 0.0,
1894
+ "batch_soft_sft_frac": 0.0,
1895
+ "batch_tf_frac": 0.0,
1896
+ "ce_loss": 0.0,
1897
+ "epoch": 1.6869333333333332,
1898
+ "grad_norm": 0.255859375,
1899
+ "kd_loss": 0.16427693545232777,
1900
+ "learning_rate": 3e-06,
1901
+ "loss": 0.1489,
1902
+ "masked_tokens": 113.5375,
1903
+ "mean_t": 0.4993515375303105,
1904
+ "step": 790,
1905
+ "student_masked_tokens": 113.5375
1906
+ },
1907
+ {
1908
+ "avg_mask_ratio": 0.4635992758907378,
1909
+ "avg_response_length": 224.15,
1910
+ "avg_student_mask_ratio": 0.4635992758907378,
1911
+ "batch_ainp_frac": 0.0,
1912
+ "batch_inp_frac": 0.0,
1913
+ "batch_inp_oh_frac": 0.0,
1914
+ "batch_inp_par_frac": 1.0,
1915
+ "batch_inp_par_reverse_frac": 0.0,
1916
+ "batch_rl_frac": 0.0,
1917
+ "batch_sft_frac": 0.0,
1918
+ "batch_soft_sft_frac": 0.0,
1919
+ "batch_tf_frac": 0.0,
1920
+ "ce_loss": 0.0,
1921
+ "epoch": 1.7082666666666668,
1922
+ "grad_norm": 0.197265625,
1923
+ "kd_loss": 0.15171801659575976,
1924
+ "learning_rate": 3e-06,
1925
+ "loss": 0.1526,
1926
+ "masked_tokens": 107.5,
1927
+ "mean_t": 0.500370389316231,
1928
+ "step": 800,
1929
+ "student_masked_tokens": 107.5
1930
+ },
1931
+ {
1932
+ "avg_mask_ratio": 0.5018501321552321,
1933
+ "avg_response_length": 235.1625,
1934
+ "avg_student_mask_ratio": 0.5018501321552321,
1935
+ "batch_ainp_frac": 0.0,
1936
+ "batch_inp_frac": 0.0,
1937
+ "batch_inp_oh_frac": 0.0,
1938
+ "batch_inp_par_frac": 1.0,
1939
+ "batch_inp_par_reverse_frac": 0.0,
1940
+ "batch_rl_frac": 0.0,
1941
+ "batch_sft_frac": 0.0,
1942
+ "batch_soft_sft_frac": 0.0,
1943
+ "batch_tf_frac": 0.0,
1944
+ "ce_loss": 0.0,
1945
+ "epoch": 1.7296,
1946
+ "grad_norm": 0.1630859375,
1947
+ "kd_loss": 0.18931926304685476,
1948
+ "learning_rate": 3e-06,
1949
+ "loss": 0.2031,
1950
+ "masked_tokens": 116.6125,
1951
+ "mean_t": 0.5317009104182944,
1952
+ "step": 810,
1953
+ "student_masked_tokens": 116.6125
1954
+ },
1955
+ {
1956
+ "avg_mask_ratio": 0.5050785383209586,
1957
+ "avg_response_length": 207.6875,
1958
+ "avg_student_mask_ratio": 0.5050785383209586,
1959
+ "batch_ainp_frac": 0.0,
1960
+ "batch_inp_frac": 0.0,
1961
+ "batch_inp_oh_frac": 0.0,
1962
+ "batch_inp_par_frac": 1.0,
1963
+ "batch_inp_par_reverse_frac": 0.0,
1964
+ "batch_rl_frac": 0.0,
1965
+ "batch_sft_frac": 0.0,
1966
+ "batch_soft_sft_frac": 0.0,
1967
+ "batch_tf_frac": 0.0,
1968
+ "ce_loss": 0.0,
1969
+ "epoch": 1.7509333333333332,
1970
+ "grad_norm": 0.5546875,
1971
+ "kd_loss": 0.24824176937403308,
1972
+ "learning_rate": 3e-06,
1973
+ "loss": 0.2566,
1974
+ "masked_tokens": 119.175,
1975
+ "mean_t": 0.5392061032878701,
1976
+ "step": 820,
1977
+ "student_masked_tokens": 119.175
1978
+ },
1979
+ {
1980
+ "avg_mask_ratio": 0.4980328972451389,
1981
+ "avg_response_length": 270.325,
1982
+ "avg_student_mask_ratio": 0.4980328972451389,
1983
+ "batch_ainp_frac": 0.0,
1984
+ "batch_inp_frac": 0.0,
1985
+ "batch_inp_oh_frac": 0.0,
1986
+ "batch_inp_par_frac": 1.0,
1987
+ "batch_inp_par_reverse_frac": 0.0,
1988
+ "batch_rl_frac": 0.0,
1989
+ "batch_sft_frac": 0.0,
1990
+ "batch_soft_sft_frac": 0.0,
1991
+ "batch_tf_frac": 0.0,
1992
+ "ce_loss": 0.0,
1993
+ "epoch": 1.7722666666666667,
1994
+ "grad_norm": 0.09912109375,
1995
+ "kd_loss": 0.1924194690429431,
1996
+ "learning_rate": 3e-06,
1997
+ "loss": 0.2006,
1998
+ "masked_tokens": 141.925,
1999
+ "mean_t": 0.5238314627087675,
2000
+ "step": 830,
2001
+ "student_masked_tokens": 141.925
2002
+ },
2003
+ {
2004
+ "avg_mask_ratio": 0.493249478796497,
2005
+ "avg_response_length": 226.025,
2006
+ "avg_student_mask_ratio": 0.493249478796497,
2007
+ "batch_ainp_frac": 0.0,
2008
+ "batch_inp_frac": 0.0,
2009
+ "batch_inp_oh_frac": 0.0,
2010
+ "batch_inp_par_frac": 1.0,
2011
+ "batch_inp_par_reverse_frac": 0.0,
2012
+ "batch_rl_frac": 0.0,
2013
+ "batch_sft_frac": 0.0,
2014
+ "batch_soft_sft_frac": 0.0,
2015
+ "batch_tf_frac": 0.0,
2016
+ "ce_loss": 0.0,
2017
+ "epoch": 1.7936,
2018
+ "grad_norm": 0.158203125,
2019
+ "kd_loss": 0.15751813794203606,
2020
+ "learning_rate": 3e-06,
2021
+ "loss": 0.1624,
2022
+ "masked_tokens": 118.825,
2023
+ "mean_t": 0.5301066277665086,
2024
+ "step": 840,
2025
+ "student_masked_tokens": 118.825
2026
+ },
2027
+ {
2028
+ "avg_mask_ratio": 0.5009000841644593,
2029
+ "avg_response_length": 233.025,
2030
+ "avg_student_mask_ratio": 0.5009000841644593,
2031
+ "batch_ainp_frac": 0.0,
2032
+ "batch_inp_frac": 0.0,
2033
+ "batch_inp_oh_frac": 0.0,
2034
+ "batch_inp_par_frac": 1.0,
2035
+ "batch_inp_par_reverse_frac": 0.0,
2036
+ "batch_rl_frac": 0.0,
2037
+ "batch_sft_frac": 0.0,
2038
+ "batch_soft_sft_frac": 0.0,
2039
+ "batch_tf_frac": 0.0,
2040
+ "ce_loss": 0.0,
2041
+ "epoch": 1.8149333333333333,
2042
+ "grad_norm": 0.10205078125,
2043
+ "kd_loss": 0.1860738446495816,
2044
+ "learning_rate": 3e-06,
2045
+ "loss": 0.1737,
2046
+ "masked_tokens": 117.3125,
2047
+ "mean_t": 0.5343429344706238,
2048
+ "step": 850,
2049
+ "student_masked_tokens": 117.3125
2050
+ },
2051
+ {
2052
+ "avg_mask_ratio": 0.46293387678451836,
2053
+ "avg_response_length": 231.3625,
2054
+ "avg_student_mask_ratio": 0.46293387678451836,
2055
+ "batch_ainp_frac": 0.0,
2056
+ "batch_inp_frac": 0.0,
2057
+ "batch_inp_oh_frac": 0.0,
2058
+ "batch_inp_par_frac": 1.0,
2059
+ "batch_inp_par_reverse_frac": 0.0,
2060
+ "batch_rl_frac": 0.0,
2061
+ "batch_sft_frac": 0.0,
2062
+ "batch_soft_sft_frac": 0.0,
2063
+ "batch_tf_frac": 0.0,
2064
+ "ce_loss": 0.0,
2065
+ "epoch": 1.8362666666666667,
2066
+ "grad_norm": 0.1103515625,
2067
+ "kd_loss": 0.19740513321539765,
2068
+ "learning_rate": 3e-06,
2069
+ "loss": 0.1841,
2070
+ "masked_tokens": 110.5625,
2071
+ "mean_t": 0.4791536889737472,
2072
+ "step": 860,
2073
+ "student_masked_tokens": 110.5625
2074
+ },
2075
+ {
2076
+ "avg_mask_ratio": 0.4846805231412873,
2077
+ "avg_response_length": 220.7375,
2078
+ "avg_student_mask_ratio": 0.4846805231412873,
2079
+ "batch_ainp_frac": 0.0,
2080
+ "batch_inp_frac": 0.0,
2081
+ "batch_inp_oh_frac": 0.0,
2082
+ "batch_inp_par_frac": 1.0,
2083
+ "batch_inp_par_reverse_frac": 0.0,
2084
+ "batch_rl_frac": 0.0,
2085
+ "batch_sft_frac": 0.0,
2086
+ "batch_soft_sft_frac": 0.0,
2087
+ "batch_tf_frac": 0.0,
2088
+ "ce_loss": 0.0,
2089
+ "epoch": 1.8576000000000001,
2090
+ "grad_norm": 0.228515625,
2091
+ "kd_loss": 0.19436422403705364,
2092
+ "learning_rate": 3e-06,
2093
+ "loss": 0.2012,
2094
+ "masked_tokens": 113.7625,
2095
+ "mean_t": 0.5203817339061061,
2096
+ "step": 870,
2097
+ "student_masked_tokens": 113.7625
2098
+ },
2099
+ {
2100
+ "avg_mask_ratio": 0.4508363194297999,
2101
+ "avg_response_length": 203.2875,
2102
+ "avg_student_mask_ratio": 0.4508363194297999,
2103
+ "batch_ainp_frac": 0.0,
2104
+ "batch_inp_frac": 0.0,
2105
+ "batch_inp_oh_frac": 0.0,
2106
+ "batch_inp_par_frac": 1.0,
2107
+ "batch_inp_par_reverse_frac": 0.0,
2108
+ "batch_rl_frac": 0.0,
2109
+ "batch_sft_frac": 0.0,
2110
+ "batch_soft_sft_frac": 0.0,
2111
+ "batch_tf_frac": 0.0,
2112
+ "ce_loss": 0.0,
2113
+ "epoch": 1.8789333333333333,
2114
+ "grad_norm": 0.1962890625,
2115
+ "kd_loss": 0.16288868402702406,
2116
+ "learning_rate": 3e-06,
2117
+ "loss": 0.1845,
2118
+ "masked_tokens": 95.3875,
2119
+ "mean_t": 0.4875184997683391,
2120
+ "step": 880,
2121
+ "student_masked_tokens": 95.3875
2122
+ },
2123
+ {
2124
+ "avg_mask_ratio": 0.43862658384023234,
2125
+ "avg_response_length": 227.9375,
2126
+ "avg_student_mask_ratio": 0.43862658384023234,
2127
+ "batch_ainp_frac": 0.0,
2128
+ "batch_inp_frac": 0.0,
2129
+ "batch_inp_oh_frac": 0.0,
2130
+ "batch_inp_par_frac": 1.0,
2131
+ "batch_inp_par_reverse_frac": 0.0,
2132
+ "batch_rl_frac": 0.0,
2133
+ "batch_sft_frac": 0.0,
2134
+ "batch_soft_sft_frac": 0.0,
2135
+ "batch_tf_frac": 0.0,
2136
+ "ce_loss": 0.0,
2137
+ "epoch": 1.9002666666666665,
2138
+ "grad_norm": 0.08251953125,
2139
+ "kd_loss": 0.11281866748422545,
2140
+ "learning_rate": 3e-06,
2141
+ "loss": 0.142,
2142
+ "masked_tokens": 101.2625,
2143
+ "mean_t": 0.4766692223958671,
2144
+ "step": 890,
2145
+ "student_masked_tokens": 101.2625
2146
+ },
2147
+ {
2148
+ "avg_mask_ratio": 0.44909207145101393,
2149
+ "avg_response_length": 237.2375,
2150
+ "avg_student_mask_ratio": 0.44909207145101393,
2151
+ "batch_ainp_frac": 0.0,
2152
+ "batch_inp_frac": 0.0,
2153
+ "batch_inp_oh_frac": 0.0,
2154
+ "batch_inp_par_frac": 1.0,
2155
+ "batch_inp_par_reverse_frac": 0.0,
2156
+ "batch_rl_frac": 0.0,
2157
+ "batch_sft_frac": 0.0,
2158
+ "batch_soft_sft_frac": 0.0,
2159
+ "batch_tf_frac": 0.0,
2160
+ "ce_loss": 0.0,
2161
+ "epoch": 1.9216,
2162
+ "grad_norm": 0.1591796875,
2163
+ "kd_loss": 0.15684176656744747,
2164
+ "learning_rate": 3e-06,
2165
+ "loss": 0.1737,
2166
+ "masked_tokens": 103.7,
2167
+ "mean_t": 0.487134758150205,
2168
+ "step": 900,
2169
+ "student_masked_tokens": 103.7
2170
+ },
2171
+ {
2172
+ "avg_mask_ratio": 0.47512493645772336,
2173
+ "avg_response_length": 246.175,
2174
+ "avg_student_mask_ratio": 0.47512493645772336,
2175
+ "batch_ainp_frac": 0.0,
2176
+ "batch_inp_frac": 0.0,
2177
+ "batch_inp_oh_frac": 0.0,
2178
+ "batch_inp_par_frac": 1.0,
2179
+ "batch_inp_par_reverse_frac": 0.0,
2180
+ "batch_rl_frac": 0.0,
2181
+ "batch_sft_frac": 0.0,
2182
+ "batch_soft_sft_frac": 0.0,
2183
+ "batch_tf_frac": 0.0,
2184
+ "ce_loss": 0.0,
2185
+ "epoch": 1.9429333333333334,
2186
+ "grad_norm": 0.1982421875,
2187
+ "kd_loss": 0.16909069961529893,
2188
+ "learning_rate": 3e-06,
2189
+ "loss": 0.1717,
2190
+ "masked_tokens": 117.2125,
2191
+ "mean_t": 0.5027793228859082,
2192
+ "step": 910,
2193
+ "student_masked_tokens": 117.2125
2194
+ },
2195
+ {
2196
+ "avg_mask_ratio": 0.46720519906957636,
2197
+ "avg_response_length": 229.0875,
2198
+ "avg_student_mask_ratio": 0.46720519906957636,
2199
+ "batch_ainp_frac": 0.0,
2200
+ "batch_inp_frac": 0.0,
2201
+ "batch_inp_oh_frac": 0.0,
2202
+ "batch_inp_par_frac": 1.0,
2203
+ "batch_inp_par_reverse_frac": 0.0,
2204
+ "batch_rl_frac": 0.0,
2205
+ "batch_sft_frac": 0.0,
2206
+ "batch_soft_sft_frac": 0.0,
2207
+ "batch_tf_frac": 0.0,
2208
+ "ce_loss": 0.0,
2209
+ "epoch": 1.9642666666666666,
2210
+ "grad_norm": 0.203125,
2211
+ "kd_loss": 0.17078794931742322,
2212
+ "learning_rate": 3e-06,
2213
+ "loss": 0.1844,
2214
+ "masked_tokens": 115.325,
2215
+ "mean_t": 0.49417946098838,
2216
+ "step": 920,
2217
+ "student_masked_tokens": 115.325
2218
+ },
2219
+ {
2220
+ "avg_mask_ratio": 0.4805813999380916,
2221
+ "avg_response_length": 236.1625,
2222
+ "avg_student_mask_ratio": 0.4805813999380916,
2223
+ "batch_ainp_frac": 0.0,
2224
+ "batch_inp_frac": 0.0,
2225
+ "batch_inp_oh_frac": 0.0,
2226
+ "batch_inp_par_frac": 1.0,
2227
+ "batch_inp_par_reverse_frac": 0.0,
2228
+ "batch_rl_frac": 0.0,
2229
+ "batch_sft_frac": 0.0,
2230
+ "batch_soft_sft_frac": 0.0,
2231
+ "batch_tf_frac": 0.0,
2232
+ "ce_loss": 0.0,
2233
+ "epoch": 1.9856,
2234
+ "grad_norm": 0.21484375,
2235
+ "kd_loss": 0.20637793739751942,
2236
+ "learning_rate": 3e-06,
2237
+ "loss": 0.176,
2238
+ "masked_tokens": 119.65,
2239
+ "mean_t": 0.5045580042526125,
2240
+ "step": 930,
2241
+ "student_masked_tokens": 119.65
2242
+ },
2243
+ {
2244
+ "avg_mask_ratio": 0.49833715503059683,
2245
+ "avg_response_length": 229.39285714285714,
2246
+ "avg_student_mask_ratio": 0.49833715503059683,
2247
+ "batch_ainp_frac": 0.0,
2248
+ "batch_inp_frac": 0.0,
2249
+ "batch_inp_oh_frac": 0.0,
2250
+ "batch_inp_par_frac": 1.0,
2251
+ "batch_inp_par_reverse_frac": 0.0,
2252
+ "batch_rl_frac": 0.0,
2253
+ "batch_sft_frac": 0.0,
2254
+ "batch_soft_sft_frac": 0.0,
2255
+ "batch_tf_frac": 0.0,
2256
+ "ce_loss": 0.0,
2257
+ "epoch": 2.0085333333333333,
2258
+ "grad_norm": 0.259765625,
2259
+ "kd_loss": 0.2117233988543549,
2260
+ "learning_rate": 3e-06,
2261
+ "loss": 0.2023,
2262
+ "masked_tokens": 121.5952380952381,
2263
+ "mean_t": 0.5321138524893849,
2264
+ "step": 940,
2265
+ "student_masked_tokens": 121.5952380952381
2266
+ },
2267
+ {
2268
+ "avg_mask_ratio": 0.4347397161007393,
2269
+ "avg_response_length": 226.925,
2270
+ "avg_student_mask_ratio": 0.4347397161007393,
2271
+ "batch_ainp_frac": 0.0,
2272
+ "batch_inp_frac": 0.0,
2273
+ "batch_inp_oh_frac": 0.0,
2274
+ "batch_inp_par_frac": 1.0,
2275
+ "batch_inp_par_reverse_frac": 0.0,
2276
+ "batch_rl_frac": 0.0,
2277
+ "batch_sft_frac": 0.0,
2278
+ "batch_soft_sft_frac": 0.0,
2279
+ "batch_tf_frac": 0.0,
2280
+ "ce_loss": 0.0,
2281
+ "epoch": 2.0298666666666665,
2282
+ "grad_norm": 0.15234375,
2283
+ "kd_loss": 0.16830119832033005,
2284
+ "learning_rate": 3e-06,
2285
+ "loss": 0.1549,
2286
+ "masked_tokens": 103.4125,
2287
+ "mean_t": 0.4632946296595037,
2288
+ "step": 950,
2289
+ "student_masked_tokens": 103.4125
2290
+ },
2291
+ {
2292
+ "avg_mask_ratio": 0.4989688004134223,
2293
+ "avg_response_length": 273.3125,
2294
+ "avg_student_mask_ratio": 0.4989688004134223,
2295
+ "batch_ainp_frac": 0.0,
2296
+ "batch_inp_frac": 0.0,
2297
+ "batch_inp_oh_frac": 0.0,
2298
+ "batch_inp_par_frac": 1.0,
2299
+ "batch_inp_par_reverse_frac": 0.0,
2300
+ "batch_rl_frac": 0.0,
2301
+ "batch_sft_frac": 0.0,
2302
+ "batch_soft_sft_frac": 0.0,
2303
+ "batch_tf_frac": 0.0,
2304
+ "ce_loss": 0.0,
2305
+ "epoch": 2.0512,
2306
+ "grad_norm": 0.123046875,
2307
+ "kd_loss": 0.2304358719712809,
2308
+ "learning_rate": 3e-06,
2309
+ "loss": 0.22,
2310
+ "masked_tokens": 137.125,
2311
+ "mean_t": 0.5198000721400604,
2312
+ "step": 960,
2313
+ "student_masked_tokens": 137.125
2314
+ },
2315
+ {
2316
+ "avg_mask_ratio": 0.4374056361732073,
2317
+ "avg_response_length": 239.125,
2318
+ "avg_student_mask_ratio": 0.4374056361732073,
2319
+ "batch_ainp_frac": 0.0,
2320
+ "batch_inp_frac": 0.0,
2321
+ "batch_inp_oh_frac": 0.0,
2322
+ "batch_inp_par_frac": 1.0,
2323
+ "batch_inp_par_reverse_frac": 0.0,
2324
+ "batch_rl_frac": 0.0,
2325
+ "batch_sft_frac": 0.0,
2326
+ "batch_soft_sft_frac": 0.0,
2327
+ "batch_tf_frac": 0.0,
2328
+ "ce_loss": 0.0,
2329
+ "epoch": 2.0725333333333333,
2330
+ "grad_norm": 0.1962890625,
2331
+ "kd_loss": 0.158095879086882,
2332
+ "learning_rate": 3e-06,
2333
+ "loss": 0.1573,
2334
+ "masked_tokens": 107.4625,
2335
+ "mean_t": 0.4703940597362816,
2336
+ "step": 970,
2337
+ "student_masked_tokens": 107.4625
2338
+ },
2339
+ {
2340
+ "avg_mask_ratio": 0.4216216350672767,
2341
+ "avg_response_length": 236.0125,
2342
+ "avg_student_mask_ratio": 0.4216216350672767,
2343
+ "batch_ainp_frac": 0.0,
2344
+ "batch_inp_frac": 0.0,
2345
+ "batch_inp_oh_frac": 0.0,
2346
+ "batch_inp_par_frac": 1.0,
2347
+ "batch_inp_par_reverse_frac": 0.0,
2348
+ "batch_rl_frac": 0.0,
2349
+ "batch_sft_frac": 0.0,
2350
+ "batch_soft_sft_frac": 0.0,
2351
+ "batch_tf_frac": 0.0,
2352
+ "ce_loss": 0.0,
2353
+ "epoch": 2.0938666666666665,
2354
+ "grad_norm": 0.177734375,
2355
+ "kd_loss": 0.15213593625503563,
2356
+ "learning_rate": 3e-06,
2357
+ "loss": 0.1305,
2358
+ "masked_tokens": 103.05,
2359
+ "mean_t": 0.4511947895749472,
2360
+ "step": 980,
2361
+ "student_masked_tokens": 103.05
2362
+ },
2363
+ {
2364
+ "avg_mask_ratio": 0.46076959141064433,
2365
+ "avg_response_length": 257.375,
2366
+ "avg_student_mask_ratio": 0.46076959141064433,
2367
+ "batch_ainp_frac": 0.0,
2368
+ "batch_inp_frac": 0.0,
2369
+ "batch_inp_oh_frac": 0.0,
2370
+ "batch_inp_par_frac": 1.0,
2371
+ "batch_inp_par_reverse_frac": 0.0,
2372
+ "batch_rl_frac": 0.0,
2373
+ "batch_sft_frac": 0.0,
2374
+ "batch_soft_sft_frac": 0.0,
2375
+ "batch_tf_frac": 0.0,
2376
+ "ce_loss": 0.0,
2377
+ "epoch": 2.1152,
2378
+ "grad_norm": 0.1484375,
2379
+ "kd_loss": 0.18778659292394478,
2380
+ "learning_rate": 3e-06,
2381
+ "loss": 0.1727,
2382
+ "masked_tokens": 121.2125,
2383
+ "mean_t": 0.4923786667350214,
2384
+ "step": 990,
2385
+ "student_masked_tokens": 121.2125
2386
+ },
2387
+ {
2388
+ "avg_mask_ratio": 0.45376153094694016,
2389
+ "avg_response_length": 223.4125,
2390
+ "avg_student_mask_ratio": 0.45376153094694016,
2391
+ "batch_ainp_frac": 0.0,
2392
+ "batch_inp_frac": 0.0,
2393
+ "batch_inp_oh_frac": 0.0,
2394
+ "batch_inp_par_frac": 1.0,
2395
+ "batch_inp_par_reverse_frac": 0.0,
2396
+ "batch_rl_frac": 0.0,
2397
+ "batch_sft_frac": 0.0,
2398
+ "batch_soft_sft_frac": 0.0,
2399
+ "batch_tf_frac": 0.0,
2400
+ "ce_loss": 0.0,
2401
+ "epoch": 2.1365333333333334,
2402
+ "grad_norm": 0.10498046875,
2403
+ "kd_loss": 0.16195435594947866,
2404
+ "learning_rate": 3e-06,
2405
+ "loss": 0.1554,
2406
+ "masked_tokens": 102.175,
2407
+ "mean_t": 0.4773523230338469,
2408
+ "step": 1000,
2409
+ "student_masked_tokens": 102.175
2410
+ }
2411
+ ],
2412
+ "logging_steps": 10,
2413
+ "max_steps": 1404,
2414
+ "num_input_tokens_seen": 0,
2415
+ "num_train_epochs": 3,
2416
+ "save_steps": 100,
2417
+ "stateful_callbacks": {
2418
+ "TrainerControl": {
2419
+ "args": {
2420
+ "should_epoch_stop": false,
2421
+ "should_evaluate": false,
2422
+ "should_log": false,
2423
+ "should_save": true,
2424
+ "should_training_stop": false
2425
+ },
2426
+ "attributes": {}
2427
+ }
2428
+ },
2429
+ "total_flos": 0.0,
2430
+ "train_batch_size": 1,
2431
+ "trial_name": null,
2432
+ "trial_params": null
2433
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c347df37da6e348160afc3fbb65d12595e5064bd8e5bfd591004e5e86a703f42
3
+ size 7992
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: GSAI-ML/LLaDA-8B-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 0.15.1
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/adapter_config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "GSAI-ML/LLaDA-8B-Instruct",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 64,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "r": 128,
24
+ "rank_pattern": {},
25
+ "revision": null,
26
+ "target_modules": [
27
+ "down_proj",
28
+ "up_proj",
29
+ "gate_proj",
30
+ "v_proj",
31
+ "k_proj",
32
+ "o_proj",
33
+ "q_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca8cf9c4c96d199e2c6fc1014d956d44848bb3e22b70c3c1f8147514eca399ae
3
+ size 2406624648
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c3ec26485c65c40a39c48574ff60fee7ab396303873ba5b68ff02a9708afca38
3
+ size 671304442
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:907e39dff0cf7ad1a1affaa1e7047653794ab16e25c6977ce7b5524769fdf799
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f7938e18b41a6ad67840013f5aac6c4576f92425b90ea79226bacc627193b9e
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25795e3b7374d0f6abdd7ab4b34fbf7ab0447ba73c04014500c2ab8b5acec5b4
3
+ size 1064
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/trainer_state.json ADDED
@@ -0,0 +1,2673 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.3498666666666668,
5
+ "eval_steps": 500,
6
+ "global_step": 1100,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "avg_mask_ratio": 0.4933756332669873,
13
+ "avg_response_length": 252.0625,
14
+ "avg_student_mask_ratio": 0.4933756332669873,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 0.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 1.0,
19
+ "batch_inp_par_reverse_frac": 0.0,
20
+ "batch_rl_frac": 0.0,
21
+ "batch_sft_frac": 0.0,
22
+ "batch_soft_sft_frac": 0.0,
23
+ "batch_tf_frac": 0.0,
24
+ "ce_loss": 0.0,
25
+ "epoch": 0.021333333333333333,
26
+ "grad_norm": 0.1650390625,
27
+ "kd_loss": 0.24795629351958723,
28
+ "learning_rate": 3e-06,
29
+ "loss": 0.2758,
30
+ "masked_tokens": 120.975,
31
+ "mean_t": 0.5145528071501758,
32
+ "step": 10,
33
+ "student_masked_tokens": 120.975
34
+ },
35
+ {
36
+ "avg_mask_ratio": 0.41923869140446185,
37
+ "avg_response_length": 221.7125,
38
+ "avg_student_mask_ratio": 0.41923869140446185,
39
+ "batch_ainp_frac": 0.0,
40
+ "batch_inp_frac": 0.0,
41
+ "batch_inp_oh_frac": 0.0,
42
+ "batch_inp_par_frac": 1.0,
43
+ "batch_inp_par_reverse_frac": 0.0,
44
+ "batch_rl_frac": 0.0,
45
+ "batch_sft_frac": 0.0,
46
+ "batch_soft_sft_frac": 0.0,
47
+ "batch_tf_frac": 0.0,
48
+ "ce_loss": 0.0,
49
+ "epoch": 0.042666666666666665,
50
+ "grad_norm": 0.388671875,
51
+ "kd_loss": 0.21509853232191353,
52
+ "learning_rate": 3e-06,
53
+ "loss": 0.2046,
54
+ "masked_tokens": 86.1875,
55
+ "mean_t": 0.43874448732240123,
56
+ "step": 20,
57
+ "student_masked_tokens": 86.1875
58
+ },
59
+ {
60
+ "avg_mask_ratio": 0.4542569225654006,
61
+ "avg_response_length": 231.45,
62
+ "avg_student_mask_ratio": 0.4542569225654006,
63
+ "batch_ainp_frac": 0.0,
64
+ "batch_inp_frac": 0.0,
65
+ "batch_inp_oh_frac": 0.0,
66
+ "batch_inp_par_frac": 1.0,
67
+ "batch_inp_par_reverse_frac": 0.0,
68
+ "batch_rl_frac": 0.0,
69
+ "batch_sft_frac": 0.0,
70
+ "batch_soft_sft_frac": 0.0,
71
+ "batch_tf_frac": 0.0,
72
+ "ce_loss": 0.0,
73
+ "epoch": 0.064,
74
+ "grad_norm": 0.52734375,
75
+ "kd_loss": 0.19756361616970963,
76
+ "learning_rate": 3e-06,
77
+ "loss": 0.1976,
78
+ "masked_tokens": 116.2875,
79
+ "mean_t": 0.4803953981841914,
80
+ "step": 30,
81
+ "student_masked_tokens": 116.2875
82
+ },
83
+ {
84
+ "avg_mask_ratio": 0.41855402445653456,
85
+ "avg_response_length": 214.125,
86
+ "avg_student_mask_ratio": 0.41855402445653456,
87
+ "batch_ainp_frac": 0.0,
88
+ "batch_inp_frac": 0.0,
89
+ "batch_inp_oh_frac": 0.0,
90
+ "batch_inp_par_frac": 1.0,
91
+ "batch_inp_par_reverse_frac": 0.0,
92
+ "batch_rl_frac": 0.0,
93
+ "batch_sft_frac": 0.0,
94
+ "batch_soft_sft_frac": 0.0,
95
+ "batch_tf_frac": 0.0,
96
+ "ce_loss": 0.0,
97
+ "epoch": 0.08533333333333333,
98
+ "grad_norm": 0.263671875,
99
+ "kd_loss": 0.1576978968325534,
100
+ "learning_rate": 3e-06,
101
+ "loss": 0.1551,
102
+ "masked_tokens": 94.225,
103
+ "mean_t": 0.4569831106782658,
104
+ "step": 40,
105
+ "student_masked_tokens": 94.225
106
+ },
107
+ {
108
+ "avg_mask_ratio": 0.4331560767372139,
109
+ "avg_response_length": 222.225,
110
+ "avg_student_mask_ratio": 0.4331560767372139,
111
+ "batch_ainp_frac": 0.0,
112
+ "batch_inp_frac": 0.0,
113
+ "batch_inp_oh_frac": 0.0,
114
+ "batch_inp_par_frac": 1.0,
115
+ "batch_inp_par_reverse_frac": 0.0,
116
+ "batch_rl_frac": 0.0,
117
+ "batch_sft_frac": 0.0,
118
+ "batch_soft_sft_frac": 0.0,
119
+ "batch_tf_frac": 0.0,
120
+ "ce_loss": 0.0,
121
+ "epoch": 0.10666666666666667,
122
+ "grad_norm": 0.125,
123
+ "kd_loss": 0.17712681048956255,
124
+ "learning_rate": 3e-06,
125
+ "loss": 0.1648,
126
+ "masked_tokens": 97.825,
127
+ "mean_t": 0.4612453707959503,
128
+ "step": 50,
129
+ "student_masked_tokens": 97.825
130
+ },
131
+ {
132
+ "avg_mask_ratio": 0.4547638618387282,
133
+ "avg_response_length": 242.9,
134
+ "avg_student_mask_ratio": 0.4547638618387282,
135
+ "batch_ainp_frac": 0.0,
136
+ "batch_inp_frac": 0.0,
137
+ "batch_inp_oh_frac": 0.0,
138
+ "batch_inp_par_frac": 1.0,
139
+ "batch_inp_par_reverse_frac": 0.0,
140
+ "batch_rl_frac": 0.0,
141
+ "batch_sft_frac": 0.0,
142
+ "batch_soft_sft_frac": 0.0,
143
+ "batch_tf_frac": 0.0,
144
+ "ce_loss": 0.0,
145
+ "epoch": 0.128,
146
+ "grad_norm": 0.130859375,
147
+ "kd_loss": 0.1745696667137963,
148
+ "learning_rate": 3e-06,
149
+ "loss": 0.1745,
150
+ "masked_tokens": 119.125,
151
+ "mean_t": 0.48597636765334756,
152
+ "step": 60,
153
+ "student_masked_tokens": 119.125
154
+ },
155
+ {
156
+ "avg_mask_ratio": 0.5151988173020072,
157
+ "avg_response_length": 214.3375,
158
+ "avg_student_mask_ratio": 0.5151988173020072,
159
+ "batch_ainp_frac": 0.0,
160
+ "batch_inp_frac": 0.0,
161
+ "batch_inp_oh_frac": 0.0,
162
+ "batch_inp_par_frac": 1.0,
163
+ "batch_inp_par_reverse_frac": 0.0,
164
+ "batch_rl_frac": 0.0,
165
+ "batch_sft_frac": 0.0,
166
+ "batch_soft_sft_frac": 0.0,
167
+ "batch_tf_frac": 0.0,
168
+ "ce_loss": 0.0,
169
+ "epoch": 0.14933333333333335,
170
+ "grad_norm": 0.17578125,
171
+ "kd_loss": 0.1657758254527316,
172
+ "learning_rate": 3e-06,
173
+ "loss": 0.1972,
174
+ "masked_tokens": 111.7875,
175
+ "mean_t": 0.5459650319069624,
176
+ "step": 70,
177
+ "student_masked_tokens": 111.7875
178
+ },
179
+ {
180
+ "avg_mask_ratio": 0.37937068473547697,
181
+ "avg_response_length": 250.2,
182
+ "avg_student_mask_ratio": 0.37937068473547697,
183
+ "batch_ainp_frac": 0.0,
184
+ "batch_inp_frac": 0.0,
185
+ "batch_inp_oh_frac": 0.0,
186
+ "batch_inp_par_frac": 1.0,
187
+ "batch_inp_par_reverse_frac": 0.0,
188
+ "batch_rl_frac": 0.0,
189
+ "batch_sft_frac": 0.0,
190
+ "batch_soft_sft_frac": 0.0,
191
+ "batch_tf_frac": 0.0,
192
+ "ce_loss": 0.0,
193
+ "epoch": 0.17066666666666666,
194
+ "grad_norm": 0.051513671875,
195
+ "kd_loss": 0.13968983994418097,
196
+ "learning_rate": 3e-06,
197
+ "loss": 0.1369,
198
+ "masked_tokens": 95.1875,
199
+ "mean_t": 0.40758824030635876,
200
+ "step": 80,
201
+ "student_masked_tokens": 95.1875
202
+ },
203
+ {
204
+ "avg_mask_ratio": 0.5006106478627771,
205
+ "avg_response_length": 242.1125,
206
+ "avg_student_mask_ratio": 0.5006106478627771,
207
+ "batch_ainp_frac": 0.0,
208
+ "batch_inp_frac": 0.0,
209
+ "batch_inp_oh_frac": 0.0,
210
+ "batch_inp_par_frac": 1.0,
211
+ "batch_inp_par_reverse_frac": 0.0,
212
+ "batch_rl_frac": 0.0,
213
+ "batch_sft_frac": 0.0,
214
+ "batch_soft_sft_frac": 0.0,
215
+ "batch_tf_frac": 0.0,
216
+ "ce_loss": 0.0,
217
+ "epoch": 0.192,
218
+ "grad_norm": 0.22265625,
219
+ "kd_loss": 0.20869405062871707,
220
+ "learning_rate": 3e-06,
221
+ "loss": 0.2007,
222
+ "masked_tokens": 116.3875,
223
+ "mean_t": 0.5297661645396147,
224
+ "step": 90,
225
+ "student_masked_tokens": 116.3875
226
+ },
227
+ {
228
+ "avg_mask_ratio": 0.4596128500183113,
229
+ "avg_response_length": 229.0625,
230
+ "avg_student_mask_ratio": 0.4596128500183113,
231
+ "batch_ainp_frac": 0.0,
232
+ "batch_inp_frac": 0.0,
233
+ "batch_inp_oh_frac": 0.0,
234
+ "batch_inp_par_frac": 1.0,
235
+ "batch_inp_par_reverse_frac": 0.0,
236
+ "batch_rl_frac": 0.0,
237
+ "batch_sft_frac": 0.0,
238
+ "batch_soft_sft_frac": 0.0,
239
+ "batch_tf_frac": 0.0,
240
+ "ce_loss": 0.0,
241
+ "epoch": 0.21333333333333335,
242
+ "grad_norm": 0.1865234375,
243
+ "kd_loss": 0.17640120884607313,
244
+ "learning_rate": 3e-06,
245
+ "loss": 0.1864,
246
+ "masked_tokens": 109.7125,
247
+ "mean_t": 0.4886587227345444,
248
+ "step": 100,
249
+ "student_masked_tokens": 109.7125
250
+ },
251
+ {
252
+ "avg_mask_ratio": 0.4920400592498481,
253
+ "avg_response_length": 229.9875,
254
+ "avg_student_mask_ratio": 0.4920400592498481,
255
+ "batch_ainp_frac": 0.0,
256
+ "batch_inp_frac": 0.0,
257
+ "batch_inp_oh_frac": 0.0,
258
+ "batch_inp_par_frac": 1.0,
259
+ "batch_inp_par_reverse_frac": 0.0,
260
+ "batch_rl_frac": 0.0,
261
+ "batch_sft_frac": 0.0,
262
+ "batch_soft_sft_frac": 0.0,
263
+ "batch_tf_frac": 0.0,
264
+ "ce_loss": 0.0,
265
+ "epoch": 0.23466666666666666,
266
+ "grad_norm": 0.326171875,
267
+ "kd_loss": 0.2011610215539008,
268
+ "learning_rate": 3e-06,
269
+ "loss": 0.2334,
270
+ "masked_tokens": 109.4,
271
+ "mean_t": 0.5246987929102034,
272
+ "step": 110,
273
+ "student_masked_tokens": 109.4
274
+ },
275
+ {
276
+ "avg_mask_ratio": 0.450224511185661,
277
+ "avg_response_length": 229.225,
278
+ "avg_student_mask_ratio": 0.450224511185661,
279
+ "batch_ainp_frac": 0.0,
280
+ "batch_inp_frac": 0.0,
281
+ "batch_inp_oh_frac": 0.0,
282
+ "batch_inp_par_frac": 1.0,
283
+ "batch_inp_par_reverse_frac": 0.0,
284
+ "batch_rl_frac": 0.0,
285
+ "batch_sft_frac": 0.0,
286
+ "batch_soft_sft_frac": 0.0,
287
+ "batch_tf_frac": 0.0,
288
+ "ce_loss": 0.0,
289
+ "epoch": 0.256,
290
+ "grad_norm": 0.171875,
291
+ "kd_loss": 0.21365654302464918,
292
+ "learning_rate": 3e-06,
293
+ "loss": 0.2021,
294
+ "masked_tokens": 102.6375,
295
+ "mean_t": 0.4805434140143916,
296
+ "step": 120,
297
+ "student_masked_tokens": 102.6375
298
+ },
299
+ {
300
+ "avg_mask_ratio": 0.4397759020910598,
301
+ "avg_response_length": 214.15,
302
+ "avg_student_mask_ratio": 0.4397759020910598,
303
+ "batch_ainp_frac": 0.0,
304
+ "batch_inp_frac": 0.0,
305
+ "batch_inp_oh_frac": 0.0,
306
+ "batch_inp_par_frac": 1.0,
307
+ "batch_inp_par_reverse_frac": 0.0,
308
+ "batch_rl_frac": 0.0,
309
+ "batch_sft_frac": 0.0,
310
+ "batch_soft_sft_frac": 0.0,
311
+ "batch_tf_frac": 0.0,
312
+ "ce_loss": 0.0,
313
+ "epoch": 0.2773333333333333,
314
+ "grad_norm": 0.173828125,
315
+ "kd_loss": 0.13860440934267615,
316
+ "learning_rate": 3e-06,
317
+ "loss": 0.1362,
318
+ "masked_tokens": 98.4,
319
+ "mean_t": 0.47522516988683494,
320
+ "step": 130,
321
+ "student_masked_tokens": 98.4
322
+ },
323
+ {
324
+ "avg_mask_ratio": 0.4890626976499334,
325
+ "avg_response_length": 242.1625,
326
+ "avg_student_mask_ratio": 0.4890626976499334,
327
+ "batch_ainp_frac": 0.0,
328
+ "batch_inp_frac": 0.0,
329
+ "batch_inp_oh_frac": 0.0,
330
+ "batch_inp_par_frac": 1.0,
331
+ "batch_inp_par_reverse_frac": 0.0,
332
+ "batch_rl_frac": 0.0,
333
+ "batch_sft_frac": 0.0,
334
+ "batch_soft_sft_frac": 0.0,
335
+ "batch_tf_frac": 0.0,
336
+ "ce_loss": 0.0,
337
+ "epoch": 0.2986666666666667,
338
+ "grad_norm": 0.27734375,
339
+ "kd_loss": 0.2106460814328841,
340
+ "learning_rate": 3e-06,
341
+ "loss": 0.2054,
342
+ "masked_tokens": 129.725,
343
+ "mean_t": 0.5204090005659964,
344
+ "step": 140,
345
+ "student_masked_tokens": 129.725
346
+ },
347
+ {
348
+ "avg_mask_ratio": 0.47298577734036373,
349
+ "avg_response_length": 262.9875,
350
+ "avg_student_mask_ratio": 0.47298577734036373,
351
+ "batch_ainp_frac": 0.0,
352
+ "batch_inp_frac": 0.0,
353
+ "batch_inp_oh_frac": 0.0,
354
+ "batch_inp_par_frac": 1.0,
355
+ "batch_inp_par_reverse_frac": 0.0,
356
+ "batch_rl_frac": 0.0,
357
+ "batch_sft_frac": 0.0,
358
+ "batch_soft_sft_frac": 0.0,
359
+ "batch_tf_frac": 0.0,
360
+ "ce_loss": 0.0,
361
+ "epoch": 0.32,
362
+ "grad_norm": 0.0673828125,
363
+ "kd_loss": 0.2408599746217078,
364
+ "learning_rate": 3e-06,
365
+ "loss": 0.2276,
366
+ "masked_tokens": 128.375,
367
+ "mean_t": 0.5012552456930279,
368
+ "step": 150,
369
+ "student_masked_tokens": 128.375
370
+ },
371
+ {
372
+ "avg_mask_ratio": 0.5043223856599071,
373
+ "avg_response_length": 217.5,
374
+ "avg_student_mask_ratio": 0.5043223856599071,
375
+ "batch_ainp_frac": 0.0,
376
+ "batch_inp_frac": 0.0,
377
+ "batch_inp_oh_frac": 0.0,
378
+ "batch_inp_par_frac": 1.0,
379
+ "batch_inp_par_reverse_frac": 0.0,
380
+ "batch_rl_frac": 0.0,
381
+ "batch_sft_frac": 0.0,
382
+ "batch_soft_sft_frac": 0.0,
383
+ "batch_tf_frac": 0.0,
384
+ "ce_loss": 0.0,
385
+ "epoch": 0.3413333333333333,
386
+ "grad_norm": 0.2236328125,
387
+ "kd_loss": 0.20484722793833043,
388
+ "learning_rate": 3e-06,
389
+ "loss": 0.2154,
390
+ "masked_tokens": 106.025,
391
+ "mean_t": 0.5408745193795766,
392
+ "step": 160,
393
+ "student_masked_tokens": 106.025
394
+ },
395
+ {
396
+ "avg_mask_ratio": 0.48419030708028005,
397
+ "avg_response_length": 196.2625,
398
+ "avg_student_mask_ratio": 0.48419030708028005,
399
+ "batch_ainp_frac": 0.0,
400
+ "batch_inp_frac": 0.0,
401
+ "batch_inp_oh_frac": 0.0,
402
+ "batch_inp_par_frac": 1.0,
403
+ "batch_inp_par_reverse_frac": 0.0,
404
+ "batch_rl_frac": 0.0,
405
+ "batch_sft_frac": 0.0,
406
+ "batch_soft_sft_frac": 0.0,
407
+ "batch_tf_frac": 0.0,
408
+ "ce_loss": 0.0,
409
+ "epoch": 0.3626666666666667,
410
+ "grad_norm": 0.1611328125,
411
+ "kd_loss": 0.20407032655223248,
412
+ "learning_rate": 3e-06,
413
+ "loss": 0.2007,
414
+ "masked_tokens": 96.2625,
415
+ "mean_t": 0.5134547733236104,
416
+ "step": 170,
417
+ "student_masked_tokens": 96.2625
418
+ },
419
+ {
420
+ "avg_mask_ratio": 0.4485494759515859,
421
+ "avg_response_length": 228.8625,
422
+ "avg_student_mask_ratio": 0.4485494759515859,
423
+ "batch_ainp_frac": 0.0,
424
+ "batch_inp_frac": 0.0,
425
+ "batch_inp_oh_frac": 0.0,
426
+ "batch_inp_par_frac": 1.0,
427
+ "batch_inp_par_reverse_frac": 0.0,
428
+ "batch_rl_frac": 0.0,
429
+ "batch_sft_frac": 0.0,
430
+ "batch_soft_sft_frac": 0.0,
431
+ "batch_tf_frac": 0.0,
432
+ "ce_loss": 0.0,
433
+ "epoch": 0.384,
434
+ "grad_norm": 0.14453125,
435
+ "kd_loss": 0.16957379382825621,
436
+ "learning_rate": 3e-06,
437
+ "loss": 0.1796,
438
+ "masked_tokens": 101.275,
439
+ "mean_t": 0.48226988823735156,
440
+ "step": 180,
441
+ "student_masked_tokens": 101.275
442
+ },
443
+ {
444
+ "avg_mask_ratio": 0.48665878190658984,
445
+ "avg_response_length": 230.5,
446
+ "avg_student_mask_ratio": 0.48665878190658984,
447
+ "batch_ainp_frac": 0.0,
448
+ "batch_inp_frac": 0.0,
449
+ "batch_inp_oh_frac": 0.0,
450
+ "batch_inp_par_frac": 1.0,
451
+ "batch_inp_par_reverse_frac": 0.0,
452
+ "batch_rl_frac": 0.0,
453
+ "batch_sft_frac": 0.0,
454
+ "batch_soft_sft_frac": 0.0,
455
+ "batch_tf_frac": 0.0,
456
+ "ce_loss": 0.0,
457
+ "epoch": 0.4053333333333333,
458
+ "grad_norm": 0.3515625,
459
+ "kd_loss": 0.22313492714965832,
460
+ "learning_rate": 3e-06,
461
+ "loss": 0.2112,
462
+ "masked_tokens": 107.975,
463
+ "mean_t": 0.5165087037021294,
464
+ "step": 190,
465
+ "student_masked_tokens": 107.975
466
+ },
467
+ {
468
+ "avg_mask_ratio": 0.4670982737792656,
469
+ "avg_response_length": 210.9125,
470
+ "avg_student_mask_ratio": 0.4670982737792656,
471
+ "batch_ainp_frac": 0.0,
472
+ "batch_inp_frac": 0.0,
473
+ "batch_inp_oh_frac": 0.0,
474
+ "batch_inp_par_frac": 1.0,
475
+ "batch_inp_par_reverse_frac": 0.0,
476
+ "batch_rl_frac": 0.0,
477
+ "batch_sft_frac": 0.0,
478
+ "batch_soft_sft_frac": 0.0,
479
+ "batch_tf_frac": 0.0,
480
+ "ce_loss": 0.0,
481
+ "epoch": 0.4266666666666667,
482
+ "grad_norm": 0.142578125,
483
+ "kd_loss": 0.15220829088375468,
484
+ "learning_rate": 3e-06,
485
+ "loss": 0.1609,
486
+ "masked_tokens": 98.2125,
487
+ "mean_t": 0.5073627714533359,
488
+ "step": 200,
489
+ "student_masked_tokens": 98.2125
490
+ },
491
+ {
492
+ "avg_mask_ratio": 0.4568137794849463,
493
+ "avg_response_length": 218.7375,
494
+ "avg_student_mask_ratio": 0.4568137794849463,
495
+ "batch_ainp_frac": 0.0,
496
+ "batch_inp_frac": 0.0,
497
+ "batch_inp_oh_frac": 0.0,
498
+ "batch_inp_par_frac": 1.0,
499
+ "batch_inp_par_reverse_frac": 0.0,
500
+ "batch_rl_frac": 0.0,
501
+ "batch_sft_frac": 0.0,
502
+ "batch_soft_sft_frac": 0.0,
503
+ "batch_tf_frac": 0.0,
504
+ "ce_loss": 0.0,
505
+ "epoch": 0.448,
506
+ "grad_norm": 0.1826171875,
507
+ "kd_loss": 0.1580252643583208,
508
+ "learning_rate": 3e-06,
509
+ "loss": 0.1798,
510
+ "masked_tokens": 99.15,
511
+ "mean_t": 0.48430291628465055,
512
+ "step": 210,
513
+ "student_masked_tokens": 99.15
514
+ },
515
+ {
516
+ "avg_mask_ratio": 0.5379857187625021,
517
+ "avg_response_length": 252.1375,
518
+ "avg_student_mask_ratio": 0.5379857187625021,
519
+ "batch_ainp_frac": 0.0,
520
+ "batch_inp_frac": 0.0,
521
+ "batch_inp_oh_frac": 0.0,
522
+ "batch_inp_par_frac": 1.0,
523
+ "batch_inp_par_reverse_frac": 0.0,
524
+ "batch_rl_frac": 0.0,
525
+ "batch_sft_frac": 0.0,
526
+ "batch_soft_sft_frac": 0.0,
527
+ "batch_tf_frac": 0.0,
528
+ "ce_loss": 0.0,
529
+ "epoch": 0.4693333333333333,
530
+ "grad_norm": 0.3203125,
531
+ "kd_loss": 0.2511090909683844,
532
+ "learning_rate": 3e-06,
533
+ "loss": 0.2583,
534
+ "masked_tokens": 135.4,
535
+ "mean_t": 0.570199209311977,
536
+ "step": 220,
537
+ "student_masked_tokens": 135.4
538
+ },
539
+ {
540
+ "avg_mask_ratio": 0.43395056116278286,
541
+ "avg_response_length": 245.2625,
542
+ "avg_student_mask_ratio": 0.43395056116278286,
543
+ "batch_ainp_frac": 0.0,
544
+ "batch_inp_frac": 0.0,
545
+ "batch_inp_oh_frac": 0.0,
546
+ "batch_inp_par_frac": 1.0,
547
+ "batch_inp_par_reverse_frac": 0.0,
548
+ "batch_rl_frac": 0.0,
549
+ "batch_sft_frac": 0.0,
550
+ "batch_soft_sft_frac": 0.0,
551
+ "batch_tf_frac": 0.0,
552
+ "ce_loss": 0.0,
553
+ "epoch": 0.49066666666666664,
554
+ "grad_norm": 0.1767578125,
555
+ "kd_loss": 0.14414861655371852,
556
+ "learning_rate": 3e-06,
557
+ "loss": 0.1964,
558
+ "masked_tokens": 102.5125,
559
+ "mean_t": 0.46891279935371133,
560
+ "step": 230,
561
+ "student_masked_tokens": 102.5125
562
+ },
563
+ {
564
+ "avg_mask_ratio": 0.46948411157354714,
565
+ "avg_response_length": 202.975,
566
+ "avg_student_mask_ratio": 0.46948411157354714,
567
+ "batch_ainp_frac": 0.0,
568
+ "batch_inp_frac": 0.0,
569
+ "batch_inp_oh_frac": 0.0,
570
+ "batch_inp_par_frac": 1.0,
571
+ "batch_inp_par_reverse_frac": 0.0,
572
+ "batch_rl_frac": 0.0,
573
+ "batch_sft_frac": 0.0,
574
+ "batch_soft_sft_frac": 0.0,
575
+ "batch_tf_frac": 0.0,
576
+ "ce_loss": 0.0,
577
+ "epoch": 0.512,
578
+ "grad_norm": 0.1181640625,
579
+ "kd_loss": 0.2197965504183493,
580
+ "learning_rate": 3e-06,
581
+ "loss": 0.243,
582
+ "masked_tokens": 97.0625,
583
+ "mean_t": 0.502228345896583,
584
+ "step": 240,
585
+ "student_masked_tokens": 97.0625
586
+ },
587
+ {
588
+ "avg_mask_ratio": 0.44631263689370826,
589
+ "avg_response_length": 243.425,
590
+ "avg_student_mask_ratio": 0.44631263689370826,
591
+ "batch_ainp_frac": 0.0,
592
+ "batch_inp_frac": 0.0,
593
+ "batch_inp_oh_frac": 0.0,
594
+ "batch_inp_par_frac": 1.0,
595
+ "batch_inp_par_reverse_frac": 0.0,
596
+ "batch_rl_frac": 0.0,
597
+ "batch_sft_frac": 0.0,
598
+ "batch_soft_sft_frac": 0.0,
599
+ "batch_tf_frac": 0.0,
600
+ "ce_loss": 0.0,
601
+ "epoch": 0.5333333333333333,
602
+ "grad_norm": 0.1064453125,
603
+ "kd_loss": 0.2151024747882957,
604
+ "learning_rate": 3e-06,
605
+ "loss": 0.1892,
606
+ "masked_tokens": 107.4625,
607
+ "mean_t": 0.4733429416548461,
608
+ "step": 250,
609
+ "student_masked_tokens": 107.4625
610
+ },
611
+ {
612
+ "avg_mask_ratio": 0.4607314572727773,
613
+ "avg_response_length": 243.375,
614
+ "avg_student_mask_ratio": 0.4607314572727773,
615
+ "batch_ainp_frac": 0.0,
616
+ "batch_inp_frac": 0.0,
617
+ "batch_inp_oh_frac": 0.0,
618
+ "batch_inp_par_frac": 1.0,
619
+ "batch_inp_par_reverse_frac": 0.0,
620
+ "batch_rl_frac": 0.0,
621
+ "batch_sft_frac": 0.0,
622
+ "batch_soft_sft_frac": 0.0,
623
+ "batch_tf_frac": 0.0,
624
+ "ce_loss": 0.0,
625
+ "epoch": 0.5546666666666666,
626
+ "grad_norm": 0.203125,
627
+ "kd_loss": 0.176242933875335,
628
+ "learning_rate": 3e-06,
629
+ "loss": 0.1885,
630
+ "masked_tokens": 110.8875,
631
+ "mean_t": 0.4843149524240289,
632
+ "step": 260,
633
+ "student_masked_tokens": 110.8875
634
+ },
635
+ {
636
+ "avg_mask_ratio": 0.5309946444118395,
637
+ "avg_response_length": 231.6875,
638
+ "avg_student_mask_ratio": 0.5309946444118395,
639
+ "batch_ainp_frac": 0.0,
640
+ "batch_inp_frac": 0.0,
641
+ "batch_inp_oh_frac": 0.0,
642
+ "batch_inp_par_frac": 1.0,
643
+ "batch_inp_par_reverse_frac": 0.0,
644
+ "batch_rl_frac": 0.0,
645
+ "batch_sft_frac": 0.0,
646
+ "batch_soft_sft_frac": 0.0,
647
+ "batch_tf_frac": 0.0,
648
+ "ce_loss": 0.0,
649
+ "epoch": 0.576,
650
+ "grad_norm": 0.208984375,
651
+ "kd_loss": 0.26651088480309115,
652
+ "learning_rate": 3e-06,
653
+ "loss": 0.2828,
654
+ "masked_tokens": 123.2875,
655
+ "mean_t": 0.5643589949700981,
656
+ "step": 270,
657
+ "student_masked_tokens": 123.2875
658
+ },
659
+ {
660
+ "avg_mask_ratio": 0.45879559536697345,
661
+ "avg_response_length": 251.8,
662
+ "avg_student_mask_ratio": 0.45879559536697345,
663
+ "batch_ainp_frac": 0.0,
664
+ "batch_inp_frac": 0.0,
665
+ "batch_inp_oh_frac": 0.0,
666
+ "batch_inp_par_frac": 1.0,
667
+ "batch_inp_par_reverse_frac": 0.0,
668
+ "batch_rl_frac": 0.0,
669
+ "batch_sft_frac": 0.0,
670
+ "batch_soft_sft_frac": 0.0,
671
+ "batch_tf_frac": 0.0,
672
+ "ce_loss": 0.0,
673
+ "epoch": 0.5973333333333334,
674
+ "grad_norm": 0.11474609375,
675
+ "kd_loss": 0.145786481295454,
676
+ "learning_rate": 3e-06,
677
+ "loss": 0.1439,
678
+ "masked_tokens": 125.425,
679
+ "mean_t": 0.47818811538163575,
680
+ "step": 280,
681
+ "student_masked_tokens": 125.425
682
+ },
683
+ {
684
+ "avg_mask_ratio": 0.3955249205115251,
685
+ "avg_response_length": 238.675,
686
+ "avg_student_mask_ratio": 0.3955249205115251,
687
+ "batch_ainp_frac": 0.0,
688
+ "batch_inp_frac": 0.0,
689
+ "batch_inp_oh_frac": 0.0,
690
+ "batch_inp_par_frac": 1.0,
691
+ "batch_inp_par_reverse_frac": 0.0,
692
+ "batch_rl_frac": 0.0,
693
+ "batch_sft_frac": 0.0,
694
+ "batch_soft_sft_frac": 0.0,
695
+ "batch_tf_frac": 0.0,
696
+ "ce_loss": 0.0,
697
+ "epoch": 0.6186666666666667,
698
+ "grad_norm": 0.057373046875,
699
+ "kd_loss": 0.15104623195453543,
700
+ "learning_rate": 3e-06,
701
+ "loss": 0.1578,
702
+ "masked_tokens": 95.2125,
703
+ "mean_t": 0.4088635521940887,
704
+ "step": 290,
705
+ "student_masked_tokens": 95.2125
706
+ },
707
+ {
708
+ "avg_mask_ratio": 0.4504710016073659,
709
+ "avg_response_length": 202.575,
710
+ "avg_student_mask_ratio": 0.4504710016073659,
711
+ "batch_ainp_frac": 0.0,
712
+ "batch_inp_frac": 0.0,
713
+ "batch_inp_oh_frac": 0.0,
714
+ "batch_inp_par_frac": 1.0,
715
+ "batch_inp_par_reverse_frac": 0.0,
716
+ "batch_rl_frac": 0.0,
717
+ "batch_sft_frac": 0.0,
718
+ "batch_soft_sft_frac": 0.0,
719
+ "batch_tf_frac": 0.0,
720
+ "ce_loss": 0.0,
721
+ "epoch": 0.64,
722
+ "grad_norm": 0.150390625,
723
+ "kd_loss": 0.16011972622800386,
724
+ "learning_rate": 3e-06,
725
+ "loss": 0.179,
726
+ "masked_tokens": 90.3375,
727
+ "mean_t": 0.47984200695063917,
728
+ "step": 300,
729
+ "student_masked_tokens": 90.3375
730
+ },
731
+ {
732
+ "avg_mask_ratio": 0.4822002159198746,
733
+ "avg_response_length": 189.9875,
734
+ "avg_student_mask_ratio": 0.4822002159198746,
735
+ "batch_ainp_frac": 0.0,
736
+ "batch_inp_frac": 0.0,
737
+ "batch_inp_oh_frac": 0.0,
738
+ "batch_inp_par_frac": 1.0,
739
+ "batch_inp_par_reverse_frac": 0.0,
740
+ "batch_rl_frac": 0.0,
741
+ "batch_sft_frac": 0.0,
742
+ "batch_soft_sft_frac": 0.0,
743
+ "batch_tf_frac": 0.0,
744
+ "ce_loss": 0.0,
745
+ "epoch": 0.6613333333333333,
746
+ "grad_norm": 0.1630859375,
747
+ "kd_loss": 0.21744939284749734,
748
+ "learning_rate": 3e-06,
749
+ "loss": 0.201,
750
+ "masked_tokens": 92.7,
751
+ "mean_t": 0.5026606284547597,
752
+ "step": 310,
753
+ "student_masked_tokens": 92.7
754
+ },
755
+ {
756
+ "avg_mask_ratio": 0.5024422638700343,
757
+ "avg_response_length": 237.2625,
758
+ "avg_student_mask_ratio": 0.5024422638700343,
759
+ "batch_ainp_frac": 0.0,
760
+ "batch_inp_frac": 0.0,
761
+ "batch_inp_oh_frac": 0.0,
762
+ "batch_inp_par_frac": 1.0,
763
+ "batch_inp_par_reverse_frac": 0.0,
764
+ "batch_rl_frac": 0.0,
765
+ "batch_sft_frac": 0.0,
766
+ "batch_soft_sft_frac": 0.0,
767
+ "batch_tf_frac": 0.0,
768
+ "ce_loss": 0.0,
769
+ "epoch": 0.6826666666666666,
770
+ "grad_norm": 0.1259765625,
771
+ "kd_loss": 0.21489343987664142,
772
+ "learning_rate": 3e-06,
773
+ "loss": 0.2268,
774
+ "masked_tokens": 123.8125,
775
+ "mean_t": 0.5293499688967132,
776
+ "step": 320,
777
+ "student_masked_tokens": 123.8125
778
+ },
779
+ {
780
+ "avg_mask_ratio": 0.510216062690597,
781
+ "avg_response_length": 257.525,
782
+ "avg_student_mask_ratio": 0.510216062690597,
783
+ "batch_ainp_frac": 0.0,
784
+ "batch_inp_frac": 0.0,
785
+ "batch_inp_oh_frac": 0.0,
786
+ "batch_inp_par_frac": 1.0,
787
+ "batch_inp_par_reverse_frac": 0.0,
788
+ "batch_rl_frac": 0.0,
789
+ "batch_sft_frac": 0.0,
790
+ "batch_soft_sft_frac": 0.0,
791
+ "batch_tf_frac": 0.0,
792
+ "ce_loss": 0.0,
793
+ "epoch": 0.704,
794
+ "grad_norm": 0.1337890625,
795
+ "kd_loss": 0.17950079924535203,
796
+ "learning_rate": 3e-06,
797
+ "loss": 0.2007,
798
+ "masked_tokens": 142.8125,
799
+ "mean_t": 0.5426030711154454,
800
+ "step": 330,
801
+ "student_masked_tokens": 142.8125
802
+ },
803
+ {
804
+ "avg_mask_ratio": 0.45074162067612633,
805
+ "avg_response_length": 245.525,
806
+ "avg_student_mask_ratio": 0.45074162067612633,
807
+ "batch_ainp_frac": 0.0,
808
+ "batch_inp_frac": 0.0,
809
+ "batch_inp_oh_frac": 0.0,
810
+ "batch_inp_par_frac": 1.0,
811
+ "batch_inp_par_reverse_frac": 0.0,
812
+ "batch_rl_frac": 0.0,
813
+ "batch_sft_frac": 0.0,
814
+ "batch_soft_sft_frac": 0.0,
815
+ "batch_tf_frac": 0.0,
816
+ "ce_loss": 0.0,
817
+ "epoch": 0.7253333333333334,
818
+ "grad_norm": 0.310546875,
819
+ "kd_loss": 0.14043198096701417,
820
+ "learning_rate": 3e-06,
821
+ "loss": 0.1669,
822
+ "masked_tokens": 116.0875,
823
+ "mean_t": 0.47635243807453664,
824
+ "step": 340,
825
+ "student_masked_tokens": 116.0875
826
+ },
827
+ {
828
+ "avg_mask_ratio": 0.4926802407717332,
829
+ "avg_response_length": 238.2,
830
+ "avg_student_mask_ratio": 0.4926802407717332,
831
+ "batch_ainp_frac": 0.0,
832
+ "batch_inp_frac": 0.0,
833
+ "batch_inp_oh_frac": 0.0,
834
+ "batch_inp_par_frac": 1.0,
835
+ "batch_inp_par_reverse_frac": 0.0,
836
+ "batch_rl_frac": 0.0,
837
+ "batch_sft_frac": 0.0,
838
+ "batch_soft_sft_frac": 0.0,
839
+ "batch_tf_frac": 0.0,
840
+ "ce_loss": 0.0,
841
+ "epoch": 0.7466666666666667,
842
+ "grad_norm": 0.330078125,
843
+ "kd_loss": 0.21977804936059328,
844
+ "learning_rate": 3e-06,
845
+ "loss": 0.2497,
846
+ "masked_tokens": 127.4875,
847
+ "mean_t": 0.5268881446914747,
848
+ "step": 350,
849
+ "student_masked_tokens": 127.4875
850
+ },
851
+ {
852
+ "avg_mask_ratio": 0.4706261330051348,
853
+ "avg_response_length": 251.7375,
854
+ "avg_student_mask_ratio": 0.4706261330051348,
855
+ "batch_ainp_frac": 0.0,
856
+ "batch_inp_frac": 0.0,
857
+ "batch_inp_oh_frac": 0.0,
858
+ "batch_inp_par_frac": 1.0,
859
+ "batch_inp_par_reverse_frac": 0.0,
860
+ "batch_rl_frac": 0.0,
861
+ "batch_sft_frac": 0.0,
862
+ "batch_soft_sft_frac": 0.0,
863
+ "batch_tf_frac": 0.0,
864
+ "ce_loss": 0.0,
865
+ "epoch": 0.768,
866
+ "grad_norm": 0.056640625,
867
+ "kd_loss": 0.29630907970476983,
868
+ "learning_rate": 3e-06,
869
+ "loss": 0.2329,
870
+ "masked_tokens": 116.8625,
871
+ "mean_t": 0.5040419134311378,
872
+ "step": 360,
873
+ "student_masked_tokens": 116.8625
874
+ },
875
+ {
876
+ "avg_mask_ratio": 0.48096118308603764,
877
+ "avg_response_length": 262.25,
878
+ "avg_student_mask_ratio": 0.48096118308603764,
879
+ "batch_ainp_frac": 0.0,
880
+ "batch_inp_frac": 0.0,
881
+ "batch_inp_oh_frac": 0.0,
882
+ "batch_inp_par_frac": 1.0,
883
+ "batch_inp_par_reverse_frac": 0.0,
884
+ "batch_rl_frac": 0.0,
885
+ "batch_sft_frac": 0.0,
886
+ "batch_soft_sft_frac": 0.0,
887
+ "batch_tf_frac": 0.0,
888
+ "ce_loss": 0.0,
889
+ "epoch": 0.7893333333333333,
890
+ "grad_norm": 0.12255859375,
891
+ "kd_loss": 0.20822112379132704,
892
+ "learning_rate": 3e-06,
893
+ "loss": 0.186,
894
+ "masked_tokens": 132.2,
895
+ "mean_t": 0.5127181728370488,
896
+ "step": 370,
897
+ "student_masked_tokens": 132.2
898
+ },
899
+ {
900
+ "avg_mask_ratio": 0.4433969090110622,
901
+ "avg_response_length": 209.7125,
902
+ "avg_student_mask_ratio": 0.4433969090110622,
903
+ "batch_ainp_frac": 0.0,
904
+ "batch_inp_frac": 0.0,
905
+ "batch_inp_oh_frac": 0.0,
906
+ "batch_inp_par_frac": 1.0,
907
+ "batch_inp_par_reverse_frac": 0.0,
908
+ "batch_rl_frac": 0.0,
909
+ "batch_sft_frac": 0.0,
910
+ "batch_soft_sft_frac": 0.0,
911
+ "batch_tf_frac": 0.0,
912
+ "ce_loss": 0.0,
913
+ "epoch": 0.8106666666666666,
914
+ "grad_norm": 0.123046875,
915
+ "kd_loss": 0.14427866181035826,
916
+ "learning_rate": 3e-06,
917
+ "loss": 0.159,
918
+ "masked_tokens": 95.8625,
919
+ "mean_t": 0.4825185665744357,
920
+ "step": 380,
921
+ "student_masked_tokens": 95.8625
922
+ },
923
+ {
924
+ "avg_mask_ratio": 0.4764250977139454,
925
+ "avg_response_length": 226.3875,
926
+ "avg_student_mask_ratio": 0.4764250977139454,
927
+ "batch_ainp_frac": 0.0,
928
+ "batch_inp_frac": 0.0,
929
+ "batch_inp_oh_frac": 0.0,
930
+ "batch_inp_par_frac": 1.0,
931
+ "batch_inp_par_reverse_frac": 0.0,
932
+ "batch_rl_frac": 0.0,
933
+ "batch_sft_frac": 0.0,
934
+ "batch_soft_sft_frac": 0.0,
935
+ "batch_tf_frac": 0.0,
936
+ "ce_loss": 0.0,
937
+ "epoch": 0.832,
938
+ "grad_norm": 0.185546875,
939
+ "kd_loss": 0.18584371552193701,
940
+ "learning_rate": 3e-06,
941
+ "loss": 0.1823,
942
+ "masked_tokens": 113.95,
943
+ "mean_t": 0.5053101469413377,
944
+ "step": 390,
945
+ "student_masked_tokens": 113.95
946
+ },
947
+ {
948
+ "avg_mask_ratio": 0.47088071387261154,
949
+ "avg_response_length": 233.2125,
950
+ "avg_student_mask_ratio": 0.47088071387261154,
951
+ "batch_ainp_frac": 0.0,
952
+ "batch_inp_frac": 0.0,
953
+ "batch_inp_oh_frac": 0.0,
954
+ "batch_inp_par_frac": 1.0,
955
+ "batch_inp_par_reverse_frac": 0.0,
956
+ "batch_rl_frac": 0.0,
957
+ "batch_sft_frac": 0.0,
958
+ "batch_soft_sft_frac": 0.0,
959
+ "batch_tf_frac": 0.0,
960
+ "ce_loss": 0.0,
961
+ "epoch": 0.8533333333333334,
962
+ "grad_norm": 0.2734375,
963
+ "kd_loss": 0.22232839192147366,
964
+ "learning_rate": 3e-06,
965
+ "loss": 0.1961,
966
+ "masked_tokens": 116.675,
967
+ "mean_t": 0.5052250675857067,
968
+ "step": 400,
969
+ "student_masked_tokens": 116.675
970
+ },
971
+ {
972
+ "avg_mask_ratio": 0.4870647343515884,
973
+ "avg_response_length": 228.3875,
974
+ "avg_student_mask_ratio": 0.4870647343515884,
975
+ "batch_ainp_frac": 0.0,
976
+ "batch_inp_frac": 0.0,
977
+ "batch_inp_oh_frac": 0.0,
978
+ "batch_inp_par_frac": 1.0,
979
+ "batch_inp_par_reverse_frac": 0.0,
980
+ "batch_rl_frac": 0.0,
981
+ "batch_sft_frac": 0.0,
982
+ "batch_soft_sft_frac": 0.0,
983
+ "batch_tf_frac": 0.0,
984
+ "ce_loss": 0.0,
985
+ "epoch": 0.8746666666666667,
986
+ "grad_norm": 0.25390625,
987
+ "kd_loss": 0.25729746209006665,
988
+ "learning_rate": 3e-06,
989
+ "loss": 0.2306,
990
+ "masked_tokens": 114.3625,
991
+ "mean_t": 0.5050956419203431,
992
+ "step": 410,
993
+ "student_masked_tokens": 114.3625
994
+ },
995
+ {
996
+ "avg_mask_ratio": 0.440834702400025,
997
+ "avg_response_length": 209.85,
998
+ "avg_student_mask_ratio": 0.440834702400025,
999
+ "batch_ainp_frac": 0.0,
1000
+ "batch_inp_frac": 0.0,
1001
+ "batch_inp_oh_frac": 0.0,
1002
+ "batch_inp_par_frac": 1.0,
1003
+ "batch_inp_par_reverse_frac": 0.0,
1004
+ "batch_rl_frac": 0.0,
1005
+ "batch_sft_frac": 0.0,
1006
+ "batch_soft_sft_frac": 0.0,
1007
+ "batch_tf_frac": 0.0,
1008
+ "ce_loss": 0.0,
1009
+ "epoch": 0.896,
1010
+ "grad_norm": 0.2275390625,
1011
+ "kd_loss": 0.15747290870124503,
1012
+ "learning_rate": 3e-06,
1013
+ "loss": 0.1583,
1014
+ "masked_tokens": 87.575,
1015
+ "mean_t": 0.4769687672611326,
1016
+ "step": 420,
1017
+ "student_masked_tokens": 87.575
1018
+ },
1019
+ {
1020
+ "avg_mask_ratio": 0.4660509963519871,
1021
+ "avg_response_length": 250.9125,
1022
+ "avg_student_mask_ratio": 0.4660509963519871,
1023
+ "batch_ainp_frac": 0.0,
1024
+ "batch_inp_frac": 0.0,
1025
+ "batch_inp_oh_frac": 0.0,
1026
+ "batch_inp_par_frac": 1.0,
1027
+ "batch_inp_par_reverse_frac": 0.0,
1028
+ "batch_rl_frac": 0.0,
1029
+ "batch_sft_frac": 0.0,
1030
+ "batch_soft_sft_frac": 0.0,
1031
+ "batch_tf_frac": 0.0,
1032
+ "ce_loss": 0.0,
1033
+ "epoch": 0.9173333333333333,
1034
+ "grad_norm": 0.392578125,
1035
+ "kd_loss": 0.17299866449352522,
1036
+ "learning_rate": 3e-06,
1037
+ "loss": 0.178,
1038
+ "masked_tokens": 109.675,
1039
+ "mean_t": 0.49262027950026094,
1040
+ "step": 430,
1041
+ "student_masked_tokens": 109.675
1042
+ },
1043
+ {
1044
+ "avg_mask_ratio": 0.4507861359976232,
1045
+ "avg_response_length": 235.875,
1046
+ "avg_student_mask_ratio": 0.4507861359976232,
1047
+ "batch_ainp_frac": 0.0,
1048
+ "batch_inp_frac": 0.0,
1049
+ "batch_inp_oh_frac": 0.0,
1050
+ "batch_inp_par_frac": 1.0,
1051
+ "batch_inp_par_reverse_frac": 0.0,
1052
+ "batch_rl_frac": 0.0,
1053
+ "batch_sft_frac": 0.0,
1054
+ "batch_soft_sft_frac": 0.0,
1055
+ "batch_tf_frac": 0.0,
1056
+ "ce_loss": 0.0,
1057
+ "epoch": 0.9386666666666666,
1058
+ "grad_norm": 0.15234375,
1059
+ "kd_loss": 0.2113636662043291,
1060
+ "learning_rate": 3e-06,
1061
+ "loss": 0.1795,
1062
+ "masked_tokens": 106.95,
1063
+ "mean_t": 0.47046207524836064,
1064
+ "step": 440,
1065
+ "student_masked_tokens": 106.95
1066
+ },
1067
+ {
1068
+ "avg_mask_ratio": 0.4283985076006502,
1069
+ "avg_response_length": 230.95,
1070
+ "avg_student_mask_ratio": 0.4283985076006502,
1071
+ "batch_ainp_frac": 0.0,
1072
+ "batch_inp_frac": 0.0,
1073
+ "batch_inp_oh_frac": 0.0,
1074
+ "batch_inp_par_frac": 1.0,
1075
+ "batch_inp_par_reverse_frac": 0.0,
1076
+ "batch_rl_frac": 0.0,
1077
+ "batch_sft_frac": 0.0,
1078
+ "batch_soft_sft_frac": 0.0,
1079
+ "batch_tf_frac": 0.0,
1080
+ "ce_loss": 0.0,
1081
+ "epoch": 0.96,
1082
+ "grad_norm": 0.16015625,
1083
+ "kd_loss": 0.15304818171161969,
1084
+ "learning_rate": 3e-06,
1085
+ "loss": 0.1724,
1086
+ "masked_tokens": 101.15,
1087
+ "mean_t": 0.45781184462830427,
1088
+ "step": 450,
1089
+ "student_masked_tokens": 101.15
1090
+ },
1091
+ {
1092
+ "avg_mask_ratio": 0.47474822774529457,
1093
+ "avg_response_length": 233.1,
1094
+ "avg_student_mask_ratio": 0.47474822774529457,
1095
+ "batch_ainp_frac": 0.0,
1096
+ "batch_inp_frac": 0.0,
1097
+ "batch_inp_oh_frac": 0.0,
1098
+ "batch_inp_par_frac": 1.0,
1099
+ "batch_inp_par_reverse_frac": 0.0,
1100
+ "batch_rl_frac": 0.0,
1101
+ "batch_sft_frac": 0.0,
1102
+ "batch_soft_sft_frac": 0.0,
1103
+ "batch_tf_frac": 0.0,
1104
+ "ce_loss": 0.0,
1105
+ "epoch": 0.9813333333333333,
1106
+ "grad_norm": 0.06689453125,
1107
+ "kd_loss": 0.1363761811357108,
1108
+ "learning_rate": 3e-06,
1109
+ "loss": 0.171,
1110
+ "masked_tokens": 112.725,
1111
+ "mean_t": 0.5075790266972036,
1112
+ "step": 460,
1113
+ "student_masked_tokens": 112.725
1114
+ },
1115
+ {
1116
+ "avg_mask_ratio": 0.4808142688901474,
1117
+ "avg_response_length": 238.54761904761904,
1118
+ "avg_student_mask_ratio": 0.4808142688901474,
1119
+ "batch_ainp_frac": 0.0,
1120
+ "batch_inp_frac": 0.0,
1121
+ "batch_inp_oh_frac": 0.0,
1122
+ "batch_inp_par_frac": 1.0,
1123
+ "batch_inp_par_reverse_frac": 0.0,
1124
+ "batch_rl_frac": 0.0,
1125
+ "batch_sft_frac": 0.0,
1126
+ "batch_soft_sft_frac": 0.0,
1127
+ "batch_tf_frac": 0.0,
1128
+ "ce_loss": 0.0,
1129
+ "epoch": 1.0042666666666666,
1130
+ "grad_norm": 0.201171875,
1131
+ "kd_loss": 0.22379483340171732,
1132
+ "learning_rate": 3e-06,
1133
+ "loss": 0.2466,
1134
+ "masked_tokens": 113.67857142857143,
1135
+ "mean_t": 0.5031429776822084,
1136
+ "step": 470,
1137
+ "student_masked_tokens": 113.67857142857143
1138
+ },
1139
+ {
1140
+ "avg_mask_ratio": 0.47175657459301873,
1141
+ "avg_response_length": 249.9625,
1142
+ "avg_student_mask_ratio": 0.47175657459301873,
1143
+ "batch_ainp_frac": 0.0,
1144
+ "batch_inp_frac": 0.0,
1145
+ "batch_inp_oh_frac": 0.0,
1146
+ "batch_inp_par_frac": 1.0,
1147
+ "batch_inp_par_reverse_frac": 0.0,
1148
+ "batch_rl_frac": 0.0,
1149
+ "batch_sft_frac": 0.0,
1150
+ "batch_soft_sft_frac": 0.0,
1151
+ "batch_tf_frac": 0.0,
1152
+ "ce_loss": 0.0,
1153
+ "epoch": 1.0256,
1154
+ "grad_norm": 0.166015625,
1155
+ "kd_loss": 0.17491777570117745,
1156
+ "learning_rate": 3e-06,
1157
+ "loss": 0.2029,
1158
+ "masked_tokens": 119.4625,
1159
+ "mean_t": 0.504472183593316,
1160
+ "step": 480,
1161
+ "student_masked_tokens": 119.4625
1162
+ },
1163
+ {
1164
+ "avg_mask_ratio": 0.4564988439786248,
1165
+ "avg_response_length": 238.8875,
1166
+ "avg_student_mask_ratio": 0.4564988439786248,
1167
+ "batch_ainp_frac": 0.0,
1168
+ "batch_inp_frac": 0.0,
1169
+ "batch_inp_oh_frac": 0.0,
1170
+ "batch_inp_par_frac": 1.0,
1171
+ "batch_inp_par_reverse_frac": 0.0,
1172
+ "batch_rl_frac": 0.0,
1173
+ "batch_sft_frac": 0.0,
1174
+ "batch_soft_sft_frac": 0.0,
1175
+ "batch_tf_frac": 0.0,
1176
+ "ce_loss": 0.0,
1177
+ "epoch": 1.0469333333333333,
1178
+ "grad_norm": 0.1279296875,
1179
+ "kd_loss": 0.12884608846077866,
1180
+ "learning_rate": 3e-06,
1181
+ "loss": 0.1536,
1182
+ "masked_tokens": 104.0,
1183
+ "mean_t": 0.4888980514719151,
1184
+ "step": 490,
1185
+ "student_masked_tokens": 104.0
1186
+ },
1187
+ {
1188
+ "avg_mask_ratio": 0.5083174118888565,
1189
+ "avg_response_length": 258.1375,
1190
+ "avg_student_mask_ratio": 0.5083174118888565,
1191
+ "batch_ainp_frac": 0.0,
1192
+ "batch_inp_frac": 0.0,
1193
+ "batch_inp_oh_frac": 0.0,
1194
+ "batch_inp_par_frac": 1.0,
1195
+ "batch_inp_par_reverse_frac": 0.0,
1196
+ "batch_rl_frac": 0.0,
1197
+ "batch_sft_frac": 0.0,
1198
+ "batch_soft_sft_frac": 0.0,
1199
+ "batch_tf_frac": 0.0,
1200
+ "ce_loss": 0.0,
1201
+ "epoch": 1.0682666666666667,
1202
+ "grad_norm": 0.1357421875,
1203
+ "kd_loss": 0.18128383785248586,
1204
+ "learning_rate": 3e-06,
1205
+ "loss": 0.1811,
1206
+ "masked_tokens": 133.7125,
1207
+ "mean_t": 0.5469163245841628,
1208
+ "step": 500,
1209
+ "student_masked_tokens": 133.7125
1210
+ },
1211
+ {
1212
+ "avg_mask_ratio": 0.5130103683215566,
1213
+ "avg_response_length": 246.4875,
1214
+ "avg_student_mask_ratio": 0.5130103683215566,
1215
+ "batch_ainp_frac": 0.0,
1216
+ "batch_inp_frac": 0.0,
1217
+ "batch_inp_oh_frac": 0.0,
1218
+ "batch_inp_par_frac": 1.0,
1219
+ "batch_inp_par_reverse_frac": 0.0,
1220
+ "batch_rl_frac": 0.0,
1221
+ "batch_sft_frac": 0.0,
1222
+ "batch_soft_sft_frac": 0.0,
1223
+ "batch_tf_frac": 0.0,
1224
+ "ce_loss": 0.0,
1225
+ "epoch": 1.0896,
1226
+ "grad_norm": 0.11181640625,
1227
+ "kd_loss": 0.23847924997493805,
1228
+ "learning_rate": 3e-06,
1229
+ "loss": 0.2289,
1230
+ "masked_tokens": 132.5625,
1231
+ "mean_t": 0.5275314710394013,
1232
+ "step": 510,
1233
+ "student_masked_tokens": 132.5625
1234
+ },
1235
+ {
1236
+ "avg_mask_ratio": 0.4543681625276804,
1237
+ "avg_response_length": 199.65,
1238
+ "avg_student_mask_ratio": 0.4543681625276804,
1239
+ "batch_ainp_frac": 0.0,
1240
+ "batch_inp_frac": 0.0,
1241
+ "batch_inp_oh_frac": 0.0,
1242
+ "batch_inp_par_frac": 1.0,
1243
+ "batch_inp_par_reverse_frac": 0.0,
1244
+ "batch_rl_frac": 0.0,
1245
+ "batch_sft_frac": 0.0,
1246
+ "batch_soft_sft_frac": 0.0,
1247
+ "batch_tf_frac": 0.0,
1248
+ "ce_loss": 0.0,
1249
+ "epoch": 1.1109333333333333,
1250
+ "grad_norm": 0.1533203125,
1251
+ "kd_loss": 0.1353249137787543,
1252
+ "learning_rate": 3e-06,
1253
+ "loss": 0.1403,
1254
+ "masked_tokens": 87.9875,
1255
+ "mean_t": 0.49193521235138177,
1256
+ "step": 520,
1257
+ "student_masked_tokens": 87.9875
1258
+ },
1259
+ {
1260
+ "avg_mask_ratio": 0.46718079667771234,
1261
+ "avg_response_length": 218.0875,
1262
+ "avg_student_mask_ratio": 0.46718079667771234,
1263
+ "batch_ainp_frac": 0.0,
1264
+ "batch_inp_frac": 0.0,
1265
+ "batch_inp_oh_frac": 0.0,
1266
+ "batch_inp_par_frac": 1.0,
1267
+ "batch_inp_par_reverse_frac": 0.0,
1268
+ "batch_rl_frac": 0.0,
1269
+ "batch_sft_frac": 0.0,
1270
+ "batch_soft_sft_frac": 0.0,
1271
+ "batch_tf_frac": 0.0,
1272
+ "ce_loss": 0.0,
1273
+ "epoch": 1.1322666666666668,
1274
+ "grad_norm": 0.2109375,
1275
+ "kd_loss": 0.15268151032492625,
1276
+ "learning_rate": 3e-06,
1277
+ "loss": 0.1789,
1278
+ "masked_tokens": 101.3875,
1279
+ "mean_t": 0.49621942077938,
1280
+ "step": 530,
1281
+ "student_masked_tokens": 101.3875
1282
+ },
1283
+ {
1284
+ "avg_mask_ratio": 0.4890203754650429,
1285
+ "avg_response_length": 244.3875,
1286
+ "avg_student_mask_ratio": 0.4890203754650429,
1287
+ "batch_ainp_frac": 0.0,
1288
+ "batch_inp_frac": 0.0,
1289
+ "batch_inp_oh_frac": 0.0,
1290
+ "batch_inp_par_frac": 1.0,
1291
+ "batch_inp_par_reverse_frac": 0.0,
1292
+ "batch_rl_frac": 0.0,
1293
+ "batch_sft_frac": 0.0,
1294
+ "batch_soft_sft_frac": 0.0,
1295
+ "batch_tf_frac": 0.0,
1296
+ "ce_loss": 0.0,
1297
+ "epoch": 1.1536,
1298
+ "grad_norm": 0.27734375,
1299
+ "kd_loss": 0.17835129436630268,
1300
+ "learning_rate": 3e-06,
1301
+ "loss": 0.2173,
1302
+ "masked_tokens": 116.175,
1303
+ "mean_t": 0.5208023569080978,
1304
+ "step": 540,
1305
+ "student_masked_tokens": 116.175
1306
+ },
1307
+ {
1308
+ "avg_mask_ratio": 0.45064474650425834,
1309
+ "avg_response_length": 217.8375,
1310
+ "avg_student_mask_ratio": 0.45064474650425834,
1311
+ "batch_ainp_frac": 0.0,
1312
+ "batch_inp_frac": 0.0,
1313
+ "batch_inp_oh_frac": 0.0,
1314
+ "batch_inp_par_frac": 1.0,
1315
+ "batch_inp_par_reverse_frac": 0.0,
1316
+ "batch_rl_frac": 0.0,
1317
+ "batch_sft_frac": 0.0,
1318
+ "batch_soft_sft_frac": 0.0,
1319
+ "batch_tf_frac": 0.0,
1320
+ "ce_loss": 0.0,
1321
+ "epoch": 1.1749333333333334,
1322
+ "grad_norm": 0.17578125,
1323
+ "kd_loss": 0.16049880692362706,
1324
+ "learning_rate": 3e-06,
1325
+ "loss": 0.1855,
1326
+ "masked_tokens": 98.6375,
1327
+ "mean_t": 0.4767197913257405,
1328
+ "step": 550,
1329
+ "student_masked_tokens": 98.6375
1330
+ },
1331
+ {
1332
+ "avg_mask_ratio": 0.3844255942822201,
1333
+ "avg_response_length": 240.825,
1334
+ "avg_student_mask_ratio": 0.3844255942822201,
1335
+ "batch_ainp_frac": 0.0,
1336
+ "batch_inp_frac": 0.0,
1337
+ "batch_inp_oh_frac": 0.0,
1338
+ "batch_inp_par_frac": 1.0,
1339
+ "batch_inp_par_reverse_frac": 0.0,
1340
+ "batch_rl_frac": 0.0,
1341
+ "batch_sft_frac": 0.0,
1342
+ "batch_soft_sft_frac": 0.0,
1343
+ "batch_tf_frac": 0.0,
1344
+ "ce_loss": 0.0,
1345
+ "epoch": 1.1962666666666666,
1346
+ "grad_norm": 0.5859375,
1347
+ "kd_loss": 0.17605857607457268,
1348
+ "learning_rate": 3e-06,
1349
+ "loss": 0.1885,
1350
+ "masked_tokens": 96.85,
1351
+ "mean_t": 0.4111072298779618,
1352
+ "step": 560,
1353
+ "student_masked_tokens": 96.85
1354
+ },
1355
+ {
1356
+ "avg_mask_ratio": 0.45103558609262107,
1357
+ "avg_response_length": 231.025,
1358
+ "avg_student_mask_ratio": 0.45103558609262107,
1359
+ "batch_ainp_frac": 0.0,
1360
+ "batch_inp_frac": 0.0,
1361
+ "batch_inp_oh_frac": 0.0,
1362
+ "batch_inp_par_frac": 1.0,
1363
+ "batch_inp_par_reverse_frac": 0.0,
1364
+ "batch_rl_frac": 0.0,
1365
+ "batch_sft_frac": 0.0,
1366
+ "batch_soft_sft_frac": 0.0,
1367
+ "batch_tf_frac": 0.0,
1368
+ "ce_loss": 0.0,
1369
+ "epoch": 1.2176,
1370
+ "grad_norm": 0.234375,
1371
+ "kd_loss": 0.1386162672638477,
1372
+ "learning_rate": 3e-06,
1373
+ "loss": 0.1681,
1374
+ "masked_tokens": 105.6875,
1375
+ "mean_t": 0.48533305872697385,
1376
+ "step": 570,
1377
+ "student_masked_tokens": 105.6875
1378
+ },
1379
+ {
1380
+ "avg_mask_ratio": 0.47033366551622746,
1381
+ "avg_response_length": 248.3875,
1382
+ "avg_student_mask_ratio": 0.47033366551622746,
1383
+ "batch_ainp_frac": 0.0,
1384
+ "batch_inp_frac": 0.0,
1385
+ "batch_inp_oh_frac": 0.0,
1386
+ "batch_inp_par_frac": 1.0,
1387
+ "batch_inp_par_reverse_frac": 0.0,
1388
+ "batch_rl_frac": 0.0,
1389
+ "batch_sft_frac": 0.0,
1390
+ "batch_soft_sft_frac": 0.0,
1391
+ "batch_tf_frac": 0.0,
1392
+ "ce_loss": 0.0,
1393
+ "epoch": 1.2389333333333332,
1394
+ "grad_norm": 0.26953125,
1395
+ "kd_loss": 0.17702910760917803,
1396
+ "learning_rate": 3e-06,
1397
+ "loss": 0.1899,
1398
+ "masked_tokens": 125.05,
1399
+ "mean_t": 0.49976949762785805,
1400
+ "step": 580,
1401
+ "student_masked_tokens": 125.05
1402
+ },
1403
+ {
1404
+ "avg_mask_ratio": 0.5230229062028229,
1405
+ "avg_response_length": 241.8125,
1406
+ "avg_student_mask_ratio": 0.5230229062028229,
1407
+ "batch_ainp_frac": 0.0,
1408
+ "batch_inp_frac": 0.0,
1409
+ "batch_inp_oh_frac": 0.0,
1410
+ "batch_inp_par_frac": 1.0,
1411
+ "batch_inp_par_reverse_frac": 0.0,
1412
+ "batch_rl_frac": 0.0,
1413
+ "batch_sft_frac": 0.0,
1414
+ "batch_soft_sft_frac": 0.0,
1415
+ "batch_tf_frac": 0.0,
1416
+ "ce_loss": 0.0,
1417
+ "epoch": 1.2602666666666666,
1418
+ "grad_norm": 0.20703125,
1419
+ "kd_loss": 0.22054996666956866,
1420
+ "learning_rate": 3e-06,
1421
+ "loss": 0.2233,
1422
+ "masked_tokens": 129.0,
1423
+ "mean_t": 0.5521843038732186,
1424
+ "step": 590,
1425
+ "student_masked_tokens": 129.0
1426
+ },
1427
+ {
1428
+ "avg_mask_ratio": 0.44929012526990847,
1429
+ "avg_response_length": 246.7375,
1430
+ "avg_student_mask_ratio": 0.44929012526990847,
1431
+ "batch_ainp_frac": 0.0,
1432
+ "batch_inp_frac": 0.0,
1433
+ "batch_inp_oh_frac": 0.0,
1434
+ "batch_inp_par_frac": 1.0,
1435
+ "batch_inp_par_reverse_frac": 0.0,
1436
+ "batch_rl_frac": 0.0,
1437
+ "batch_sft_frac": 0.0,
1438
+ "batch_soft_sft_frac": 0.0,
1439
+ "batch_tf_frac": 0.0,
1440
+ "ce_loss": 0.0,
1441
+ "epoch": 1.2816,
1442
+ "grad_norm": 0.212890625,
1443
+ "kd_loss": 0.15257543138572202,
1444
+ "learning_rate": 3e-06,
1445
+ "loss": 0.1581,
1446
+ "masked_tokens": 113.9375,
1447
+ "mean_t": 0.4788527532829903,
1448
+ "step": 600,
1449
+ "student_masked_tokens": 113.9375
1450
+ },
1451
+ {
1452
+ "avg_mask_ratio": 0.5061312943696976,
1453
+ "avg_response_length": 237.175,
1454
+ "avg_student_mask_ratio": 0.5061312943696976,
1455
+ "batch_ainp_frac": 0.0,
1456
+ "batch_inp_frac": 0.0,
1457
+ "batch_inp_oh_frac": 0.0,
1458
+ "batch_inp_par_frac": 1.0,
1459
+ "batch_inp_par_reverse_frac": 0.0,
1460
+ "batch_rl_frac": 0.0,
1461
+ "batch_sft_frac": 0.0,
1462
+ "batch_soft_sft_frac": 0.0,
1463
+ "batch_tf_frac": 0.0,
1464
+ "ce_loss": 0.0,
1465
+ "epoch": 1.3029333333333333,
1466
+ "grad_norm": 0.06884765625,
1467
+ "kd_loss": 0.21158275993875578,
1468
+ "learning_rate": 3e-06,
1469
+ "loss": 0.1942,
1470
+ "masked_tokens": 116.675,
1471
+ "mean_t": 0.5445419924799353,
1472
+ "step": 610,
1473
+ "student_masked_tokens": 116.675
1474
+ },
1475
+ {
1476
+ "avg_mask_ratio": 0.4925117701757699,
1477
+ "avg_response_length": 233.2625,
1478
+ "avg_student_mask_ratio": 0.4925117701757699,
1479
+ "batch_ainp_frac": 0.0,
1480
+ "batch_inp_frac": 0.0,
1481
+ "batch_inp_oh_frac": 0.0,
1482
+ "batch_inp_par_frac": 1.0,
1483
+ "batch_inp_par_reverse_frac": 0.0,
1484
+ "batch_rl_frac": 0.0,
1485
+ "batch_sft_frac": 0.0,
1486
+ "batch_soft_sft_frac": 0.0,
1487
+ "batch_tf_frac": 0.0,
1488
+ "ce_loss": 0.0,
1489
+ "epoch": 1.3242666666666667,
1490
+ "grad_norm": 0.1787109375,
1491
+ "kd_loss": 0.1805886138310143,
1492
+ "learning_rate": 3e-06,
1493
+ "loss": 0.1744,
1494
+ "masked_tokens": 111.6875,
1495
+ "mean_t": 0.52490478400141,
1496
+ "step": 620,
1497
+ "student_masked_tokens": 111.6875
1498
+ },
1499
+ {
1500
+ "avg_mask_ratio": 0.5307835865532979,
1501
+ "avg_response_length": 249.65,
1502
+ "avg_student_mask_ratio": 0.5307835865532979,
1503
+ "batch_ainp_frac": 0.0,
1504
+ "batch_inp_frac": 0.0,
1505
+ "batch_inp_oh_frac": 0.0,
1506
+ "batch_inp_par_frac": 1.0,
1507
+ "batch_inp_par_reverse_frac": 0.0,
1508
+ "batch_rl_frac": 0.0,
1509
+ "batch_sft_frac": 0.0,
1510
+ "batch_soft_sft_frac": 0.0,
1511
+ "batch_tf_frac": 0.0,
1512
+ "ce_loss": 0.0,
1513
+ "epoch": 1.3456000000000001,
1514
+ "grad_norm": 0.28125,
1515
+ "kd_loss": 0.25403604302136956,
1516
+ "learning_rate": 3e-06,
1517
+ "loss": 0.2627,
1518
+ "masked_tokens": 126.6875,
1519
+ "mean_t": 0.5564947265549562,
1520
+ "step": 630,
1521
+ "student_masked_tokens": 126.6875
1522
+ },
1523
+ {
1524
+ "avg_mask_ratio": 0.48631439574528484,
1525
+ "avg_response_length": 241.475,
1526
+ "avg_student_mask_ratio": 0.48631439574528484,
1527
+ "batch_ainp_frac": 0.0,
1528
+ "batch_inp_frac": 0.0,
1529
+ "batch_inp_oh_frac": 0.0,
1530
+ "batch_inp_par_frac": 1.0,
1531
+ "batch_inp_par_reverse_frac": 0.0,
1532
+ "batch_rl_frac": 0.0,
1533
+ "batch_sft_frac": 0.0,
1534
+ "batch_soft_sft_frac": 0.0,
1535
+ "batch_tf_frac": 0.0,
1536
+ "ce_loss": 0.0,
1537
+ "epoch": 1.3669333333333333,
1538
+ "grad_norm": 0.2412109375,
1539
+ "kd_loss": 0.1642333460577902,
1540
+ "learning_rate": 3e-06,
1541
+ "loss": 0.1731,
1542
+ "masked_tokens": 122.1125,
1543
+ "mean_t": 0.5093393943971023,
1544
+ "step": 640,
1545
+ "student_masked_tokens": 122.1125
1546
+ },
1547
+ {
1548
+ "avg_mask_ratio": 0.5248487972887232,
1549
+ "avg_response_length": 231.5375,
1550
+ "avg_student_mask_ratio": 0.5248487972887232,
1551
+ "batch_ainp_frac": 0.0,
1552
+ "batch_inp_frac": 0.0,
1553
+ "batch_inp_oh_frac": 0.0,
1554
+ "batch_inp_par_frac": 1.0,
1555
+ "batch_inp_par_reverse_frac": 0.0,
1556
+ "batch_rl_frac": 0.0,
1557
+ "batch_sft_frac": 0.0,
1558
+ "batch_soft_sft_frac": 0.0,
1559
+ "batch_tf_frac": 0.0,
1560
+ "ce_loss": 0.0,
1561
+ "epoch": 1.3882666666666665,
1562
+ "grad_norm": 0.193359375,
1563
+ "kd_loss": 0.2508082524812494,
1564
+ "learning_rate": 3e-06,
1565
+ "loss": 0.2054,
1566
+ "masked_tokens": 127.175,
1567
+ "mean_t": 0.5558586571365595,
1568
+ "step": 650,
1569
+ "student_masked_tokens": 127.175
1570
+ },
1571
+ {
1572
+ "avg_mask_ratio": 0.48489007767057046,
1573
+ "avg_response_length": 246.0125,
1574
+ "avg_student_mask_ratio": 0.48489007767057046,
1575
+ "batch_ainp_frac": 0.0,
1576
+ "batch_inp_frac": 0.0,
1577
+ "batch_inp_oh_frac": 0.0,
1578
+ "batch_inp_par_frac": 1.0,
1579
+ "batch_inp_par_reverse_frac": 0.0,
1580
+ "batch_rl_frac": 0.0,
1581
+ "batch_sft_frac": 0.0,
1582
+ "batch_soft_sft_frac": 0.0,
1583
+ "batch_tf_frac": 0.0,
1584
+ "ce_loss": 0.0,
1585
+ "epoch": 1.4096,
1586
+ "grad_norm": 0.28125,
1587
+ "kd_loss": 0.18781521328146483,
1588
+ "learning_rate": 3e-06,
1589
+ "loss": 0.2045,
1590
+ "masked_tokens": 125.225,
1591
+ "mean_t": 0.51307404555846,
1592
+ "step": 660,
1593
+ "student_masked_tokens": 125.225
1594
+ },
1595
+ {
1596
+ "avg_mask_ratio": 0.4605832444969565,
1597
+ "avg_response_length": 244.5,
1598
+ "avg_student_mask_ratio": 0.4605832444969565,
1599
+ "batch_ainp_frac": 0.0,
1600
+ "batch_inp_frac": 0.0,
1601
+ "batch_inp_oh_frac": 0.0,
1602
+ "batch_inp_par_frac": 1.0,
1603
+ "batch_inp_par_reverse_frac": 0.0,
1604
+ "batch_rl_frac": 0.0,
1605
+ "batch_sft_frac": 0.0,
1606
+ "batch_soft_sft_frac": 0.0,
1607
+ "batch_tf_frac": 0.0,
1608
+ "ce_loss": 0.0,
1609
+ "epoch": 1.4309333333333334,
1610
+ "grad_norm": 0.19140625,
1611
+ "kd_loss": 0.1806626110754223,
1612
+ "learning_rate": 3e-06,
1613
+ "loss": 0.1702,
1614
+ "masked_tokens": 120.825,
1615
+ "mean_t": 0.4907285622088239,
1616
+ "step": 670,
1617
+ "student_masked_tokens": 120.825
1618
+ },
1619
+ {
1620
+ "avg_mask_ratio": 0.4662455078505445,
1621
+ "avg_response_length": 244.6125,
1622
+ "avg_student_mask_ratio": 0.4662455078505445,
1623
+ "batch_ainp_frac": 0.0,
1624
+ "batch_inp_frac": 0.0,
1625
+ "batch_inp_oh_frac": 0.0,
1626
+ "batch_inp_par_frac": 1.0,
1627
+ "batch_inp_par_reverse_frac": 0.0,
1628
+ "batch_rl_frac": 0.0,
1629
+ "batch_sft_frac": 0.0,
1630
+ "batch_soft_sft_frac": 0.0,
1631
+ "batch_tf_frac": 0.0,
1632
+ "ce_loss": 0.0,
1633
+ "epoch": 1.4522666666666666,
1634
+ "grad_norm": 0.16796875,
1635
+ "kd_loss": 0.20038694294766798,
1636
+ "learning_rate": 3e-06,
1637
+ "loss": 0.1824,
1638
+ "masked_tokens": 114.2,
1639
+ "mean_t": 0.4913603452499956,
1640
+ "step": 680,
1641
+ "student_masked_tokens": 114.2
1642
+ },
1643
+ {
1644
+ "avg_mask_ratio": 0.4820630593923852,
1645
+ "avg_response_length": 217.3,
1646
+ "avg_student_mask_ratio": 0.4820630593923852,
1647
+ "batch_ainp_frac": 0.0,
1648
+ "batch_inp_frac": 0.0,
1649
+ "batch_inp_oh_frac": 0.0,
1650
+ "batch_inp_par_frac": 1.0,
1651
+ "batch_inp_par_reverse_frac": 0.0,
1652
+ "batch_rl_frac": 0.0,
1653
+ "batch_sft_frac": 0.0,
1654
+ "batch_soft_sft_frac": 0.0,
1655
+ "batch_tf_frac": 0.0,
1656
+ "ce_loss": 0.0,
1657
+ "epoch": 1.4736,
1658
+ "grad_norm": 0.11279296875,
1659
+ "kd_loss": 0.16563009086588637,
1660
+ "learning_rate": 3e-06,
1661
+ "loss": 0.17,
1662
+ "masked_tokens": 118.4875,
1663
+ "mean_t": 0.5133644798654131,
1664
+ "step": 690,
1665
+ "student_masked_tokens": 118.4875
1666
+ },
1667
+ {
1668
+ "avg_mask_ratio": 0.5206489040749147,
1669
+ "avg_response_length": 216.45,
1670
+ "avg_student_mask_ratio": 0.5206489040749147,
1671
+ "batch_ainp_frac": 0.0,
1672
+ "batch_inp_frac": 0.0,
1673
+ "batch_inp_oh_frac": 0.0,
1674
+ "batch_inp_par_frac": 1.0,
1675
+ "batch_inp_par_reverse_frac": 0.0,
1676
+ "batch_rl_frac": 0.0,
1677
+ "batch_sft_frac": 0.0,
1678
+ "batch_soft_sft_frac": 0.0,
1679
+ "batch_tf_frac": 0.0,
1680
+ "ce_loss": 0.0,
1681
+ "epoch": 1.4949333333333334,
1682
+ "grad_norm": 0.236328125,
1683
+ "kd_loss": 0.23649522811845144,
1684
+ "learning_rate": 3e-06,
1685
+ "loss": 0.2241,
1686
+ "masked_tokens": 118.4375,
1687
+ "mean_t": 0.5625698395539075,
1688
+ "step": 700,
1689
+ "student_masked_tokens": 118.4375
1690
+ },
1691
+ {
1692
+ "avg_mask_ratio": 0.4955552547937259,
1693
+ "avg_response_length": 211.175,
1694
+ "avg_student_mask_ratio": 0.4955552547937259,
1695
+ "batch_ainp_frac": 0.0,
1696
+ "batch_inp_frac": 0.0,
1697
+ "batch_inp_oh_frac": 0.0,
1698
+ "batch_inp_par_frac": 1.0,
1699
+ "batch_inp_par_reverse_frac": 0.0,
1700
+ "batch_rl_frac": 0.0,
1701
+ "batch_sft_frac": 0.0,
1702
+ "batch_soft_sft_frac": 0.0,
1703
+ "batch_tf_frac": 0.0,
1704
+ "ce_loss": 0.0,
1705
+ "epoch": 1.5162666666666667,
1706
+ "grad_norm": 0.275390625,
1707
+ "kd_loss": 0.23970817765721222,
1708
+ "learning_rate": 3e-06,
1709
+ "loss": 0.2184,
1710
+ "masked_tokens": 112.3375,
1711
+ "mean_t": 0.5135623761918395,
1712
+ "step": 710,
1713
+ "student_masked_tokens": 112.3375
1714
+ },
1715
+ {
1716
+ "avg_mask_ratio": 0.4820543818641454,
1717
+ "avg_response_length": 229.1375,
1718
+ "avg_student_mask_ratio": 0.4820543818641454,
1719
+ "batch_ainp_frac": 0.0,
1720
+ "batch_inp_frac": 0.0,
1721
+ "batch_inp_oh_frac": 0.0,
1722
+ "batch_inp_par_frac": 1.0,
1723
+ "batch_inp_par_reverse_frac": 0.0,
1724
+ "batch_rl_frac": 0.0,
1725
+ "batch_sft_frac": 0.0,
1726
+ "batch_soft_sft_frac": 0.0,
1727
+ "batch_tf_frac": 0.0,
1728
+ "ce_loss": 0.0,
1729
+ "epoch": 1.5375999999999999,
1730
+ "grad_norm": 0.435546875,
1731
+ "kd_loss": 0.18955910701470202,
1732
+ "learning_rate": 3e-06,
1733
+ "loss": 0.1978,
1734
+ "masked_tokens": 114.55,
1735
+ "mean_t": 0.5117021896177902,
1736
+ "step": 720,
1737
+ "student_masked_tokens": 114.55
1738
+ },
1739
+ {
1740
+ "avg_mask_ratio": 0.4605119539948646,
1741
+ "avg_response_length": 245.5625,
1742
+ "avg_student_mask_ratio": 0.4605119539948646,
1743
+ "batch_ainp_frac": 0.0,
1744
+ "batch_inp_frac": 0.0,
1745
+ "batch_inp_oh_frac": 0.0,
1746
+ "batch_inp_par_frac": 1.0,
1747
+ "batch_inp_par_reverse_frac": 0.0,
1748
+ "batch_rl_frac": 0.0,
1749
+ "batch_sft_frac": 0.0,
1750
+ "batch_soft_sft_frac": 0.0,
1751
+ "batch_tf_frac": 0.0,
1752
+ "ce_loss": 0.0,
1753
+ "epoch": 1.5589333333333333,
1754
+ "grad_norm": 0.1103515625,
1755
+ "kd_loss": 0.16859328244926958,
1756
+ "learning_rate": 3e-06,
1757
+ "loss": 0.1779,
1758
+ "masked_tokens": 113.85,
1759
+ "mean_t": 0.49050743713742123,
1760
+ "step": 730,
1761
+ "student_masked_tokens": 113.85
1762
+ },
1763
+ {
1764
+ "avg_mask_ratio": 0.5134038798511028,
1765
+ "avg_response_length": 194.125,
1766
+ "avg_student_mask_ratio": 0.5134038798511028,
1767
+ "batch_ainp_frac": 0.0,
1768
+ "batch_inp_frac": 0.0,
1769
+ "batch_inp_oh_frac": 0.0,
1770
+ "batch_inp_par_frac": 1.0,
1771
+ "batch_inp_par_reverse_frac": 0.0,
1772
+ "batch_rl_frac": 0.0,
1773
+ "batch_sft_frac": 0.0,
1774
+ "batch_soft_sft_frac": 0.0,
1775
+ "batch_tf_frac": 0.0,
1776
+ "ce_loss": 0.0,
1777
+ "epoch": 1.5802666666666667,
1778
+ "grad_norm": 0.0830078125,
1779
+ "kd_loss": 0.17122714665274544,
1780
+ "learning_rate": 3e-06,
1781
+ "loss": 0.2018,
1782
+ "masked_tokens": 102.9375,
1783
+ "mean_t": 0.5505168779753149,
1784
+ "step": 740,
1785
+ "student_masked_tokens": 102.9375
1786
+ },
1787
+ {
1788
+ "avg_mask_ratio": 0.4201362137740944,
1789
+ "avg_response_length": 223.55,
1790
+ "avg_student_mask_ratio": 0.4201362137740944,
1791
+ "batch_ainp_frac": 0.0,
1792
+ "batch_inp_frac": 0.0,
1793
+ "batch_inp_oh_frac": 0.0,
1794
+ "batch_inp_par_frac": 1.0,
1795
+ "batch_inp_par_reverse_frac": 0.0,
1796
+ "batch_rl_frac": 0.0,
1797
+ "batch_sft_frac": 0.0,
1798
+ "batch_soft_sft_frac": 0.0,
1799
+ "batch_tf_frac": 0.0,
1800
+ "ce_loss": 0.0,
1801
+ "epoch": 1.6016,
1802
+ "grad_norm": 0.16796875,
1803
+ "kd_loss": 0.19197621339357057,
1804
+ "learning_rate": 3e-06,
1805
+ "loss": 0.1792,
1806
+ "masked_tokens": 94.7125,
1807
+ "mean_t": 0.4437690361432033,
1808
+ "step": 750,
1809
+ "student_masked_tokens": 94.7125
1810
+ },
1811
+ {
1812
+ "avg_mask_ratio": 0.46924527404480615,
1813
+ "avg_response_length": 227.35,
1814
+ "avg_student_mask_ratio": 0.46924527404480615,
1815
+ "batch_ainp_frac": 0.0,
1816
+ "batch_inp_frac": 0.0,
1817
+ "batch_inp_oh_frac": 0.0,
1818
+ "batch_inp_par_frac": 1.0,
1819
+ "batch_inp_par_reverse_frac": 0.0,
1820
+ "batch_rl_frac": 0.0,
1821
+ "batch_sft_frac": 0.0,
1822
+ "batch_soft_sft_frac": 0.0,
1823
+ "batch_tf_frac": 0.0,
1824
+ "ce_loss": 0.0,
1825
+ "epoch": 1.6229333333333333,
1826
+ "grad_norm": 0.8125,
1827
+ "kd_loss": 0.24938117066874382,
1828
+ "learning_rate": 3e-06,
1829
+ "loss": 0.2591,
1830
+ "masked_tokens": 107.5375,
1831
+ "mean_t": 0.49132869170280175,
1832
+ "step": 760,
1833
+ "student_masked_tokens": 107.5375
1834
+ },
1835
+ {
1836
+ "avg_mask_ratio": 0.445551612455165,
1837
+ "avg_response_length": 268.6625,
1838
+ "avg_student_mask_ratio": 0.445551612455165,
1839
+ "batch_ainp_frac": 0.0,
1840
+ "batch_inp_frac": 0.0,
1841
+ "batch_inp_oh_frac": 0.0,
1842
+ "batch_inp_par_frac": 1.0,
1843
+ "batch_inp_par_reverse_frac": 0.0,
1844
+ "batch_rl_frac": 0.0,
1845
+ "batch_sft_frac": 0.0,
1846
+ "batch_soft_sft_frac": 0.0,
1847
+ "batch_tf_frac": 0.0,
1848
+ "ce_loss": 0.0,
1849
+ "epoch": 1.6442666666666668,
1850
+ "grad_norm": 0.1435546875,
1851
+ "kd_loss": 0.1928954417056957,
1852
+ "learning_rate": 3e-06,
1853
+ "loss": 0.1563,
1854
+ "masked_tokens": 124.0875,
1855
+ "mean_t": 0.4667695587326307,
1856
+ "step": 770,
1857
+ "student_masked_tokens": 124.0875
1858
+ },
1859
+ {
1860
+ "avg_mask_ratio": 0.4321410794305848,
1861
+ "avg_response_length": 256.1625,
1862
+ "avg_student_mask_ratio": 0.4321410794305848,
1863
+ "batch_ainp_frac": 0.0,
1864
+ "batch_inp_frac": 0.0,
1865
+ "batch_inp_oh_frac": 0.0,
1866
+ "batch_inp_par_frac": 1.0,
1867
+ "batch_inp_par_reverse_frac": 0.0,
1868
+ "batch_rl_frac": 0.0,
1869
+ "batch_sft_frac": 0.0,
1870
+ "batch_soft_sft_frac": 0.0,
1871
+ "batch_tf_frac": 0.0,
1872
+ "ce_loss": 0.0,
1873
+ "epoch": 1.6656,
1874
+ "grad_norm": 0.2216796875,
1875
+ "kd_loss": 0.15059620087446887,
1876
+ "learning_rate": 3e-06,
1877
+ "loss": 0.1534,
1878
+ "masked_tokens": 117.7,
1879
+ "mean_t": 0.44942845597106496,
1880
+ "step": 780,
1881
+ "student_masked_tokens": 117.7
1882
+ },
1883
+ {
1884
+ "avg_mask_ratio": 0.4697655299096368,
1885
+ "avg_response_length": 240.6125,
1886
+ "avg_student_mask_ratio": 0.4697655299096368,
1887
+ "batch_ainp_frac": 0.0,
1888
+ "batch_inp_frac": 0.0,
1889
+ "batch_inp_oh_frac": 0.0,
1890
+ "batch_inp_par_frac": 1.0,
1891
+ "batch_inp_par_reverse_frac": 0.0,
1892
+ "batch_rl_frac": 0.0,
1893
+ "batch_sft_frac": 0.0,
1894
+ "batch_soft_sft_frac": 0.0,
1895
+ "batch_tf_frac": 0.0,
1896
+ "ce_loss": 0.0,
1897
+ "epoch": 1.6869333333333332,
1898
+ "grad_norm": 0.255859375,
1899
+ "kd_loss": 0.16427693545232777,
1900
+ "learning_rate": 3e-06,
1901
+ "loss": 0.1489,
1902
+ "masked_tokens": 113.5375,
1903
+ "mean_t": 0.4993515375303105,
1904
+ "step": 790,
1905
+ "student_masked_tokens": 113.5375
1906
+ },
1907
+ {
1908
+ "avg_mask_ratio": 0.4635992758907378,
1909
+ "avg_response_length": 224.15,
1910
+ "avg_student_mask_ratio": 0.4635992758907378,
1911
+ "batch_ainp_frac": 0.0,
1912
+ "batch_inp_frac": 0.0,
1913
+ "batch_inp_oh_frac": 0.0,
1914
+ "batch_inp_par_frac": 1.0,
1915
+ "batch_inp_par_reverse_frac": 0.0,
1916
+ "batch_rl_frac": 0.0,
1917
+ "batch_sft_frac": 0.0,
1918
+ "batch_soft_sft_frac": 0.0,
1919
+ "batch_tf_frac": 0.0,
1920
+ "ce_loss": 0.0,
1921
+ "epoch": 1.7082666666666668,
1922
+ "grad_norm": 0.197265625,
1923
+ "kd_loss": 0.15171801659575976,
1924
+ "learning_rate": 3e-06,
1925
+ "loss": 0.1526,
1926
+ "masked_tokens": 107.5,
1927
+ "mean_t": 0.500370389316231,
1928
+ "step": 800,
1929
+ "student_masked_tokens": 107.5
1930
+ },
1931
+ {
1932
+ "avg_mask_ratio": 0.5018501321552321,
1933
+ "avg_response_length": 235.1625,
1934
+ "avg_student_mask_ratio": 0.5018501321552321,
1935
+ "batch_ainp_frac": 0.0,
1936
+ "batch_inp_frac": 0.0,
1937
+ "batch_inp_oh_frac": 0.0,
1938
+ "batch_inp_par_frac": 1.0,
1939
+ "batch_inp_par_reverse_frac": 0.0,
1940
+ "batch_rl_frac": 0.0,
1941
+ "batch_sft_frac": 0.0,
1942
+ "batch_soft_sft_frac": 0.0,
1943
+ "batch_tf_frac": 0.0,
1944
+ "ce_loss": 0.0,
1945
+ "epoch": 1.7296,
1946
+ "grad_norm": 0.1630859375,
1947
+ "kd_loss": 0.18931926304685476,
1948
+ "learning_rate": 3e-06,
1949
+ "loss": 0.2031,
1950
+ "masked_tokens": 116.6125,
1951
+ "mean_t": 0.5317009104182944,
1952
+ "step": 810,
1953
+ "student_masked_tokens": 116.6125
1954
+ },
1955
+ {
1956
+ "avg_mask_ratio": 0.5050785383209586,
1957
+ "avg_response_length": 207.6875,
1958
+ "avg_student_mask_ratio": 0.5050785383209586,
1959
+ "batch_ainp_frac": 0.0,
1960
+ "batch_inp_frac": 0.0,
1961
+ "batch_inp_oh_frac": 0.0,
1962
+ "batch_inp_par_frac": 1.0,
1963
+ "batch_inp_par_reverse_frac": 0.0,
1964
+ "batch_rl_frac": 0.0,
1965
+ "batch_sft_frac": 0.0,
1966
+ "batch_soft_sft_frac": 0.0,
1967
+ "batch_tf_frac": 0.0,
1968
+ "ce_loss": 0.0,
1969
+ "epoch": 1.7509333333333332,
1970
+ "grad_norm": 0.5546875,
1971
+ "kd_loss": 0.24824176937403308,
1972
+ "learning_rate": 3e-06,
1973
+ "loss": 0.2566,
1974
+ "masked_tokens": 119.175,
1975
+ "mean_t": 0.5392061032878701,
1976
+ "step": 820,
1977
+ "student_masked_tokens": 119.175
1978
+ },
1979
+ {
1980
+ "avg_mask_ratio": 0.4980328972451389,
1981
+ "avg_response_length": 270.325,
1982
+ "avg_student_mask_ratio": 0.4980328972451389,
1983
+ "batch_ainp_frac": 0.0,
1984
+ "batch_inp_frac": 0.0,
1985
+ "batch_inp_oh_frac": 0.0,
1986
+ "batch_inp_par_frac": 1.0,
1987
+ "batch_inp_par_reverse_frac": 0.0,
1988
+ "batch_rl_frac": 0.0,
1989
+ "batch_sft_frac": 0.0,
1990
+ "batch_soft_sft_frac": 0.0,
1991
+ "batch_tf_frac": 0.0,
1992
+ "ce_loss": 0.0,
1993
+ "epoch": 1.7722666666666667,
1994
+ "grad_norm": 0.09912109375,
1995
+ "kd_loss": 0.1924194690429431,
1996
+ "learning_rate": 3e-06,
1997
+ "loss": 0.2006,
1998
+ "masked_tokens": 141.925,
1999
+ "mean_t": 0.5238314627087675,
2000
+ "step": 830,
2001
+ "student_masked_tokens": 141.925
2002
+ },
2003
+ {
2004
+ "avg_mask_ratio": 0.493249478796497,
2005
+ "avg_response_length": 226.025,
2006
+ "avg_student_mask_ratio": 0.493249478796497,
2007
+ "batch_ainp_frac": 0.0,
2008
+ "batch_inp_frac": 0.0,
2009
+ "batch_inp_oh_frac": 0.0,
2010
+ "batch_inp_par_frac": 1.0,
2011
+ "batch_inp_par_reverse_frac": 0.0,
2012
+ "batch_rl_frac": 0.0,
2013
+ "batch_sft_frac": 0.0,
2014
+ "batch_soft_sft_frac": 0.0,
2015
+ "batch_tf_frac": 0.0,
2016
+ "ce_loss": 0.0,
2017
+ "epoch": 1.7936,
2018
+ "grad_norm": 0.158203125,
2019
+ "kd_loss": 0.15751813794203606,
2020
+ "learning_rate": 3e-06,
2021
+ "loss": 0.1624,
2022
+ "masked_tokens": 118.825,
2023
+ "mean_t": 0.5301066277665086,
2024
+ "step": 840,
2025
+ "student_masked_tokens": 118.825
2026
+ },
2027
+ {
2028
+ "avg_mask_ratio": 0.5009000841644593,
2029
+ "avg_response_length": 233.025,
2030
+ "avg_student_mask_ratio": 0.5009000841644593,
2031
+ "batch_ainp_frac": 0.0,
2032
+ "batch_inp_frac": 0.0,
2033
+ "batch_inp_oh_frac": 0.0,
2034
+ "batch_inp_par_frac": 1.0,
2035
+ "batch_inp_par_reverse_frac": 0.0,
2036
+ "batch_rl_frac": 0.0,
2037
+ "batch_sft_frac": 0.0,
2038
+ "batch_soft_sft_frac": 0.0,
2039
+ "batch_tf_frac": 0.0,
2040
+ "ce_loss": 0.0,
2041
+ "epoch": 1.8149333333333333,
2042
+ "grad_norm": 0.10205078125,
2043
+ "kd_loss": 0.1860738446495816,
2044
+ "learning_rate": 3e-06,
2045
+ "loss": 0.1737,
2046
+ "masked_tokens": 117.3125,
2047
+ "mean_t": 0.5343429344706238,
2048
+ "step": 850,
2049
+ "student_masked_tokens": 117.3125
2050
+ },
2051
+ {
2052
+ "avg_mask_ratio": 0.46293387678451836,
2053
+ "avg_response_length": 231.3625,
2054
+ "avg_student_mask_ratio": 0.46293387678451836,
2055
+ "batch_ainp_frac": 0.0,
2056
+ "batch_inp_frac": 0.0,
2057
+ "batch_inp_oh_frac": 0.0,
2058
+ "batch_inp_par_frac": 1.0,
2059
+ "batch_inp_par_reverse_frac": 0.0,
2060
+ "batch_rl_frac": 0.0,
2061
+ "batch_sft_frac": 0.0,
2062
+ "batch_soft_sft_frac": 0.0,
2063
+ "batch_tf_frac": 0.0,
2064
+ "ce_loss": 0.0,
2065
+ "epoch": 1.8362666666666667,
2066
+ "grad_norm": 0.1103515625,
2067
+ "kd_loss": 0.19740513321539765,
2068
+ "learning_rate": 3e-06,
2069
+ "loss": 0.1841,
2070
+ "masked_tokens": 110.5625,
2071
+ "mean_t": 0.4791536889737472,
2072
+ "step": 860,
2073
+ "student_masked_tokens": 110.5625
2074
+ },
2075
+ {
2076
+ "avg_mask_ratio": 0.4846805231412873,
2077
+ "avg_response_length": 220.7375,
2078
+ "avg_student_mask_ratio": 0.4846805231412873,
2079
+ "batch_ainp_frac": 0.0,
2080
+ "batch_inp_frac": 0.0,
2081
+ "batch_inp_oh_frac": 0.0,
2082
+ "batch_inp_par_frac": 1.0,
2083
+ "batch_inp_par_reverse_frac": 0.0,
2084
+ "batch_rl_frac": 0.0,
2085
+ "batch_sft_frac": 0.0,
2086
+ "batch_soft_sft_frac": 0.0,
2087
+ "batch_tf_frac": 0.0,
2088
+ "ce_loss": 0.0,
2089
+ "epoch": 1.8576000000000001,
2090
+ "grad_norm": 0.228515625,
2091
+ "kd_loss": 0.19436422403705364,
2092
+ "learning_rate": 3e-06,
2093
+ "loss": 0.2012,
2094
+ "masked_tokens": 113.7625,
2095
+ "mean_t": 0.5203817339061061,
2096
+ "step": 870,
2097
+ "student_masked_tokens": 113.7625
2098
+ },
2099
+ {
2100
+ "avg_mask_ratio": 0.4508363194297999,
2101
+ "avg_response_length": 203.2875,
2102
+ "avg_student_mask_ratio": 0.4508363194297999,
2103
+ "batch_ainp_frac": 0.0,
2104
+ "batch_inp_frac": 0.0,
2105
+ "batch_inp_oh_frac": 0.0,
2106
+ "batch_inp_par_frac": 1.0,
2107
+ "batch_inp_par_reverse_frac": 0.0,
2108
+ "batch_rl_frac": 0.0,
2109
+ "batch_sft_frac": 0.0,
2110
+ "batch_soft_sft_frac": 0.0,
2111
+ "batch_tf_frac": 0.0,
2112
+ "ce_loss": 0.0,
2113
+ "epoch": 1.8789333333333333,
2114
+ "grad_norm": 0.1962890625,
2115
+ "kd_loss": 0.16288868402702406,
2116
+ "learning_rate": 3e-06,
2117
+ "loss": 0.1845,
2118
+ "masked_tokens": 95.3875,
2119
+ "mean_t": 0.4875184997683391,
2120
+ "step": 880,
2121
+ "student_masked_tokens": 95.3875
2122
+ },
2123
+ {
2124
+ "avg_mask_ratio": 0.43862658384023234,
2125
+ "avg_response_length": 227.9375,
2126
+ "avg_student_mask_ratio": 0.43862658384023234,
2127
+ "batch_ainp_frac": 0.0,
2128
+ "batch_inp_frac": 0.0,
2129
+ "batch_inp_oh_frac": 0.0,
2130
+ "batch_inp_par_frac": 1.0,
2131
+ "batch_inp_par_reverse_frac": 0.0,
2132
+ "batch_rl_frac": 0.0,
2133
+ "batch_sft_frac": 0.0,
2134
+ "batch_soft_sft_frac": 0.0,
2135
+ "batch_tf_frac": 0.0,
2136
+ "ce_loss": 0.0,
2137
+ "epoch": 1.9002666666666665,
2138
+ "grad_norm": 0.08251953125,
2139
+ "kd_loss": 0.11281866748422545,
2140
+ "learning_rate": 3e-06,
2141
+ "loss": 0.142,
2142
+ "masked_tokens": 101.2625,
2143
+ "mean_t": 0.4766692223958671,
2144
+ "step": 890,
2145
+ "student_masked_tokens": 101.2625
2146
+ },
2147
+ {
2148
+ "avg_mask_ratio": 0.44909207145101393,
2149
+ "avg_response_length": 237.2375,
2150
+ "avg_student_mask_ratio": 0.44909207145101393,
2151
+ "batch_ainp_frac": 0.0,
2152
+ "batch_inp_frac": 0.0,
2153
+ "batch_inp_oh_frac": 0.0,
2154
+ "batch_inp_par_frac": 1.0,
2155
+ "batch_inp_par_reverse_frac": 0.0,
2156
+ "batch_rl_frac": 0.0,
2157
+ "batch_sft_frac": 0.0,
2158
+ "batch_soft_sft_frac": 0.0,
2159
+ "batch_tf_frac": 0.0,
2160
+ "ce_loss": 0.0,
2161
+ "epoch": 1.9216,
2162
+ "grad_norm": 0.1591796875,
2163
+ "kd_loss": 0.15684176656744747,
2164
+ "learning_rate": 3e-06,
2165
+ "loss": 0.1737,
2166
+ "masked_tokens": 103.7,
2167
+ "mean_t": 0.487134758150205,
2168
+ "step": 900,
2169
+ "student_masked_tokens": 103.7
2170
+ },
2171
+ {
2172
+ "avg_mask_ratio": 0.47512493645772336,
2173
+ "avg_response_length": 246.175,
2174
+ "avg_student_mask_ratio": 0.47512493645772336,
2175
+ "batch_ainp_frac": 0.0,
2176
+ "batch_inp_frac": 0.0,
2177
+ "batch_inp_oh_frac": 0.0,
2178
+ "batch_inp_par_frac": 1.0,
2179
+ "batch_inp_par_reverse_frac": 0.0,
2180
+ "batch_rl_frac": 0.0,
2181
+ "batch_sft_frac": 0.0,
2182
+ "batch_soft_sft_frac": 0.0,
2183
+ "batch_tf_frac": 0.0,
2184
+ "ce_loss": 0.0,
2185
+ "epoch": 1.9429333333333334,
2186
+ "grad_norm": 0.1982421875,
2187
+ "kd_loss": 0.16909069961529893,
2188
+ "learning_rate": 3e-06,
2189
+ "loss": 0.1717,
2190
+ "masked_tokens": 117.2125,
2191
+ "mean_t": 0.5027793228859082,
2192
+ "step": 910,
2193
+ "student_masked_tokens": 117.2125
2194
+ },
2195
+ {
2196
+ "avg_mask_ratio": 0.46720519906957636,
2197
+ "avg_response_length": 229.0875,
2198
+ "avg_student_mask_ratio": 0.46720519906957636,
2199
+ "batch_ainp_frac": 0.0,
2200
+ "batch_inp_frac": 0.0,
2201
+ "batch_inp_oh_frac": 0.0,
2202
+ "batch_inp_par_frac": 1.0,
2203
+ "batch_inp_par_reverse_frac": 0.0,
2204
+ "batch_rl_frac": 0.0,
2205
+ "batch_sft_frac": 0.0,
2206
+ "batch_soft_sft_frac": 0.0,
2207
+ "batch_tf_frac": 0.0,
2208
+ "ce_loss": 0.0,
2209
+ "epoch": 1.9642666666666666,
2210
+ "grad_norm": 0.203125,
2211
+ "kd_loss": 0.17078794931742322,
2212
+ "learning_rate": 3e-06,
2213
+ "loss": 0.1844,
2214
+ "masked_tokens": 115.325,
2215
+ "mean_t": 0.49417946098838,
2216
+ "step": 920,
2217
+ "student_masked_tokens": 115.325
2218
+ },
2219
+ {
2220
+ "avg_mask_ratio": 0.4805813999380916,
2221
+ "avg_response_length": 236.1625,
2222
+ "avg_student_mask_ratio": 0.4805813999380916,
2223
+ "batch_ainp_frac": 0.0,
2224
+ "batch_inp_frac": 0.0,
2225
+ "batch_inp_oh_frac": 0.0,
2226
+ "batch_inp_par_frac": 1.0,
2227
+ "batch_inp_par_reverse_frac": 0.0,
2228
+ "batch_rl_frac": 0.0,
2229
+ "batch_sft_frac": 0.0,
2230
+ "batch_soft_sft_frac": 0.0,
2231
+ "batch_tf_frac": 0.0,
2232
+ "ce_loss": 0.0,
2233
+ "epoch": 1.9856,
2234
+ "grad_norm": 0.21484375,
2235
+ "kd_loss": 0.20637793739751942,
2236
+ "learning_rate": 3e-06,
2237
+ "loss": 0.176,
2238
+ "masked_tokens": 119.65,
2239
+ "mean_t": 0.5045580042526125,
2240
+ "step": 930,
2241
+ "student_masked_tokens": 119.65
2242
+ },
2243
+ {
2244
+ "avg_mask_ratio": 0.49833715503059683,
2245
+ "avg_response_length": 229.39285714285714,
2246
+ "avg_student_mask_ratio": 0.49833715503059683,
2247
+ "batch_ainp_frac": 0.0,
2248
+ "batch_inp_frac": 0.0,
2249
+ "batch_inp_oh_frac": 0.0,
2250
+ "batch_inp_par_frac": 1.0,
2251
+ "batch_inp_par_reverse_frac": 0.0,
2252
+ "batch_rl_frac": 0.0,
2253
+ "batch_sft_frac": 0.0,
2254
+ "batch_soft_sft_frac": 0.0,
2255
+ "batch_tf_frac": 0.0,
2256
+ "ce_loss": 0.0,
2257
+ "epoch": 2.0085333333333333,
2258
+ "grad_norm": 0.259765625,
2259
+ "kd_loss": 0.2117233988543549,
2260
+ "learning_rate": 3e-06,
2261
+ "loss": 0.2023,
2262
+ "masked_tokens": 121.5952380952381,
2263
+ "mean_t": 0.5321138524893849,
2264
+ "step": 940,
2265
+ "student_masked_tokens": 121.5952380952381
2266
+ },
2267
+ {
2268
+ "avg_mask_ratio": 0.4347397161007393,
2269
+ "avg_response_length": 226.925,
2270
+ "avg_student_mask_ratio": 0.4347397161007393,
2271
+ "batch_ainp_frac": 0.0,
2272
+ "batch_inp_frac": 0.0,
2273
+ "batch_inp_oh_frac": 0.0,
2274
+ "batch_inp_par_frac": 1.0,
2275
+ "batch_inp_par_reverse_frac": 0.0,
2276
+ "batch_rl_frac": 0.0,
2277
+ "batch_sft_frac": 0.0,
2278
+ "batch_soft_sft_frac": 0.0,
2279
+ "batch_tf_frac": 0.0,
2280
+ "ce_loss": 0.0,
2281
+ "epoch": 2.0298666666666665,
2282
+ "grad_norm": 0.15234375,
2283
+ "kd_loss": 0.16830119832033005,
2284
+ "learning_rate": 3e-06,
2285
+ "loss": 0.1549,
2286
+ "masked_tokens": 103.4125,
2287
+ "mean_t": 0.4632946296595037,
2288
+ "step": 950,
2289
+ "student_masked_tokens": 103.4125
2290
+ },
2291
+ {
2292
+ "avg_mask_ratio": 0.4989688004134223,
2293
+ "avg_response_length": 273.3125,
2294
+ "avg_student_mask_ratio": 0.4989688004134223,
2295
+ "batch_ainp_frac": 0.0,
2296
+ "batch_inp_frac": 0.0,
2297
+ "batch_inp_oh_frac": 0.0,
2298
+ "batch_inp_par_frac": 1.0,
2299
+ "batch_inp_par_reverse_frac": 0.0,
2300
+ "batch_rl_frac": 0.0,
2301
+ "batch_sft_frac": 0.0,
2302
+ "batch_soft_sft_frac": 0.0,
2303
+ "batch_tf_frac": 0.0,
2304
+ "ce_loss": 0.0,
2305
+ "epoch": 2.0512,
2306
+ "grad_norm": 0.123046875,
2307
+ "kd_loss": 0.2304358719712809,
2308
+ "learning_rate": 3e-06,
2309
+ "loss": 0.22,
2310
+ "masked_tokens": 137.125,
2311
+ "mean_t": 0.5198000721400604,
2312
+ "step": 960,
2313
+ "student_masked_tokens": 137.125
2314
+ },
2315
+ {
2316
+ "avg_mask_ratio": 0.4374056361732073,
2317
+ "avg_response_length": 239.125,
2318
+ "avg_student_mask_ratio": 0.4374056361732073,
2319
+ "batch_ainp_frac": 0.0,
2320
+ "batch_inp_frac": 0.0,
2321
+ "batch_inp_oh_frac": 0.0,
2322
+ "batch_inp_par_frac": 1.0,
2323
+ "batch_inp_par_reverse_frac": 0.0,
2324
+ "batch_rl_frac": 0.0,
2325
+ "batch_sft_frac": 0.0,
2326
+ "batch_soft_sft_frac": 0.0,
2327
+ "batch_tf_frac": 0.0,
2328
+ "ce_loss": 0.0,
2329
+ "epoch": 2.0725333333333333,
2330
+ "grad_norm": 0.1962890625,
2331
+ "kd_loss": 0.158095879086882,
2332
+ "learning_rate": 3e-06,
2333
+ "loss": 0.1573,
2334
+ "masked_tokens": 107.4625,
2335
+ "mean_t": 0.4703940597362816,
2336
+ "step": 970,
2337
+ "student_masked_tokens": 107.4625
2338
+ },
2339
+ {
2340
+ "avg_mask_ratio": 0.4216216350672767,
2341
+ "avg_response_length": 236.0125,
2342
+ "avg_student_mask_ratio": 0.4216216350672767,
2343
+ "batch_ainp_frac": 0.0,
2344
+ "batch_inp_frac": 0.0,
2345
+ "batch_inp_oh_frac": 0.0,
2346
+ "batch_inp_par_frac": 1.0,
2347
+ "batch_inp_par_reverse_frac": 0.0,
2348
+ "batch_rl_frac": 0.0,
2349
+ "batch_sft_frac": 0.0,
2350
+ "batch_soft_sft_frac": 0.0,
2351
+ "batch_tf_frac": 0.0,
2352
+ "ce_loss": 0.0,
2353
+ "epoch": 2.0938666666666665,
2354
+ "grad_norm": 0.177734375,
2355
+ "kd_loss": 0.15213593625503563,
2356
+ "learning_rate": 3e-06,
2357
+ "loss": 0.1305,
2358
+ "masked_tokens": 103.05,
2359
+ "mean_t": 0.4511947895749472,
2360
+ "step": 980,
2361
+ "student_masked_tokens": 103.05
2362
+ },
2363
+ {
2364
+ "avg_mask_ratio": 0.46076959141064433,
2365
+ "avg_response_length": 257.375,
2366
+ "avg_student_mask_ratio": 0.46076959141064433,
2367
+ "batch_ainp_frac": 0.0,
2368
+ "batch_inp_frac": 0.0,
2369
+ "batch_inp_oh_frac": 0.0,
2370
+ "batch_inp_par_frac": 1.0,
2371
+ "batch_inp_par_reverse_frac": 0.0,
2372
+ "batch_rl_frac": 0.0,
2373
+ "batch_sft_frac": 0.0,
2374
+ "batch_soft_sft_frac": 0.0,
2375
+ "batch_tf_frac": 0.0,
2376
+ "ce_loss": 0.0,
2377
+ "epoch": 2.1152,
2378
+ "grad_norm": 0.1484375,
2379
+ "kd_loss": 0.18778659292394478,
2380
+ "learning_rate": 3e-06,
2381
+ "loss": 0.1727,
2382
+ "masked_tokens": 121.2125,
2383
+ "mean_t": 0.4923786667350214,
2384
+ "step": 990,
2385
+ "student_masked_tokens": 121.2125
2386
+ },
2387
+ {
2388
+ "avg_mask_ratio": 0.45376153094694016,
2389
+ "avg_response_length": 223.4125,
2390
+ "avg_student_mask_ratio": 0.45376153094694016,
2391
+ "batch_ainp_frac": 0.0,
2392
+ "batch_inp_frac": 0.0,
2393
+ "batch_inp_oh_frac": 0.0,
2394
+ "batch_inp_par_frac": 1.0,
2395
+ "batch_inp_par_reverse_frac": 0.0,
2396
+ "batch_rl_frac": 0.0,
2397
+ "batch_sft_frac": 0.0,
2398
+ "batch_soft_sft_frac": 0.0,
2399
+ "batch_tf_frac": 0.0,
2400
+ "ce_loss": 0.0,
2401
+ "epoch": 2.1365333333333334,
2402
+ "grad_norm": 0.10498046875,
2403
+ "kd_loss": 0.16195435594947866,
2404
+ "learning_rate": 3e-06,
2405
+ "loss": 0.1554,
2406
+ "masked_tokens": 102.175,
2407
+ "mean_t": 0.4773523230338469,
2408
+ "step": 1000,
2409
+ "student_masked_tokens": 102.175
2410
+ },
2411
+ {
2412
+ "avg_mask_ratio": 0.4360893844394013,
2413
+ "avg_response_length": 230.8375,
2414
+ "avg_student_mask_ratio": 0.4360893844394013,
2415
+ "batch_ainp_frac": 0.0,
2416
+ "batch_inp_frac": 0.0,
2417
+ "batch_inp_oh_frac": 0.0,
2418
+ "batch_inp_par_frac": 1.0,
2419
+ "batch_inp_par_reverse_frac": 0.0,
2420
+ "batch_rl_frac": 0.0,
2421
+ "batch_sft_frac": 0.0,
2422
+ "batch_soft_sft_frac": 0.0,
2423
+ "batch_tf_frac": 0.0,
2424
+ "ce_loss": 0.0,
2425
+ "epoch": 2.1578666666666666,
2426
+ "grad_norm": 0.09130859375,
2427
+ "kd_loss": 0.15129900982686878,
2428
+ "learning_rate": 3e-06,
2429
+ "loss": 0.1525,
2430
+ "masked_tokens": 95.2875,
2431
+ "mean_t": 0.4648138735938119,
2432
+ "step": 1010,
2433
+ "student_masked_tokens": 95.2875
2434
+ },
2435
+ {
2436
+ "avg_mask_ratio": 0.5085378198185936,
2437
+ "avg_response_length": 228.975,
2438
+ "avg_student_mask_ratio": 0.5085378198185936,
2439
+ "batch_ainp_frac": 0.0,
2440
+ "batch_inp_frac": 0.0,
2441
+ "batch_inp_oh_frac": 0.0,
2442
+ "batch_inp_par_frac": 1.0,
2443
+ "batch_inp_par_reverse_frac": 0.0,
2444
+ "batch_rl_frac": 0.0,
2445
+ "batch_sft_frac": 0.0,
2446
+ "batch_soft_sft_frac": 0.0,
2447
+ "batch_tf_frac": 0.0,
2448
+ "ce_loss": 0.0,
2449
+ "epoch": 2.1792,
2450
+ "grad_norm": 0.1279296875,
2451
+ "kd_loss": 0.1835148020709127,
2452
+ "learning_rate": 3e-06,
2453
+ "loss": 0.2059,
2454
+ "masked_tokens": 118.825,
2455
+ "mean_t": 0.5327763411332853,
2456
+ "step": 1020,
2457
+ "student_masked_tokens": 118.825
2458
+ },
2459
+ {
2460
+ "avg_mask_ratio": 0.4720251938910224,
2461
+ "avg_response_length": 237.7125,
2462
+ "avg_student_mask_ratio": 0.4720251938910224,
2463
+ "batch_ainp_frac": 0.0,
2464
+ "batch_inp_frac": 0.0,
2465
+ "batch_inp_oh_frac": 0.0,
2466
+ "batch_inp_par_frac": 1.0,
2467
+ "batch_inp_par_reverse_frac": 0.0,
2468
+ "batch_rl_frac": 0.0,
2469
+ "batch_sft_frac": 0.0,
2470
+ "batch_soft_sft_frac": 0.0,
2471
+ "batch_tf_frac": 0.0,
2472
+ "ce_loss": 0.0,
2473
+ "epoch": 2.2005333333333335,
2474
+ "grad_norm": 0.27734375,
2475
+ "kd_loss": 0.19297441139430588,
2476
+ "learning_rate": 3e-06,
2477
+ "loss": 0.195,
2478
+ "masked_tokens": 113.1,
2479
+ "mean_t": 0.5033508580760099,
2480
+ "step": 1030,
2481
+ "student_masked_tokens": 113.1
2482
+ },
2483
+ {
2484
+ "avg_mask_ratio": 0.4877920758561231,
2485
+ "avg_response_length": 206.2875,
2486
+ "avg_student_mask_ratio": 0.4877920758561231,
2487
+ "batch_ainp_frac": 0.0,
2488
+ "batch_inp_frac": 0.0,
2489
+ "batch_inp_oh_frac": 0.0,
2490
+ "batch_inp_par_frac": 1.0,
2491
+ "batch_inp_par_reverse_frac": 0.0,
2492
+ "batch_rl_frac": 0.0,
2493
+ "batch_sft_frac": 0.0,
2494
+ "batch_soft_sft_frac": 0.0,
2495
+ "batch_tf_frac": 0.0,
2496
+ "ce_loss": 0.0,
2497
+ "epoch": 2.2218666666666667,
2498
+ "grad_norm": 0.1865234375,
2499
+ "kd_loss": 0.18438395577541086,
2500
+ "learning_rate": 3e-06,
2501
+ "loss": 0.1806,
2502
+ "masked_tokens": 99.8625,
2503
+ "mean_t": 0.5349024560535327,
2504
+ "step": 1040,
2505
+ "student_masked_tokens": 99.8625
2506
+ },
2507
+ {
2508
+ "avg_mask_ratio": 0.5081706921104342,
2509
+ "avg_response_length": 237.05,
2510
+ "avg_student_mask_ratio": 0.5081706921104342,
2511
+ "batch_ainp_frac": 0.0,
2512
+ "batch_inp_frac": 0.0,
2513
+ "batch_inp_oh_frac": 0.0,
2514
+ "batch_inp_par_frac": 1.0,
2515
+ "batch_inp_par_reverse_frac": 0.0,
2516
+ "batch_rl_frac": 0.0,
2517
+ "batch_sft_frac": 0.0,
2518
+ "batch_soft_sft_frac": 0.0,
2519
+ "batch_tf_frac": 0.0,
2520
+ "ce_loss": 0.0,
2521
+ "epoch": 2.2432,
2522
+ "grad_norm": 0.158203125,
2523
+ "kd_loss": 0.1950792422781717,
2524
+ "learning_rate": 3e-06,
2525
+ "loss": 0.1941,
2526
+ "masked_tokens": 122.7375,
2527
+ "mean_t": 0.5457118917722255,
2528
+ "step": 1050,
2529
+ "student_masked_tokens": 122.7375
2530
+ },
2531
+ {
2532
+ "avg_mask_ratio": 0.4656533743953332,
2533
+ "avg_response_length": 259.25,
2534
+ "avg_student_mask_ratio": 0.4656533743953332,
2535
+ "batch_ainp_frac": 0.0,
2536
+ "batch_inp_frac": 0.0,
2537
+ "batch_inp_oh_frac": 0.0,
2538
+ "batch_inp_par_frac": 1.0,
2539
+ "batch_inp_par_reverse_frac": 0.0,
2540
+ "batch_rl_frac": 0.0,
2541
+ "batch_sft_frac": 0.0,
2542
+ "batch_soft_sft_frac": 0.0,
2543
+ "batch_tf_frac": 0.0,
2544
+ "ce_loss": 0.0,
2545
+ "epoch": 2.2645333333333335,
2546
+ "grad_norm": 0.1552734375,
2547
+ "kd_loss": 0.14457174434564876,
2548
+ "learning_rate": 3e-06,
2549
+ "loss": 0.1245,
2550
+ "masked_tokens": 124.0875,
2551
+ "mean_t": 0.48194136443780733,
2552
+ "step": 1060,
2553
+ "student_masked_tokens": 124.0875
2554
+ },
2555
+ {
2556
+ "avg_mask_ratio": 0.4821103142516222,
2557
+ "avg_response_length": 265.35,
2558
+ "avg_student_mask_ratio": 0.4821103142516222,
2559
+ "batch_ainp_frac": 0.0,
2560
+ "batch_inp_frac": 0.0,
2561
+ "batch_inp_oh_frac": 0.0,
2562
+ "batch_inp_par_frac": 1.0,
2563
+ "batch_inp_par_reverse_frac": 0.0,
2564
+ "batch_rl_frac": 0.0,
2565
+ "batch_sft_frac": 0.0,
2566
+ "batch_soft_sft_frac": 0.0,
2567
+ "batch_tf_frac": 0.0,
2568
+ "ce_loss": 0.0,
2569
+ "epoch": 2.2858666666666667,
2570
+ "grad_norm": 0.146484375,
2571
+ "kd_loss": 0.1856000915321232,
2572
+ "learning_rate": 3e-06,
2573
+ "loss": 0.1583,
2574
+ "masked_tokens": 142.1875,
2575
+ "mean_t": 0.5015889146190602,
2576
+ "step": 1070,
2577
+ "student_masked_tokens": 142.1875
2578
+ },
2579
+ {
2580
+ "avg_mask_ratio": 0.461884257895872,
2581
+ "avg_response_length": 232.4,
2582
+ "avg_student_mask_ratio": 0.461884257895872,
2583
+ "batch_ainp_frac": 0.0,
2584
+ "batch_inp_frac": 0.0,
2585
+ "batch_inp_oh_frac": 0.0,
2586
+ "batch_inp_par_frac": 1.0,
2587
+ "batch_inp_par_reverse_frac": 0.0,
2588
+ "batch_rl_frac": 0.0,
2589
+ "batch_sft_frac": 0.0,
2590
+ "batch_soft_sft_frac": 0.0,
2591
+ "batch_tf_frac": 0.0,
2592
+ "ce_loss": 0.0,
2593
+ "epoch": 2.3072,
2594
+ "grad_norm": 0.09033203125,
2595
+ "kd_loss": 0.16016541003627935,
2596
+ "learning_rate": 3e-06,
2597
+ "loss": 0.1633,
2598
+ "masked_tokens": 106.725,
2599
+ "mean_t": 0.4983203248586506,
2600
+ "step": 1080,
2601
+ "student_masked_tokens": 106.725
2602
+ },
2603
+ {
2604
+ "avg_mask_ratio": 0.4401097826776095,
2605
+ "avg_response_length": 207.325,
2606
+ "avg_student_mask_ratio": 0.4401097826776095,
2607
+ "batch_ainp_frac": 0.0,
2608
+ "batch_inp_frac": 0.0,
2609
+ "batch_inp_oh_frac": 0.0,
2610
+ "batch_inp_par_frac": 1.0,
2611
+ "batch_inp_par_reverse_frac": 0.0,
2612
+ "batch_rl_frac": 0.0,
2613
+ "batch_sft_frac": 0.0,
2614
+ "batch_soft_sft_frac": 0.0,
2615
+ "batch_tf_frac": 0.0,
2616
+ "ce_loss": 0.0,
2617
+ "epoch": 2.3285333333333336,
2618
+ "grad_norm": 0.0908203125,
2619
+ "kd_loss": 0.1443735401110633,
2620
+ "learning_rate": 3e-06,
2621
+ "loss": 0.1676,
2622
+ "masked_tokens": 88.5375,
2623
+ "mean_t": 0.47094749807147307,
2624
+ "step": 1090,
2625
+ "student_masked_tokens": 88.5375
2626
+ },
2627
+ {
2628
+ "avg_mask_ratio": 0.5224281516042538,
2629
+ "avg_response_length": 257.675,
2630
+ "avg_student_mask_ratio": 0.5224281516042538,
2631
+ "batch_ainp_frac": 0.0,
2632
+ "batch_inp_frac": 0.0,
2633
+ "batch_inp_oh_frac": 0.0,
2634
+ "batch_inp_par_frac": 1.0,
2635
+ "batch_inp_par_reverse_frac": 0.0,
2636
+ "batch_rl_frac": 0.0,
2637
+ "batch_sft_frac": 0.0,
2638
+ "batch_soft_sft_frac": 0.0,
2639
+ "batch_tf_frac": 0.0,
2640
+ "ce_loss": 0.0,
2641
+ "epoch": 2.3498666666666668,
2642
+ "grad_norm": 0.1953125,
2643
+ "kd_loss": 0.18995505797054618,
2644
+ "learning_rate": 3e-06,
2645
+ "loss": 0.2181,
2646
+ "masked_tokens": 142.15,
2647
+ "mean_t": 0.5531192034482956,
2648
+ "step": 1100,
2649
+ "student_masked_tokens": 142.15
2650
+ }
2651
+ ],
2652
+ "logging_steps": 10,
2653
+ "max_steps": 1404,
2654
+ "num_input_tokens_seen": 0,
2655
+ "num_train_epochs": 3,
2656
+ "save_steps": 100,
2657
+ "stateful_callbacks": {
2658
+ "TrainerControl": {
2659
+ "args": {
2660
+ "should_epoch_stop": false,
2661
+ "should_evaluate": false,
2662
+ "should_log": false,
2663
+ "should_save": true,
2664
+ "should_training_stop": false
2665
+ },
2666
+ "attributes": {}
2667
+ }
2668
+ },
2669
+ "total_flos": 0.0,
2670
+ "train_batch_size": 1,
2671
+ "trial_name": null,
2672
+ "trial_params": null
2673
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1100/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c347df37da6e348160afc3fbb65d12595e5064bd8e5bfd591004e5e86a703f42
3
+ size 7992
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: GSAI-ML/LLaDA-8B-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 0.15.1
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/adapter_config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "GSAI-ML/LLaDA-8B-Instruct",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 64,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "r": 128,
24
+ "rank_pattern": {},
25
+ "revision": null,
26
+ "target_modules": [
27
+ "down_proj",
28
+ "up_proj",
29
+ "gate_proj",
30
+ "v_proj",
31
+ "k_proj",
32
+ "o_proj",
33
+ "q_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2dca70abd85b31b564a1982f92388cdf4292993f9da9e8b65efe47e5ccbc6db
3
+ size 2406624648
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfe31031f6599e908b2bf7a71f0b3239c56a319ecf746fa7cae7cd9230e52f9a
3
+ size 671304442
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1394463a46489e6dce7c0369a296b9effad20c6a87b30dbb892b34b73b5d6365
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c4761cd452b58b0c04d9b6cd8ff1e6d128acb7cabdb300e86daf48f850e7e941
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f8c95a6d9085dfcee1e6620c88ede526366d3a02c5018932b1bc04809c0e0c7
3
+ size 1064
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/trainer_state.json ADDED
@@ -0,0 +1,2913 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.5632,
5
+ "eval_steps": 500,
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
+ "avg_mask_ratio": 0.4933756332669873,
13
+ "avg_response_length": 252.0625,
14
+ "avg_student_mask_ratio": 0.4933756332669873,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 0.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 1.0,
19
+ "batch_inp_par_reverse_frac": 0.0,
20
+ "batch_rl_frac": 0.0,
21
+ "batch_sft_frac": 0.0,
22
+ "batch_soft_sft_frac": 0.0,
23
+ "batch_tf_frac": 0.0,
24
+ "ce_loss": 0.0,
25
+ "epoch": 0.021333333333333333,
26
+ "grad_norm": 0.1650390625,
27
+ "kd_loss": 0.24795629351958723,
28
+ "learning_rate": 3e-06,
29
+ "loss": 0.2758,
30
+ "masked_tokens": 120.975,
31
+ "mean_t": 0.5145528071501758,
32
+ "step": 10,
33
+ "student_masked_tokens": 120.975
34
+ },
35
+ {
36
+ "avg_mask_ratio": 0.41923869140446185,
37
+ "avg_response_length": 221.7125,
38
+ "avg_student_mask_ratio": 0.41923869140446185,
39
+ "batch_ainp_frac": 0.0,
40
+ "batch_inp_frac": 0.0,
41
+ "batch_inp_oh_frac": 0.0,
42
+ "batch_inp_par_frac": 1.0,
43
+ "batch_inp_par_reverse_frac": 0.0,
44
+ "batch_rl_frac": 0.0,
45
+ "batch_sft_frac": 0.0,
46
+ "batch_soft_sft_frac": 0.0,
47
+ "batch_tf_frac": 0.0,
48
+ "ce_loss": 0.0,
49
+ "epoch": 0.042666666666666665,
50
+ "grad_norm": 0.388671875,
51
+ "kd_loss": 0.21509853232191353,
52
+ "learning_rate": 3e-06,
53
+ "loss": 0.2046,
54
+ "masked_tokens": 86.1875,
55
+ "mean_t": 0.43874448732240123,
56
+ "step": 20,
57
+ "student_masked_tokens": 86.1875
58
+ },
59
+ {
60
+ "avg_mask_ratio": 0.4542569225654006,
61
+ "avg_response_length": 231.45,
62
+ "avg_student_mask_ratio": 0.4542569225654006,
63
+ "batch_ainp_frac": 0.0,
64
+ "batch_inp_frac": 0.0,
65
+ "batch_inp_oh_frac": 0.0,
66
+ "batch_inp_par_frac": 1.0,
67
+ "batch_inp_par_reverse_frac": 0.0,
68
+ "batch_rl_frac": 0.0,
69
+ "batch_sft_frac": 0.0,
70
+ "batch_soft_sft_frac": 0.0,
71
+ "batch_tf_frac": 0.0,
72
+ "ce_loss": 0.0,
73
+ "epoch": 0.064,
74
+ "grad_norm": 0.52734375,
75
+ "kd_loss": 0.19756361616970963,
76
+ "learning_rate": 3e-06,
77
+ "loss": 0.1976,
78
+ "masked_tokens": 116.2875,
79
+ "mean_t": 0.4803953981841914,
80
+ "step": 30,
81
+ "student_masked_tokens": 116.2875
82
+ },
83
+ {
84
+ "avg_mask_ratio": 0.41855402445653456,
85
+ "avg_response_length": 214.125,
86
+ "avg_student_mask_ratio": 0.41855402445653456,
87
+ "batch_ainp_frac": 0.0,
88
+ "batch_inp_frac": 0.0,
89
+ "batch_inp_oh_frac": 0.0,
90
+ "batch_inp_par_frac": 1.0,
91
+ "batch_inp_par_reverse_frac": 0.0,
92
+ "batch_rl_frac": 0.0,
93
+ "batch_sft_frac": 0.0,
94
+ "batch_soft_sft_frac": 0.0,
95
+ "batch_tf_frac": 0.0,
96
+ "ce_loss": 0.0,
97
+ "epoch": 0.08533333333333333,
98
+ "grad_norm": 0.263671875,
99
+ "kd_loss": 0.1576978968325534,
100
+ "learning_rate": 3e-06,
101
+ "loss": 0.1551,
102
+ "masked_tokens": 94.225,
103
+ "mean_t": 0.4569831106782658,
104
+ "step": 40,
105
+ "student_masked_tokens": 94.225
106
+ },
107
+ {
108
+ "avg_mask_ratio": 0.4331560767372139,
109
+ "avg_response_length": 222.225,
110
+ "avg_student_mask_ratio": 0.4331560767372139,
111
+ "batch_ainp_frac": 0.0,
112
+ "batch_inp_frac": 0.0,
113
+ "batch_inp_oh_frac": 0.0,
114
+ "batch_inp_par_frac": 1.0,
115
+ "batch_inp_par_reverse_frac": 0.0,
116
+ "batch_rl_frac": 0.0,
117
+ "batch_sft_frac": 0.0,
118
+ "batch_soft_sft_frac": 0.0,
119
+ "batch_tf_frac": 0.0,
120
+ "ce_loss": 0.0,
121
+ "epoch": 0.10666666666666667,
122
+ "grad_norm": 0.125,
123
+ "kd_loss": 0.17712681048956255,
124
+ "learning_rate": 3e-06,
125
+ "loss": 0.1648,
126
+ "masked_tokens": 97.825,
127
+ "mean_t": 0.4612453707959503,
128
+ "step": 50,
129
+ "student_masked_tokens": 97.825
130
+ },
131
+ {
132
+ "avg_mask_ratio": 0.4547638618387282,
133
+ "avg_response_length": 242.9,
134
+ "avg_student_mask_ratio": 0.4547638618387282,
135
+ "batch_ainp_frac": 0.0,
136
+ "batch_inp_frac": 0.0,
137
+ "batch_inp_oh_frac": 0.0,
138
+ "batch_inp_par_frac": 1.0,
139
+ "batch_inp_par_reverse_frac": 0.0,
140
+ "batch_rl_frac": 0.0,
141
+ "batch_sft_frac": 0.0,
142
+ "batch_soft_sft_frac": 0.0,
143
+ "batch_tf_frac": 0.0,
144
+ "ce_loss": 0.0,
145
+ "epoch": 0.128,
146
+ "grad_norm": 0.130859375,
147
+ "kd_loss": 0.1745696667137963,
148
+ "learning_rate": 3e-06,
149
+ "loss": 0.1745,
150
+ "masked_tokens": 119.125,
151
+ "mean_t": 0.48597636765334756,
152
+ "step": 60,
153
+ "student_masked_tokens": 119.125
154
+ },
155
+ {
156
+ "avg_mask_ratio": 0.5151988173020072,
157
+ "avg_response_length": 214.3375,
158
+ "avg_student_mask_ratio": 0.5151988173020072,
159
+ "batch_ainp_frac": 0.0,
160
+ "batch_inp_frac": 0.0,
161
+ "batch_inp_oh_frac": 0.0,
162
+ "batch_inp_par_frac": 1.0,
163
+ "batch_inp_par_reverse_frac": 0.0,
164
+ "batch_rl_frac": 0.0,
165
+ "batch_sft_frac": 0.0,
166
+ "batch_soft_sft_frac": 0.0,
167
+ "batch_tf_frac": 0.0,
168
+ "ce_loss": 0.0,
169
+ "epoch": 0.14933333333333335,
170
+ "grad_norm": 0.17578125,
171
+ "kd_loss": 0.1657758254527316,
172
+ "learning_rate": 3e-06,
173
+ "loss": 0.1972,
174
+ "masked_tokens": 111.7875,
175
+ "mean_t": 0.5459650319069624,
176
+ "step": 70,
177
+ "student_masked_tokens": 111.7875
178
+ },
179
+ {
180
+ "avg_mask_ratio": 0.37937068473547697,
181
+ "avg_response_length": 250.2,
182
+ "avg_student_mask_ratio": 0.37937068473547697,
183
+ "batch_ainp_frac": 0.0,
184
+ "batch_inp_frac": 0.0,
185
+ "batch_inp_oh_frac": 0.0,
186
+ "batch_inp_par_frac": 1.0,
187
+ "batch_inp_par_reverse_frac": 0.0,
188
+ "batch_rl_frac": 0.0,
189
+ "batch_sft_frac": 0.0,
190
+ "batch_soft_sft_frac": 0.0,
191
+ "batch_tf_frac": 0.0,
192
+ "ce_loss": 0.0,
193
+ "epoch": 0.17066666666666666,
194
+ "grad_norm": 0.051513671875,
195
+ "kd_loss": 0.13968983994418097,
196
+ "learning_rate": 3e-06,
197
+ "loss": 0.1369,
198
+ "masked_tokens": 95.1875,
199
+ "mean_t": 0.40758824030635876,
200
+ "step": 80,
201
+ "student_masked_tokens": 95.1875
202
+ },
203
+ {
204
+ "avg_mask_ratio": 0.5006106478627771,
205
+ "avg_response_length": 242.1125,
206
+ "avg_student_mask_ratio": 0.5006106478627771,
207
+ "batch_ainp_frac": 0.0,
208
+ "batch_inp_frac": 0.0,
209
+ "batch_inp_oh_frac": 0.0,
210
+ "batch_inp_par_frac": 1.0,
211
+ "batch_inp_par_reverse_frac": 0.0,
212
+ "batch_rl_frac": 0.0,
213
+ "batch_sft_frac": 0.0,
214
+ "batch_soft_sft_frac": 0.0,
215
+ "batch_tf_frac": 0.0,
216
+ "ce_loss": 0.0,
217
+ "epoch": 0.192,
218
+ "grad_norm": 0.22265625,
219
+ "kd_loss": 0.20869405062871707,
220
+ "learning_rate": 3e-06,
221
+ "loss": 0.2007,
222
+ "masked_tokens": 116.3875,
223
+ "mean_t": 0.5297661645396147,
224
+ "step": 90,
225
+ "student_masked_tokens": 116.3875
226
+ },
227
+ {
228
+ "avg_mask_ratio": 0.4596128500183113,
229
+ "avg_response_length": 229.0625,
230
+ "avg_student_mask_ratio": 0.4596128500183113,
231
+ "batch_ainp_frac": 0.0,
232
+ "batch_inp_frac": 0.0,
233
+ "batch_inp_oh_frac": 0.0,
234
+ "batch_inp_par_frac": 1.0,
235
+ "batch_inp_par_reverse_frac": 0.0,
236
+ "batch_rl_frac": 0.0,
237
+ "batch_sft_frac": 0.0,
238
+ "batch_soft_sft_frac": 0.0,
239
+ "batch_tf_frac": 0.0,
240
+ "ce_loss": 0.0,
241
+ "epoch": 0.21333333333333335,
242
+ "grad_norm": 0.1865234375,
243
+ "kd_loss": 0.17640120884607313,
244
+ "learning_rate": 3e-06,
245
+ "loss": 0.1864,
246
+ "masked_tokens": 109.7125,
247
+ "mean_t": 0.4886587227345444,
248
+ "step": 100,
249
+ "student_masked_tokens": 109.7125
250
+ },
251
+ {
252
+ "avg_mask_ratio": 0.4920400592498481,
253
+ "avg_response_length": 229.9875,
254
+ "avg_student_mask_ratio": 0.4920400592498481,
255
+ "batch_ainp_frac": 0.0,
256
+ "batch_inp_frac": 0.0,
257
+ "batch_inp_oh_frac": 0.0,
258
+ "batch_inp_par_frac": 1.0,
259
+ "batch_inp_par_reverse_frac": 0.0,
260
+ "batch_rl_frac": 0.0,
261
+ "batch_sft_frac": 0.0,
262
+ "batch_soft_sft_frac": 0.0,
263
+ "batch_tf_frac": 0.0,
264
+ "ce_loss": 0.0,
265
+ "epoch": 0.23466666666666666,
266
+ "grad_norm": 0.326171875,
267
+ "kd_loss": 0.2011610215539008,
268
+ "learning_rate": 3e-06,
269
+ "loss": 0.2334,
270
+ "masked_tokens": 109.4,
271
+ "mean_t": 0.5246987929102034,
272
+ "step": 110,
273
+ "student_masked_tokens": 109.4
274
+ },
275
+ {
276
+ "avg_mask_ratio": 0.450224511185661,
277
+ "avg_response_length": 229.225,
278
+ "avg_student_mask_ratio": 0.450224511185661,
279
+ "batch_ainp_frac": 0.0,
280
+ "batch_inp_frac": 0.0,
281
+ "batch_inp_oh_frac": 0.0,
282
+ "batch_inp_par_frac": 1.0,
283
+ "batch_inp_par_reverse_frac": 0.0,
284
+ "batch_rl_frac": 0.0,
285
+ "batch_sft_frac": 0.0,
286
+ "batch_soft_sft_frac": 0.0,
287
+ "batch_tf_frac": 0.0,
288
+ "ce_loss": 0.0,
289
+ "epoch": 0.256,
290
+ "grad_norm": 0.171875,
291
+ "kd_loss": 0.21365654302464918,
292
+ "learning_rate": 3e-06,
293
+ "loss": 0.2021,
294
+ "masked_tokens": 102.6375,
295
+ "mean_t": 0.4805434140143916,
296
+ "step": 120,
297
+ "student_masked_tokens": 102.6375
298
+ },
299
+ {
300
+ "avg_mask_ratio": 0.4397759020910598,
301
+ "avg_response_length": 214.15,
302
+ "avg_student_mask_ratio": 0.4397759020910598,
303
+ "batch_ainp_frac": 0.0,
304
+ "batch_inp_frac": 0.0,
305
+ "batch_inp_oh_frac": 0.0,
306
+ "batch_inp_par_frac": 1.0,
307
+ "batch_inp_par_reverse_frac": 0.0,
308
+ "batch_rl_frac": 0.0,
309
+ "batch_sft_frac": 0.0,
310
+ "batch_soft_sft_frac": 0.0,
311
+ "batch_tf_frac": 0.0,
312
+ "ce_loss": 0.0,
313
+ "epoch": 0.2773333333333333,
314
+ "grad_norm": 0.173828125,
315
+ "kd_loss": 0.13860440934267615,
316
+ "learning_rate": 3e-06,
317
+ "loss": 0.1362,
318
+ "masked_tokens": 98.4,
319
+ "mean_t": 0.47522516988683494,
320
+ "step": 130,
321
+ "student_masked_tokens": 98.4
322
+ },
323
+ {
324
+ "avg_mask_ratio": 0.4890626976499334,
325
+ "avg_response_length": 242.1625,
326
+ "avg_student_mask_ratio": 0.4890626976499334,
327
+ "batch_ainp_frac": 0.0,
328
+ "batch_inp_frac": 0.0,
329
+ "batch_inp_oh_frac": 0.0,
330
+ "batch_inp_par_frac": 1.0,
331
+ "batch_inp_par_reverse_frac": 0.0,
332
+ "batch_rl_frac": 0.0,
333
+ "batch_sft_frac": 0.0,
334
+ "batch_soft_sft_frac": 0.0,
335
+ "batch_tf_frac": 0.0,
336
+ "ce_loss": 0.0,
337
+ "epoch": 0.2986666666666667,
338
+ "grad_norm": 0.27734375,
339
+ "kd_loss": 0.2106460814328841,
340
+ "learning_rate": 3e-06,
341
+ "loss": 0.2054,
342
+ "masked_tokens": 129.725,
343
+ "mean_t": 0.5204090005659964,
344
+ "step": 140,
345
+ "student_masked_tokens": 129.725
346
+ },
347
+ {
348
+ "avg_mask_ratio": 0.47298577734036373,
349
+ "avg_response_length": 262.9875,
350
+ "avg_student_mask_ratio": 0.47298577734036373,
351
+ "batch_ainp_frac": 0.0,
352
+ "batch_inp_frac": 0.0,
353
+ "batch_inp_oh_frac": 0.0,
354
+ "batch_inp_par_frac": 1.0,
355
+ "batch_inp_par_reverse_frac": 0.0,
356
+ "batch_rl_frac": 0.0,
357
+ "batch_sft_frac": 0.0,
358
+ "batch_soft_sft_frac": 0.0,
359
+ "batch_tf_frac": 0.0,
360
+ "ce_loss": 0.0,
361
+ "epoch": 0.32,
362
+ "grad_norm": 0.0673828125,
363
+ "kd_loss": 0.2408599746217078,
364
+ "learning_rate": 3e-06,
365
+ "loss": 0.2276,
366
+ "masked_tokens": 128.375,
367
+ "mean_t": 0.5012552456930279,
368
+ "step": 150,
369
+ "student_masked_tokens": 128.375
370
+ },
371
+ {
372
+ "avg_mask_ratio": 0.5043223856599071,
373
+ "avg_response_length": 217.5,
374
+ "avg_student_mask_ratio": 0.5043223856599071,
375
+ "batch_ainp_frac": 0.0,
376
+ "batch_inp_frac": 0.0,
377
+ "batch_inp_oh_frac": 0.0,
378
+ "batch_inp_par_frac": 1.0,
379
+ "batch_inp_par_reverse_frac": 0.0,
380
+ "batch_rl_frac": 0.0,
381
+ "batch_sft_frac": 0.0,
382
+ "batch_soft_sft_frac": 0.0,
383
+ "batch_tf_frac": 0.0,
384
+ "ce_loss": 0.0,
385
+ "epoch": 0.3413333333333333,
386
+ "grad_norm": 0.2236328125,
387
+ "kd_loss": 0.20484722793833043,
388
+ "learning_rate": 3e-06,
389
+ "loss": 0.2154,
390
+ "masked_tokens": 106.025,
391
+ "mean_t": 0.5408745193795766,
392
+ "step": 160,
393
+ "student_masked_tokens": 106.025
394
+ },
395
+ {
396
+ "avg_mask_ratio": 0.48419030708028005,
397
+ "avg_response_length": 196.2625,
398
+ "avg_student_mask_ratio": 0.48419030708028005,
399
+ "batch_ainp_frac": 0.0,
400
+ "batch_inp_frac": 0.0,
401
+ "batch_inp_oh_frac": 0.0,
402
+ "batch_inp_par_frac": 1.0,
403
+ "batch_inp_par_reverse_frac": 0.0,
404
+ "batch_rl_frac": 0.0,
405
+ "batch_sft_frac": 0.0,
406
+ "batch_soft_sft_frac": 0.0,
407
+ "batch_tf_frac": 0.0,
408
+ "ce_loss": 0.0,
409
+ "epoch": 0.3626666666666667,
410
+ "grad_norm": 0.1611328125,
411
+ "kd_loss": 0.20407032655223248,
412
+ "learning_rate": 3e-06,
413
+ "loss": 0.2007,
414
+ "masked_tokens": 96.2625,
415
+ "mean_t": 0.5134547733236104,
416
+ "step": 170,
417
+ "student_masked_tokens": 96.2625
418
+ },
419
+ {
420
+ "avg_mask_ratio": 0.4485494759515859,
421
+ "avg_response_length": 228.8625,
422
+ "avg_student_mask_ratio": 0.4485494759515859,
423
+ "batch_ainp_frac": 0.0,
424
+ "batch_inp_frac": 0.0,
425
+ "batch_inp_oh_frac": 0.0,
426
+ "batch_inp_par_frac": 1.0,
427
+ "batch_inp_par_reverse_frac": 0.0,
428
+ "batch_rl_frac": 0.0,
429
+ "batch_sft_frac": 0.0,
430
+ "batch_soft_sft_frac": 0.0,
431
+ "batch_tf_frac": 0.0,
432
+ "ce_loss": 0.0,
433
+ "epoch": 0.384,
434
+ "grad_norm": 0.14453125,
435
+ "kd_loss": 0.16957379382825621,
436
+ "learning_rate": 3e-06,
437
+ "loss": 0.1796,
438
+ "masked_tokens": 101.275,
439
+ "mean_t": 0.48226988823735156,
440
+ "step": 180,
441
+ "student_masked_tokens": 101.275
442
+ },
443
+ {
444
+ "avg_mask_ratio": 0.48665878190658984,
445
+ "avg_response_length": 230.5,
446
+ "avg_student_mask_ratio": 0.48665878190658984,
447
+ "batch_ainp_frac": 0.0,
448
+ "batch_inp_frac": 0.0,
449
+ "batch_inp_oh_frac": 0.0,
450
+ "batch_inp_par_frac": 1.0,
451
+ "batch_inp_par_reverse_frac": 0.0,
452
+ "batch_rl_frac": 0.0,
453
+ "batch_sft_frac": 0.0,
454
+ "batch_soft_sft_frac": 0.0,
455
+ "batch_tf_frac": 0.0,
456
+ "ce_loss": 0.0,
457
+ "epoch": 0.4053333333333333,
458
+ "grad_norm": 0.3515625,
459
+ "kd_loss": 0.22313492714965832,
460
+ "learning_rate": 3e-06,
461
+ "loss": 0.2112,
462
+ "masked_tokens": 107.975,
463
+ "mean_t": 0.5165087037021294,
464
+ "step": 190,
465
+ "student_masked_tokens": 107.975
466
+ },
467
+ {
468
+ "avg_mask_ratio": 0.4670982737792656,
469
+ "avg_response_length": 210.9125,
470
+ "avg_student_mask_ratio": 0.4670982737792656,
471
+ "batch_ainp_frac": 0.0,
472
+ "batch_inp_frac": 0.0,
473
+ "batch_inp_oh_frac": 0.0,
474
+ "batch_inp_par_frac": 1.0,
475
+ "batch_inp_par_reverse_frac": 0.0,
476
+ "batch_rl_frac": 0.0,
477
+ "batch_sft_frac": 0.0,
478
+ "batch_soft_sft_frac": 0.0,
479
+ "batch_tf_frac": 0.0,
480
+ "ce_loss": 0.0,
481
+ "epoch": 0.4266666666666667,
482
+ "grad_norm": 0.142578125,
483
+ "kd_loss": 0.15220829088375468,
484
+ "learning_rate": 3e-06,
485
+ "loss": 0.1609,
486
+ "masked_tokens": 98.2125,
487
+ "mean_t": 0.5073627714533359,
488
+ "step": 200,
489
+ "student_masked_tokens": 98.2125
490
+ },
491
+ {
492
+ "avg_mask_ratio": 0.4568137794849463,
493
+ "avg_response_length": 218.7375,
494
+ "avg_student_mask_ratio": 0.4568137794849463,
495
+ "batch_ainp_frac": 0.0,
496
+ "batch_inp_frac": 0.0,
497
+ "batch_inp_oh_frac": 0.0,
498
+ "batch_inp_par_frac": 1.0,
499
+ "batch_inp_par_reverse_frac": 0.0,
500
+ "batch_rl_frac": 0.0,
501
+ "batch_sft_frac": 0.0,
502
+ "batch_soft_sft_frac": 0.0,
503
+ "batch_tf_frac": 0.0,
504
+ "ce_loss": 0.0,
505
+ "epoch": 0.448,
506
+ "grad_norm": 0.1826171875,
507
+ "kd_loss": 0.1580252643583208,
508
+ "learning_rate": 3e-06,
509
+ "loss": 0.1798,
510
+ "masked_tokens": 99.15,
511
+ "mean_t": 0.48430291628465055,
512
+ "step": 210,
513
+ "student_masked_tokens": 99.15
514
+ },
515
+ {
516
+ "avg_mask_ratio": 0.5379857187625021,
517
+ "avg_response_length": 252.1375,
518
+ "avg_student_mask_ratio": 0.5379857187625021,
519
+ "batch_ainp_frac": 0.0,
520
+ "batch_inp_frac": 0.0,
521
+ "batch_inp_oh_frac": 0.0,
522
+ "batch_inp_par_frac": 1.0,
523
+ "batch_inp_par_reverse_frac": 0.0,
524
+ "batch_rl_frac": 0.0,
525
+ "batch_sft_frac": 0.0,
526
+ "batch_soft_sft_frac": 0.0,
527
+ "batch_tf_frac": 0.0,
528
+ "ce_loss": 0.0,
529
+ "epoch": 0.4693333333333333,
530
+ "grad_norm": 0.3203125,
531
+ "kd_loss": 0.2511090909683844,
532
+ "learning_rate": 3e-06,
533
+ "loss": 0.2583,
534
+ "masked_tokens": 135.4,
535
+ "mean_t": 0.570199209311977,
536
+ "step": 220,
537
+ "student_masked_tokens": 135.4
538
+ },
539
+ {
540
+ "avg_mask_ratio": 0.43395056116278286,
541
+ "avg_response_length": 245.2625,
542
+ "avg_student_mask_ratio": 0.43395056116278286,
543
+ "batch_ainp_frac": 0.0,
544
+ "batch_inp_frac": 0.0,
545
+ "batch_inp_oh_frac": 0.0,
546
+ "batch_inp_par_frac": 1.0,
547
+ "batch_inp_par_reverse_frac": 0.0,
548
+ "batch_rl_frac": 0.0,
549
+ "batch_sft_frac": 0.0,
550
+ "batch_soft_sft_frac": 0.0,
551
+ "batch_tf_frac": 0.0,
552
+ "ce_loss": 0.0,
553
+ "epoch": 0.49066666666666664,
554
+ "grad_norm": 0.1767578125,
555
+ "kd_loss": 0.14414861655371852,
556
+ "learning_rate": 3e-06,
557
+ "loss": 0.1964,
558
+ "masked_tokens": 102.5125,
559
+ "mean_t": 0.46891279935371133,
560
+ "step": 230,
561
+ "student_masked_tokens": 102.5125
562
+ },
563
+ {
564
+ "avg_mask_ratio": 0.46948411157354714,
565
+ "avg_response_length": 202.975,
566
+ "avg_student_mask_ratio": 0.46948411157354714,
567
+ "batch_ainp_frac": 0.0,
568
+ "batch_inp_frac": 0.0,
569
+ "batch_inp_oh_frac": 0.0,
570
+ "batch_inp_par_frac": 1.0,
571
+ "batch_inp_par_reverse_frac": 0.0,
572
+ "batch_rl_frac": 0.0,
573
+ "batch_sft_frac": 0.0,
574
+ "batch_soft_sft_frac": 0.0,
575
+ "batch_tf_frac": 0.0,
576
+ "ce_loss": 0.0,
577
+ "epoch": 0.512,
578
+ "grad_norm": 0.1181640625,
579
+ "kd_loss": 0.2197965504183493,
580
+ "learning_rate": 3e-06,
581
+ "loss": 0.243,
582
+ "masked_tokens": 97.0625,
583
+ "mean_t": 0.502228345896583,
584
+ "step": 240,
585
+ "student_masked_tokens": 97.0625
586
+ },
587
+ {
588
+ "avg_mask_ratio": 0.44631263689370826,
589
+ "avg_response_length": 243.425,
590
+ "avg_student_mask_ratio": 0.44631263689370826,
591
+ "batch_ainp_frac": 0.0,
592
+ "batch_inp_frac": 0.0,
593
+ "batch_inp_oh_frac": 0.0,
594
+ "batch_inp_par_frac": 1.0,
595
+ "batch_inp_par_reverse_frac": 0.0,
596
+ "batch_rl_frac": 0.0,
597
+ "batch_sft_frac": 0.0,
598
+ "batch_soft_sft_frac": 0.0,
599
+ "batch_tf_frac": 0.0,
600
+ "ce_loss": 0.0,
601
+ "epoch": 0.5333333333333333,
602
+ "grad_norm": 0.1064453125,
603
+ "kd_loss": 0.2151024747882957,
604
+ "learning_rate": 3e-06,
605
+ "loss": 0.1892,
606
+ "masked_tokens": 107.4625,
607
+ "mean_t": 0.4733429416548461,
608
+ "step": 250,
609
+ "student_masked_tokens": 107.4625
610
+ },
611
+ {
612
+ "avg_mask_ratio": 0.4607314572727773,
613
+ "avg_response_length": 243.375,
614
+ "avg_student_mask_ratio": 0.4607314572727773,
615
+ "batch_ainp_frac": 0.0,
616
+ "batch_inp_frac": 0.0,
617
+ "batch_inp_oh_frac": 0.0,
618
+ "batch_inp_par_frac": 1.0,
619
+ "batch_inp_par_reverse_frac": 0.0,
620
+ "batch_rl_frac": 0.0,
621
+ "batch_sft_frac": 0.0,
622
+ "batch_soft_sft_frac": 0.0,
623
+ "batch_tf_frac": 0.0,
624
+ "ce_loss": 0.0,
625
+ "epoch": 0.5546666666666666,
626
+ "grad_norm": 0.203125,
627
+ "kd_loss": 0.176242933875335,
628
+ "learning_rate": 3e-06,
629
+ "loss": 0.1885,
630
+ "masked_tokens": 110.8875,
631
+ "mean_t": 0.4843149524240289,
632
+ "step": 260,
633
+ "student_masked_tokens": 110.8875
634
+ },
635
+ {
636
+ "avg_mask_ratio": 0.5309946444118395,
637
+ "avg_response_length": 231.6875,
638
+ "avg_student_mask_ratio": 0.5309946444118395,
639
+ "batch_ainp_frac": 0.0,
640
+ "batch_inp_frac": 0.0,
641
+ "batch_inp_oh_frac": 0.0,
642
+ "batch_inp_par_frac": 1.0,
643
+ "batch_inp_par_reverse_frac": 0.0,
644
+ "batch_rl_frac": 0.0,
645
+ "batch_sft_frac": 0.0,
646
+ "batch_soft_sft_frac": 0.0,
647
+ "batch_tf_frac": 0.0,
648
+ "ce_loss": 0.0,
649
+ "epoch": 0.576,
650
+ "grad_norm": 0.208984375,
651
+ "kd_loss": 0.26651088480309115,
652
+ "learning_rate": 3e-06,
653
+ "loss": 0.2828,
654
+ "masked_tokens": 123.2875,
655
+ "mean_t": 0.5643589949700981,
656
+ "step": 270,
657
+ "student_masked_tokens": 123.2875
658
+ },
659
+ {
660
+ "avg_mask_ratio": 0.45879559536697345,
661
+ "avg_response_length": 251.8,
662
+ "avg_student_mask_ratio": 0.45879559536697345,
663
+ "batch_ainp_frac": 0.0,
664
+ "batch_inp_frac": 0.0,
665
+ "batch_inp_oh_frac": 0.0,
666
+ "batch_inp_par_frac": 1.0,
667
+ "batch_inp_par_reverse_frac": 0.0,
668
+ "batch_rl_frac": 0.0,
669
+ "batch_sft_frac": 0.0,
670
+ "batch_soft_sft_frac": 0.0,
671
+ "batch_tf_frac": 0.0,
672
+ "ce_loss": 0.0,
673
+ "epoch": 0.5973333333333334,
674
+ "grad_norm": 0.11474609375,
675
+ "kd_loss": 0.145786481295454,
676
+ "learning_rate": 3e-06,
677
+ "loss": 0.1439,
678
+ "masked_tokens": 125.425,
679
+ "mean_t": 0.47818811538163575,
680
+ "step": 280,
681
+ "student_masked_tokens": 125.425
682
+ },
683
+ {
684
+ "avg_mask_ratio": 0.3955249205115251,
685
+ "avg_response_length": 238.675,
686
+ "avg_student_mask_ratio": 0.3955249205115251,
687
+ "batch_ainp_frac": 0.0,
688
+ "batch_inp_frac": 0.0,
689
+ "batch_inp_oh_frac": 0.0,
690
+ "batch_inp_par_frac": 1.0,
691
+ "batch_inp_par_reverse_frac": 0.0,
692
+ "batch_rl_frac": 0.0,
693
+ "batch_sft_frac": 0.0,
694
+ "batch_soft_sft_frac": 0.0,
695
+ "batch_tf_frac": 0.0,
696
+ "ce_loss": 0.0,
697
+ "epoch": 0.6186666666666667,
698
+ "grad_norm": 0.057373046875,
699
+ "kd_loss": 0.15104623195453543,
700
+ "learning_rate": 3e-06,
701
+ "loss": 0.1578,
702
+ "masked_tokens": 95.2125,
703
+ "mean_t": 0.4088635521940887,
704
+ "step": 290,
705
+ "student_masked_tokens": 95.2125
706
+ },
707
+ {
708
+ "avg_mask_ratio": 0.4504710016073659,
709
+ "avg_response_length": 202.575,
710
+ "avg_student_mask_ratio": 0.4504710016073659,
711
+ "batch_ainp_frac": 0.0,
712
+ "batch_inp_frac": 0.0,
713
+ "batch_inp_oh_frac": 0.0,
714
+ "batch_inp_par_frac": 1.0,
715
+ "batch_inp_par_reverse_frac": 0.0,
716
+ "batch_rl_frac": 0.0,
717
+ "batch_sft_frac": 0.0,
718
+ "batch_soft_sft_frac": 0.0,
719
+ "batch_tf_frac": 0.0,
720
+ "ce_loss": 0.0,
721
+ "epoch": 0.64,
722
+ "grad_norm": 0.150390625,
723
+ "kd_loss": 0.16011972622800386,
724
+ "learning_rate": 3e-06,
725
+ "loss": 0.179,
726
+ "masked_tokens": 90.3375,
727
+ "mean_t": 0.47984200695063917,
728
+ "step": 300,
729
+ "student_masked_tokens": 90.3375
730
+ },
731
+ {
732
+ "avg_mask_ratio": 0.4822002159198746,
733
+ "avg_response_length": 189.9875,
734
+ "avg_student_mask_ratio": 0.4822002159198746,
735
+ "batch_ainp_frac": 0.0,
736
+ "batch_inp_frac": 0.0,
737
+ "batch_inp_oh_frac": 0.0,
738
+ "batch_inp_par_frac": 1.0,
739
+ "batch_inp_par_reverse_frac": 0.0,
740
+ "batch_rl_frac": 0.0,
741
+ "batch_sft_frac": 0.0,
742
+ "batch_soft_sft_frac": 0.0,
743
+ "batch_tf_frac": 0.0,
744
+ "ce_loss": 0.0,
745
+ "epoch": 0.6613333333333333,
746
+ "grad_norm": 0.1630859375,
747
+ "kd_loss": 0.21744939284749734,
748
+ "learning_rate": 3e-06,
749
+ "loss": 0.201,
750
+ "masked_tokens": 92.7,
751
+ "mean_t": 0.5026606284547597,
752
+ "step": 310,
753
+ "student_masked_tokens": 92.7
754
+ },
755
+ {
756
+ "avg_mask_ratio": 0.5024422638700343,
757
+ "avg_response_length": 237.2625,
758
+ "avg_student_mask_ratio": 0.5024422638700343,
759
+ "batch_ainp_frac": 0.0,
760
+ "batch_inp_frac": 0.0,
761
+ "batch_inp_oh_frac": 0.0,
762
+ "batch_inp_par_frac": 1.0,
763
+ "batch_inp_par_reverse_frac": 0.0,
764
+ "batch_rl_frac": 0.0,
765
+ "batch_sft_frac": 0.0,
766
+ "batch_soft_sft_frac": 0.0,
767
+ "batch_tf_frac": 0.0,
768
+ "ce_loss": 0.0,
769
+ "epoch": 0.6826666666666666,
770
+ "grad_norm": 0.1259765625,
771
+ "kd_loss": 0.21489343987664142,
772
+ "learning_rate": 3e-06,
773
+ "loss": 0.2268,
774
+ "masked_tokens": 123.8125,
775
+ "mean_t": 0.5293499688967132,
776
+ "step": 320,
777
+ "student_masked_tokens": 123.8125
778
+ },
779
+ {
780
+ "avg_mask_ratio": 0.510216062690597,
781
+ "avg_response_length": 257.525,
782
+ "avg_student_mask_ratio": 0.510216062690597,
783
+ "batch_ainp_frac": 0.0,
784
+ "batch_inp_frac": 0.0,
785
+ "batch_inp_oh_frac": 0.0,
786
+ "batch_inp_par_frac": 1.0,
787
+ "batch_inp_par_reverse_frac": 0.0,
788
+ "batch_rl_frac": 0.0,
789
+ "batch_sft_frac": 0.0,
790
+ "batch_soft_sft_frac": 0.0,
791
+ "batch_tf_frac": 0.0,
792
+ "ce_loss": 0.0,
793
+ "epoch": 0.704,
794
+ "grad_norm": 0.1337890625,
795
+ "kd_loss": 0.17950079924535203,
796
+ "learning_rate": 3e-06,
797
+ "loss": 0.2007,
798
+ "masked_tokens": 142.8125,
799
+ "mean_t": 0.5426030711154454,
800
+ "step": 330,
801
+ "student_masked_tokens": 142.8125
802
+ },
803
+ {
804
+ "avg_mask_ratio": 0.45074162067612633,
805
+ "avg_response_length": 245.525,
806
+ "avg_student_mask_ratio": 0.45074162067612633,
807
+ "batch_ainp_frac": 0.0,
808
+ "batch_inp_frac": 0.0,
809
+ "batch_inp_oh_frac": 0.0,
810
+ "batch_inp_par_frac": 1.0,
811
+ "batch_inp_par_reverse_frac": 0.0,
812
+ "batch_rl_frac": 0.0,
813
+ "batch_sft_frac": 0.0,
814
+ "batch_soft_sft_frac": 0.0,
815
+ "batch_tf_frac": 0.0,
816
+ "ce_loss": 0.0,
817
+ "epoch": 0.7253333333333334,
818
+ "grad_norm": 0.310546875,
819
+ "kd_loss": 0.14043198096701417,
820
+ "learning_rate": 3e-06,
821
+ "loss": 0.1669,
822
+ "masked_tokens": 116.0875,
823
+ "mean_t": 0.47635243807453664,
824
+ "step": 340,
825
+ "student_masked_tokens": 116.0875
826
+ },
827
+ {
828
+ "avg_mask_ratio": 0.4926802407717332,
829
+ "avg_response_length": 238.2,
830
+ "avg_student_mask_ratio": 0.4926802407717332,
831
+ "batch_ainp_frac": 0.0,
832
+ "batch_inp_frac": 0.0,
833
+ "batch_inp_oh_frac": 0.0,
834
+ "batch_inp_par_frac": 1.0,
835
+ "batch_inp_par_reverse_frac": 0.0,
836
+ "batch_rl_frac": 0.0,
837
+ "batch_sft_frac": 0.0,
838
+ "batch_soft_sft_frac": 0.0,
839
+ "batch_tf_frac": 0.0,
840
+ "ce_loss": 0.0,
841
+ "epoch": 0.7466666666666667,
842
+ "grad_norm": 0.330078125,
843
+ "kd_loss": 0.21977804936059328,
844
+ "learning_rate": 3e-06,
845
+ "loss": 0.2497,
846
+ "masked_tokens": 127.4875,
847
+ "mean_t": 0.5268881446914747,
848
+ "step": 350,
849
+ "student_masked_tokens": 127.4875
850
+ },
851
+ {
852
+ "avg_mask_ratio": 0.4706261330051348,
853
+ "avg_response_length": 251.7375,
854
+ "avg_student_mask_ratio": 0.4706261330051348,
855
+ "batch_ainp_frac": 0.0,
856
+ "batch_inp_frac": 0.0,
857
+ "batch_inp_oh_frac": 0.0,
858
+ "batch_inp_par_frac": 1.0,
859
+ "batch_inp_par_reverse_frac": 0.0,
860
+ "batch_rl_frac": 0.0,
861
+ "batch_sft_frac": 0.0,
862
+ "batch_soft_sft_frac": 0.0,
863
+ "batch_tf_frac": 0.0,
864
+ "ce_loss": 0.0,
865
+ "epoch": 0.768,
866
+ "grad_norm": 0.056640625,
867
+ "kd_loss": 0.29630907970476983,
868
+ "learning_rate": 3e-06,
869
+ "loss": 0.2329,
870
+ "masked_tokens": 116.8625,
871
+ "mean_t": 0.5040419134311378,
872
+ "step": 360,
873
+ "student_masked_tokens": 116.8625
874
+ },
875
+ {
876
+ "avg_mask_ratio": 0.48096118308603764,
877
+ "avg_response_length": 262.25,
878
+ "avg_student_mask_ratio": 0.48096118308603764,
879
+ "batch_ainp_frac": 0.0,
880
+ "batch_inp_frac": 0.0,
881
+ "batch_inp_oh_frac": 0.0,
882
+ "batch_inp_par_frac": 1.0,
883
+ "batch_inp_par_reverse_frac": 0.0,
884
+ "batch_rl_frac": 0.0,
885
+ "batch_sft_frac": 0.0,
886
+ "batch_soft_sft_frac": 0.0,
887
+ "batch_tf_frac": 0.0,
888
+ "ce_loss": 0.0,
889
+ "epoch": 0.7893333333333333,
890
+ "grad_norm": 0.12255859375,
891
+ "kd_loss": 0.20822112379132704,
892
+ "learning_rate": 3e-06,
893
+ "loss": 0.186,
894
+ "masked_tokens": 132.2,
895
+ "mean_t": 0.5127181728370488,
896
+ "step": 370,
897
+ "student_masked_tokens": 132.2
898
+ },
899
+ {
900
+ "avg_mask_ratio": 0.4433969090110622,
901
+ "avg_response_length": 209.7125,
902
+ "avg_student_mask_ratio": 0.4433969090110622,
903
+ "batch_ainp_frac": 0.0,
904
+ "batch_inp_frac": 0.0,
905
+ "batch_inp_oh_frac": 0.0,
906
+ "batch_inp_par_frac": 1.0,
907
+ "batch_inp_par_reverse_frac": 0.0,
908
+ "batch_rl_frac": 0.0,
909
+ "batch_sft_frac": 0.0,
910
+ "batch_soft_sft_frac": 0.0,
911
+ "batch_tf_frac": 0.0,
912
+ "ce_loss": 0.0,
913
+ "epoch": 0.8106666666666666,
914
+ "grad_norm": 0.123046875,
915
+ "kd_loss": 0.14427866181035826,
916
+ "learning_rate": 3e-06,
917
+ "loss": 0.159,
918
+ "masked_tokens": 95.8625,
919
+ "mean_t": 0.4825185665744357,
920
+ "step": 380,
921
+ "student_masked_tokens": 95.8625
922
+ },
923
+ {
924
+ "avg_mask_ratio": 0.4764250977139454,
925
+ "avg_response_length": 226.3875,
926
+ "avg_student_mask_ratio": 0.4764250977139454,
927
+ "batch_ainp_frac": 0.0,
928
+ "batch_inp_frac": 0.0,
929
+ "batch_inp_oh_frac": 0.0,
930
+ "batch_inp_par_frac": 1.0,
931
+ "batch_inp_par_reverse_frac": 0.0,
932
+ "batch_rl_frac": 0.0,
933
+ "batch_sft_frac": 0.0,
934
+ "batch_soft_sft_frac": 0.0,
935
+ "batch_tf_frac": 0.0,
936
+ "ce_loss": 0.0,
937
+ "epoch": 0.832,
938
+ "grad_norm": 0.185546875,
939
+ "kd_loss": 0.18584371552193701,
940
+ "learning_rate": 3e-06,
941
+ "loss": 0.1823,
942
+ "masked_tokens": 113.95,
943
+ "mean_t": 0.5053101469413377,
944
+ "step": 390,
945
+ "student_masked_tokens": 113.95
946
+ },
947
+ {
948
+ "avg_mask_ratio": 0.47088071387261154,
949
+ "avg_response_length": 233.2125,
950
+ "avg_student_mask_ratio": 0.47088071387261154,
951
+ "batch_ainp_frac": 0.0,
952
+ "batch_inp_frac": 0.0,
953
+ "batch_inp_oh_frac": 0.0,
954
+ "batch_inp_par_frac": 1.0,
955
+ "batch_inp_par_reverse_frac": 0.0,
956
+ "batch_rl_frac": 0.0,
957
+ "batch_sft_frac": 0.0,
958
+ "batch_soft_sft_frac": 0.0,
959
+ "batch_tf_frac": 0.0,
960
+ "ce_loss": 0.0,
961
+ "epoch": 0.8533333333333334,
962
+ "grad_norm": 0.2734375,
963
+ "kd_loss": 0.22232839192147366,
964
+ "learning_rate": 3e-06,
965
+ "loss": 0.1961,
966
+ "masked_tokens": 116.675,
967
+ "mean_t": 0.5052250675857067,
968
+ "step": 400,
969
+ "student_masked_tokens": 116.675
970
+ },
971
+ {
972
+ "avg_mask_ratio": 0.4870647343515884,
973
+ "avg_response_length": 228.3875,
974
+ "avg_student_mask_ratio": 0.4870647343515884,
975
+ "batch_ainp_frac": 0.0,
976
+ "batch_inp_frac": 0.0,
977
+ "batch_inp_oh_frac": 0.0,
978
+ "batch_inp_par_frac": 1.0,
979
+ "batch_inp_par_reverse_frac": 0.0,
980
+ "batch_rl_frac": 0.0,
981
+ "batch_sft_frac": 0.0,
982
+ "batch_soft_sft_frac": 0.0,
983
+ "batch_tf_frac": 0.0,
984
+ "ce_loss": 0.0,
985
+ "epoch": 0.8746666666666667,
986
+ "grad_norm": 0.25390625,
987
+ "kd_loss": 0.25729746209006665,
988
+ "learning_rate": 3e-06,
989
+ "loss": 0.2306,
990
+ "masked_tokens": 114.3625,
991
+ "mean_t": 0.5050956419203431,
992
+ "step": 410,
993
+ "student_masked_tokens": 114.3625
994
+ },
995
+ {
996
+ "avg_mask_ratio": 0.440834702400025,
997
+ "avg_response_length": 209.85,
998
+ "avg_student_mask_ratio": 0.440834702400025,
999
+ "batch_ainp_frac": 0.0,
1000
+ "batch_inp_frac": 0.0,
1001
+ "batch_inp_oh_frac": 0.0,
1002
+ "batch_inp_par_frac": 1.0,
1003
+ "batch_inp_par_reverse_frac": 0.0,
1004
+ "batch_rl_frac": 0.0,
1005
+ "batch_sft_frac": 0.0,
1006
+ "batch_soft_sft_frac": 0.0,
1007
+ "batch_tf_frac": 0.0,
1008
+ "ce_loss": 0.0,
1009
+ "epoch": 0.896,
1010
+ "grad_norm": 0.2275390625,
1011
+ "kd_loss": 0.15747290870124503,
1012
+ "learning_rate": 3e-06,
1013
+ "loss": 0.1583,
1014
+ "masked_tokens": 87.575,
1015
+ "mean_t": 0.4769687672611326,
1016
+ "step": 420,
1017
+ "student_masked_tokens": 87.575
1018
+ },
1019
+ {
1020
+ "avg_mask_ratio": 0.4660509963519871,
1021
+ "avg_response_length": 250.9125,
1022
+ "avg_student_mask_ratio": 0.4660509963519871,
1023
+ "batch_ainp_frac": 0.0,
1024
+ "batch_inp_frac": 0.0,
1025
+ "batch_inp_oh_frac": 0.0,
1026
+ "batch_inp_par_frac": 1.0,
1027
+ "batch_inp_par_reverse_frac": 0.0,
1028
+ "batch_rl_frac": 0.0,
1029
+ "batch_sft_frac": 0.0,
1030
+ "batch_soft_sft_frac": 0.0,
1031
+ "batch_tf_frac": 0.0,
1032
+ "ce_loss": 0.0,
1033
+ "epoch": 0.9173333333333333,
1034
+ "grad_norm": 0.392578125,
1035
+ "kd_loss": 0.17299866449352522,
1036
+ "learning_rate": 3e-06,
1037
+ "loss": 0.178,
1038
+ "masked_tokens": 109.675,
1039
+ "mean_t": 0.49262027950026094,
1040
+ "step": 430,
1041
+ "student_masked_tokens": 109.675
1042
+ },
1043
+ {
1044
+ "avg_mask_ratio": 0.4507861359976232,
1045
+ "avg_response_length": 235.875,
1046
+ "avg_student_mask_ratio": 0.4507861359976232,
1047
+ "batch_ainp_frac": 0.0,
1048
+ "batch_inp_frac": 0.0,
1049
+ "batch_inp_oh_frac": 0.0,
1050
+ "batch_inp_par_frac": 1.0,
1051
+ "batch_inp_par_reverse_frac": 0.0,
1052
+ "batch_rl_frac": 0.0,
1053
+ "batch_sft_frac": 0.0,
1054
+ "batch_soft_sft_frac": 0.0,
1055
+ "batch_tf_frac": 0.0,
1056
+ "ce_loss": 0.0,
1057
+ "epoch": 0.9386666666666666,
1058
+ "grad_norm": 0.15234375,
1059
+ "kd_loss": 0.2113636662043291,
1060
+ "learning_rate": 3e-06,
1061
+ "loss": 0.1795,
1062
+ "masked_tokens": 106.95,
1063
+ "mean_t": 0.47046207524836064,
1064
+ "step": 440,
1065
+ "student_masked_tokens": 106.95
1066
+ },
1067
+ {
1068
+ "avg_mask_ratio": 0.4283985076006502,
1069
+ "avg_response_length": 230.95,
1070
+ "avg_student_mask_ratio": 0.4283985076006502,
1071
+ "batch_ainp_frac": 0.0,
1072
+ "batch_inp_frac": 0.0,
1073
+ "batch_inp_oh_frac": 0.0,
1074
+ "batch_inp_par_frac": 1.0,
1075
+ "batch_inp_par_reverse_frac": 0.0,
1076
+ "batch_rl_frac": 0.0,
1077
+ "batch_sft_frac": 0.0,
1078
+ "batch_soft_sft_frac": 0.0,
1079
+ "batch_tf_frac": 0.0,
1080
+ "ce_loss": 0.0,
1081
+ "epoch": 0.96,
1082
+ "grad_norm": 0.16015625,
1083
+ "kd_loss": 0.15304818171161969,
1084
+ "learning_rate": 3e-06,
1085
+ "loss": 0.1724,
1086
+ "masked_tokens": 101.15,
1087
+ "mean_t": 0.45781184462830427,
1088
+ "step": 450,
1089
+ "student_masked_tokens": 101.15
1090
+ },
1091
+ {
1092
+ "avg_mask_ratio": 0.47474822774529457,
1093
+ "avg_response_length": 233.1,
1094
+ "avg_student_mask_ratio": 0.47474822774529457,
1095
+ "batch_ainp_frac": 0.0,
1096
+ "batch_inp_frac": 0.0,
1097
+ "batch_inp_oh_frac": 0.0,
1098
+ "batch_inp_par_frac": 1.0,
1099
+ "batch_inp_par_reverse_frac": 0.0,
1100
+ "batch_rl_frac": 0.0,
1101
+ "batch_sft_frac": 0.0,
1102
+ "batch_soft_sft_frac": 0.0,
1103
+ "batch_tf_frac": 0.0,
1104
+ "ce_loss": 0.0,
1105
+ "epoch": 0.9813333333333333,
1106
+ "grad_norm": 0.06689453125,
1107
+ "kd_loss": 0.1363761811357108,
1108
+ "learning_rate": 3e-06,
1109
+ "loss": 0.171,
1110
+ "masked_tokens": 112.725,
1111
+ "mean_t": 0.5075790266972036,
1112
+ "step": 460,
1113
+ "student_masked_tokens": 112.725
1114
+ },
1115
+ {
1116
+ "avg_mask_ratio": 0.4808142688901474,
1117
+ "avg_response_length": 238.54761904761904,
1118
+ "avg_student_mask_ratio": 0.4808142688901474,
1119
+ "batch_ainp_frac": 0.0,
1120
+ "batch_inp_frac": 0.0,
1121
+ "batch_inp_oh_frac": 0.0,
1122
+ "batch_inp_par_frac": 1.0,
1123
+ "batch_inp_par_reverse_frac": 0.0,
1124
+ "batch_rl_frac": 0.0,
1125
+ "batch_sft_frac": 0.0,
1126
+ "batch_soft_sft_frac": 0.0,
1127
+ "batch_tf_frac": 0.0,
1128
+ "ce_loss": 0.0,
1129
+ "epoch": 1.0042666666666666,
1130
+ "grad_norm": 0.201171875,
1131
+ "kd_loss": 0.22379483340171732,
1132
+ "learning_rate": 3e-06,
1133
+ "loss": 0.2466,
1134
+ "masked_tokens": 113.67857142857143,
1135
+ "mean_t": 0.5031429776822084,
1136
+ "step": 470,
1137
+ "student_masked_tokens": 113.67857142857143
1138
+ },
1139
+ {
1140
+ "avg_mask_ratio": 0.47175657459301873,
1141
+ "avg_response_length": 249.9625,
1142
+ "avg_student_mask_ratio": 0.47175657459301873,
1143
+ "batch_ainp_frac": 0.0,
1144
+ "batch_inp_frac": 0.0,
1145
+ "batch_inp_oh_frac": 0.0,
1146
+ "batch_inp_par_frac": 1.0,
1147
+ "batch_inp_par_reverse_frac": 0.0,
1148
+ "batch_rl_frac": 0.0,
1149
+ "batch_sft_frac": 0.0,
1150
+ "batch_soft_sft_frac": 0.0,
1151
+ "batch_tf_frac": 0.0,
1152
+ "ce_loss": 0.0,
1153
+ "epoch": 1.0256,
1154
+ "grad_norm": 0.166015625,
1155
+ "kd_loss": 0.17491777570117745,
1156
+ "learning_rate": 3e-06,
1157
+ "loss": 0.2029,
1158
+ "masked_tokens": 119.4625,
1159
+ "mean_t": 0.504472183593316,
1160
+ "step": 480,
1161
+ "student_masked_tokens": 119.4625
1162
+ },
1163
+ {
1164
+ "avg_mask_ratio": 0.4564988439786248,
1165
+ "avg_response_length": 238.8875,
1166
+ "avg_student_mask_ratio": 0.4564988439786248,
1167
+ "batch_ainp_frac": 0.0,
1168
+ "batch_inp_frac": 0.0,
1169
+ "batch_inp_oh_frac": 0.0,
1170
+ "batch_inp_par_frac": 1.0,
1171
+ "batch_inp_par_reverse_frac": 0.0,
1172
+ "batch_rl_frac": 0.0,
1173
+ "batch_sft_frac": 0.0,
1174
+ "batch_soft_sft_frac": 0.0,
1175
+ "batch_tf_frac": 0.0,
1176
+ "ce_loss": 0.0,
1177
+ "epoch": 1.0469333333333333,
1178
+ "grad_norm": 0.1279296875,
1179
+ "kd_loss": 0.12884608846077866,
1180
+ "learning_rate": 3e-06,
1181
+ "loss": 0.1536,
1182
+ "masked_tokens": 104.0,
1183
+ "mean_t": 0.4888980514719151,
1184
+ "step": 490,
1185
+ "student_masked_tokens": 104.0
1186
+ },
1187
+ {
1188
+ "avg_mask_ratio": 0.5083174118888565,
1189
+ "avg_response_length": 258.1375,
1190
+ "avg_student_mask_ratio": 0.5083174118888565,
1191
+ "batch_ainp_frac": 0.0,
1192
+ "batch_inp_frac": 0.0,
1193
+ "batch_inp_oh_frac": 0.0,
1194
+ "batch_inp_par_frac": 1.0,
1195
+ "batch_inp_par_reverse_frac": 0.0,
1196
+ "batch_rl_frac": 0.0,
1197
+ "batch_sft_frac": 0.0,
1198
+ "batch_soft_sft_frac": 0.0,
1199
+ "batch_tf_frac": 0.0,
1200
+ "ce_loss": 0.0,
1201
+ "epoch": 1.0682666666666667,
1202
+ "grad_norm": 0.1357421875,
1203
+ "kd_loss": 0.18128383785248586,
1204
+ "learning_rate": 3e-06,
1205
+ "loss": 0.1811,
1206
+ "masked_tokens": 133.7125,
1207
+ "mean_t": 0.5469163245841628,
1208
+ "step": 500,
1209
+ "student_masked_tokens": 133.7125
1210
+ },
1211
+ {
1212
+ "avg_mask_ratio": 0.5130103683215566,
1213
+ "avg_response_length": 246.4875,
1214
+ "avg_student_mask_ratio": 0.5130103683215566,
1215
+ "batch_ainp_frac": 0.0,
1216
+ "batch_inp_frac": 0.0,
1217
+ "batch_inp_oh_frac": 0.0,
1218
+ "batch_inp_par_frac": 1.0,
1219
+ "batch_inp_par_reverse_frac": 0.0,
1220
+ "batch_rl_frac": 0.0,
1221
+ "batch_sft_frac": 0.0,
1222
+ "batch_soft_sft_frac": 0.0,
1223
+ "batch_tf_frac": 0.0,
1224
+ "ce_loss": 0.0,
1225
+ "epoch": 1.0896,
1226
+ "grad_norm": 0.11181640625,
1227
+ "kd_loss": 0.23847924997493805,
1228
+ "learning_rate": 3e-06,
1229
+ "loss": 0.2289,
1230
+ "masked_tokens": 132.5625,
1231
+ "mean_t": 0.5275314710394013,
1232
+ "step": 510,
1233
+ "student_masked_tokens": 132.5625
1234
+ },
1235
+ {
1236
+ "avg_mask_ratio": 0.4543681625276804,
1237
+ "avg_response_length": 199.65,
1238
+ "avg_student_mask_ratio": 0.4543681625276804,
1239
+ "batch_ainp_frac": 0.0,
1240
+ "batch_inp_frac": 0.0,
1241
+ "batch_inp_oh_frac": 0.0,
1242
+ "batch_inp_par_frac": 1.0,
1243
+ "batch_inp_par_reverse_frac": 0.0,
1244
+ "batch_rl_frac": 0.0,
1245
+ "batch_sft_frac": 0.0,
1246
+ "batch_soft_sft_frac": 0.0,
1247
+ "batch_tf_frac": 0.0,
1248
+ "ce_loss": 0.0,
1249
+ "epoch": 1.1109333333333333,
1250
+ "grad_norm": 0.1533203125,
1251
+ "kd_loss": 0.1353249137787543,
1252
+ "learning_rate": 3e-06,
1253
+ "loss": 0.1403,
1254
+ "masked_tokens": 87.9875,
1255
+ "mean_t": 0.49193521235138177,
1256
+ "step": 520,
1257
+ "student_masked_tokens": 87.9875
1258
+ },
1259
+ {
1260
+ "avg_mask_ratio": 0.46718079667771234,
1261
+ "avg_response_length": 218.0875,
1262
+ "avg_student_mask_ratio": 0.46718079667771234,
1263
+ "batch_ainp_frac": 0.0,
1264
+ "batch_inp_frac": 0.0,
1265
+ "batch_inp_oh_frac": 0.0,
1266
+ "batch_inp_par_frac": 1.0,
1267
+ "batch_inp_par_reverse_frac": 0.0,
1268
+ "batch_rl_frac": 0.0,
1269
+ "batch_sft_frac": 0.0,
1270
+ "batch_soft_sft_frac": 0.0,
1271
+ "batch_tf_frac": 0.0,
1272
+ "ce_loss": 0.0,
1273
+ "epoch": 1.1322666666666668,
1274
+ "grad_norm": 0.2109375,
1275
+ "kd_loss": 0.15268151032492625,
1276
+ "learning_rate": 3e-06,
1277
+ "loss": 0.1789,
1278
+ "masked_tokens": 101.3875,
1279
+ "mean_t": 0.49621942077938,
1280
+ "step": 530,
1281
+ "student_masked_tokens": 101.3875
1282
+ },
1283
+ {
1284
+ "avg_mask_ratio": 0.4890203754650429,
1285
+ "avg_response_length": 244.3875,
1286
+ "avg_student_mask_ratio": 0.4890203754650429,
1287
+ "batch_ainp_frac": 0.0,
1288
+ "batch_inp_frac": 0.0,
1289
+ "batch_inp_oh_frac": 0.0,
1290
+ "batch_inp_par_frac": 1.0,
1291
+ "batch_inp_par_reverse_frac": 0.0,
1292
+ "batch_rl_frac": 0.0,
1293
+ "batch_sft_frac": 0.0,
1294
+ "batch_soft_sft_frac": 0.0,
1295
+ "batch_tf_frac": 0.0,
1296
+ "ce_loss": 0.0,
1297
+ "epoch": 1.1536,
1298
+ "grad_norm": 0.27734375,
1299
+ "kd_loss": 0.17835129436630268,
1300
+ "learning_rate": 3e-06,
1301
+ "loss": 0.2173,
1302
+ "masked_tokens": 116.175,
1303
+ "mean_t": 0.5208023569080978,
1304
+ "step": 540,
1305
+ "student_masked_tokens": 116.175
1306
+ },
1307
+ {
1308
+ "avg_mask_ratio": 0.45064474650425834,
1309
+ "avg_response_length": 217.8375,
1310
+ "avg_student_mask_ratio": 0.45064474650425834,
1311
+ "batch_ainp_frac": 0.0,
1312
+ "batch_inp_frac": 0.0,
1313
+ "batch_inp_oh_frac": 0.0,
1314
+ "batch_inp_par_frac": 1.0,
1315
+ "batch_inp_par_reverse_frac": 0.0,
1316
+ "batch_rl_frac": 0.0,
1317
+ "batch_sft_frac": 0.0,
1318
+ "batch_soft_sft_frac": 0.0,
1319
+ "batch_tf_frac": 0.0,
1320
+ "ce_loss": 0.0,
1321
+ "epoch": 1.1749333333333334,
1322
+ "grad_norm": 0.17578125,
1323
+ "kd_loss": 0.16049880692362706,
1324
+ "learning_rate": 3e-06,
1325
+ "loss": 0.1855,
1326
+ "masked_tokens": 98.6375,
1327
+ "mean_t": 0.4767197913257405,
1328
+ "step": 550,
1329
+ "student_masked_tokens": 98.6375
1330
+ },
1331
+ {
1332
+ "avg_mask_ratio": 0.3844255942822201,
1333
+ "avg_response_length": 240.825,
1334
+ "avg_student_mask_ratio": 0.3844255942822201,
1335
+ "batch_ainp_frac": 0.0,
1336
+ "batch_inp_frac": 0.0,
1337
+ "batch_inp_oh_frac": 0.0,
1338
+ "batch_inp_par_frac": 1.0,
1339
+ "batch_inp_par_reverse_frac": 0.0,
1340
+ "batch_rl_frac": 0.0,
1341
+ "batch_sft_frac": 0.0,
1342
+ "batch_soft_sft_frac": 0.0,
1343
+ "batch_tf_frac": 0.0,
1344
+ "ce_loss": 0.0,
1345
+ "epoch": 1.1962666666666666,
1346
+ "grad_norm": 0.5859375,
1347
+ "kd_loss": 0.17605857607457268,
1348
+ "learning_rate": 3e-06,
1349
+ "loss": 0.1885,
1350
+ "masked_tokens": 96.85,
1351
+ "mean_t": 0.4111072298779618,
1352
+ "step": 560,
1353
+ "student_masked_tokens": 96.85
1354
+ },
1355
+ {
1356
+ "avg_mask_ratio": 0.45103558609262107,
1357
+ "avg_response_length": 231.025,
1358
+ "avg_student_mask_ratio": 0.45103558609262107,
1359
+ "batch_ainp_frac": 0.0,
1360
+ "batch_inp_frac": 0.0,
1361
+ "batch_inp_oh_frac": 0.0,
1362
+ "batch_inp_par_frac": 1.0,
1363
+ "batch_inp_par_reverse_frac": 0.0,
1364
+ "batch_rl_frac": 0.0,
1365
+ "batch_sft_frac": 0.0,
1366
+ "batch_soft_sft_frac": 0.0,
1367
+ "batch_tf_frac": 0.0,
1368
+ "ce_loss": 0.0,
1369
+ "epoch": 1.2176,
1370
+ "grad_norm": 0.234375,
1371
+ "kd_loss": 0.1386162672638477,
1372
+ "learning_rate": 3e-06,
1373
+ "loss": 0.1681,
1374
+ "masked_tokens": 105.6875,
1375
+ "mean_t": 0.48533305872697385,
1376
+ "step": 570,
1377
+ "student_masked_tokens": 105.6875
1378
+ },
1379
+ {
1380
+ "avg_mask_ratio": 0.47033366551622746,
1381
+ "avg_response_length": 248.3875,
1382
+ "avg_student_mask_ratio": 0.47033366551622746,
1383
+ "batch_ainp_frac": 0.0,
1384
+ "batch_inp_frac": 0.0,
1385
+ "batch_inp_oh_frac": 0.0,
1386
+ "batch_inp_par_frac": 1.0,
1387
+ "batch_inp_par_reverse_frac": 0.0,
1388
+ "batch_rl_frac": 0.0,
1389
+ "batch_sft_frac": 0.0,
1390
+ "batch_soft_sft_frac": 0.0,
1391
+ "batch_tf_frac": 0.0,
1392
+ "ce_loss": 0.0,
1393
+ "epoch": 1.2389333333333332,
1394
+ "grad_norm": 0.26953125,
1395
+ "kd_loss": 0.17702910760917803,
1396
+ "learning_rate": 3e-06,
1397
+ "loss": 0.1899,
1398
+ "masked_tokens": 125.05,
1399
+ "mean_t": 0.49976949762785805,
1400
+ "step": 580,
1401
+ "student_masked_tokens": 125.05
1402
+ },
1403
+ {
1404
+ "avg_mask_ratio": 0.5230229062028229,
1405
+ "avg_response_length": 241.8125,
1406
+ "avg_student_mask_ratio": 0.5230229062028229,
1407
+ "batch_ainp_frac": 0.0,
1408
+ "batch_inp_frac": 0.0,
1409
+ "batch_inp_oh_frac": 0.0,
1410
+ "batch_inp_par_frac": 1.0,
1411
+ "batch_inp_par_reverse_frac": 0.0,
1412
+ "batch_rl_frac": 0.0,
1413
+ "batch_sft_frac": 0.0,
1414
+ "batch_soft_sft_frac": 0.0,
1415
+ "batch_tf_frac": 0.0,
1416
+ "ce_loss": 0.0,
1417
+ "epoch": 1.2602666666666666,
1418
+ "grad_norm": 0.20703125,
1419
+ "kd_loss": 0.22054996666956866,
1420
+ "learning_rate": 3e-06,
1421
+ "loss": 0.2233,
1422
+ "masked_tokens": 129.0,
1423
+ "mean_t": 0.5521843038732186,
1424
+ "step": 590,
1425
+ "student_masked_tokens": 129.0
1426
+ },
1427
+ {
1428
+ "avg_mask_ratio": 0.44929012526990847,
1429
+ "avg_response_length": 246.7375,
1430
+ "avg_student_mask_ratio": 0.44929012526990847,
1431
+ "batch_ainp_frac": 0.0,
1432
+ "batch_inp_frac": 0.0,
1433
+ "batch_inp_oh_frac": 0.0,
1434
+ "batch_inp_par_frac": 1.0,
1435
+ "batch_inp_par_reverse_frac": 0.0,
1436
+ "batch_rl_frac": 0.0,
1437
+ "batch_sft_frac": 0.0,
1438
+ "batch_soft_sft_frac": 0.0,
1439
+ "batch_tf_frac": 0.0,
1440
+ "ce_loss": 0.0,
1441
+ "epoch": 1.2816,
1442
+ "grad_norm": 0.212890625,
1443
+ "kd_loss": 0.15257543138572202,
1444
+ "learning_rate": 3e-06,
1445
+ "loss": 0.1581,
1446
+ "masked_tokens": 113.9375,
1447
+ "mean_t": 0.4788527532829903,
1448
+ "step": 600,
1449
+ "student_masked_tokens": 113.9375
1450
+ },
1451
+ {
1452
+ "avg_mask_ratio": 0.5061312943696976,
1453
+ "avg_response_length": 237.175,
1454
+ "avg_student_mask_ratio": 0.5061312943696976,
1455
+ "batch_ainp_frac": 0.0,
1456
+ "batch_inp_frac": 0.0,
1457
+ "batch_inp_oh_frac": 0.0,
1458
+ "batch_inp_par_frac": 1.0,
1459
+ "batch_inp_par_reverse_frac": 0.0,
1460
+ "batch_rl_frac": 0.0,
1461
+ "batch_sft_frac": 0.0,
1462
+ "batch_soft_sft_frac": 0.0,
1463
+ "batch_tf_frac": 0.0,
1464
+ "ce_loss": 0.0,
1465
+ "epoch": 1.3029333333333333,
1466
+ "grad_norm": 0.06884765625,
1467
+ "kd_loss": 0.21158275993875578,
1468
+ "learning_rate": 3e-06,
1469
+ "loss": 0.1942,
1470
+ "masked_tokens": 116.675,
1471
+ "mean_t": 0.5445419924799353,
1472
+ "step": 610,
1473
+ "student_masked_tokens": 116.675
1474
+ },
1475
+ {
1476
+ "avg_mask_ratio": 0.4925117701757699,
1477
+ "avg_response_length": 233.2625,
1478
+ "avg_student_mask_ratio": 0.4925117701757699,
1479
+ "batch_ainp_frac": 0.0,
1480
+ "batch_inp_frac": 0.0,
1481
+ "batch_inp_oh_frac": 0.0,
1482
+ "batch_inp_par_frac": 1.0,
1483
+ "batch_inp_par_reverse_frac": 0.0,
1484
+ "batch_rl_frac": 0.0,
1485
+ "batch_sft_frac": 0.0,
1486
+ "batch_soft_sft_frac": 0.0,
1487
+ "batch_tf_frac": 0.0,
1488
+ "ce_loss": 0.0,
1489
+ "epoch": 1.3242666666666667,
1490
+ "grad_norm": 0.1787109375,
1491
+ "kd_loss": 0.1805886138310143,
1492
+ "learning_rate": 3e-06,
1493
+ "loss": 0.1744,
1494
+ "masked_tokens": 111.6875,
1495
+ "mean_t": 0.52490478400141,
1496
+ "step": 620,
1497
+ "student_masked_tokens": 111.6875
1498
+ },
1499
+ {
1500
+ "avg_mask_ratio": 0.5307835865532979,
1501
+ "avg_response_length": 249.65,
1502
+ "avg_student_mask_ratio": 0.5307835865532979,
1503
+ "batch_ainp_frac": 0.0,
1504
+ "batch_inp_frac": 0.0,
1505
+ "batch_inp_oh_frac": 0.0,
1506
+ "batch_inp_par_frac": 1.0,
1507
+ "batch_inp_par_reverse_frac": 0.0,
1508
+ "batch_rl_frac": 0.0,
1509
+ "batch_sft_frac": 0.0,
1510
+ "batch_soft_sft_frac": 0.0,
1511
+ "batch_tf_frac": 0.0,
1512
+ "ce_loss": 0.0,
1513
+ "epoch": 1.3456000000000001,
1514
+ "grad_norm": 0.28125,
1515
+ "kd_loss": 0.25403604302136956,
1516
+ "learning_rate": 3e-06,
1517
+ "loss": 0.2627,
1518
+ "masked_tokens": 126.6875,
1519
+ "mean_t": 0.5564947265549562,
1520
+ "step": 630,
1521
+ "student_masked_tokens": 126.6875
1522
+ },
1523
+ {
1524
+ "avg_mask_ratio": 0.48631439574528484,
1525
+ "avg_response_length": 241.475,
1526
+ "avg_student_mask_ratio": 0.48631439574528484,
1527
+ "batch_ainp_frac": 0.0,
1528
+ "batch_inp_frac": 0.0,
1529
+ "batch_inp_oh_frac": 0.0,
1530
+ "batch_inp_par_frac": 1.0,
1531
+ "batch_inp_par_reverse_frac": 0.0,
1532
+ "batch_rl_frac": 0.0,
1533
+ "batch_sft_frac": 0.0,
1534
+ "batch_soft_sft_frac": 0.0,
1535
+ "batch_tf_frac": 0.0,
1536
+ "ce_loss": 0.0,
1537
+ "epoch": 1.3669333333333333,
1538
+ "grad_norm": 0.2412109375,
1539
+ "kd_loss": 0.1642333460577902,
1540
+ "learning_rate": 3e-06,
1541
+ "loss": 0.1731,
1542
+ "masked_tokens": 122.1125,
1543
+ "mean_t": 0.5093393943971023,
1544
+ "step": 640,
1545
+ "student_masked_tokens": 122.1125
1546
+ },
1547
+ {
1548
+ "avg_mask_ratio": 0.5248487972887232,
1549
+ "avg_response_length": 231.5375,
1550
+ "avg_student_mask_ratio": 0.5248487972887232,
1551
+ "batch_ainp_frac": 0.0,
1552
+ "batch_inp_frac": 0.0,
1553
+ "batch_inp_oh_frac": 0.0,
1554
+ "batch_inp_par_frac": 1.0,
1555
+ "batch_inp_par_reverse_frac": 0.0,
1556
+ "batch_rl_frac": 0.0,
1557
+ "batch_sft_frac": 0.0,
1558
+ "batch_soft_sft_frac": 0.0,
1559
+ "batch_tf_frac": 0.0,
1560
+ "ce_loss": 0.0,
1561
+ "epoch": 1.3882666666666665,
1562
+ "grad_norm": 0.193359375,
1563
+ "kd_loss": 0.2508082524812494,
1564
+ "learning_rate": 3e-06,
1565
+ "loss": 0.2054,
1566
+ "masked_tokens": 127.175,
1567
+ "mean_t": 0.5558586571365595,
1568
+ "step": 650,
1569
+ "student_masked_tokens": 127.175
1570
+ },
1571
+ {
1572
+ "avg_mask_ratio": 0.48489007767057046,
1573
+ "avg_response_length": 246.0125,
1574
+ "avg_student_mask_ratio": 0.48489007767057046,
1575
+ "batch_ainp_frac": 0.0,
1576
+ "batch_inp_frac": 0.0,
1577
+ "batch_inp_oh_frac": 0.0,
1578
+ "batch_inp_par_frac": 1.0,
1579
+ "batch_inp_par_reverse_frac": 0.0,
1580
+ "batch_rl_frac": 0.0,
1581
+ "batch_sft_frac": 0.0,
1582
+ "batch_soft_sft_frac": 0.0,
1583
+ "batch_tf_frac": 0.0,
1584
+ "ce_loss": 0.0,
1585
+ "epoch": 1.4096,
1586
+ "grad_norm": 0.28125,
1587
+ "kd_loss": 0.18781521328146483,
1588
+ "learning_rate": 3e-06,
1589
+ "loss": 0.2045,
1590
+ "masked_tokens": 125.225,
1591
+ "mean_t": 0.51307404555846,
1592
+ "step": 660,
1593
+ "student_masked_tokens": 125.225
1594
+ },
1595
+ {
1596
+ "avg_mask_ratio": 0.4605832444969565,
1597
+ "avg_response_length": 244.5,
1598
+ "avg_student_mask_ratio": 0.4605832444969565,
1599
+ "batch_ainp_frac": 0.0,
1600
+ "batch_inp_frac": 0.0,
1601
+ "batch_inp_oh_frac": 0.0,
1602
+ "batch_inp_par_frac": 1.0,
1603
+ "batch_inp_par_reverse_frac": 0.0,
1604
+ "batch_rl_frac": 0.0,
1605
+ "batch_sft_frac": 0.0,
1606
+ "batch_soft_sft_frac": 0.0,
1607
+ "batch_tf_frac": 0.0,
1608
+ "ce_loss": 0.0,
1609
+ "epoch": 1.4309333333333334,
1610
+ "grad_norm": 0.19140625,
1611
+ "kd_loss": 0.1806626110754223,
1612
+ "learning_rate": 3e-06,
1613
+ "loss": 0.1702,
1614
+ "masked_tokens": 120.825,
1615
+ "mean_t": 0.4907285622088239,
1616
+ "step": 670,
1617
+ "student_masked_tokens": 120.825
1618
+ },
1619
+ {
1620
+ "avg_mask_ratio": 0.4662455078505445,
1621
+ "avg_response_length": 244.6125,
1622
+ "avg_student_mask_ratio": 0.4662455078505445,
1623
+ "batch_ainp_frac": 0.0,
1624
+ "batch_inp_frac": 0.0,
1625
+ "batch_inp_oh_frac": 0.0,
1626
+ "batch_inp_par_frac": 1.0,
1627
+ "batch_inp_par_reverse_frac": 0.0,
1628
+ "batch_rl_frac": 0.0,
1629
+ "batch_sft_frac": 0.0,
1630
+ "batch_soft_sft_frac": 0.0,
1631
+ "batch_tf_frac": 0.0,
1632
+ "ce_loss": 0.0,
1633
+ "epoch": 1.4522666666666666,
1634
+ "grad_norm": 0.16796875,
1635
+ "kd_loss": 0.20038694294766798,
1636
+ "learning_rate": 3e-06,
1637
+ "loss": 0.1824,
1638
+ "masked_tokens": 114.2,
1639
+ "mean_t": 0.4913603452499956,
1640
+ "step": 680,
1641
+ "student_masked_tokens": 114.2
1642
+ },
1643
+ {
1644
+ "avg_mask_ratio": 0.4820630593923852,
1645
+ "avg_response_length": 217.3,
1646
+ "avg_student_mask_ratio": 0.4820630593923852,
1647
+ "batch_ainp_frac": 0.0,
1648
+ "batch_inp_frac": 0.0,
1649
+ "batch_inp_oh_frac": 0.0,
1650
+ "batch_inp_par_frac": 1.0,
1651
+ "batch_inp_par_reverse_frac": 0.0,
1652
+ "batch_rl_frac": 0.0,
1653
+ "batch_sft_frac": 0.0,
1654
+ "batch_soft_sft_frac": 0.0,
1655
+ "batch_tf_frac": 0.0,
1656
+ "ce_loss": 0.0,
1657
+ "epoch": 1.4736,
1658
+ "grad_norm": 0.11279296875,
1659
+ "kd_loss": 0.16563009086588637,
1660
+ "learning_rate": 3e-06,
1661
+ "loss": 0.17,
1662
+ "masked_tokens": 118.4875,
1663
+ "mean_t": 0.5133644798654131,
1664
+ "step": 690,
1665
+ "student_masked_tokens": 118.4875
1666
+ },
1667
+ {
1668
+ "avg_mask_ratio": 0.5206489040749147,
1669
+ "avg_response_length": 216.45,
1670
+ "avg_student_mask_ratio": 0.5206489040749147,
1671
+ "batch_ainp_frac": 0.0,
1672
+ "batch_inp_frac": 0.0,
1673
+ "batch_inp_oh_frac": 0.0,
1674
+ "batch_inp_par_frac": 1.0,
1675
+ "batch_inp_par_reverse_frac": 0.0,
1676
+ "batch_rl_frac": 0.0,
1677
+ "batch_sft_frac": 0.0,
1678
+ "batch_soft_sft_frac": 0.0,
1679
+ "batch_tf_frac": 0.0,
1680
+ "ce_loss": 0.0,
1681
+ "epoch": 1.4949333333333334,
1682
+ "grad_norm": 0.236328125,
1683
+ "kd_loss": 0.23649522811845144,
1684
+ "learning_rate": 3e-06,
1685
+ "loss": 0.2241,
1686
+ "masked_tokens": 118.4375,
1687
+ "mean_t": 0.5625698395539075,
1688
+ "step": 700,
1689
+ "student_masked_tokens": 118.4375
1690
+ },
1691
+ {
1692
+ "avg_mask_ratio": 0.4955552547937259,
1693
+ "avg_response_length": 211.175,
1694
+ "avg_student_mask_ratio": 0.4955552547937259,
1695
+ "batch_ainp_frac": 0.0,
1696
+ "batch_inp_frac": 0.0,
1697
+ "batch_inp_oh_frac": 0.0,
1698
+ "batch_inp_par_frac": 1.0,
1699
+ "batch_inp_par_reverse_frac": 0.0,
1700
+ "batch_rl_frac": 0.0,
1701
+ "batch_sft_frac": 0.0,
1702
+ "batch_soft_sft_frac": 0.0,
1703
+ "batch_tf_frac": 0.0,
1704
+ "ce_loss": 0.0,
1705
+ "epoch": 1.5162666666666667,
1706
+ "grad_norm": 0.275390625,
1707
+ "kd_loss": 0.23970817765721222,
1708
+ "learning_rate": 3e-06,
1709
+ "loss": 0.2184,
1710
+ "masked_tokens": 112.3375,
1711
+ "mean_t": 0.5135623761918395,
1712
+ "step": 710,
1713
+ "student_masked_tokens": 112.3375
1714
+ },
1715
+ {
1716
+ "avg_mask_ratio": 0.4820543818641454,
1717
+ "avg_response_length": 229.1375,
1718
+ "avg_student_mask_ratio": 0.4820543818641454,
1719
+ "batch_ainp_frac": 0.0,
1720
+ "batch_inp_frac": 0.0,
1721
+ "batch_inp_oh_frac": 0.0,
1722
+ "batch_inp_par_frac": 1.0,
1723
+ "batch_inp_par_reverse_frac": 0.0,
1724
+ "batch_rl_frac": 0.0,
1725
+ "batch_sft_frac": 0.0,
1726
+ "batch_soft_sft_frac": 0.0,
1727
+ "batch_tf_frac": 0.0,
1728
+ "ce_loss": 0.0,
1729
+ "epoch": 1.5375999999999999,
1730
+ "grad_norm": 0.435546875,
1731
+ "kd_loss": 0.18955910701470202,
1732
+ "learning_rate": 3e-06,
1733
+ "loss": 0.1978,
1734
+ "masked_tokens": 114.55,
1735
+ "mean_t": 0.5117021896177902,
1736
+ "step": 720,
1737
+ "student_masked_tokens": 114.55
1738
+ },
1739
+ {
1740
+ "avg_mask_ratio": 0.4605119539948646,
1741
+ "avg_response_length": 245.5625,
1742
+ "avg_student_mask_ratio": 0.4605119539948646,
1743
+ "batch_ainp_frac": 0.0,
1744
+ "batch_inp_frac": 0.0,
1745
+ "batch_inp_oh_frac": 0.0,
1746
+ "batch_inp_par_frac": 1.0,
1747
+ "batch_inp_par_reverse_frac": 0.0,
1748
+ "batch_rl_frac": 0.0,
1749
+ "batch_sft_frac": 0.0,
1750
+ "batch_soft_sft_frac": 0.0,
1751
+ "batch_tf_frac": 0.0,
1752
+ "ce_loss": 0.0,
1753
+ "epoch": 1.5589333333333333,
1754
+ "grad_norm": 0.1103515625,
1755
+ "kd_loss": 0.16859328244926958,
1756
+ "learning_rate": 3e-06,
1757
+ "loss": 0.1779,
1758
+ "masked_tokens": 113.85,
1759
+ "mean_t": 0.49050743713742123,
1760
+ "step": 730,
1761
+ "student_masked_tokens": 113.85
1762
+ },
1763
+ {
1764
+ "avg_mask_ratio": 0.5134038798511028,
1765
+ "avg_response_length": 194.125,
1766
+ "avg_student_mask_ratio": 0.5134038798511028,
1767
+ "batch_ainp_frac": 0.0,
1768
+ "batch_inp_frac": 0.0,
1769
+ "batch_inp_oh_frac": 0.0,
1770
+ "batch_inp_par_frac": 1.0,
1771
+ "batch_inp_par_reverse_frac": 0.0,
1772
+ "batch_rl_frac": 0.0,
1773
+ "batch_sft_frac": 0.0,
1774
+ "batch_soft_sft_frac": 0.0,
1775
+ "batch_tf_frac": 0.0,
1776
+ "ce_loss": 0.0,
1777
+ "epoch": 1.5802666666666667,
1778
+ "grad_norm": 0.0830078125,
1779
+ "kd_loss": 0.17122714665274544,
1780
+ "learning_rate": 3e-06,
1781
+ "loss": 0.2018,
1782
+ "masked_tokens": 102.9375,
1783
+ "mean_t": 0.5505168779753149,
1784
+ "step": 740,
1785
+ "student_masked_tokens": 102.9375
1786
+ },
1787
+ {
1788
+ "avg_mask_ratio": 0.4201362137740944,
1789
+ "avg_response_length": 223.55,
1790
+ "avg_student_mask_ratio": 0.4201362137740944,
1791
+ "batch_ainp_frac": 0.0,
1792
+ "batch_inp_frac": 0.0,
1793
+ "batch_inp_oh_frac": 0.0,
1794
+ "batch_inp_par_frac": 1.0,
1795
+ "batch_inp_par_reverse_frac": 0.0,
1796
+ "batch_rl_frac": 0.0,
1797
+ "batch_sft_frac": 0.0,
1798
+ "batch_soft_sft_frac": 0.0,
1799
+ "batch_tf_frac": 0.0,
1800
+ "ce_loss": 0.0,
1801
+ "epoch": 1.6016,
1802
+ "grad_norm": 0.16796875,
1803
+ "kd_loss": 0.19197621339357057,
1804
+ "learning_rate": 3e-06,
1805
+ "loss": 0.1792,
1806
+ "masked_tokens": 94.7125,
1807
+ "mean_t": 0.4437690361432033,
1808
+ "step": 750,
1809
+ "student_masked_tokens": 94.7125
1810
+ },
1811
+ {
1812
+ "avg_mask_ratio": 0.46924527404480615,
1813
+ "avg_response_length": 227.35,
1814
+ "avg_student_mask_ratio": 0.46924527404480615,
1815
+ "batch_ainp_frac": 0.0,
1816
+ "batch_inp_frac": 0.0,
1817
+ "batch_inp_oh_frac": 0.0,
1818
+ "batch_inp_par_frac": 1.0,
1819
+ "batch_inp_par_reverse_frac": 0.0,
1820
+ "batch_rl_frac": 0.0,
1821
+ "batch_sft_frac": 0.0,
1822
+ "batch_soft_sft_frac": 0.0,
1823
+ "batch_tf_frac": 0.0,
1824
+ "ce_loss": 0.0,
1825
+ "epoch": 1.6229333333333333,
1826
+ "grad_norm": 0.8125,
1827
+ "kd_loss": 0.24938117066874382,
1828
+ "learning_rate": 3e-06,
1829
+ "loss": 0.2591,
1830
+ "masked_tokens": 107.5375,
1831
+ "mean_t": 0.49132869170280175,
1832
+ "step": 760,
1833
+ "student_masked_tokens": 107.5375
1834
+ },
1835
+ {
1836
+ "avg_mask_ratio": 0.445551612455165,
1837
+ "avg_response_length": 268.6625,
1838
+ "avg_student_mask_ratio": 0.445551612455165,
1839
+ "batch_ainp_frac": 0.0,
1840
+ "batch_inp_frac": 0.0,
1841
+ "batch_inp_oh_frac": 0.0,
1842
+ "batch_inp_par_frac": 1.0,
1843
+ "batch_inp_par_reverse_frac": 0.0,
1844
+ "batch_rl_frac": 0.0,
1845
+ "batch_sft_frac": 0.0,
1846
+ "batch_soft_sft_frac": 0.0,
1847
+ "batch_tf_frac": 0.0,
1848
+ "ce_loss": 0.0,
1849
+ "epoch": 1.6442666666666668,
1850
+ "grad_norm": 0.1435546875,
1851
+ "kd_loss": 0.1928954417056957,
1852
+ "learning_rate": 3e-06,
1853
+ "loss": 0.1563,
1854
+ "masked_tokens": 124.0875,
1855
+ "mean_t": 0.4667695587326307,
1856
+ "step": 770,
1857
+ "student_masked_tokens": 124.0875
1858
+ },
1859
+ {
1860
+ "avg_mask_ratio": 0.4321410794305848,
1861
+ "avg_response_length": 256.1625,
1862
+ "avg_student_mask_ratio": 0.4321410794305848,
1863
+ "batch_ainp_frac": 0.0,
1864
+ "batch_inp_frac": 0.0,
1865
+ "batch_inp_oh_frac": 0.0,
1866
+ "batch_inp_par_frac": 1.0,
1867
+ "batch_inp_par_reverse_frac": 0.0,
1868
+ "batch_rl_frac": 0.0,
1869
+ "batch_sft_frac": 0.0,
1870
+ "batch_soft_sft_frac": 0.0,
1871
+ "batch_tf_frac": 0.0,
1872
+ "ce_loss": 0.0,
1873
+ "epoch": 1.6656,
1874
+ "grad_norm": 0.2216796875,
1875
+ "kd_loss": 0.15059620087446887,
1876
+ "learning_rate": 3e-06,
1877
+ "loss": 0.1534,
1878
+ "masked_tokens": 117.7,
1879
+ "mean_t": 0.44942845597106496,
1880
+ "step": 780,
1881
+ "student_masked_tokens": 117.7
1882
+ },
1883
+ {
1884
+ "avg_mask_ratio": 0.4697655299096368,
1885
+ "avg_response_length": 240.6125,
1886
+ "avg_student_mask_ratio": 0.4697655299096368,
1887
+ "batch_ainp_frac": 0.0,
1888
+ "batch_inp_frac": 0.0,
1889
+ "batch_inp_oh_frac": 0.0,
1890
+ "batch_inp_par_frac": 1.0,
1891
+ "batch_inp_par_reverse_frac": 0.0,
1892
+ "batch_rl_frac": 0.0,
1893
+ "batch_sft_frac": 0.0,
1894
+ "batch_soft_sft_frac": 0.0,
1895
+ "batch_tf_frac": 0.0,
1896
+ "ce_loss": 0.0,
1897
+ "epoch": 1.6869333333333332,
1898
+ "grad_norm": 0.255859375,
1899
+ "kd_loss": 0.16427693545232777,
1900
+ "learning_rate": 3e-06,
1901
+ "loss": 0.1489,
1902
+ "masked_tokens": 113.5375,
1903
+ "mean_t": 0.4993515375303105,
1904
+ "step": 790,
1905
+ "student_masked_tokens": 113.5375
1906
+ },
1907
+ {
1908
+ "avg_mask_ratio": 0.4635992758907378,
1909
+ "avg_response_length": 224.15,
1910
+ "avg_student_mask_ratio": 0.4635992758907378,
1911
+ "batch_ainp_frac": 0.0,
1912
+ "batch_inp_frac": 0.0,
1913
+ "batch_inp_oh_frac": 0.0,
1914
+ "batch_inp_par_frac": 1.0,
1915
+ "batch_inp_par_reverse_frac": 0.0,
1916
+ "batch_rl_frac": 0.0,
1917
+ "batch_sft_frac": 0.0,
1918
+ "batch_soft_sft_frac": 0.0,
1919
+ "batch_tf_frac": 0.0,
1920
+ "ce_loss": 0.0,
1921
+ "epoch": 1.7082666666666668,
1922
+ "grad_norm": 0.197265625,
1923
+ "kd_loss": 0.15171801659575976,
1924
+ "learning_rate": 3e-06,
1925
+ "loss": 0.1526,
1926
+ "masked_tokens": 107.5,
1927
+ "mean_t": 0.500370389316231,
1928
+ "step": 800,
1929
+ "student_masked_tokens": 107.5
1930
+ },
1931
+ {
1932
+ "avg_mask_ratio": 0.5018501321552321,
1933
+ "avg_response_length": 235.1625,
1934
+ "avg_student_mask_ratio": 0.5018501321552321,
1935
+ "batch_ainp_frac": 0.0,
1936
+ "batch_inp_frac": 0.0,
1937
+ "batch_inp_oh_frac": 0.0,
1938
+ "batch_inp_par_frac": 1.0,
1939
+ "batch_inp_par_reverse_frac": 0.0,
1940
+ "batch_rl_frac": 0.0,
1941
+ "batch_sft_frac": 0.0,
1942
+ "batch_soft_sft_frac": 0.0,
1943
+ "batch_tf_frac": 0.0,
1944
+ "ce_loss": 0.0,
1945
+ "epoch": 1.7296,
1946
+ "grad_norm": 0.1630859375,
1947
+ "kd_loss": 0.18931926304685476,
1948
+ "learning_rate": 3e-06,
1949
+ "loss": 0.2031,
1950
+ "masked_tokens": 116.6125,
1951
+ "mean_t": 0.5317009104182944,
1952
+ "step": 810,
1953
+ "student_masked_tokens": 116.6125
1954
+ },
1955
+ {
1956
+ "avg_mask_ratio": 0.5050785383209586,
1957
+ "avg_response_length": 207.6875,
1958
+ "avg_student_mask_ratio": 0.5050785383209586,
1959
+ "batch_ainp_frac": 0.0,
1960
+ "batch_inp_frac": 0.0,
1961
+ "batch_inp_oh_frac": 0.0,
1962
+ "batch_inp_par_frac": 1.0,
1963
+ "batch_inp_par_reverse_frac": 0.0,
1964
+ "batch_rl_frac": 0.0,
1965
+ "batch_sft_frac": 0.0,
1966
+ "batch_soft_sft_frac": 0.0,
1967
+ "batch_tf_frac": 0.0,
1968
+ "ce_loss": 0.0,
1969
+ "epoch": 1.7509333333333332,
1970
+ "grad_norm": 0.5546875,
1971
+ "kd_loss": 0.24824176937403308,
1972
+ "learning_rate": 3e-06,
1973
+ "loss": 0.2566,
1974
+ "masked_tokens": 119.175,
1975
+ "mean_t": 0.5392061032878701,
1976
+ "step": 820,
1977
+ "student_masked_tokens": 119.175
1978
+ },
1979
+ {
1980
+ "avg_mask_ratio": 0.4980328972451389,
1981
+ "avg_response_length": 270.325,
1982
+ "avg_student_mask_ratio": 0.4980328972451389,
1983
+ "batch_ainp_frac": 0.0,
1984
+ "batch_inp_frac": 0.0,
1985
+ "batch_inp_oh_frac": 0.0,
1986
+ "batch_inp_par_frac": 1.0,
1987
+ "batch_inp_par_reverse_frac": 0.0,
1988
+ "batch_rl_frac": 0.0,
1989
+ "batch_sft_frac": 0.0,
1990
+ "batch_soft_sft_frac": 0.0,
1991
+ "batch_tf_frac": 0.0,
1992
+ "ce_loss": 0.0,
1993
+ "epoch": 1.7722666666666667,
1994
+ "grad_norm": 0.09912109375,
1995
+ "kd_loss": 0.1924194690429431,
1996
+ "learning_rate": 3e-06,
1997
+ "loss": 0.2006,
1998
+ "masked_tokens": 141.925,
1999
+ "mean_t": 0.5238314627087675,
2000
+ "step": 830,
2001
+ "student_masked_tokens": 141.925
2002
+ },
2003
+ {
2004
+ "avg_mask_ratio": 0.493249478796497,
2005
+ "avg_response_length": 226.025,
2006
+ "avg_student_mask_ratio": 0.493249478796497,
2007
+ "batch_ainp_frac": 0.0,
2008
+ "batch_inp_frac": 0.0,
2009
+ "batch_inp_oh_frac": 0.0,
2010
+ "batch_inp_par_frac": 1.0,
2011
+ "batch_inp_par_reverse_frac": 0.0,
2012
+ "batch_rl_frac": 0.0,
2013
+ "batch_sft_frac": 0.0,
2014
+ "batch_soft_sft_frac": 0.0,
2015
+ "batch_tf_frac": 0.0,
2016
+ "ce_loss": 0.0,
2017
+ "epoch": 1.7936,
2018
+ "grad_norm": 0.158203125,
2019
+ "kd_loss": 0.15751813794203606,
2020
+ "learning_rate": 3e-06,
2021
+ "loss": 0.1624,
2022
+ "masked_tokens": 118.825,
2023
+ "mean_t": 0.5301066277665086,
2024
+ "step": 840,
2025
+ "student_masked_tokens": 118.825
2026
+ },
2027
+ {
2028
+ "avg_mask_ratio": 0.5009000841644593,
2029
+ "avg_response_length": 233.025,
2030
+ "avg_student_mask_ratio": 0.5009000841644593,
2031
+ "batch_ainp_frac": 0.0,
2032
+ "batch_inp_frac": 0.0,
2033
+ "batch_inp_oh_frac": 0.0,
2034
+ "batch_inp_par_frac": 1.0,
2035
+ "batch_inp_par_reverse_frac": 0.0,
2036
+ "batch_rl_frac": 0.0,
2037
+ "batch_sft_frac": 0.0,
2038
+ "batch_soft_sft_frac": 0.0,
2039
+ "batch_tf_frac": 0.0,
2040
+ "ce_loss": 0.0,
2041
+ "epoch": 1.8149333333333333,
2042
+ "grad_norm": 0.10205078125,
2043
+ "kd_loss": 0.1860738446495816,
2044
+ "learning_rate": 3e-06,
2045
+ "loss": 0.1737,
2046
+ "masked_tokens": 117.3125,
2047
+ "mean_t": 0.5343429344706238,
2048
+ "step": 850,
2049
+ "student_masked_tokens": 117.3125
2050
+ },
2051
+ {
2052
+ "avg_mask_ratio": 0.46293387678451836,
2053
+ "avg_response_length": 231.3625,
2054
+ "avg_student_mask_ratio": 0.46293387678451836,
2055
+ "batch_ainp_frac": 0.0,
2056
+ "batch_inp_frac": 0.0,
2057
+ "batch_inp_oh_frac": 0.0,
2058
+ "batch_inp_par_frac": 1.0,
2059
+ "batch_inp_par_reverse_frac": 0.0,
2060
+ "batch_rl_frac": 0.0,
2061
+ "batch_sft_frac": 0.0,
2062
+ "batch_soft_sft_frac": 0.0,
2063
+ "batch_tf_frac": 0.0,
2064
+ "ce_loss": 0.0,
2065
+ "epoch": 1.8362666666666667,
2066
+ "grad_norm": 0.1103515625,
2067
+ "kd_loss": 0.19740513321539765,
2068
+ "learning_rate": 3e-06,
2069
+ "loss": 0.1841,
2070
+ "masked_tokens": 110.5625,
2071
+ "mean_t": 0.4791536889737472,
2072
+ "step": 860,
2073
+ "student_masked_tokens": 110.5625
2074
+ },
2075
+ {
2076
+ "avg_mask_ratio": 0.4846805231412873,
2077
+ "avg_response_length": 220.7375,
2078
+ "avg_student_mask_ratio": 0.4846805231412873,
2079
+ "batch_ainp_frac": 0.0,
2080
+ "batch_inp_frac": 0.0,
2081
+ "batch_inp_oh_frac": 0.0,
2082
+ "batch_inp_par_frac": 1.0,
2083
+ "batch_inp_par_reverse_frac": 0.0,
2084
+ "batch_rl_frac": 0.0,
2085
+ "batch_sft_frac": 0.0,
2086
+ "batch_soft_sft_frac": 0.0,
2087
+ "batch_tf_frac": 0.0,
2088
+ "ce_loss": 0.0,
2089
+ "epoch": 1.8576000000000001,
2090
+ "grad_norm": 0.228515625,
2091
+ "kd_loss": 0.19436422403705364,
2092
+ "learning_rate": 3e-06,
2093
+ "loss": 0.2012,
2094
+ "masked_tokens": 113.7625,
2095
+ "mean_t": 0.5203817339061061,
2096
+ "step": 870,
2097
+ "student_masked_tokens": 113.7625
2098
+ },
2099
+ {
2100
+ "avg_mask_ratio": 0.4508363194297999,
2101
+ "avg_response_length": 203.2875,
2102
+ "avg_student_mask_ratio": 0.4508363194297999,
2103
+ "batch_ainp_frac": 0.0,
2104
+ "batch_inp_frac": 0.0,
2105
+ "batch_inp_oh_frac": 0.0,
2106
+ "batch_inp_par_frac": 1.0,
2107
+ "batch_inp_par_reverse_frac": 0.0,
2108
+ "batch_rl_frac": 0.0,
2109
+ "batch_sft_frac": 0.0,
2110
+ "batch_soft_sft_frac": 0.0,
2111
+ "batch_tf_frac": 0.0,
2112
+ "ce_loss": 0.0,
2113
+ "epoch": 1.8789333333333333,
2114
+ "grad_norm": 0.1962890625,
2115
+ "kd_loss": 0.16288868402702406,
2116
+ "learning_rate": 3e-06,
2117
+ "loss": 0.1845,
2118
+ "masked_tokens": 95.3875,
2119
+ "mean_t": 0.4875184997683391,
2120
+ "step": 880,
2121
+ "student_masked_tokens": 95.3875
2122
+ },
2123
+ {
2124
+ "avg_mask_ratio": 0.43862658384023234,
2125
+ "avg_response_length": 227.9375,
2126
+ "avg_student_mask_ratio": 0.43862658384023234,
2127
+ "batch_ainp_frac": 0.0,
2128
+ "batch_inp_frac": 0.0,
2129
+ "batch_inp_oh_frac": 0.0,
2130
+ "batch_inp_par_frac": 1.0,
2131
+ "batch_inp_par_reverse_frac": 0.0,
2132
+ "batch_rl_frac": 0.0,
2133
+ "batch_sft_frac": 0.0,
2134
+ "batch_soft_sft_frac": 0.0,
2135
+ "batch_tf_frac": 0.0,
2136
+ "ce_loss": 0.0,
2137
+ "epoch": 1.9002666666666665,
2138
+ "grad_norm": 0.08251953125,
2139
+ "kd_loss": 0.11281866748422545,
2140
+ "learning_rate": 3e-06,
2141
+ "loss": 0.142,
2142
+ "masked_tokens": 101.2625,
2143
+ "mean_t": 0.4766692223958671,
2144
+ "step": 890,
2145
+ "student_masked_tokens": 101.2625
2146
+ },
2147
+ {
2148
+ "avg_mask_ratio": 0.44909207145101393,
2149
+ "avg_response_length": 237.2375,
2150
+ "avg_student_mask_ratio": 0.44909207145101393,
2151
+ "batch_ainp_frac": 0.0,
2152
+ "batch_inp_frac": 0.0,
2153
+ "batch_inp_oh_frac": 0.0,
2154
+ "batch_inp_par_frac": 1.0,
2155
+ "batch_inp_par_reverse_frac": 0.0,
2156
+ "batch_rl_frac": 0.0,
2157
+ "batch_sft_frac": 0.0,
2158
+ "batch_soft_sft_frac": 0.0,
2159
+ "batch_tf_frac": 0.0,
2160
+ "ce_loss": 0.0,
2161
+ "epoch": 1.9216,
2162
+ "grad_norm": 0.1591796875,
2163
+ "kd_loss": 0.15684176656744747,
2164
+ "learning_rate": 3e-06,
2165
+ "loss": 0.1737,
2166
+ "masked_tokens": 103.7,
2167
+ "mean_t": 0.487134758150205,
2168
+ "step": 900,
2169
+ "student_masked_tokens": 103.7
2170
+ },
2171
+ {
2172
+ "avg_mask_ratio": 0.47512493645772336,
2173
+ "avg_response_length": 246.175,
2174
+ "avg_student_mask_ratio": 0.47512493645772336,
2175
+ "batch_ainp_frac": 0.0,
2176
+ "batch_inp_frac": 0.0,
2177
+ "batch_inp_oh_frac": 0.0,
2178
+ "batch_inp_par_frac": 1.0,
2179
+ "batch_inp_par_reverse_frac": 0.0,
2180
+ "batch_rl_frac": 0.0,
2181
+ "batch_sft_frac": 0.0,
2182
+ "batch_soft_sft_frac": 0.0,
2183
+ "batch_tf_frac": 0.0,
2184
+ "ce_loss": 0.0,
2185
+ "epoch": 1.9429333333333334,
2186
+ "grad_norm": 0.1982421875,
2187
+ "kd_loss": 0.16909069961529893,
2188
+ "learning_rate": 3e-06,
2189
+ "loss": 0.1717,
2190
+ "masked_tokens": 117.2125,
2191
+ "mean_t": 0.5027793228859082,
2192
+ "step": 910,
2193
+ "student_masked_tokens": 117.2125
2194
+ },
2195
+ {
2196
+ "avg_mask_ratio": 0.46720519906957636,
2197
+ "avg_response_length": 229.0875,
2198
+ "avg_student_mask_ratio": 0.46720519906957636,
2199
+ "batch_ainp_frac": 0.0,
2200
+ "batch_inp_frac": 0.0,
2201
+ "batch_inp_oh_frac": 0.0,
2202
+ "batch_inp_par_frac": 1.0,
2203
+ "batch_inp_par_reverse_frac": 0.0,
2204
+ "batch_rl_frac": 0.0,
2205
+ "batch_sft_frac": 0.0,
2206
+ "batch_soft_sft_frac": 0.0,
2207
+ "batch_tf_frac": 0.0,
2208
+ "ce_loss": 0.0,
2209
+ "epoch": 1.9642666666666666,
2210
+ "grad_norm": 0.203125,
2211
+ "kd_loss": 0.17078794931742322,
2212
+ "learning_rate": 3e-06,
2213
+ "loss": 0.1844,
2214
+ "masked_tokens": 115.325,
2215
+ "mean_t": 0.49417946098838,
2216
+ "step": 920,
2217
+ "student_masked_tokens": 115.325
2218
+ },
2219
+ {
2220
+ "avg_mask_ratio": 0.4805813999380916,
2221
+ "avg_response_length": 236.1625,
2222
+ "avg_student_mask_ratio": 0.4805813999380916,
2223
+ "batch_ainp_frac": 0.0,
2224
+ "batch_inp_frac": 0.0,
2225
+ "batch_inp_oh_frac": 0.0,
2226
+ "batch_inp_par_frac": 1.0,
2227
+ "batch_inp_par_reverse_frac": 0.0,
2228
+ "batch_rl_frac": 0.0,
2229
+ "batch_sft_frac": 0.0,
2230
+ "batch_soft_sft_frac": 0.0,
2231
+ "batch_tf_frac": 0.0,
2232
+ "ce_loss": 0.0,
2233
+ "epoch": 1.9856,
2234
+ "grad_norm": 0.21484375,
2235
+ "kd_loss": 0.20637793739751942,
2236
+ "learning_rate": 3e-06,
2237
+ "loss": 0.176,
2238
+ "masked_tokens": 119.65,
2239
+ "mean_t": 0.5045580042526125,
2240
+ "step": 930,
2241
+ "student_masked_tokens": 119.65
2242
+ },
2243
+ {
2244
+ "avg_mask_ratio": 0.49833715503059683,
2245
+ "avg_response_length": 229.39285714285714,
2246
+ "avg_student_mask_ratio": 0.49833715503059683,
2247
+ "batch_ainp_frac": 0.0,
2248
+ "batch_inp_frac": 0.0,
2249
+ "batch_inp_oh_frac": 0.0,
2250
+ "batch_inp_par_frac": 1.0,
2251
+ "batch_inp_par_reverse_frac": 0.0,
2252
+ "batch_rl_frac": 0.0,
2253
+ "batch_sft_frac": 0.0,
2254
+ "batch_soft_sft_frac": 0.0,
2255
+ "batch_tf_frac": 0.0,
2256
+ "ce_loss": 0.0,
2257
+ "epoch": 2.0085333333333333,
2258
+ "grad_norm": 0.259765625,
2259
+ "kd_loss": 0.2117233988543549,
2260
+ "learning_rate": 3e-06,
2261
+ "loss": 0.2023,
2262
+ "masked_tokens": 121.5952380952381,
2263
+ "mean_t": 0.5321138524893849,
2264
+ "step": 940,
2265
+ "student_masked_tokens": 121.5952380952381
2266
+ },
2267
+ {
2268
+ "avg_mask_ratio": 0.4347397161007393,
2269
+ "avg_response_length": 226.925,
2270
+ "avg_student_mask_ratio": 0.4347397161007393,
2271
+ "batch_ainp_frac": 0.0,
2272
+ "batch_inp_frac": 0.0,
2273
+ "batch_inp_oh_frac": 0.0,
2274
+ "batch_inp_par_frac": 1.0,
2275
+ "batch_inp_par_reverse_frac": 0.0,
2276
+ "batch_rl_frac": 0.0,
2277
+ "batch_sft_frac": 0.0,
2278
+ "batch_soft_sft_frac": 0.0,
2279
+ "batch_tf_frac": 0.0,
2280
+ "ce_loss": 0.0,
2281
+ "epoch": 2.0298666666666665,
2282
+ "grad_norm": 0.15234375,
2283
+ "kd_loss": 0.16830119832033005,
2284
+ "learning_rate": 3e-06,
2285
+ "loss": 0.1549,
2286
+ "masked_tokens": 103.4125,
2287
+ "mean_t": 0.4632946296595037,
2288
+ "step": 950,
2289
+ "student_masked_tokens": 103.4125
2290
+ },
2291
+ {
2292
+ "avg_mask_ratio": 0.4989688004134223,
2293
+ "avg_response_length": 273.3125,
2294
+ "avg_student_mask_ratio": 0.4989688004134223,
2295
+ "batch_ainp_frac": 0.0,
2296
+ "batch_inp_frac": 0.0,
2297
+ "batch_inp_oh_frac": 0.0,
2298
+ "batch_inp_par_frac": 1.0,
2299
+ "batch_inp_par_reverse_frac": 0.0,
2300
+ "batch_rl_frac": 0.0,
2301
+ "batch_sft_frac": 0.0,
2302
+ "batch_soft_sft_frac": 0.0,
2303
+ "batch_tf_frac": 0.0,
2304
+ "ce_loss": 0.0,
2305
+ "epoch": 2.0512,
2306
+ "grad_norm": 0.123046875,
2307
+ "kd_loss": 0.2304358719712809,
2308
+ "learning_rate": 3e-06,
2309
+ "loss": 0.22,
2310
+ "masked_tokens": 137.125,
2311
+ "mean_t": 0.5198000721400604,
2312
+ "step": 960,
2313
+ "student_masked_tokens": 137.125
2314
+ },
2315
+ {
2316
+ "avg_mask_ratio": 0.4374056361732073,
2317
+ "avg_response_length": 239.125,
2318
+ "avg_student_mask_ratio": 0.4374056361732073,
2319
+ "batch_ainp_frac": 0.0,
2320
+ "batch_inp_frac": 0.0,
2321
+ "batch_inp_oh_frac": 0.0,
2322
+ "batch_inp_par_frac": 1.0,
2323
+ "batch_inp_par_reverse_frac": 0.0,
2324
+ "batch_rl_frac": 0.0,
2325
+ "batch_sft_frac": 0.0,
2326
+ "batch_soft_sft_frac": 0.0,
2327
+ "batch_tf_frac": 0.0,
2328
+ "ce_loss": 0.0,
2329
+ "epoch": 2.0725333333333333,
2330
+ "grad_norm": 0.1962890625,
2331
+ "kd_loss": 0.158095879086882,
2332
+ "learning_rate": 3e-06,
2333
+ "loss": 0.1573,
2334
+ "masked_tokens": 107.4625,
2335
+ "mean_t": 0.4703940597362816,
2336
+ "step": 970,
2337
+ "student_masked_tokens": 107.4625
2338
+ },
2339
+ {
2340
+ "avg_mask_ratio": 0.4216216350672767,
2341
+ "avg_response_length": 236.0125,
2342
+ "avg_student_mask_ratio": 0.4216216350672767,
2343
+ "batch_ainp_frac": 0.0,
2344
+ "batch_inp_frac": 0.0,
2345
+ "batch_inp_oh_frac": 0.0,
2346
+ "batch_inp_par_frac": 1.0,
2347
+ "batch_inp_par_reverse_frac": 0.0,
2348
+ "batch_rl_frac": 0.0,
2349
+ "batch_sft_frac": 0.0,
2350
+ "batch_soft_sft_frac": 0.0,
2351
+ "batch_tf_frac": 0.0,
2352
+ "ce_loss": 0.0,
2353
+ "epoch": 2.0938666666666665,
2354
+ "grad_norm": 0.177734375,
2355
+ "kd_loss": 0.15213593625503563,
2356
+ "learning_rate": 3e-06,
2357
+ "loss": 0.1305,
2358
+ "masked_tokens": 103.05,
2359
+ "mean_t": 0.4511947895749472,
2360
+ "step": 980,
2361
+ "student_masked_tokens": 103.05
2362
+ },
2363
+ {
2364
+ "avg_mask_ratio": 0.46076959141064433,
2365
+ "avg_response_length": 257.375,
2366
+ "avg_student_mask_ratio": 0.46076959141064433,
2367
+ "batch_ainp_frac": 0.0,
2368
+ "batch_inp_frac": 0.0,
2369
+ "batch_inp_oh_frac": 0.0,
2370
+ "batch_inp_par_frac": 1.0,
2371
+ "batch_inp_par_reverse_frac": 0.0,
2372
+ "batch_rl_frac": 0.0,
2373
+ "batch_sft_frac": 0.0,
2374
+ "batch_soft_sft_frac": 0.0,
2375
+ "batch_tf_frac": 0.0,
2376
+ "ce_loss": 0.0,
2377
+ "epoch": 2.1152,
2378
+ "grad_norm": 0.1484375,
2379
+ "kd_loss": 0.18778659292394478,
2380
+ "learning_rate": 3e-06,
2381
+ "loss": 0.1727,
2382
+ "masked_tokens": 121.2125,
2383
+ "mean_t": 0.4923786667350214,
2384
+ "step": 990,
2385
+ "student_masked_tokens": 121.2125
2386
+ },
2387
+ {
2388
+ "avg_mask_ratio": 0.45376153094694016,
2389
+ "avg_response_length": 223.4125,
2390
+ "avg_student_mask_ratio": 0.45376153094694016,
2391
+ "batch_ainp_frac": 0.0,
2392
+ "batch_inp_frac": 0.0,
2393
+ "batch_inp_oh_frac": 0.0,
2394
+ "batch_inp_par_frac": 1.0,
2395
+ "batch_inp_par_reverse_frac": 0.0,
2396
+ "batch_rl_frac": 0.0,
2397
+ "batch_sft_frac": 0.0,
2398
+ "batch_soft_sft_frac": 0.0,
2399
+ "batch_tf_frac": 0.0,
2400
+ "ce_loss": 0.0,
2401
+ "epoch": 2.1365333333333334,
2402
+ "grad_norm": 0.10498046875,
2403
+ "kd_loss": 0.16195435594947866,
2404
+ "learning_rate": 3e-06,
2405
+ "loss": 0.1554,
2406
+ "masked_tokens": 102.175,
2407
+ "mean_t": 0.4773523230338469,
2408
+ "step": 1000,
2409
+ "student_masked_tokens": 102.175
2410
+ },
2411
+ {
2412
+ "avg_mask_ratio": 0.4360893844394013,
2413
+ "avg_response_length": 230.8375,
2414
+ "avg_student_mask_ratio": 0.4360893844394013,
2415
+ "batch_ainp_frac": 0.0,
2416
+ "batch_inp_frac": 0.0,
2417
+ "batch_inp_oh_frac": 0.0,
2418
+ "batch_inp_par_frac": 1.0,
2419
+ "batch_inp_par_reverse_frac": 0.0,
2420
+ "batch_rl_frac": 0.0,
2421
+ "batch_sft_frac": 0.0,
2422
+ "batch_soft_sft_frac": 0.0,
2423
+ "batch_tf_frac": 0.0,
2424
+ "ce_loss": 0.0,
2425
+ "epoch": 2.1578666666666666,
2426
+ "grad_norm": 0.09130859375,
2427
+ "kd_loss": 0.15129900982686878,
2428
+ "learning_rate": 3e-06,
2429
+ "loss": 0.1525,
2430
+ "masked_tokens": 95.2875,
2431
+ "mean_t": 0.4648138735938119,
2432
+ "step": 1010,
2433
+ "student_masked_tokens": 95.2875
2434
+ },
2435
+ {
2436
+ "avg_mask_ratio": 0.5085378198185936,
2437
+ "avg_response_length": 228.975,
2438
+ "avg_student_mask_ratio": 0.5085378198185936,
2439
+ "batch_ainp_frac": 0.0,
2440
+ "batch_inp_frac": 0.0,
2441
+ "batch_inp_oh_frac": 0.0,
2442
+ "batch_inp_par_frac": 1.0,
2443
+ "batch_inp_par_reverse_frac": 0.0,
2444
+ "batch_rl_frac": 0.0,
2445
+ "batch_sft_frac": 0.0,
2446
+ "batch_soft_sft_frac": 0.0,
2447
+ "batch_tf_frac": 0.0,
2448
+ "ce_loss": 0.0,
2449
+ "epoch": 2.1792,
2450
+ "grad_norm": 0.1279296875,
2451
+ "kd_loss": 0.1835148020709127,
2452
+ "learning_rate": 3e-06,
2453
+ "loss": 0.2059,
2454
+ "masked_tokens": 118.825,
2455
+ "mean_t": 0.5327763411332853,
2456
+ "step": 1020,
2457
+ "student_masked_tokens": 118.825
2458
+ },
2459
+ {
2460
+ "avg_mask_ratio": 0.4720251938910224,
2461
+ "avg_response_length": 237.7125,
2462
+ "avg_student_mask_ratio": 0.4720251938910224,
2463
+ "batch_ainp_frac": 0.0,
2464
+ "batch_inp_frac": 0.0,
2465
+ "batch_inp_oh_frac": 0.0,
2466
+ "batch_inp_par_frac": 1.0,
2467
+ "batch_inp_par_reverse_frac": 0.0,
2468
+ "batch_rl_frac": 0.0,
2469
+ "batch_sft_frac": 0.0,
2470
+ "batch_soft_sft_frac": 0.0,
2471
+ "batch_tf_frac": 0.0,
2472
+ "ce_loss": 0.0,
2473
+ "epoch": 2.2005333333333335,
2474
+ "grad_norm": 0.27734375,
2475
+ "kd_loss": 0.19297441139430588,
2476
+ "learning_rate": 3e-06,
2477
+ "loss": 0.195,
2478
+ "masked_tokens": 113.1,
2479
+ "mean_t": 0.5033508580760099,
2480
+ "step": 1030,
2481
+ "student_masked_tokens": 113.1
2482
+ },
2483
+ {
2484
+ "avg_mask_ratio": 0.4877920758561231,
2485
+ "avg_response_length": 206.2875,
2486
+ "avg_student_mask_ratio": 0.4877920758561231,
2487
+ "batch_ainp_frac": 0.0,
2488
+ "batch_inp_frac": 0.0,
2489
+ "batch_inp_oh_frac": 0.0,
2490
+ "batch_inp_par_frac": 1.0,
2491
+ "batch_inp_par_reverse_frac": 0.0,
2492
+ "batch_rl_frac": 0.0,
2493
+ "batch_sft_frac": 0.0,
2494
+ "batch_soft_sft_frac": 0.0,
2495
+ "batch_tf_frac": 0.0,
2496
+ "ce_loss": 0.0,
2497
+ "epoch": 2.2218666666666667,
2498
+ "grad_norm": 0.1865234375,
2499
+ "kd_loss": 0.18438395577541086,
2500
+ "learning_rate": 3e-06,
2501
+ "loss": 0.1806,
2502
+ "masked_tokens": 99.8625,
2503
+ "mean_t": 0.5349024560535327,
2504
+ "step": 1040,
2505
+ "student_masked_tokens": 99.8625
2506
+ },
2507
+ {
2508
+ "avg_mask_ratio": 0.5081706921104342,
2509
+ "avg_response_length": 237.05,
2510
+ "avg_student_mask_ratio": 0.5081706921104342,
2511
+ "batch_ainp_frac": 0.0,
2512
+ "batch_inp_frac": 0.0,
2513
+ "batch_inp_oh_frac": 0.0,
2514
+ "batch_inp_par_frac": 1.0,
2515
+ "batch_inp_par_reverse_frac": 0.0,
2516
+ "batch_rl_frac": 0.0,
2517
+ "batch_sft_frac": 0.0,
2518
+ "batch_soft_sft_frac": 0.0,
2519
+ "batch_tf_frac": 0.0,
2520
+ "ce_loss": 0.0,
2521
+ "epoch": 2.2432,
2522
+ "grad_norm": 0.158203125,
2523
+ "kd_loss": 0.1950792422781717,
2524
+ "learning_rate": 3e-06,
2525
+ "loss": 0.1941,
2526
+ "masked_tokens": 122.7375,
2527
+ "mean_t": 0.5457118917722255,
2528
+ "step": 1050,
2529
+ "student_masked_tokens": 122.7375
2530
+ },
2531
+ {
2532
+ "avg_mask_ratio": 0.4656533743953332,
2533
+ "avg_response_length": 259.25,
2534
+ "avg_student_mask_ratio": 0.4656533743953332,
2535
+ "batch_ainp_frac": 0.0,
2536
+ "batch_inp_frac": 0.0,
2537
+ "batch_inp_oh_frac": 0.0,
2538
+ "batch_inp_par_frac": 1.0,
2539
+ "batch_inp_par_reverse_frac": 0.0,
2540
+ "batch_rl_frac": 0.0,
2541
+ "batch_sft_frac": 0.0,
2542
+ "batch_soft_sft_frac": 0.0,
2543
+ "batch_tf_frac": 0.0,
2544
+ "ce_loss": 0.0,
2545
+ "epoch": 2.2645333333333335,
2546
+ "grad_norm": 0.1552734375,
2547
+ "kd_loss": 0.14457174434564876,
2548
+ "learning_rate": 3e-06,
2549
+ "loss": 0.1245,
2550
+ "masked_tokens": 124.0875,
2551
+ "mean_t": 0.48194136443780733,
2552
+ "step": 1060,
2553
+ "student_masked_tokens": 124.0875
2554
+ },
2555
+ {
2556
+ "avg_mask_ratio": 0.4821103142516222,
2557
+ "avg_response_length": 265.35,
2558
+ "avg_student_mask_ratio": 0.4821103142516222,
2559
+ "batch_ainp_frac": 0.0,
2560
+ "batch_inp_frac": 0.0,
2561
+ "batch_inp_oh_frac": 0.0,
2562
+ "batch_inp_par_frac": 1.0,
2563
+ "batch_inp_par_reverse_frac": 0.0,
2564
+ "batch_rl_frac": 0.0,
2565
+ "batch_sft_frac": 0.0,
2566
+ "batch_soft_sft_frac": 0.0,
2567
+ "batch_tf_frac": 0.0,
2568
+ "ce_loss": 0.0,
2569
+ "epoch": 2.2858666666666667,
2570
+ "grad_norm": 0.146484375,
2571
+ "kd_loss": 0.1856000915321232,
2572
+ "learning_rate": 3e-06,
2573
+ "loss": 0.1583,
2574
+ "masked_tokens": 142.1875,
2575
+ "mean_t": 0.5015889146190602,
2576
+ "step": 1070,
2577
+ "student_masked_tokens": 142.1875
2578
+ },
2579
+ {
2580
+ "avg_mask_ratio": 0.461884257895872,
2581
+ "avg_response_length": 232.4,
2582
+ "avg_student_mask_ratio": 0.461884257895872,
2583
+ "batch_ainp_frac": 0.0,
2584
+ "batch_inp_frac": 0.0,
2585
+ "batch_inp_oh_frac": 0.0,
2586
+ "batch_inp_par_frac": 1.0,
2587
+ "batch_inp_par_reverse_frac": 0.0,
2588
+ "batch_rl_frac": 0.0,
2589
+ "batch_sft_frac": 0.0,
2590
+ "batch_soft_sft_frac": 0.0,
2591
+ "batch_tf_frac": 0.0,
2592
+ "ce_loss": 0.0,
2593
+ "epoch": 2.3072,
2594
+ "grad_norm": 0.09033203125,
2595
+ "kd_loss": 0.16016541003627935,
2596
+ "learning_rate": 3e-06,
2597
+ "loss": 0.1633,
2598
+ "masked_tokens": 106.725,
2599
+ "mean_t": 0.4983203248586506,
2600
+ "step": 1080,
2601
+ "student_masked_tokens": 106.725
2602
+ },
2603
+ {
2604
+ "avg_mask_ratio": 0.4401097826776095,
2605
+ "avg_response_length": 207.325,
2606
+ "avg_student_mask_ratio": 0.4401097826776095,
2607
+ "batch_ainp_frac": 0.0,
2608
+ "batch_inp_frac": 0.0,
2609
+ "batch_inp_oh_frac": 0.0,
2610
+ "batch_inp_par_frac": 1.0,
2611
+ "batch_inp_par_reverse_frac": 0.0,
2612
+ "batch_rl_frac": 0.0,
2613
+ "batch_sft_frac": 0.0,
2614
+ "batch_soft_sft_frac": 0.0,
2615
+ "batch_tf_frac": 0.0,
2616
+ "ce_loss": 0.0,
2617
+ "epoch": 2.3285333333333336,
2618
+ "grad_norm": 0.0908203125,
2619
+ "kd_loss": 0.1443735401110633,
2620
+ "learning_rate": 3e-06,
2621
+ "loss": 0.1676,
2622
+ "masked_tokens": 88.5375,
2623
+ "mean_t": 0.47094749807147307,
2624
+ "step": 1090,
2625
+ "student_masked_tokens": 88.5375
2626
+ },
2627
+ {
2628
+ "avg_mask_ratio": 0.5224281516042538,
2629
+ "avg_response_length": 257.675,
2630
+ "avg_student_mask_ratio": 0.5224281516042538,
2631
+ "batch_ainp_frac": 0.0,
2632
+ "batch_inp_frac": 0.0,
2633
+ "batch_inp_oh_frac": 0.0,
2634
+ "batch_inp_par_frac": 1.0,
2635
+ "batch_inp_par_reverse_frac": 0.0,
2636
+ "batch_rl_frac": 0.0,
2637
+ "batch_sft_frac": 0.0,
2638
+ "batch_soft_sft_frac": 0.0,
2639
+ "batch_tf_frac": 0.0,
2640
+ "ce_loss": 0.0,
2641
+ "epoch": 2.3498666666666668,
2642
+ "grad_norm": 0.1953125,
2643
+ "kd_loss": 0.18995505797054618,
2644
+ "learning_rate": 3e-06,
2645
+ "loss": 0.2181,
2646
+ "masked_tokens": 142.15,
2647
+ "mean_t": 0.5531192034482956,
2648
+ "step": 1100,
2649
+ "student_masked_tokens": 142.15
2650
+ },
2651
+ {
2652
+ "avg_mask_ratio": 0.4417481235635933,
2653
+ "avg_response_length": 225.2625,
2654
+ "avg_student_mask_ratio": 0.4417481235635933,
2655
+ "batch_ainp_frac": 0.0,
2656
+ "batch_inp_frac": 0.0,
2657
+ "batch_inp_oh_frac": 0.0,
2658
+ "batch_inp_par_frac": 1.0,
2659
+ "batch_inp_par_reverse_frac": 0.0,
2660
+ "batch_rl_frac": 0.0,
2661
+ "batch_sft_frac": 0.0,
2662
+ "batch_soft_sft_frac": 0.0,
2663
+ "batch_tf_frac": 0.0,
2664
+ "ce_loss": 0.0,
2665
+ "epoch": 2.3712,
2666
+ "grad_norm": 0.56640625,
2667
+ "kd_loss": 0.14014514066888067,
2668
+ "learning_rate": 3e-06,
2669
+ "loss": 0.1626,
2670
+ "masked_tokens": 101.725,
2671
+ "mean_t": 0.4757364276825683,
2672
+ "step": 1110,
2673
+ "student_masked_tokens": 101.725
2674
+ },
2675
+ {
2676
+ "avg_mask_ratio": 0.4719216389814392,
2677
+ "avg_response_length": 245.7875,
2678
+ "avg_student_mask_ratio": 0.4719216389814392,
2679
+ "batch_ainp_frac": 0.0,
2680
+ "batch_inp_frac": 0.0,
2681
+ "batch_inp_oh_frac": 0.0,
2682
+ "batch_inp_par_frac": 1.0,
2683
+ "batch_inp_par_reverse_frac": 0.0,
2684
+ "batch_rl_frac": 0.0,
2685
+ "batch_sft_frac": 0.0,
2686
+ "batch_soft_sft_frac": 0.0,
2687
+ "batch_tf_frac": 0.0,
2688
+ "ce_loss": 0.0,
2689
+ "epoch": 2.392533333333333,
2690
+ "grad_norm": 0.2109375,
2691
+ "kd_loss": 0.16249904381531924,
2692
+ "learning_rate": 3e-06,
2693
+ "loss": 0.1673,
2694
+ "masked_tokens": 112.05,
2695
+ "mean_t": 0.5013068238971755,
2696
+ "step": 1120,
2697
+ "student_masked_tokens": 112.05
2698
+ },
2699
+ {
2700
+ "avg_mask_ratio": 0.5034189606667496,
2701
+ "avg_response_length": 249.6625,
2702
+ "avg_student_mask_ratio": 0.5034189606667496,
2703
+ "batch_ainp_frac": 0.0,
2704
+ "batch_inp_frac": 0.0,
2705
+ "batch_inp_oh_frac": 0.0,
2706
+ "batch_inp_par_frac": 1.0,
2707
+ "batch_inp_par_reverse_frac": 0.0,
2708
+ "batch_rl_frac": 0.0,
2709
+ "batch_sft_frac": 0.0,
2710
+ "batch_soft_sft_frac": 0.0,
2711
+ "batch_tf_frac": 0.0,
2712
+ "ce_loss": 0.0,
2713
+ "epoch": 2.413866666666667,
2714
+ "grad_norm": 0.28125,
2715
+ "kd_loss": 0.18220705530329723,
2716
+ "learning_rate": 3e-06,
2717
+ "loss": 0.1841,
2718
+ "masked_tokens": 129.0125,
2719
+ "mean_t": 0.5303254407714121,
2720
+ "step": 1130,
2721
+ "student_masked_tokens": 129.0125
2722
+ },
2723
+ {
2724
+ "avg_mask_ratio": 0.4614338358165696,
2725
+ "avg_response_length": 210.7125,
2726
+ "avg_student_mask_ratio": 0.4614338358165696,
2727
+ "batch_ainp_frac": 0.0,
2728
+ "batch_inp_frac": 0.0,
2729
+ "batch_inp_oh_frac": 0.0,
2730
+ "batch_inp_par_frac": 1.0,
2731
+ "batch_inp_par_reverse_frac": 0.0,
2732
+ "batch_rl_frac": 0.0,
2733
+ "batch_sft_frac": 0.0,
2734
+ "batch_soft_sft_frac": 0.0,
2735
+ "batch_tf_frac": 0.0,
2736
+ "ce_loss": 0.0,
2737
+ "epoch": 2.4352,
2738
+ "grad_norm": 0.41015625,
2739
+ "kd_loss": 0.15378471011499642,
2740
+ "learning_rate": 3e-06,
2741
+ "loss": 0.1687,
2742
+ "masked_tokens": 102.1875,
2743
+ "mean_t": 0.4845335395424627,
2744
+ "step": 1140,
2745
+ "student_masked_tokens": 102.1875
2746
+ },
2747
+ {
2748
+ "avg_mask_ratio": 0.5357818282442167,
2749
+ "avg_response_length": 229.425,
2750
+ "avg_student_mask_ratio": 0.5357818282442167,
2751
+ "batch_ainp_frac": 0.0,
2752
+ "batch_inp_frac": 0.0,
2753
+ "batch_inp_oh_frac": 0.0,
2754
+ "batch_inp_par_frac": 1.0,
2755
+ "batch_inp_par_reverse_frac": 0.0,
2756
+ "batch_rl_frac": 0.0,
2757
+ "batch_sft_frac": 0.0,
2758
+ "batch_soft_sft_frac": 0.0,
2759
+ "batch_tf_frac": 0.0,
2760
+ "ce_loss": 0.0,
2761
+ "epoch": 2.4565333333333332,
2762
+ "grad_norm": 0.2197265625,
2763
+ "kd_loss": 0.2261572911098483,
2764
+ "learning_rate": 3e-06,
2765
+ "loss": 0.2287,
2766
+ "masked_tokens": 123.675,
2767
+ "mean_t": 0.5690932425903157,
2768
+ "step": 1150,
2769
+ "student_masked_tokens": 123.675
2770
+ },
2771
+ {
2772
+ "avg_mask_ratio": 0.47289583514211697,
2773
+ "avg_response_length": 244.3,
2774
+ "avg_student_mask_ratio": 0.47289583514211697,
2775
+ "batch_ainp_frac": 0.0,
2776
+ "batch_inp_frac": 0.0,
2777
+ "batch_inp_oh_frac": 0.0,
2778
+ "batch_inp_par_frac": 1.0,
2779
+ "batch_inp_par_reverse_frac": 0.0,
2780
+ "batch_rl_frac": 0.0,
2781
+ "batch_sft_frac": 0.0,
2782
+ "batch_soft_sft_frac": 0.0,
2783
+ "batch_tf_frac": 0.0,
2784
+ "ce_loss": 0.0,
2785
+ "epoch": 2.4778666666666664,
2786
+ "grad_norm": 0.205078125,
2787
+ "kd_loss": 0.1848451584734164,
2788
+ "learning_rate": 3e-06,
2789
+ "loss": 0.158,
2790
+ "masked_tokens": 107.175,
2791
+ "mean_t": 0.5040684466948733,
2792
+ "step": 1160,
2793
+ "student_masked_tokens": 107.175
2794
+ },
2795
+ {
2796
+ "avg_mask_ratio": 0.48331039408221843,
2797
+ "avg_response_length": 254.9,
2798
+ "avg_student_mask_ratio": 0.48331039408221843,
2799
+ "batch_ainp_frac": 0.0,
2800
+ "batch_inp_frac": 0.0,
2801
+ "batch_inp_oh_frac": 0.0,
2802
+ "batch_inp_par_frac": 1.0,
2803
+ "batch_inp_par_reverse_frac": 0.0,
2804
+ "batch_rl_frac": 0.0,
2805
+ "batch_sft_frac": 0.0,
2806
+ "batch_soft_sft_frac": 0.0,
2807
+ "batch_tf_frac": 0.0,
2808
+ "ce_loss": 0.0,
2809
+ "epoch": 2.4992,
2810
+ "grad_norm": 0.27734375,
2811
+ "kd_loss": 0.18950699392153184,
2812
+ "learning_rate": 3e-06,
2813
+ "loss": 0.1895,
2814
+ "masked_tokens": 120.575,
2815
+ "mean_t": 0.5114516971167177,
2816
+ "step": 1170,
2817
+ "student_masked_tokens": 120.575
2818
+ },
2819
+ {
2820
+ "avg_mask_ratio": 0.4191652584122494,
2821
+ "avg_response_length": 214.3,
2822
+ "avg_student_mask_ratio": 0.4191652584122494,
2823
+ "batch_ainp_frac": 0.0,
2824
+ "batch_inp_frac": 0.0,
2825
+ "batch_inp_oh_frac": 0.0,
2826
+ "batch_inp_par_frac": 1.0,
2827
+ "batch_inp_par_reverse_frac": 0.0,
2828
+ "batch_rl_frac": 0.0,
2829
+ "batch_sft_frac": 0.0,
2830
+ "batch_soft_sft_frac": 0.0,
2831
+ "batch_tf_frac": 0.0,
2832
+ "ce_loss": 0.0,
2833
+ "epoch": 2.5205333333333333,
2834
+ "grad_norm": 0.16015625,
2835
+ "kd_loss": 0.12068161078326511,
2836
+ "learning_rate": 3e-06,
2837
+ "loss": 0.143,
2838
+ "masked_tokens": 86.8625,
2839
+ "mean_t": 0.4491677140351385,
2840
+ "step": 1180,
2841
+ "student_masked_tokens": 86.8625
2842
+ },
2843
+ {
2844
+ "avg_mask_ratio": 0.520819503068924,
2845
+ "avg_response_length": 234.775,
2846
+ "avg_student_mask_ratio": 0.520819503068924,
2847
+ "batch_ainp_frac": 0.0,
2848
+ "batch_inp_frac": 0.0,
2849
+ "batch_inp_oh_frac": 0.0,
2850
+ "batch_inp_par_frac": 1.0,
2851
+ "batch_inp_par_reverse_frac": 0.0,
2852
+ "batch_rl_frac": 0.0,
2853
+ "batch_sft_frac": 0.0,
2854
+ "batch_soft_sft_frac": 0.0,
2855
+ "batch_tf_frac": 0.0,
2856
+ "ce_loss": 0.0,
2857
+ "epoch": 2.5418666666666665,
2858
+ "grad_norm": 0.271484375,
2859
+ "kd_loss": 0.2008784212745013,
2860
+ "learning_rate": 3e-06,
2861
+ "loss": 0.2037,
2862
+ "masked_tokens": 131.8,
2863
+ "mean_t": 0.5590635397238657,
2864
+ "step": 1190,
2865
+ "student_masked_tokens": 131.8
2866
+ },
2867
+ {
2868
+ "avg_mask_ratio": 0.4767530772660393,
2869
+ "avg_response_length": 233.275,
2870
+ "avg_student_mask_ratio": 0.4767530772660393,
2871
+ "batch_ainp_frac": 0.0,
2872
+ "batch_inp_frac": 0.0,
2873
+ "batch_inp_oh_frac": 0.0,
2874
+ "batch_inp_par_frac": 1.0,
2875
+ "batch_inp_par_reverse_frac": 0.0,
2876
+ "batch_rl_frac": 0.0,
2877
+ "batch_sft_frac": 0.0,
2878
+ "batch_soft_sft_frac": 0.0,
2879
+ "batch_tf_frac": 0.0,
2880
+ "ce_loss": 0.0,
2881
+ "epoch": 2.5632,
2882
+ "grad_norm": 0.1396484375,
2883
+ "kd_loss": 0.16206236403125446,
2884
+ "learning_rate": 3e-06,
2885
+ "loss": 0.1771,
2886
+ "masked_tokens": 106.375,
2887
+ "mean_t": 0.506370971655997,
2888
+ "step": 1200,
2889
+ "student_masked_tokens": 106.375
2890
+ }
2891
+ ],
2892
+ "logging_steps": 10,
2893
+ "max_steps": 1404,
2894
+ "num_input_tokens_seen": 0,
2895
+ "num_train_epochs": 3,
2896
+ "save_steps": 100,
2897
+ "stateful_callbacks": {
2898
+ "TrainerControl": {
2899
+ "args": {
2900
+ "should_epoch_stop": false,
2901
+ "should_evaluate": false,
2902
+ "should_log": false,
2903
+ "should_save": true,
2904
+ "should_training_stop": false
2905
+ },
2906
+ "attributes": {}
2907
+ }
2908
+ },
2909
+ "total_flos": 0.0,
2910
+ "train_batch_size": 1,
2911
+ "trial_name": null,
2912
+ "trial_params": null
2913
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-1200/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c347df37da6e348160afc3fbb65d12595e5064bd8e5bfd591004e5e86a703f42
3
+ size 7992
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: GSAI-ML/LLaDA-8B-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 0.15.1
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/adapter_config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "GSAI-ML/LLaDA-8B-Instruct",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 64,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "r": 128,
24
+ "rank_pattern": {},
25
+ "revision": null,
26
+ "target_modules": [
27
+ "down_proj",
28
+ "up_proj",
29
+ "gate_proj",
30
+ "v_proj",
31
+ "k_proj",
32
+ "o_proj",
33
+ "q_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d6152a1f978ab9113396a93370fd3f8dff0188cda94938a210d3f30daf65456e
3
+ size 2406624648
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b98b21ae1fd37544eb0e71be8c56c48ef31d5d2dc1f526a72c69358de436c2b7
3
+ size 671304442
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:976572fc3dc5a068a2dabeece6d63c59a59ca329314c851e41f3c3eb08cfd244
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60246bba8f414ee9c88a354a9ea815a89b0ca4caa98573b704dc68fdff12dea3
3
+ size 14512
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e3f0e18fd4ce38e61410a1f0e851c2762584e71a80ec7ce0bc5150325adcecc
3
+ size 1064
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/trainer_state.json ADDED
@@ -0,0 +1,513 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.4266666666666667,
5
+ "eval_steps": 500,
6
+ "global_step": 200,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "avg_mask_ratio": 0.4933756332669873,
13
+ "avg_response_length": 252.0625,
14
+ "avg_student_mask_ratio": 0.4933756332669873,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 0.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 1.0,
19
+ "batch_inp_par_reverse_frac": 0.0,
20
+ "batch_rl_frac": 0.0,
21
+ "batch_sft_frac": 0.0,
22
+ "batch_soft_sft_frac": 0.0,
23
+ "batch_tf_frac": 0.0,
24
+ "ce_loss": 0.0,
25
+ "epoch": 0.021333333333333333,
26
+ "grad_norm": 0.1650390625,
27
+ "kd_loss": 0.24795629351958723,
28
+ "learning_rate": 3e-06,
29
+ "loss": 0.2758,
30
+ "masked_tokens": 120.975,
31
+ "mean_t": 0.5145528071501758,
32
+ "step": 10,
33
+ "student_masked_tokens": 120.975
34
+ },
35
+ {
36
+ "avg_mask_ratio": 0.41923869140446185,
37
+ "avg_response_length": 221.7125,
38
+ "avg_student_mask_ratio": 0.41923869140446185,
39
+ "batch_ainp_frac": 0.0,
40
+ "batch_inp_frac": 0.0,
41
+ "batch_inp_oh_frac": 0.0,
42
+ "batch_inp_par_frac": 1.0,
43
+ "batch_inp_par_reverse_frac": 0.0,
44
+ "batch_rl_frac": 0.0,
45
+ "batch_sft_frac": 0.0,
46
+ "batch_soft_sft_frac": 0.0,
47
+ "batch_tf_frac": 0.0,
48
+ "ce_loss": 0.0,
49
+ "epoch": 0.042666666666666665,
50
+ "grad_norm": 0.388671875,
51
+ "kd_loss": 0.21509853232191353,
52
+ "learning_rate": 3e-06,
53
+ "loss": 0.2046,
54
+ "masked_tokens": 86.1875,
55
+ "mean_t": 0.43874448732240123,
56
+ "step": 20,
57
+ "student_masked_tokens": 86.1875
58
+ },
59
+ {
60
+ "avg_mask_ratio": 0.4542569225654006,
61
+ "avg_response_length": 231.45,
62
+ "avg_student_mask_ratio": 0.4542569225654006,
63
+ "batch_ainp_frac": 0.0,
64
+ "batch_inp_frac": 0.0,
65
+ "batch_inp_oh_frac": 0.0,
66
+ "batch_inp_par_frac": 1.0,
67
+ "batch_inp_par_reverse_frac": 0.0,
68
+ "batch_rl_frac": 0.0,
69
+ "batch_sft_frac": 0.0,
70
+ "batch_soft_sft_frac": 0.0,
71
+ "batch_tf_frac": 0.0,
72
+ "ce_loss": 0.0,
73
+ "epoch": 0.064,
74
+ "grad_norm": 0.52734375,
75
+ "kd_loss": 0.19756361616970963,
76
+ "learning_rate": 3e-06,
77
+ "loss": 0.1976,
78
+ "masked_tokens": 116.2875,
79
+ "mean_t": 0.4803953981841914,
80
+ "step": 30,
81
+ "student_masked_tokens": 116.2875
82
+ },
83
+ {
84
+ "avg_mask_ratio": 0.41855402445653456,
85
+ "avg_response_length": 214.125,
86
+ "avg_student_mask_ratio": 0.41855402445653456,
87
+ "batch_ainp_frac": 0.0,
88
+ "batch_inp_frac": 0.0,
89
+ "batch_inp_oh_frac": 0.0,
90
+ "batch_inp_par_frac": 1.0,
91
+ "batch_inp_par_reverse_frac": 0.0,
92
+ "batch_rl_frac": 0.0,
93
+ "batch_sft_frac": 0.0,
94
+ "batch_soft_sft_frac": 0.0,
95
+ "batch_tf_frac": 0.0,
96
+ "ce_loss": 0.0,
97
+ "epoch": 0.08533333333333333,
98
+ "grad_norm": 0.263671875,
99
+ "kd_loss": 0.1576978968325534,
100
+ "learning_rate": 3e-06,
101
+ "loss": 0.1551,
102
+ "masked_tokens": 94.225,
103
+ "mean_t": 0.4569831106782658,
104
+ "step": 40,
105
+ "student_masked_tokens": 94.225
106
+ },
107
+ {
108
+ "avg_mask_ratio": 0.4331560767372139,
109
+ "avg_response_length": 222.225,
110
+ "avg_student_mask_ratio": 0.4331560767372139,
111
+ "batch_ainp_frac": 0.0,
112
+ "batch_inp_frac": 0.0,
113
+ "batch_inp_oh_frac": 0.0,
114
+ "batch_inp_par_frac": 1.0,
115
+ "batch_inp_par_reverse_frac": 0.0,
116
+ "batch_rl_frac": 0.0,
117
+ "batch_sft_frac": 0.0,
118
+ "batch_soft_sft_frac": 0.0,
119
+ "batch_tf_frac": 0.0,
120
+ "ce_loss": 0.0,
121
+ "epoch": 0.10666666666666667,
122
+ "grad_norm": 0.125,
123
+ "kd_loss": 0.17712681048956255,
124
+ "learning_rate": 3e-06,
125
+ "loss": 0.1648,
126
+ "masked_tokens": 97.825,
127
+ "mean_t": 0.4612453707959503,
128
+ "step": 50,
129
+ "student_masked_tokens": 97.825
130
+ },
131
+ {
132
+ "avg_mask_ratio": 0.4547638618387282,
133
+ "avg_response_length": 242.9,
134
+ "avg_student_mask_ratio": 0.4547638618387282,
135
+ "batch_ainp_frac": 0.0,
136
+ "batch_inp_frac": 0.0,
137
+ "batch_inp_oh_frac": 0.0,
138
+ "batch_inp_par_frac": 1.0,
139
+ "batch_inp_par_reverse_frac": 0.0,
140
+ "batch_rl_frac": 0.0,
141
+ "batch_sft_frac": 0.0,
142
+ "batch_soft_sft_frac": 0.0,
143
+ "batch_tf_frac": 0.0,
144
+ "ce_loss": 0.0,
145
+ "epoch": 0.128,
146
+ "grad_norm": 0.130859375,
147
+ "kd_loss": 0.1745696667137963,
148
+ "learning_rate": 3e-06,
149
+ "loss": 0.1745,
150
+ "masked_tokens": 119.125,
151
+ "mean_t": 0.48597636765334756,
152
+ "step": 60,
153
+ "student_masked_tokens": 119.125
154
+ },
155
+ {
156
+ "avg_mask_ratio": 0.5151988173020072,
157
+ "avg_response_length": 214.3375,
158
+ "avg_student_mask_ratio": 0.5151988173020072,
159
+ "batch_ainp_frac": 0.0,
160
+ "batch_inp_frac": 0.0,
161
+ "batch_inp_oh_frac": 0.0,
162
+ "batch_inp_par_frac": 1.0,
163
+ "batch_inp_par_reverse_frac": 0.0,
164
+ "batch_rl_frac": 0.0,
165
+ "batch_sft_frac": 0.0,
166
+ "batch_soft_sft_frac": 0.0,
167
+ "batch_tf_frac": 0.0,
168
+ "ce_loss": 0.0,
169
+ "epoch": 0.14933333333333335,
170
+ "grad_norm": 0.17578125,
171
+ "kd_loss": 0.1657758254527316,
172
+ "learning_rate": 3e-06,
173
+ "loss": 0.1972,
174
+ "masked_tokens": 111.7875,
175
+ "mean_t": 0.5459650319069624,
176
+ "step": 70,
177
+ "student_masked_tokens": 111.7875
178
+ },
179
+ {
180
+ "avg_mask_ratio": 0.37937068473547697,
181
+ "avg_response_length": 250.2,
182
+ "avg_student_mask_ratio": 0.37937068473547697,
183
+ "batch_ainp_frac": 0.0,
184
+ "batch_inp_frac": 0.0,
185
+ "batch_inp_oh_frac": 0.0,
186
+ "batch_inp_par_frac": 1.0,
187
+ "batch_inp_par_reverse_frac": 0.0,
188
+ "batch_rl_frac": 0.0,
189
+ "batch_sft_frac": 0.0,
190
+ "batch_soft_sft_frac": 0.0,
191
+ "batch_tf_frac": 0.0,
192
+ "ce_loss": 0.0,
193
+ "epoch": 0.17066666666666666,
194
+ "grad_norm": 0.051513671875,
195
+ "kd_loss": 0.13968983994418097,
196
+ "learning_rate": 3e-06,
197
+ "loss": 0.1369,
198
+ "masked_tokens": 95.1875,
199
+ "mean_t": 0.40758824030635876,
200
+ "step": 80,
201
+ "student_masked_tokens": 95.1875
202
+ },
203
+ {
204
+ "avg_mask_ratio": 0.5006106478627771,
205
+ "avg_response_length": 242.1125,
206
+ "avg_student_mask_ratio": 0.5006106478627771,
207
+ "batch_ainp_frac": 0.0,
208
+ "batch_inp_frac": 0.0,
209
+ "batch_inp_oh_frac": 0.0,
210
+ "batch_inp_par_frac": 1.0,
211
+ "batch_inp_par_reverse_frac": 0.0,
212
+ "batch_rl_frac": 0.0,
213
+ "batch_sft_frac": 0.0,
214
+ "batch_soft_sft_frac": 0.0,
215
+ "batch_tf_frac": 0.0,
216
+ "ce_loss": 0.0,
217
+ "epoch": 0.192,
218
+ "grad_norm": 0.22265625,
219
+ "kd_loss": 0.20869405062871707,
220
+ "learning_rate": 3e-06,
221
+ "loss": 0.2007,
222
+ "masked_tokens": 116.3875,
223
+ "mean_t": 0.5297661645396147,
224
+ "step": 90,
225
+ "student_masked_tokens": 116.3875
226
+ },
227
+ {
228
+ "avg_mask_ratio": 0.4596128500183113,
229
+ "avg_response_length": 229.0625,
230
+ "avg_student_mask_ratio": 0.4596128500183113,
231
+ "batch_ainp_frac": 0.0,
232
+ "batch_inp_frac": 0.0,
233
+ "batch_inp_oh_frac": 0.0,
234
+ "batch_inp_par_frac": 1.0,
235
+ "batch_inp_par_reverse_frac": 0.0,
236
+ "batch_rl_frac": 0.0,
237
+ "batch_sft_frac": 0.0,
238
+ "batch_soft_sft_frac": 0.0,
239
+ "batch_tf_frac": 0.0,
240
+ "ce_loss": 0.0,
241
+ "epoch": 0.21333333333333335,
242
+ "grad_norm": 0.1865234375,
243
+ "kd_loss": 0.17640120884607313,
244
+ "learning_rate": 3e-06,
245
+ "loss": 0.1864,
246
+ "masked_tokens": 109.7125,
247
+ "mean_t": 0.4886587227345444,
248
+ "step": 100,
249
+ "student_masked_tokens": 109.7125
250
+ },
251
+ {
252
+ "avg_mask_ratio": 0.4920400592498481,
253
+ "avg_response_length": 229.9875,
254
+ "avg_student_mask_ratio": 0.4920400592498481,
255
+ "batch_ainp_frac": 0.0,
256
+ "batch_inp_frac": 0.0,
257
+ "batch_inp_oh_frac": 0.0,
258
+ "batch_inp_par_frac": 1.0,
259
+ "batch_inp_par_reverse_frac": 0.0,
260
+ "batch_rl_frac": 0.0,
261
+ "batch_sft_frac": 0.0,
262
+ "batch_soft_sft_frac": 0.0,
263
+ "batch_tf_frac": 0.0,
264
+ "ce_loss": 0.0,
265
+ "epoch": 0.23466666666666666,
266
+ "grad_norm": 0.326171875,
267
+ "kd_loss": 0.2011610215539008,
268
+ "learning_rate": 3e-06,
269
+ "loss": 0.2334,
270
+ "masked_tokens": 109.4,
271
+ "mean_t": 0.5246987929102034,
272
+ "step": 110,
273
+ "student_masked_tokens": 109.4
274
+ },
275
+ {
276
+ "avg_mask_ratio": 0.450224511185661,
277
+ "avg_response_length": 229.225,
278
+ "avg_student_mask_ratio": 0.450224511185661,
279
+ "batch_ainp_frac": 0.0,
280
+ "batch_inp_frac": 0.0,
281
+ "batch_inp_oh_frac": 0.0,
282
+ "batch_inp_par_frac": 1.0,
283
+ "batch_inp_par_reverse_frac": 0.0,
284
+ "batch_rl_frac": 0.0,
285
+ "batch_sft_frac": 0.0,
286
+ "batch_soft_sft_frac": 0.0,
287
+ "batch_tf_frac": 0.0,
288
+ "ce_loss": 0.0,
289
+ "epoch": 0.256,
290
+ "grad_norm": 0.171875,
291
+ "kd_loss": 0.21365654302464918,
292
+ "learning_rate": 3e-06,
293
+ "loss": 0.2021,
294
+ "masked_tokens": 102.6375,
295
+ "mean_t": 0.4805434140143916,
296
+ "step": 120,
297
+ "student_masked_tokens": 102.6375
298
+ },
299
+ {
300
+ "avg_mask_ratio": 0.4397759020910598,
301
+ "avg_response_length": 214.15,
302
+ "avg_student_mask_ratio": 0.4397759020910598,
303
+ "batch_ainp_frac": 0.0,
304
+ "batch_inp_frac": 0.0,
305
+ "batch_inp_oh_frac": 0.0,
306
+ "batch_inp_par_frac": 1.0,
307
+ "batch_inp_par_reverse_frac": 0.0,
308
+ "batch_rl_frac": 0.0,
309
+ "batch_sft_frac": 0.0,
310
+ "batch_soft_sft_frac": 0.0,
311
+ "batch_tf_frac": 0.0,
312
+ "ce_loss": 0.0,
313
+ "epoch": 0.2773333333333333,
314
+ "grad_norm": 0.173828125,
315
+ "kd_loss": 0.13860440934267615,
316
+ "learning_rate": 3e-06,
317
+ "loss": 0.1362,
318
+ "masked_tokens": 98.4,
319
+ "mean_t": 0.47522516988683494,
320
+ "step": 130,
321
+ "student_masked_tokens": 98.4
322
+ },
323
+ {
324
+ "avg_mask_ratio": 0.4890626976499334,
325
+ "avg_response_length": 242.1625,
326
+ "avg_student_mask_ratio": 0.4890626976499334,
327
+ "batch_ainp_frac": 0.0,
328
+ "batch_inp_frac": 0.0,
329
+ "batch_inp_oh_frac": 0.0,
330
+ "batch_inp_par_frac": 1.0,
331
+ "batch_inp_par_reverse_frac": 0.0,
332
+ "batch_rl_frac": 0.0,
333
+ "batch_sft_frac": 0.0,
334
+ "batch_soft_sft_frac": 0.0,
335
+ "batch_tf_frac": 0.0,
336
+ "ce_loss": 0.0,
337
+ "epoch": 0.2986666666666667,
338
+ "grad_norm": 0.27734375,
339
+ "kd_loss": 0.2106460814328841,
340
+ "learning_rate": 3e-06,
341
+ "loss": 0.2054,
342
+ "masked_tokens": 129.725,
343
+ "mean_t": 0.5204090005659964,
344
+ "step": 140,
345
+ "student_masked_tokens": 129.725
346
+ },
347
+ {
348
+ "avg_mask_ratio": 0.47298577734036373,
349
+ "avg_response_length": 262.9875,
350
+ "avg_student_mask_ratio": 0.47298577734036373,
351
+ "batch_ainp_frac": 0.0,
352
+ "batch_inp_frac": 0.0,
353
+ "batch_inp_oh_frac": 0.0,
354
+ "batch_inp_par_frac": 1.0,
355
+ "batch_inp_par_reverse_frac": 0.0,
356
+ "batch_rl_frac": 0.0,
357
+ "batch_sft_frac": 0.0,
358
+ "batch_soft_sft_frac": 0.0,
359
+ "batch_tf_frac": 0.0,
360
+ "ce_loss": 0.0,
361
+ "epoch": 0.32,
362
+ "grad_norm": 0.0673828125,
363
+ "kd_loss": 0.2408599746217078,
364
+ "learning_rate": 3e-06,
365
+ "loss": 0.2276,
366
+ "masked_tokens": 128.375,
367
+ "mean_t": 0.5012552456930279,
368
+ "step": 150,
369
+ "student_masked_tokens": 128.375
370
+ },
371
+ {
372
+ "avg_mask_ratio": 0.5043223856599071,
373
+ "avg_response_length": 217.5,
374
+ "avg_student_mask_ratio": 0.5043223856599071,
375
+ "batch_ainp_frac": 0.0,
376
+ "batch_inp_frac": 0.0,
377
+ "batch_inp_oh_frac": 0.0,
378
+ "batch_inp_par_frac": 1.0,
379
+ "batch_inp_par_reverse_frac": 0.0,
380
+ "batch_rl_frac": 0.0,
381
+ "batch_sft_frac": 0.0,
382
+ "batch_soft_sft_frac": 0.0,
383
+ "batch_tf_frac": 0.0,
384
+ "ce_loss": 0.0,
385
+ "epoch": 0.3413333333333333,
386
+ "grad_norm": 0.2236328125,
387
+ "kd_loss": 0.20484722793833043,
388
+ "learning_rate": 3e-06,
389
+ "loss": 0.2154,
390
+ "masked_tokens": 106.025,
391
+ "mean_t": 0.5408745193795766,
392
+ "step": 160,
393
+ "student_masked_tokens": 106.025
394
+ },
395
+ {
396
+ "avg_mask_ratio": 0.48419030708028005,
397
+ "avg_response_length": 196.2625,
398
+ "avg_student_mask_ratio": 0.48419030708028005,
399
+ "batch_ainp_frac": 0.0,
400
+ "batch_inp_frac": 0.0,
401
+ "batch_inp_oh_frac": 0.0,
402
+ "batch_inp_par_frac": 1.0,
403
+ "batch_inp_par_reverse_frac": 0.0,
404
+ "batch_rl_frac": 0.0,
405
+ "batch_sft_frac": 0.0,
406
+ "batch_soft_sft_frac": 0.0,
407
+ "batch_tf_frac": 0.0,
408
+ "ce_loss": 0.0,
409
+ "epoch": 0.3626666666666667,
410
+ "grad_norm": 0.1611328125,
411
+ "kd_loss": 0.20407032655223248,
412
+ "learning_rate": 3e-06,
413
+ "loss": 0.2007,
414
+ "masked_tokens": 96.2625,
415
+ "mean_t": 0.5134547733236104,
416
+ "step": 170,
417
+ "student_masked_tokens": 96.2625
418
+ },
419
+ {
420
+ "avg_mask_ratio": 0.4485494759515859,
421
+ "avg_response_length": 228.8625,
422
+ "avg_student_mask_ratio": 0.4485494759515859,
423
+ "batch_ainp_frac": 0.0,
424
+ "batch_inp_frac": 0.0,
425
+ "batch_inp_oh_frac": 0.0,
426
+ "batch_inp_par_frac": 1.0,
427
+ "batch_inp_par_reverse_frac": 0.0,
428
+ "batch_rl_frac": 0.0,
429
+ "batch_sft_frac": 0.0,
430
+ "batch_soft_sft_frac": 0.0,
431
+ "batch_tf_frac": 0.0,
432
+ "ce_loss": 0.0,
433
+ "epoch": 0.384,
434
+ "grad_norm": 0.14453125,
435
+ "kd_loss": 0.16957379382825621,
436
+ "learning_rate": 3e-06,
437
+ "loss": 0.1796,
438
+ "masked_tokens": 101.275,
439
+ "mean_t": 0.48226988823735156,
440
+ "step": 180,
441
+ "student_masked_tokens": 101.275
442
+ },
443
+ {
444
+ "avg_mask_ratio": 0.48665878190658984,
445
+ "avg_response_length": 230.5,
446
+ "avg_student_mask_ratio": 0.48665878190658984,
447
+ "batch_ainp_frac": 0.0,
448
+ "batch_inp_frac": 0.0,
449
+ "batch_inp_oh_frac": 0.0,
450
+ "batch_inp_par_frac": 1.0,
451
+ "batch_inp_par_reverse_frac": 0.0,
452
+ "batch_rl_frac": 0.0,
453
+ "batch_sft_frac": 0.0,
454
+ "batch_soft_sft_frac": 0.0,
455
+ "batch_tf_frac": 0.0,
456
+ "ce_loss": 0.0,
457
+ "epoch": 0.4053333333333333,
458
+ "grad_norm": 0.3515625,
459
+ "kd_loss": 0.22313492714965832,
460
+ "learning_rate": 3e-06,
461
+ "loss": 0.2112,
462
+ "masked_tokens": 107.975,
463
+ "mean_t": 0.5165087037021294,
464
+ "step": 190,
465
+ "student_masked_tokens": 107.975
466
+ },
467
+ {
468
+ "avg_mask_ratio": 0.4670982737792656,
469
+ "avg_response_length": 210.9125,
470
+ "avg_student_mask_ratio": 0.4670982737792656,
471
+ "batch_ainp_frac": 0.0,
472
+ "batch_inp_frac": 0.0,
473
+ "batch_inp_oh_frac": 0.0,
474
+ "batch_inp_par_frac": 1.0,
475
+ "batch_inp_par_reverse_frac": 0.0,
476
+ "batch_rl_frac": 0.0,
477
+ "batch_sft_frac": 0.0,
478
+ "batch_soft_sft_frac": 0.0,
479
+ "batch_tf_frac": 0.0,
480
+ "ce_loss": 0.0,
481
+ "epoch": 0.4266666666666667,
482
+ "grad_norm": 0.142578125,
483
+ "kd_loss": 0.15220829088375468,
484
+ "learning_rate": 3e-06,
485
+ "loss": 0.1609,
486
+ "masked_tokens": 98.2125,
487
+ "mean_t": 0.5073627714533359,
488
+ "step": 200,
489
+ "student_masked_tokens": 98.2125
490
+ }
491
+ ],
492
+ "logging_steps": 10,
493
+ "max_steps": 1404,
494
+ "num_input_tokens_seen": 0,
495
+ "num_train_epochs": 3,
496
+ "save_steps": 100,
497
+ "stateful_callbacks": {
498
+ "TrainerControl": {
499
+ "args": {
500
+ "should_epoch_stop": false,
501
+ "should_evaluate": false,
502
+ "should_log": false,
503
+ "should_save": true,
504
+ "should_training_stop": false
505
+ },
506
+ "attributes": {}
507
+ }
508
+ },
509
+ "total_flos": 0.0,
510
+ "train_batch_size": 1,
511
+ "trial_name": null,
512
+ "trial_params": null
513
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-200/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c347df37da6e348160afc3fbb65d12595e5064bd8e5bfd591004e5e86a703f42
3
+ size 7992
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: GSAI-ML/LLaDA-8B-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 0.15.1
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/adapter_config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "GSAI-ML/LLaDA-8B-Instruct",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 64,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.05,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "r": 128,
24
+ "rank_pattern": {},
25
+ "revision": null,
26
+ "target_modules": [
27
+ "down_proj",
28
+ "up_proj",
29
+ "gate_proj",
30
+ "v_proj",
31
+ "k_proj",
32
+ "o_proj",
33
+ "q_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:31580c01ba9f7e42b4e3a77ad2ffeffe859d26996818904ad40a43e9ade88006
3
+ size 2406624648
math/INP-PAR/unmask_tags_gold1_target1_ce0.0/checkpoint-300/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f64be610340e992bfcd9cce3b1bf82167c267b33aae29fe2cdc293f818f0d359
3
+ size 671304442