Hugging Face's logo

minhaeoh
/
diffu-cd

Safetensors
Model card Files
xet
 Community
minhaeoh commited on
Commit
e854ad9
·
verified ·
1 Parent(s): 4249dec

Upload checkpoint from unmask_tags_math_self_distill_INP_trefleave_last_step_u0.001-1.0_gold1_target1_ce0.5

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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/README.md +202 -0
  3. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/adapter_config.json +39 -0
  4. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/adapter_model.safetensors +3 -0
  5. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/optimizer.pt +3 -0
  6. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/rng_state_0.pth +3 -0
  7. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/rng_state_1.pth +3 -0
  8. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/scheduler.pt +3 -0
  9. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/trainer_state.json +283 -0
  10. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/training_args.bin +3 -0
  11. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/README.md +202 -0
  12. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/adapter_config.json +39 -0
  13. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/adapter_model.safetensors +3 -0
  14. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/optimizer.pt +3 -0
  15. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/rng_state_0.pth +3 -0
  16. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/rng_state_1.pth +3 -0
  17. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/scheduler.pt +3 -0
  18. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/trainer_state.json +2533 -0
  19. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/training_args.bin +3 -0
  20. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/README.md +202 -0
  21. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/adapter_config.json +39 -0
  22. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/adapter_model.safetensors +3 -0
  23. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/optimizer.pt +3 -0
  24. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/rng_state_0.pth +3 -0
  25. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/rng_state_1.pth +3 -0
  26. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/scheduler.pt +3 -0
  27. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/trainer_state.json +2783 -0
  28. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/training_args.bin +3 -0
  29. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/README.md +202 -0
  30. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/adapter_config.json +39 -0
  31. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/adapter_model.safetensors +3 -0
  32. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/optimizer.pt +3 -0
  33. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/rng_state_0.pth +3 -0
  34. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/rng_state_1.pth +3 -0
  35. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/scheduler.pt +3 -0
  36. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/trainer_state.json +3033 -0
  37. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/training_args.bin +3 -0
  38. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/README.md +202 -0
  39. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/adapter_config.json +39 -0
  40. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/adapter_model.safetensors +3 -0
  41. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/optimizer.pt +3 -0
  42. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/rng_state_0.pth +3 -0
  43. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/rng_state_1.pth +3 -0
  44. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/scheduler.pt +3 -0
  45. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/trainer_state.json +533 -0
  46. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/training_args.bin +3 -0
  47. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-300/README.md +202 -0
  48. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-300/adapter_config.json +39 -0
  49. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-300/adapter_model.safetensors +3 -0
  50. math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-300/optimizer.pt +3 -0
.gitattributes CHANGED
@@ -43,3 +43,4 @@ math/INP-PAR-REVERSE/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jso
43
  math/INP-PAR/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
44
  math/INP-OH/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
45
  math/INP/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
44
  math/INP-OH/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
45
  math/INP/unmask_tags_gold1_target1_ce0.0/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
46
+ math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/debug_training_examples.jsonl filter=lfs diff=lfs merge=lfs -text
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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
+ "gate_proj",
28
+ "down_proj",
29
+ "up_proj",
30
+ "o_proj",
31
+ "q_proj",
32
+ "v_proj",
33
+ "k_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf9df9234dc653a13ebc37a58bc5247fc241e0d9c4f7f0d2e49203ba7a8b929c
3
+ size 2406624648
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60a11a0febb706843c34d89db0943ed63a747c7a4b28b88d25900c72332a6aaf
3
+ size 671304442
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6603180b4f7b4f23eb2fefc470a07c3ec6223e2d309190662f43f30d96be9ce5
3
+ size 14512
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/trainer_state.json ADDED
@@ -0,0 +1,283 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.4931091487989761,
13
+ "avg_response_length": 225.975,
14
+ "avg_student_mask_ratio": 0.4931091487989761,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 1.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 0.0,
19
+ "batch_inp_par_par_frac": 0.0,
20
+ "batch_inp_par_reverse_frac": 0.0,
21
+ "batch_rl_frac": 0.0,
22
+ "batch_sft_frac": 0.0,
23
+ "batch_soft_sft_frac": 0.0,
24
+ "batch_tf_frac": 0.0,
25
+ "ce_loss": 0.5570551689027979,
26
+ "epoch": 0.021333333333333333,
27
+ "grad_norm": 0.404296875,
28
+ "kd_loss": 0.5375588692116253,
29
+ "learning_rate": 3e-06,
30
+ "loss": 0.8247,
31
+ "masked_tokens": 111.95,
32
+ "mean_t": 0.5145528071501758,
33
+ "step": 10,
34
+ "student_masked_tokens": 111.95
35
+ },
36
+ {
37
+ "avg_mask_ratio": 0.42058031369233506,
38
+ "avg_response_length": 255.2625,
39
+ "avg_student_mask_ratio": 0.42058031369233506,
40
+ "batch_ainp_frac": 0.0,
41
+ "batch_inp_frac": 1.0,
42
+ "batch_inp_oh_frac": 0.0,
43
+ "batch_inp_par_frac": 0.0,
44
+ "batch_inp_par_par_frac": 0.0,
45
+ "batch_inp_par_reverse_frac": 0.0,
46
+ "batch_rl_frac": 0.0,
47
+ "batch_sft_frac": 0.0,
48
+ "batch_soft_sft_frac": 0.0,
49
+ "batch_tf_frac": 0.0,
50
+ "ce_loss": 0.42826092825978324,
51
+ "epoch": 0.042666666666666665,
52
+ "grad_norm": 0.8046875,
53
+ "kd_loss": 0.4450965437417761,
54
+ "learning_rate": 3e-06,
55
+ "loss": 0.5724,
56
+ "masked_tokens": 97.35,
57
+ "mean_t": 0.43874448732240123,
58
+ "step": 20,
59
+ "student_masked_tokens": 97.35
60
+ },
61
+ {
62
+ "avg_mask_ratio": 0.4538542575784959,
63
+ "avg_response_length": 211.7625,
64
+ "avg_student_mask_ratio": 0.4538542575784959,
65
+ "batch_ainp_frac": 0.0,
66
+ "batch_inp_frac": 1.0,
67
+ "batch_inp_oh_frac": 0.0,
68
+ "batch_inp_par_frac": 0.0,
69
+ "batch_inp_par_par_frac": 0.0,
70
+ "batch_inp_par_reverse_frac": 0.0,
71
+ "batch_rl_frac": 0.0,
72
+ "batch_sft_frac": 0.0,
73
+ "batch_soft_sft_frac": 0.0,
74
+ "batch_tf_frac": 0.0,
75
+ "ce_loss": 0.4461815005188782,
76
+ "epoch": 0.064,
77
+ "grad_norm": 0.50390625,
78
+ "kd_loss": 0.5296064364436825,
79
+ "learning_rate": 3e-06,
80
+ "loss": 0.702,
81
+ "masked_tokens": 110.2,
82
+ "mean_t": 0.4803953981841914,
83
+ "step": 30,
84
+ "student_masked_tokens": 110.2
85
+ },
86
+ {
87
+ "avg_mask_ratio": 0.4207469140383182,
88
+ "avg_response_length": 224.125,
89
+ "avg_student_mask_ratio": 0.4207469140383182,
90
+ "batch_ainp_frac": 0.0,
91
+ "batch_inp_frac": 1.0,
92
+ "batch_inp_oh_frac": 0.0,
93
+ "batch_inp_par_frac": 0.0,
94
+ "batch_inp_par_par_frac": 0.0,
95
+ "batch_inp_par_reverse_frac": 0.0,
96
+ "batch_rl_frac": 0.0,
97
+ "batch_sft_frac": 0.0,
98
+ "batch_soft_sft_frac": 0.0,
99
+ "batch_tf_frac": 0.0,
100
+ "ce_loss": 0.38499989152683156,
101
+ "epoch": 0.08533333333333333,
102
+ "grad_norm": 1.671875,
103
+ "kd_loss": 0.33118802310931417,
104
+ "learning_rate": 3e-06,
105
+ "loss": 0.5529,
106
+ "masked_tokens": 98.1625,
107
+ "mean_t": 0.4569831106782658,
108
+ "step": 40,
109
+ "student_masked_tokens": 98.1625
110
+ },
111
+ {
112
+ "avg_mask_ratio": 0.43260439952719026,
113
+ "avg_response_length": 207.125,
114
+ "avg_student_mask_ratio": 0.43260439952719026,
115
+ "batch_ainp_frac": 0.0,
116
+ "batch_inp_frac": 1.0,
117
+ "batch_inp_oh_frac": 0.0,
118
+ "batch_inp_par_frac": 0.0,
119
+ "batch_inp_par_par_frac": 0.0,
120
+ "batch_inp_par_reverse_frac": 0.0,
121
+ "batch_rl_frac": 0.0,
122
+ "batch_sft_frac": 0.0,
123
+ "batch_soft_sft_frac": 0.0,
124
+ "batch_tf_frac": 0.0,
125
+ "ce_loss": 0.5240421466317912,
126
+ "epoch": 0.10666666666666667,
127
+ "grad_norm": 1.6875,
128
+ "kd_loss": 0.4270985169670894,
129
+ "learning_rate": 3e-06,
130
+ "loss": 0.671,
131
+ "masked_tokens": 85.05,
132
+ "mean_t": 0.4612453707959503,
133
+ "step": 50,
134
+ "student_masked_tokens": 85.05
135
+ },
136
+ {
137
+ "avg_mask_ratio": 0.46053453313652426,
138
+ "avg_response_length": 251.0875,
139
+ "avg_student_mask_ratio": 0.46053453313652426,
140
+ "batch_ainp_frac": 0.0,
141
+ "batch_inp_frac": 1.0,
142
+ "batch_inp_oh_frac": 0.0,
143
+ "batch_inp_par_frac": 0.0,
144
+ "batch_inp_par_par_frac": 0.0,
145
+ "batch_inp_par_reverse_frac": 0.0,
146
+ "batch_rl_frac": 0.0,
147
+ "batch_sft_frac": 0.0,
148
+ "batch_soft_sft_frac": 0.0,
149
+ "batch_tf_frac": 0.0,
150
+ "ce_loss": 0.5027546818272185,
151
+ "epoch": 0.128,
152
+ "grad_norm": 0.17578125,
153
+ "kd_loss": 0.3904111967755945,
154
+ "learning_rate": 3e-06,
155
+ "loss": 0.6672,
156
+ "masked_tokens": 120.9,
157
+ "mean_t": 0.48597636765334756,
158
+ "step": 60,
159
+ "student_masked_tokens": 120.9
160
+ },
161
+ {
162
+ "avg_mask_ratio": 0.5112146578729153,
163
+ "avg_response_length": 202.5875,
164
+ "avg_student_mask_ratio": 0.5112146578729153,
165
+ "batch_ainp_frac": 0.0,
166
+ "batch_inp_frac": 1.0,
167
+ "batch_inp_oh_frac": 0.0,
168
+ "batch_inp_par_frac": 0.0,
169
+ "batch_inp_par_par_frac": 0.0,
170
+ "batch_inp_par_reverse_frac": 0.0,
171
+ "batch_rl_frac": 0.0,
172
+ "batch_sft_frac": 0.0,
173
+ "batch_soft_sft_frac": 0.0,
174
+ "batch_tf_frac": 0.0,
175
+ "ce_loss": 0.7753003867959023,
176
+ "epoch": 0.14933333333333335,
177
+ "grad_norm": 0.953125,
178
+ "kd_loss": 0.4415664039527428,
179
+ "learning_rate": 3e-06,
180
+ "loss": 0.856,
181
+ "masked_tokens": 104.5875,
182
+ "mean_t": 0.5459650319069624,
183
+ "step": 70,
184
+ "student_masked_tokens": 104.5875
185
+ },
186
+ {
187
+ "avg_mask_ratio": 0.37548826879356056,
188
+ "avg_response_length": 225.85,
189
+ "avg_student_mask_ratio": 0.37548826879356056,
190
+ "batch_ainp_frac": 0.0,
191
+ "batch_inp_frac": 1.0,
192
+ "batch_inp_oh_frac": 0.0,
193
+ "batch_inp_par_frac": 0.0,
194
+ "batch_inp_par_par_frac": 0.0,
195
+ "batch_inp_par_reverse_frac": 0.0,
196
+ "batch_rl_frac": 0.0,
197
+ "batch_sft_frac": 0.0,
198
+ "batch_soft_sft_frac": 0.0,
199
+ "batch_tf_frac": 0.0,
200
+ "ce_loss": 0.3791731233859082,
201
+ "epoch": 0.17066666666666666,
202
+ "grad_norm": 0.1552734375,
203
+ "kd_loss": 0.31052538527774515,
204
+ "learning_rate": 3e-06,
205
+ "loss": 0.4843,
206
+ "masked_tokens": 85.0625,
207
+ "mean_t": 0.40758824030635876,
208
+ "step": 80,
209
+ "student_masked_tokens": 85.0625
210
+ },
211
+ {
212
+ "avg_mask_ratio": 0.5001560213277116,
213
+ "avg_response_length": 229.75,
214
+ "avg_student_mask_ratio": 0.5001560213277116,
215
+ "batch_ainp_frac": 0.0,
216
+ "batch_inp_frac": 1.0,
217
+ "batch_inp_oh_frac": 0.0,
218
+ "batch_inp_par_frac": 0.0,
219
+ "batch_inp_par_par_frac": 0.0,
220
+ "batch_inp_par_reverse_frac": 0.0,
221
+ "batch_rl_frac": 0.0,
222
+ "batch_sft_frac": 0.0,
223
+ "batch_soft_sft_frac": 0.0,
224
+ "batch_tf_frac": 0.0,
225
+ "ce_loss": 0.6899960007944174,
226
+ "epoch": 0.192,
227
+ "grad_norm": 1.25,
228
+ "kd_loss": 0.5995283465861896,
229
+ "learning_rate": 3e-06,
230
+ "loss": 0.9721,
231
+ "masked_tokens": 107.6625,
232
+ "mean_t": 0.5297661645396147,
233
+ "step": 90,
234
+ "student_masked_tokens": 107.6625
235
+ },
236
+ {
237
+ "avg_mask_ratio": 0.4576045103633078,
238
+ "avg_response_length": 208.0,
239
+ "avg_student_mask_ratio": 0.4576045103633078,
240
+ "batch_ainp_frac": 0.0,
241
+ "batch_inp_frac": 1.0,
242
+ "batch_inp_oh_frac": 0.0,
243
+ "batch_inp_par_frac": 0.0,
244
+ "batch_inp_par_par_frac": 0.0,
245
+ "batch_inp_par_reverse_frac": 0.0,
246
+ "batch_rl_frac": 0.0,
247
+ "batch_sft_frac": 0.0,
248
+ "batch_soft_sft_frac": 0.0,
249
+ "batch_tf_frac": 0.0,
250
+ "ce_loss": 0.41132245859021166,
251
+ "epoch": 0.21333333333333335,
252
+ "grad_norm": 0.64453125,
253
+ "kd_loss": 0.3813956479015957,
254
+ "learning_rate": 3e-06,
255
+ "loss": 0.6635,
256
+ "masked_tokens": 104.1625,
257
+ "mean_t": 0.4886587227345444,
258
+ "step": 100,
259
+ "student_masked_tokens": 104.1625
260
+ }
261
+ ],
262
+ "logging_steps": 10,
263
+ "max_steps": 1404,
264
+ "num_input_tokens_seen": 0,
265
+ "num_train_epochs": 3,
266
+ "save_steps": 100,
267
+ "stateful_callbacks": {
268
+ "TrainerControl": {
269
+ "args": {
270
+ "should_epoch_stop": false,
271
+ "should_evaluate": false,
272
+ "should_log": false,
273
+ "should_save": true,
274
+ "should_training_stop": false
275
+ },
276
+ "attributes": {}
277
+ }
278
+ },
279
+ "total_flos": 0.0,
280
+ "train_batch_size": 1,
281
+ "trial_name": null,
282
+ "trial_params": null
283
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-100/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89ae704e41a4a62f6ca56789c45ce45887326cd2f8d1e97e398e5ada4a93398c
3
+ size 8312
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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
+ "gate_proj",
28
+ "down_proj",
29
+ "up_proj",
30
+ "o_proj",
31
+ "q_proj",
32
+ "v_proj",
33
+ "k_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cb3060230d41dc9ae4e7608db56f3eea1306392dd1de03faebd841f30b160cc1
3
+ size 2406624648
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57580597ffdb6d3dea49b345f1bc5a35357a29e1c2bb2dc9e19a132c229e1d66
3
+ size 671304442
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69989ab343a15b03b088a3c9e2bb39b88ed718b675223b5e1d55890e63636453
3
+ size 14512
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/trainer_state.json ADDED
@@ -0,0 +1,2533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.4931091487989761,
13
+ "avg_response_length": 225.975,
14
+ "avg_student_mask_ratio": 0.4931091487989761,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 1.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 0.0,
19
+ "batch_inp_par_par_frac": 0.0,
20
+ "batch_inp_par_reverse_frac": 0.0,
21
+ "batch_rl_frac": 0.0,
22
+ "batch_sft_frac": 0.0,
23
+ "batch_soft_sft_frac": 0.0,
24
+ "batch_tf_frac": 0.0,
25
+ "ce_loss": 0.5570551689027979,
26
+ "epoch": 0.021333333333333333,
27
+ "grad_norm": 0.404296875,
28
+ "kd_loss": 0.5375588692116253,
29
+ "learning_rate": 3e-06,
30
+ "loss": 0.8247,
31
+ "masked_tokens": 111.95,
32
+ "mean_t": 0.5145528071501758,
33
+ "step": 10,
34
+ "student_masked_tokens": 111.95
35
+ },
36
+ {
37
+ "avg_mask_ratio": 0.42058031369233506,
38
+ "avg_response_length": 255.2625,
39
+ "avg_student_mask_ratio": 0.42058031369233506,
40
+ "batch_ainp_frac": 0.0,
41
+ "batch_inp_frac": 1.0,
42
+ "batch_inp_oh_frac": 0.0,
43
+ "batch_inp_par_frac": 0.0,
44
+ "batch_inp_par_par_frac": 0.0,
45
+ "batch_inp_par_reverse_frac": 0.0,
46
+ "batch_rl_frac": 0.0,
47
+ "batch_sft_frac": 0.0,
48
+ "batch_soft_sft_frac": 0.0,
49
+ "batch_tf_frac": 0.0,
50
+ "ce_loss": 0.42826092825978324,
51
+ "epoch": 0.042666666666666665,
52
+ "grad_norm": 0.8046875,
53
+ "kd_loss": 0.4450965437417761,
54
+ "learning_rate": 3e-06,
55
+ "loss": 0.5724,
56
+ "masked_tokens": 97.35,
57
+ "mean_t": 0.43874448732240123,
58
+ "step": 20,
59
+ "student_masked_tokens": 97.35
60
+ },
61
+ {
62
+ "avg_mask_ratio": 0.4538542575784959,
63
+ "avg_response_length": 211.7625,
64
+ "avg_student_mask_ratio": 0.4538542575784959,
65
+ "batch_ainp_frac": 0.0,
66
+ "batch_inp_frac": 1.0,
67
+ "batch_inp_oh_frac": 0.0,
68
+ "batch_inp_par_frac": 0.0,
69
+ "batch_inp_par_par_frac": 0.0,
70
+ "batch_inp_par_reverse_frac": 0.0,
71
+ "batch_rl_frac": 0.0,
72
+ "batch_sft_frac": 0.0,
73
+ "batch_soft_sft_frac": 0.0,
74
+ "batch_tf_frac": 0.0,
75
+ "ce_loss": 0.4461815005188782,
76
+ "epoch": 0.064,
77
+ "grad_norm": 0.50390625,
78
+ "kd_loss": 0.5296064364436825,
79
+ "learning_rate": 3e-06,
80
+ "loss": 0.702,
81
+ "masked_tokens": 110.2,
82
+ "mean_t": 0.4803953981841914,
83
+ "step": 30,
84
+ "student_masked_tokens": 110.2
85
+ },
86
+ {
87
+ "avg_mask_ratio": 0.4207469140383182,
88
+ "avg_response_length": 224.125,
89
+ "avg_student_mask_ratio": 0.4207469140383182,
90
+ "batch_ainp_frac": 0.0,
91
+ "batch_inp_frac": 1.0,
92
+ "batch_inp_oh_frac": 0.0,
93
+ "batch_inp_par_frac": 0.0,
94
+ "batch_inp_par_par_frac": 0.0,
95
+ "batch_inp_par_reverse_frac": 0.0,
96
+ "batch_rl_frac": 0.0,
97
+ "batch_sft_frac": 0.0,
98
+ "batch_soft_sft_frac": 0.0,
99
+ "batch_tf_frac": 0.0,
100
+ "ce_loss": 0.38499989152683156,
101
+ "epoch": 0.08533333333333333,
102
+ "grad_norm": 1.671875,
103
+ "kd_loss": 0.33118802310931417,
104
+ "learning_rate": 3e-06,
105
+ "loss": 0.5529,
106
+ "masked_tokens": 98.1625,
107
+ "mean_t": 0.4569831106782658,
108
+ "step": 40,
109
+ "student_masked_tokens": 98.1625
110
+ },
111
+ {
112
+ "avg_mask_ratio": 0.43260439952719026,
113
+ "avg_response_length": 207.125,
114
+ "avg_student_mask_ratio": 0.43260439952719026,
115
+ "batch_ainp_frac": 0.0,
116
+ "batch_inp_frac": 1.0,
117
+ "batch_inp_oh_frac": 0.0,
118
+ "batch_inp_par_frac": 0.0,
119
+ "batch_inp_par_par_frac": 0.0,
120
+ "batch_inp_par_reverse_frac": 0.0,
121
+ "batch_rl_frac": 0.0,
122
+ "batch_sft_frac": 0.0,
123
+ "batch_soft_sft_frac": 0.0,
124
+ "batch_tf_frac": 0.0,
125
+ "ce_loss": 0.5240421466317912,
126
+ "epoch": 0.10666666666666667,
127
+ "grad_norm": 1.6875,
128
+ "kd_loss": 0.4270985169670894,
129
+ "learning_rate": 3e-06,
130
+ "loss": 0.671,
131
+ "masked_tokens": 85.05,
132
+ "mean_t": 0.4612453707959503,
133
+ "step": 50,
134
+ "student_masked_tokens": 85.05
135
+ },
136
+ {
137
+ "avg_mask_ratio": 0.46053453313652426,
138
+ "avg_response_length": 251.0875,
139
+ "avg_student_mask_ratio": 0.46053453313652426,
140
+ "batch_ainp_frac": 0.0,
141
+ "batch_inp_frac": 1.0,
142
+ "batch_inp_oh_frac": 0.0,
143
+ "batch_inp_par_frac": 0.0,
144
+ "batch_inp_par_par_frac": 0.0,
145
+ "batch_inp_par_reverse_frac": 0.0,
146
+ "batch_rl_frac": 0.0,
147
+ "batch_sft_frac": 0.0,
148
+ "batch_soft_sft_frac": 0.0,
149
+ "batch_tf_frac": 0.0,
150
+ "ce_loss": 0.5027546818272185,
151
+ "epoch": 0.128,
152
+ "grad_norm": 0.17578125,
153
+ "kd_loss": 0.3904111967755945,
154
+ "learning_rate": 3e-06,
155
+ "loss": 0.6672,
156
+ "masked_tokens": 120.9,
157
+ "mean_t": 0.48597636765334756,
158
+ "step": 60,
159
+ "student_masked_tokens": 120.9
160
+ },
161
+ {
162
+ "avg_mask_ratio": 0.5112146578729153,
163
+ "avg_response_length": 202.5875,
164
+ "avg_student_mask_ratio": 0.5112146578729153,
165
+ "batch_ainp_frac": 0.0,
166
+ "batch_inp_frac": 1.0,
167
+ "batch_inp_oh_frac": 0.0,
168
+ "batch_inp_par_frac": 0.0,
169
+ "batch_inp_par_par_frac": 0.0,
170
+ "batch_inp_par_reverse_frac": 0.0,
171
+ "batch_rl_frac": 0.0,
172
+ "batch_sft_frac": 0.0,
173
+ "batch_soft_sft_frac": 0.0,
174
+ "batch_tf_frac": 0.0,
175
+ "ce_loss": 0.7753003867959023,
176
+ "epoch": 0.14933333333333335,
177
+ "grad_norm": 0.953125,
178
+ "kd_loss": 0.4415664039527428,
179
+ "learning_rate": 3e-06,
180
+ "loss": 0.856,
181
+ "masked_tokens": 104.5875,
182
+ "mean_t": 0.5459650319069624,
183
+ "step": 70,
184
+ "student_masked_tokens": 104.5875
185
+ },
186
+ {
187
+ "avg_mask_ratio": 0.37548826879356056,
188
+ "avg_response_length": 225.85,
189
+ "avg_student_mask_ratio": 0.37548826879356056,
190
+ "batch_ainp_frac": 0.0,
191
+ "batch_inp_frac": 1.0,
192
+ "batch_inp_oh_frac": 0.0,
193
+ "batch_inp_par_frac": 0.0,
194
+ "batch_inp_par_par_frac": 0.0,
195
+ "batch_inp_par_reverse_frac": 0.0,
196
+ "batch_rl_frac": 0.0,
197
+ "batch_sft_frac": 0.0,
198
+ "batch_soft_sft_frac": 0.0,
199
+ "batch_tf_frac": 0.0,
200
+ "ce_loss": 0.3791731233859082,
201
+ "epoch": 0.17066666666666666,
202
+ "grad_norm": 0.1552734375,
203
+ "kd_loss": 0.31052538527774515,
204
+ "learning_rate": 3e-06,
205
+ "loss": 0.4843,
206
+ "masked_tokens": 85.0625,
207
+ "mean_t": 0.40758824030635876,
208
+ "step": 80,
209
+ "student_masked_tokens": 85.0625
210
+ },
211
+ {
212
+ "avg_mask_ratio": 0.5001560213277116,
213
+ "avg_response_length": 229.75,
214
+ "avg_student_mask_ratio": 0.5001560213277116,
215
+ "batch_ainp_frac": 0.0,
216
+ "batch_inp_frac": 1.0,
217
+ "batch_inp_oh_frac": 0.0,
218
+ "batch_inp_par_frac": 0.0,
219
+ "batch_inp_par_par_frac": 0.0,
220
+ "batch_inp_par_reverse_frac": 0.0,
221
+ "batch_rl_frac": 0.0,
222
+ "batch_sft_frac": 0.0,
223
+ "batch_soft_sft_frac": 0.0,
224
+ "batch_tf_frac": 0.0,
225
+ "ce_loss": 0.6899960007944174,
226
+ "epoch": 0.192,
227
+ "grad_norm": 1.25,
228
+ "kd_loss": 0.5995283465861896,
229
+ "learning_rate": 3e-06,
230
+ "loss": 0.9721,
231
+ "masked_tokens": 107.6625,
232
+ "mean_t": 0.5297661645396147,
233
+ "step": 90,
234
+ "student_masked_tokens": 107.6625
235
+ },
236
+ {
237
+ "avg_mask_ratio": 0.4576045103633078,
238
+ "avg_response_length": 208.0,
239
+ "avg_student_mask_ratio": 0.4576045103633078,
240
+ "batch_ainp_frac": 0.0,
241
+ "batch_inp_frac": 1.0,
242
+ "batch_inp_oh_frac": 0.0,
243
+ "batch_inp_par_frac": 0.0,
244
+ "batch_inp_par_par_frac": 0.0,
245
+ "batch_inp_par_reverse_frac": 0.0,
246
+ "batch_rl_frac": 0.0,
247
+ "batch_sft_frac": 0.0,
248
+ "batch_soft_sft_frac": 0.0,
249
+ "batch_tf_frac": 0.0,
250
+ "ce_loss": 0.41132245859021166,
251
+ "epoch": 0.21333333333333335,
252
+ "grad_norm": 0.64453125,
253
+ "kd_loss": 0.3813956479015957,
254
+ "learning_rate": 3e-06,
255
+ "loss": 0.6635,
256
+ "masked_tokens": 104.1625,
257
+ "mean_t": 0.4886587227345444,
258
+ "step": 100,
259
+ "student_masked_tokens": 104.1625
260
+ },
261
+ {
262
+ "avg_mask_ratio": 0.4877026333590038,
263
+ "avg_response_length": 213.0875,
264
+ "avg_student_mask_ratio": 0.4877026333590038,
265
+ "batch_ainp_frac": 0.0,
266
+ "batch_inp_frac": 1.0,
267
+ "batch_inp_oh_frac": 0.0,
268
+ "batch_inp_par_frac": 0.0,
269
+ "batch_inp_par_par_frac": 0.0,
270
+ "batch_inp_par_reverse_frac": 0.0,
271
+ "batch_rl_frac": 0.0,
272
+ "batch_sft_frac": 0.0,
273
+ "batch_soft_sft_frac": 0.0,
274
+ "batch_tf_frac": 0.0,
275
+ "ce_loss": 0.4612084587922368,
276
+ "epoch": 0.23466666666666666,
277
+ "grad_norm": 0.64453125,
278
+ "kd_loss": 0.5074845846289577,
279
+ "learning_rate": 3e-06,
280
+ "loss": 0.7993,
281
+ "masked_tokens": 102.075,
282
+ "mean_t": 0.5246987929102034,
283
+ "step": 110,
284
+ "student_masked_tokens": 102.075
285
+ },
286
+ {
287
+ "avg_mask_ratio": 0.45146879020612685,
288
+ "avg_response_length": 224.1875,
289
+ "avg_student_mask_ratio": 0.45146879020612685,
290
+ "batch_ainp_frac": 0.0,
291
+ "batch_inp_frac": 1.0,
292
+ "batch_inp_oh_frac": 0.0,
293
+ "batch_inp_par_frac": 0.0,
294
+ "batch_inp_par_par_frac": 0.0,
295
+ "batch_inp_par_reverse_frac": 0.0,
296
+ "batch_rl_frac": 0.0,
297
+ "batch_sft_frac": 0.0,
298
+ "batch_soft_sft_frac": 0.0,
299
+ "batch_tf_frac": 0.0,
300
+ "ce_loss": 0.3276976759495483,
301
+ "epoch": 0.256,
302
+ "grad_norm": 0.30078125,
303
+ "kd_loss": 0.41461311469229256,
304
+ "learning_rate": 3e-06,
305
+ "loss": 0.6088,
306
+ "masked_tokens": 100.525,
307
+ "mean_t": 0.4805434140143916,
308
+ "step": 120,
309
+ "student_masked_tokens": 100.525
310
+ },
311
+ {
312
+ "avg_mask_ratio": 0.4356566035945434,
313
+ "avg_response_length": 202.7,
314
+ "avg_student_mask_ratio": 0.4356566035945434,
315
+ "batch_ainp_frac": 0.0,
316
+ "batch_inp_frac": 1.0,
317
+ "batch_inp_oh_frac": 0.0,
318
+ "batch_inp_par_frac": 0.0,
319
+ "batch_inp_par_par_frac": 0.0,
320
+ "batch_inp_par_reverse_frac": 0.0,
321
+ "batch_rl_frac": 0.0,
322
+ "batch_sft_frac": 0.0,
323
+ "batch_soft_sft_frac": 0.0,
324
+ "batch_tf_frac": 0.0,
325
+ "ce_loss": 0.12710368948505674,
326
+ "epoch": 0.2773333333333333,
327
+ "grad_norm": 0.490234375,
328
+ "kd_loss": 0.23057804748218585,
329
+ "learning_rate": 3e-06,
330
+ "loss": 0.384,
331
+ "masked_tokens": 89.5625,
332
+ "mean_t": 0.47522516988683494,
333
+ "step": 130,
334
+ "student_masked_tokens": 89.5625
335
+ },
336
+ {
337
+ "avg_mask_ratio": 0.49419954856857656,
338
+ "avg_response_length": 255.625,
339
+ "avg_student_mask_ratio": 0.49419954856857656,
340
+ "batch_ainp_frac": 0.0,
341
+ "batch_inp_frac": 1.0,
342
+ "batch_inp_oh_frac": 0.0,
343
+ "batch_inp_par_frac": 0.0,
344
+ "batch_inp_par_par_frac": 0.0,
345
+ "batch_inp_par_reverse_frac": 0.0,
346
+ "batch_rl_frac": 0.0,
347
+ "batch_sft_frac": 0.0,
348
+ "batch_soft_sft_frac": 0.0,
349
+ "batch_tf_frac": 0.0,
350
+ "ce_loss": 0.48596099013025196,
351
+ "epoch": 0.2986666666666667,
352
+ "grad_norm": 0.859375,
353
+ "kd_loss": 0.5025483015746885,
354
+ "learning_rate": 3e-06,
355
+ "loss": 0.7892,
356
+ "masked_tokens": 136.575,
357
+ "mean_t": 0.5204090005659964,
358
+ "step": 140,
359
+ "student_masked_tokens": 136.575
360
+ },
361
+ {
362
+ "avg_mask_ratio": 0.4736677930341102,
363
+ "avg_response_length": 255.375,
364
+ "avg_student_mask_ratio": 0.4736677930341102,
365
+ "batch_ainp_frac": 0.0,
366
+ "batch_inp_frac": 1.0,
367
+ "batch_inp_oh_frac": 0.0,
368
+ "batch_inp_par_frac": 0.0,
369
+ "batch_inp_par_par_frac": 0.0,
370
+ "batch_inp_par_reverse_frac": 0.0,
371
+ "batch_rl_frac": 0.0,
372
+ "batch_sft_frac": 0.0,
373
+ "batch_soft_sft_frac": 0.0,
374
+ "batch_tf_frac": 0.0,
375
+ "ce_loss": 0.5493089448234059,
376
+ "epoch": 0.32,
377
+ "grad_norm": 0.076171875,
378
+ "kd_loss": 0.4892602212316547,
379
+ "learning_rate": 3e-06,
380
+ "loss": 0.7918,
381
+ "masked_tokens": 126.575,
382
+ "mean_t": 0.5012552456930279,
383
+ "step": 150,
384
+ "student_masked_tokens": 126.575
385
+ },
386
+ {
387
+ "avg_mask_ratio": 0.5032523009285796,
388
+ "avg_response_length": 209.325,
389
+ "avg_student_mask_ratio": 0.5032523009285796,
390
+ "batch_ainp_frac": 0.0,
391
+ "batch_inp_frac": 1.0,
392
+ "batch_inp_oh_frac": 0.0,
393
+ "batch_inp_par_frac": 0.0,
394
+ "batch_inp_par_par_frac": 0.0,
395
+ "batch_inp_par_reverse_frac": 0.0,
396
+ "batch_rl_frac": 0.0,
397
+ "batch_sft_frac": 0.0,
398
+ "batch_soft_sft_frac": 0.0,
399
+ "batch_tf_frac": 0.0,
400
+ "ce_loss": 0.5929547422666019,
401
+ "epoch": 0.3413333333333333,
402
+ "grad_norm": 1.171875,
403
+ "kd_loss": 0.44711892502580214,
404
+ "learning_rate": 3e-06,
405
+ "loss": 0.7472,
406
+ "masked_tokens": 99.425,
407
+ "mean_t": 0.5408745193795766,
408
+ "step": 160,
409
+ "student_masked_tokens": 99.425
410
+ },
411
+ {
412
+ "avg_mask_ratio": 0.4806730231270194,
413
+ "avg_response_length": 190.325,
414
+ "avg_student_mask_ratio": 0.4806730231270194,
415
+ "batch_ainp_frac": 0.0,
416
+ "batch_inp_frac": 1.0,
417
+ "batch_inp_oh_frac": 0.0,
418
+ "batch_inp_par_frac": 0.0,
419
+ "batch_inp_par_par_frac": 0.0,
420
+ "batch_inp_par_reverse_frac": 0.0,
421
+ "batch_rl_frac": 0.0,
422
+ "batch_sft_frac": 0.0,
423
+ "batch_soft_sft_frac": 0.0,
424
+ "batch_tf_frac": 0.0,
425
+ "ce_loss": 0.3583432949517601,
426
+ "epoch": 0.3626666666666667,
427
+ "grad_norm": 0.6171875,
428
+ "kd_loss": 0.4521343837219092,
429
+ "learning_rate": 3e-06,
430
+ "loss": 0.6358,
431
+ "masked_tokens": 89.825,
432
+ "mean_t": 0.5134547733236104,
433
+ "step": 170,
434
+ "student_masked_tokens": 89.825
435
+ },
436
+ {
437
+ "avg_mask_ratio": 0.45829249716189224,
438
+ "avg_response_length": 244.0,
439
+ "avg_student_mask_ratio": 0.45829249716189224,
440
+ "batch_ainp_frac": 0.0,
441
+ "batch_inp_frac": 1.0,
442
+ "batch_inp_oh_frac": 0.0,
443
+ "batch_inp_par_frac": 0.0,
444
+ "batch_inp_par_par_frac": 0.0,
445
+ "batch_inp_par_reverse_frac": 0.0,
446
+ "batch_rl_frac": 0.0,
447
+ "batch_sft_frac": 0.0,
448
+ "batch_soft_sft_frac": 0.0,
449
+ "batch_tf_frac": 0.0,
450
+ "ce_loss": 0.3746713957985094,
451
+ "epoch": 0.384,
452
+ "grad_norm": 0.49609375,
453
+ "kd_loss": 0.34934306121722897,
454
+ "learning_rate": 3e-06,
455
+ "loss": 0.5747,
456
+ "masked_tokens": 110.075,
457
+ "mean_t": 0.48226988823735156,
458
+ "step": 180,
459
+ "student_masked_tokens": 110.075
460
+ },
461
+ {
462
+ "avg_mask_ratio": 0.4842760307248682,
463
+ "avg_response_length": 233.675,
464
+ "avg_student_mask_ratio": 0.4842760307248682,
465
+ "batch_ainp_frac": 0.0,
466
+ "batch_inp_frac": 1.0,
467
+ "batch_inp_oh_frac": 0.0,
468
+ "batch_inp_par_frac": 0.0,
469
+ "batch_inp_par_par_frac": 0.0,
470
+ "batch_inp_par_reverse_frac": 0.0,
471
+ "batch_rl_frac": 0.0,
472
+ "batch_sft_frac": 0.0,
473
+ "batch_soft_sft_frac": 0.0,
474
+ "batch_tf_frac": 0.0,
475
+ "ce_loss": 0.5090123614077584,
476
+ "epoch": 0.4053333333333333,
477
+ "grad_norm": 1.6171875,
478
+ "kd_loss": 0.43204482231294605,
479
+ "learning_rate": 3e-06,
480
+ "loss": 0.7055,
481
+ "masked_tokens": 109.5875,
482
+ "mean_t": 0.5165087037021294,
483
+ "step": 190,
484
+ "student_masked_tokens": 109.5875
485
+ },
486
+ {
487
+ "avg_mask_ratio": 0.4665210062637925,
488
+ "avg_response_length": 197.8,
489
+ "avg_student_mask_ratio": 0.4665210062637925,
490
+ "batch_ainp_frac": 0.0,
491
+ "batch_inp_frac": 1.0,
492
+ "batch_inp_oh_frac": 0.0,
493
+ "batch_inp_par_frac": 0.0,
494
+ "batch_inp_par_par_frac": 0.0,
495
+ "batch_inp_par_reverse_frac": 0.0,
496
+ "batch_rl_frac": 0.0,
497
+ "batch_sft_frac": 0.0,
498
+ "batch_soft_sft_frac": 0.0,
499
+ "batch_tf_frac": 0.0,
500
+ "ce_loss": 0.27265903051802526,
501
+ "epoch": 0.4266666666666667,
502
+ "grad_norm": 0.318359375,
503
+ "kd_loss": 0.357759011555504,
504
+ "learning_rate": 3e-06,
505
+ "loss": 0.5013,
506
+ "masked_tokens": 97.0125,
507
+ "mean_t": 0.5073627714533359,
508
+ "step": 200,
509
+ "student_masked_tokens": 97.0125
510
+ },
511
+ {
512
+ "avg_mask_ratio": 0.45183838629163803,
513
+ "avg_response_length": 213.7875,
514
+ "avg_student_mask_ratio": 0.45183838629163803,
515
+ "batch_ainp_frac": 0.0,
516
+ "batch_inp_frac": 1.0,
517
+ "batch_inp_oh_frac": 0.0,
518
+ "batch_inp_par_frac": 0.0,
519
+ "batch_inp_par_par_frac": 0.0,
520
+ "batch_inp_par_reverse_frac": 0.0,
521
+ "batch_rl_frac": 0.0,
522
+ "batch_sft_frac": 0.0,
523
+ "batch_soft_sft_frac": 0.0,
524
+ "batch_tf_frac": 0.0,
525
+ "ce_loss": 0.21031314329709402,
526
+ "epoch": 0.448,
527
+ "grad_norm": 0.8359375,
528
+ "kd_loss": 0.38585986606940426,
529
+ "learning_rate": 3e-06,
530
+ "loss": 0.5103,
531
+ "masked_tokens": 96.0125,
532
+ "mean_t": 0.48430291628465055,
533
+ "step": 210,
534
+ "student_masked_tokens": 96.0125
535
+ },
536
+ {
537
+ "avg_mask_ratio": 0.5396152360364794,
538
+ "avg_response_length": 236.5125,
539
+ "avg_student_mask_ratio": 0.5396152360364794,
540
+ "batch_ainp_frac": 0.0,
541
+ "batch_inp_frac": 1.0,
542
+ "batch_inp_oh_frac": 0.0,
543
+ "batch_inp_par_frac": 0.0,
544
+ "batch_inp_par_par_frac": 0.0,
545
+ "batch_inp_par_reverse_frac": 0.0,
546
+ "batch_rl_frac": 0.0,
547
+ "batch_sft_frac": 0.0,
548
+ "batch_soft_sft_frac": 0.0,
549
+ "batch_tf_frac": 0.0,
550
+ "ce_loss": 0.6178499971098063,
551
+ "epoch": 0.4693333333333333,
552
+ "grad_norm": 0.96875,
553
+ "kd_loss": 0.46674597742967305,
554
+ "learning_rate": 3e-06,
555
+ "loss": 0.8495,
556
+ "masked_tokens": 125.35,
557
+ "mean_t": 0.570199209311977,
558
+ "step": 220,
559
+ "student_masked_tokens": 125.35
560
+ },
561
+ {
562
+ "avg_mask_ratio": 0.4403991688624956,
563
+ "avg_response_length": 252.7,
564
+ "avg_student_mask_ratio": 0.4403991688624956,
565
+ "batch_ainp_frac": 0.0,
566
+ "batch_inp_frac": 1.0,
567
+ "batch_inp_oh_frac": 0.0,
568
+ "batch_inp_par_frac": 0.0,
569
+ "batch_inp_par_par_frac": 0.0,
570
+ "batch_inp_par_reverse_frac": 0.0,
571
+ "batch_rl_frac": 0.0,
572
+ "batch_sft_frac": 0.0,
573
+ "batch_soft_sft_frac": 0.0,
574
+ "batch_tf_frac": 0.0,
575
+ "ce_loss": 0.25455478885055527,
576
+ "epoch": 0.49066666666666664,
577
+ "grad_norm": 0.5703125,
578
+ "kd_loss": 0.43305868929596725,
579
+ "learning_rate": 3e-06,
580
+ "loss": 0.6428,
581
+ "masked_tokens": 107.325,
582
+ "mean_t": 0.46891279935371133,
583
+ "step": 230,
584
+ "student_masked_tokens": 107.325
585
+ },
586
+ {
587
+ "avg_mask_ratio": 0.47419991258066146,
588
+ "avg_response_length": 212.85,
589
+ "avg_student_mask_ratio": 0.47419991258066146,
590
+ "batch_ainp_frac": 0.0,
591
+ "batch_inp_frac": 1.0,
592
+ "batch_inp_oh_frac": 0.0,
593
+ "batch_inp_par_frac": 0.0,
594
+ "batch_inp_par_par_frac": 0.0,
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": 0.32057130943685763,
601
+ "epoch": 0.512,
602
+ "grad_norm": 0.43359375,
603
+ "kd_loss": 0.5083060303753086,
604
+ "learning_rate": 3e-06,
605
+ "loss": 0.6986,
606
+ "masked_tokens": 106.9,
607
+ "mean_t": 0.502228345896583,
608
+ "step": 240,
609
+ "student_masked_tokens": 106.9
610
+ },
611
+ {
612
+ "avg_mask_ratio": 0.4464209079160355,
613
+ "avg_response_length": 243.475,
614
+ "avg_student_mask_ratio": 0.4464209079160355,
615
+ "batch_ainp_frac": 0.0,
616
+ "batch_inp_frac": 1.0,
617
+ "batch_inp_oh_frac": 0.0,
618
+ "batch_inp_par_frac": 0.0,
619
+ "batch_inp_par_par_frac": 0.0,
620
+ "batch_inp_par_reverse_frac": 0.0,
621
+ "batch_rl_frac": 0.0,
622
+ "batch_sft_frac": 0.0,
623
+ "batch_soft_sft_frac": 0.0,
624
+ "batch_tf_frac": 0.0,
625
+ "ce_loss": 0.33636454603331456,
626
+ "epoch": 0.5333333333333333,
627
+ "grad_norm": 0.1142578125,
628
+ "kd_loss": 0.41649795620701296,
629
+ "learning_rate": 3e-06,
630
+ "loss": 0.5666,
631
+ "masked_tokens": 112.7375,
632
+ "mean_t": 0.4733429416548461,
633
+ "step": 250,
634
+ "student_masked_tokens": 112.7375
635
+ },
636
+ {
637
+ "avg_mask_ratio": 0.4520751796895638,
638
+ "avg_response_length": 245.55,
639
+ "avg_student_mask_ratio": 0.4520751796895638,
640
+ "batch_ainp_frac": 0.0,
641
+ "batch_inp_frac": 1.0,
642
+ "batch_inp_oh_frac": 0.0,
643
+ "batch_inp_par_frac": 0.0,
644
+ "batch_inp_par_par_frac": 0.0,
645
+ "batch_inp_par_reverse_frac": 0.0,
646
+ "batch_rl_frac": 0.0,
647
+ "batch_sft_frac": 0.0,
648
+ "batch_soft_sft_frac": 0.0,
649
+ "batch_tf_frac": 0.0,
650
+ "ce_loss": 0.37478437887749577,
651
+ "epoch": 0.5546666666666666,
652
+ "grad_norm": 0.328125,
653
+ "kd_loss": 0.31532439299670545,
654
+ "learning_rate": 3e-06,
655
+ "loss": 0.5129,
656
+ "masked_tokens": 109.6375,
657
+ "mean_t": 0.4843149524240289,
658
+ "step": 260,
659
+ "student_masked_tokens": 109.6375
660
+ },
661
+ {
662
+ "avg_mask_ratio": 0.5305180630879477,
663
+ "avg_response_length": 224.45,
664
+ "avg_student_mask_ratio": 0.5305180630879477,
665
+ "batch_ainp_frac": 0.0,
666
+ "batch_inp_frac": 1.0,
667
+ "batch_inp_oh_frac": 0.0,
668
+ "batch_inp_par_frac": 0.0,
669
+ "batch_inp_par_par_frac": 0.0,
670
+ "batch_inp_par_reverse_frac": 0.0,
671
+ "batch_rl_frac": 0.0,
672
+ "batch_sft_frac": 0.0,
673
+ "batch_soft_sft_frac": 0.0,
674
+ "batch_tf_frac": 0.0,
675
+ "ce_loss": 0.42709534656005416,
676
+ "epoch": 0.576,
677
+ "grad_norm": 0.7578125,
678
+ "kd_loss": 0.5525495689224045,
679
+ "learning_rate": 3e-06,
680
+ "loss": 0.812,
681
+ "masked_tokens": 120.475,
682
+ "mean_t": 0.5643589949700981,
683
+ "step": 270,
684
+ "student_masked_tokens": 120.475
685
+ },
686
+ {
687
+ "avg_mask_ratio": 0.46451686368091033,
688
+ "avg_response_length": 254.825,
689
+ "avg_student_mask_ratio": 0.46451686368091033,
690
+ "batch_ainp_frac": 0.0,
691
+ "batch_inp_frac": 1.0,
692
+ "batch_inp_oh_frac": 0.0,
693
+ "batch_inp_par_frac": 0.0,
694
+ "batch_inp_par_par_frac": 0.0,
695
+ "batch_inp_par_reverse_frac": 0.0,
696
+ "batch_rl_frac": 0.0,
697
+ "batch_sft_frac": 0.0,
698
+ "batch_soft_sft_frac": 0.0,
699
+ "batch_tf_frac": 0.0,
700
+ "ce_loss": 0.31382316479499084,
701
+ "epoch": 0.5973333333333334,
702
+ "grad_norm": 0.90234375,
703
+ "kd_loss": 0.3957495673693458,
704
+ "learning_rate": 3e-06,
705
+ "loss": 0.6028,
706
+ "masked_tokens": 129.225,
707
+ "mean_t": 0.47818811538163575,
708
+ "step": 280,
709
+ "student_masked_tokens": 129.225
710
+ },
711
+ {
712
+ "avg_mask_ratio": 0.389662017847877,
713
+ "avg_response_length": 245.9125,
714
+ "avg_student_mask_ratio": 0.389662017847877,
715
+ "batch_ainp_frac": 0.0,
716
+ "batch_inp_frac": 1.0,
717
+ "batch_inp_oh_frac": 0.0,
718
+ "batch_inp_par_frac": 0.0,
719
+ "batch_inp_par_par_frac": 0.0,
720
+ "batch_inp_par_reverse_frac": 0.0,
721
+ "batch_rl_frac": 0.0,
722
+ "batch_sft_frac": 0.0,
723
+ "batch_soft_sft_frac": 0.0,
724
+ "batch_tf_frac": 0.0,
725
+ "ce_loss": 0.23645576389110373,
726
+ "epoch": 0.6186666666666667,
727
+ "grad_norm": 0.302734375,
728
+ "kd_loss": 0.27728830450374853,
729
+ "learning_rate": 3e-06,
730
+ "loss": 0.4314,
731
+ "masked_tokens": 99.8625,
732
+ "mean_t": 0.4088635521940887,
733
+ "step": 290,
734
+ "student_masked_tokens": 99.8625
735
+ },
736
+ {
737
+ "avg_mask_ratio": 0.44417000194080175,
738
+ "avg_response_length": 217.0375,
739
+ "avg_student_mask_ratio": 0.44417000194080175,
740
+ "batch_ainp_frac": 0.0,
741
+ "batch_inp_frac": 1.0,
742
+ "batch_inp_oh_frac": 0.0,
743
+ "batch_inp_par_frac": 0.0,
744
+ "batch_inp_par_par_frac": 0.0,
745
+ "batch_inp_par_reverse_frac": 0.0,
746
+ "batch_rl_frac": 0.0,
747
+ "batch_sft_frac": 0.0,
748
+ "batch_soft_sft_frac": 0.0,
749
+ "batch_tf_frac": 0.0,
750
+ "ce_loss": 0.3240562055096575,
751
+ "epoch": 0.64,
752
+ "grad_norm": 1.09375,
753
+ "kd_loss": 0.31930388437995133,
754
+ "learning_rate": 3e-06,
755
+ "loss": 0.5264,
756
+ "masked_tokens": 104.625,
757
+ "mean_t": 0.47984200695063917,
758
+ "step": 300,
759
+ "student_masked_tokens": 104.625
760
+ },
761
+ {
762
+ "avg_mask_ratio": 0.4706685543409549,
763
+ "avg_response_length": 175.45,
764
+ "avg_student_mask_ratio": 0.4706685543409549,
765
+ "batch_ainp_frac": 0.0,
766
+ "batch_inp_frac": 1.0,
767
+ "batch_inp_oh_frac": 0.0,
768
+ "batch_inp_par_frac": 0.0,
769
+ "batch_inp_par_par_frac": 0.0,
770
+ "batch_inp_par_reverse_frac": 0.0,
771
+ "batch_rl_frac": 0.0,
772
+ "batch_sft_frac": 0.0,
773
+ "batch_soft_sft_frac": 0.0,
774
+ "batch_tf_frac": 0.0,
775
+ "ce_loss": 0.34333510200582396,
776
+ "epoch": 0.6613333333333333,
777
+ "grad_norm": 1.234375,
778
+ "kd_loss": 0.5067149527083984,
779
+ "learning_rate": 3e-06,
780
+ "loss": 0.6534,
781
+ "masked_tokens": 84.875,
782
+ "mean_t": 0.5026606284547597,
783
+ "step": 310,
784
+ "student_masked_tokens": 84.875
785
+ },
786
+ {
787
+ "avg_mask_ratio": 0.4974605386145413,
788
+ "avg_response_length": 234.7875,
789
+ "avg_student_mask_ratio": 0.4974605386145413,
790
+ "batch_ainp_frac": 0.0,
791
+ "batch_inp_frac": 1.0,
792
+ "batch_inp_oh_frac": 0.0,
793
+ "batch_inp_par_frac": 0.0,
794
+ "batch_inp_par_par_frac": 0.0,
795
+ "batch_inp_par_reverse_frac": 0.0,
796
+ "batch_rl_frac": 0.0,
797
+ "batch_sft_frac": 0.0,
798
+ "batch_soft_sft_frac": 0.0,
799
+ "batch_tf_frac": 0.0,
800
+ "ce_loss": 0.34462752127872137,
801
+ "epoch": 0.6826666666666666,
802
+ "grad_norm": 0.333984375,
803
+ "kd_loss": 0.3942846609736307,
804
+ "learning_rate": 3e-06,
805
+ "loss": 0.7133,
806
+ "masked_tokens": 119.6,
807
+ "mean_t": 0.5293499688967132,
808
+ "step": 320,
809
+ "student_masked_tokens": 119.6
810
+ },
811
+ {
812
+ "avg_mask_ratio": 0.5112370474264025,
813
+ "avg_response_length": 236.0625,
814
+ "avg_student_mask_ratio": 0.5112370474264025,
815
+ "batch_ainp_frac": 0.0,
816
+ "batch_inp_frac": 1.0,
817
+ "batch_inp_oh_frac": 0.0,
818
+ "batch_inp_par_frac": 0.0,
819
+ "batch_inp_par_par_frac": 0.0,
820
+ "batch_inp_par_reverse_frac": 0.0,
821
+ "batch_rl_frac": 0.0,
822
+ "batch_sft_frac": 0.0,
823
+ "batch_soft_sft_frac": 0.0,
824
+ "batch_tf_frac": 0.0,
825
+ "ce_loss": 0.2974585796398969,
826
+ "epoch": 0.704,
827
+ "grad_norm": 0.44140625,
828
+ "kd_loss": 0.4301003347501496,
829
+ "learning_rate": 3e-06,
830
+ "loss": 0.6754,
831
+ "masked_tokens": 129.425,
832
+ "mean_t": 0.5426030711154454,
833
+ "step": 330,
834
+ "student_masked_tokens": 129.425
835
+ },
836
+ {
837
+ "avg_mask_ratio": 0.44370225080056114,
838
+ "avg_response_length": 241.4875,
839
+ "avg_student_mask_ratio": 0.44370225080056114,
840
+ "batch_ainp_frac": 0.0,
841
+ "batch_inp_frac": 1.0,
842
+ "batch_inp_oh_frac": 0.0,
843
+ "batch_inp_par_frac": 0.0,
844
+ "batch_inp_par_par_frac": 0.0,
845
+ "batch_inp_par_reverse_frac": 0.0,
846
+ "batch_rl_frac": 0.0,
847
+ "batch_sft_frac": 0.0,
848
+ "batch_soft_sft_frac": 0.0,
849
+ "batch_tf_frac": 0.0,
850
+ "ce_loss": 0.3732590021626493,
851
+ "epoch": 0.7253333333333334,
852
+ "grad_norm": 0.98046875,
853
+ "kd_loss": 0.4610515360019235,
854
+ "learning_rate": 3e-06,
855
+ "loss": 0.6627,
856
+ "masked_tokens": 108.775,
857
+ "mean_t": 0.47635243807453664,
858
+ "step": 340,
859
+ "student_masked_tokens": 108.775
860
+ },
861
+ {
862
+ "avg_mask_ratio": 0.49959173843380994,
863
+ "avg_response_length": 235.6375,
864
+ "avg_student_mask_ratio": 0.49959173843380994,
865
+ "batch_ainp_frac": 0.0,
866
+ "batch_inp_frac": 1.0,
867
+ "batch_inp_oh_frac": 0.0,
868
+ "batch_inp_par_frac": 0.0,
869
+ "batch_inp_par_par_frac": 0.0,
870
+ "batch_inp_par_reverse_frac": 0.0,
871
+ "batch_rl_frac": 0.0,
872
+ "batch_sft_frac": 0.0,
873
+ "batch_soft_sft_frac": 0.0,
874
+ "batch_tf_frac": 0.0,
875
+ "ce_loss": 0.48515336151417615,
876
+ "epoch": 0.7466666666666667,
877
+ "grad_norm": 0.92578125,
878
+ "kd_loss": 0.5031771080357654,
879
+ "learning_rate": 3e-06,
880
+ "loss": 0.7668,
881
+ "masked_tokens": 125.625,
882
+ "mean_t": 0.5268881446914747,
883
+ "step": 350,
884
+ "student_masked_tokens": 125.625
885
+ },
886
+ {
887
+ "avg_mask_ratio": 0.4744729608530179,
888
+ "avg_response_length": 246.1625,
889
+ "avg_student_mask_ratio": 0.4744729608530179,
890
+ "batch_ainp_frac": 0.0,
891
+ "batch_inp_frac": 1.0,
892
+ "batch_inp_oh_frac": 0.0,
893
+ "batch_inp_par_frac": 0.0,
894
+ "batch_inp_par_par_frac": 0.0,
895
+ "batch_inp_par_reverse_frac": 0.0,
896
+ "batch_rl_frac": 0.0,
897
+ "batch_sft_frac": 0.0,
898
+ "batch_soft_sft_frac": 0.0,
899
+ "batch_tf_frac": 0.0,
900
+ "ce_loss": 0.3005135279950082,
901
+ "epoch": 0.768,
902
+ "grad_norm": 0.169921875,
903
+ "kd_loss": 0.5216399239409879,
904
+ "learning_rate": 3e-06,
905
+ "loss": 0.6077,
906
+ "masked_tokens": 116.875,
907
+ "mean_t": 0.5040419134311378,
908
+ "step": 360,
909
+ "student_masked_tokens": 116.875
910
+ },
911
+ {
912
+ "avg_mask_ratio": 0.4738045462174341,
913
+ "avg_response_length": 257.575,
914
+ "avg_student_mask_ratio": 0.4738045462174341,
915
+ "batch_ainp_frac": 0.0,
916
+ "batch_inp_frac": 1.0,
917
+ "batch_inp_oh_frac": 0.0,
918
+ "batch_inp_par_frac": 0.0,
919
+ "batch_inp_par_par_frac": 0.0,
920
+ "batch_inp_par_reverse_frac": 0.0,
921
+ "batch_rl_frac": 0.0,
922
+ "batch_sft_frac": 0.0,
923
+ "batch_soft_sft_frac": 0.0,
924
+ "batch_tf_frac": 0.0,
925
+ "ce_loss": 0.5349442186782426,
926
+ "epoch": 0.7893333333333333,
927
+ "grad_norm": 0.201171875,
928
+ "kd_loss": 0.6039233199480805,
929
+ "learning_rate": 3e-06,
930
+ "loss": 0.7196,
931
+ "masked_tokens": 127.4625,
932
+ "mean_t": 0.5127181728370488,
933
+ "step": 370,
934
+ "student_masked_tokens": 127.4625
935
+ },
936
+ {
937
+ "avg_mask_ratio": 0.4512475330149755,
938
+ "avg_response_length": 209.8,
939
+ "avg_student_mask_ratio": 0.4512475330149755,
940
+ "batch_ainp_frac": 0.0,
941
+ "batch_inp_frac": 1.0,
942
+ "batch_inp_oh_frac": 0.0,
943
+ "batch_inp_par_frac": 0.0,
944
+ "batch_inp_par_par_frac": 0.0,
945
+ "batch_inp_par_reverse_frac": 0.0,
946
+ "batch_rl_frac": 0.0,
947
+ "batch_sft_frac": 0.0,
948
+ "batch_soft_sft_frac": 0.0,
949
+ "batch_tf_frac": 0.0,
950
+ "ce_loss": 0.19145508916275275,
951
+ "epoch": 0.8106666666666666,
952
+ "grad_norm": 0.6875,
953
+ "kd_loss": 0.4029755606519984,
954
+ "learning_rate": 3e-06,
955
+ "loss": 0.5055,
956
+ "masked_tokens": 100.8375,
957
+ "mean_t": 0.4825185665744357,
958
+ "step": 380,
959
+ "student_masked_tokens": 100.8375
960
+ },
961
+ {
962
+ "avg_mask_ratio": 0.4752940105390735,
963
+ "avg_response_length": 219.5625,
964
+ "avg_student_mask_ratio": 0.4752940105390735,
965
+ "batch_ainp_frac": 0.0,
966
+ "batch_inp_frac": 1.0,
967
+ "batch_inp_oh_frac": 0.0,
968
+ "batch_inp_par_frac": 0.0,
969
+ "batch_inp_par_par_frac": 0.0,
970
+ "batch_inp_par_reverse_frac": 0.0,
971
+ "batch_rl_frac": 0.0,
972
+ "batch_sft_frac": 0.0,
973
+ "batch_soft_sft_frac": 0.0,
974
+ "batch_tf_frac": 0.0,
975
+ "ce_loss": 0.4267096655552223,
976
+ "epoch": 0.832,
977
+ "grad_norm": 0.2578125,
978
+ "kd_loss": 0.4655849843487971,
979
+ "learning_rate": 3e-06,
980
+ "loss": 0.6749,
981
+ "masked_tokens": 112.375,
982
+ "mean_t": 0.5053101469413377,
983
+ "step": 390,
984
+ "student_masked_tokens": 112.375
985
+ },
986
+ {
987
+ "avg_mask_ratio": 0.47461870914557946,
988
+ "avg_response_length": 242.6125,
989
+ "avg_student_mask_ratio": 0.47461870914557946,
990
+ "batch_ainp_frac": 0.0,
991
+ "batch_inp_frac": 1.0,
992
+ "batch_inp_oh_frac": 0.0,
993
+ "batch_inp_par_frac": 0.0,
994
+ "batch_inp_par_par_frac": 0.0,
995
+ "batch_inp_par_reverse_frac": 0.0,
996
+ "batch_rl_frac": 0.0,
997
+ "batch_sft_frac": 0.0,
998
+ "batch_soft_sft_frac": 0.0,
999
+ "batch_tf_frac": 0.0,
1000
+ "ce_loss": 0.27868834779033025,
1001
+ "epoch": 0.8533333333333334,
1002
+ "grad_norm": 0.640625,
1003
+ "kd_loss": 0.5299579592951205,
1004
+ "learning_rate": 3e-06,
1005
+ "loss": 0.6538,
1006
+ "masked_tokens": 120.4125,
1007
+ "mean_t": 0.5052250675857067,
1008
+ "step": 400,
1009
+ "student_masked_tokens": 120.4125
1010
+ },
1011
+ {
1012
+ "avg_mask_ratio": 0.48321815438685006,
1013
+ "avg_response_length": 228.15,
1014
+ "avg_student_mask_ratio": 0.48321815438685006,
1015
+ "batch_ainp_frac": 0.0,
1016
+ "batch_inp_frac": 1.0,
1017
+ "batch_inp_oh_frac": 0.0,
1018
+ "batch_inp_par_frac": 0.0,
1019
+ "batch_inp_par_par_frac": 0.0,
1020
+ "batch_inp_par_reverse_frac": 0.0,
1021
+ "batch_rl_frac": 0.0,
1022
+ "batch_sft_frac": 0.0,
1023
+ "batch_soft_sft_frac": 0.0,
1024
+ "batch_tf_frac": 0.0,
1025
+ "ce_loss": 0.43057951200541994,
1026
+ "epoch": 0.8746666666666667,
1027
+ "grad_norm": 0.5390625,
1028
+ "kd_loss": 0.504674318619719,
1029
+ "learning_rate": 3e-06,
1030
+ "loss": 0.7381,
1031
+ "masked_tokens": 119.0,
1032
+ "mean_t": 0.5050956419203431,
1033
+ "step": 410,
1034
+ "student_masked_tokens": 119.0
1035
+ },
1036
+ {
1037
+ "avg_mask_ratio": 0.4379329536575824,
1038
+ "avg_response_length": 220.225,
1039
+ "avg_student_mask_ratio": 0.4379329536575824,
1040
+ "batch_ainp_frac": 0.0,
1041
+ "batch_inp_frac": 1.0,
1042
+ "batch_inp_oh_frac": 0.0,
1043
+ "batch_inp_par_frac": 0.0,
1044
+ "batch_inp_par_par_frac": 0.0,
1045
+ "batch_inp_par_reverse_frac": 0.0,
1046
+ "batch_rl_frac": 0.0,
1047
+ "batch_sft_frac": 0.0,
1048
+ "batch_soft_sft_frac": 0.0,
1049
+ "batch_tf_frac": 0.0,
1050
+ "ce_loss": 0.132674143492045,
1051
+ "epoch": 0.896,
1052
+ "grad_norm": 1.09375,
1053
+ "kd_loss": 0.27731474525324984,
1054
+ "learning_rate": 3e-06,
1055
+ "loss": 0.3953,
1056
+ "masked_tokens": 85.525,
1057
+ "mean_t": 0.4769687672611326,
1058
+ "step": 420,
1059
+ "student_masked_tokens": 85.525
1060
+ },
1061
+ {
1062
+ "avg_mask_ratio": 0.4674084897618741,
1063
+ "avg_response_length": 249.2125,
1064
+ "avg_student_mask_ratio": 0.4674084897618741,
1065
+ "batch_ainp_frac": 0.0,
1066
+ "batch_inp_frac": 1.0,
1067
+ "batch_inp_oh_frac": 0.0,
1068
+ "batch_inp_par_frac": 0.0,
1069
+ "batch_inp_par_par_frac": 0.0,
1070
+ "batch_inp_par_reverse_frac": 0.0,
1071
+ "batch_rl_frac": 0.0,
1072
+ "batch_sft_frac": 0.0,
1073
+ "batch_soft_sft_frac": 0.0,
1074
+ "batch_tf_frac": 0.0,
1075
+ "ce_loss": 0.37605725416574387,
1076
+ "epoch": 0.9173333333333333,
1077
+ "grad_norm": 0.43359375,
1078
+ "kd_loss": 0.49442086774362226,
1079
+ "learning_rate": 3e-06,
1080
+ "loss": 0.6699,
1081
+ "masked_tokens": 104.5625,
1082
+ "mean_t": 0.49262027950026094,
1083
+ "step": 430,
1084
+ "student_masked_tokens": 104.5625
1085
+ },
1086
+ {
1087
+ "avg_mask_ratio": 0.4415457699564286,
1088
+ "avg_response_length": 241.0875,
1089
+ "avg_student_mask_ratio": 0.4415457699564286,
1090
+ "batch_ainp_frac": 0.0,
1091
+ "batch_inp_frac": 1.0,
1092
+ "batch_inp_oh_frac": 0.0,
1093
+ "batch_inp_par_frac": 0.0,
1094
+ "batch_inp_par_par_frac": 0.0,
1095
+ "batch_inp_par_reverse_frac": 0.0,
1096
+ "batch_rl_frac": 0.0,
1097
+ "batch_sft_frac": 0.0,
1098
+ "batch_soft_sft_frac": 0.0,
1099
+ "batch_tf_frac": 0.0,
1100
+ "ce_loss": 0.3754083825901603,
1101
+ "epoch": 0.9386666666666666,
1102
+ "grad_norm": 0.6328125,
1103
+ "kd_loss": 0.45159815376919143,
1104
+ "learning_rate": 3e-06,
1105
+ "loss": 0.6585,
1106
+ "masked_tokens": 113.0875,
1107
+ "mean_t": 0.47046207524836064,
1108
+ "step": 440,
1109
+ "student_masked_tokens": 113.0875
1110
+ },
1111
+ {
1112
+ "avg_mask_ratio": 0.42486972180195154,
1113
+ "avg_response_length": 231.9875,
1114
+ "avg_student_mask_ratio": 0.42486972180195154,
1115
+ "batch_ainp_frac": 0.0,
1116
+ "batch_inp_frac": 1.0,
1117
+ "batch_inp_oh_frac": 0.0,
1118
+ "batch_inp_par_frac": 0.0,
1119
+ "batch_inp_par_par_frac": 0.0,
1120
+ "batch_inp_par_reverse_frac": 0.0,
1121
+ "batch_rl_frac": 0.0,
1122
+ "batch_sft_frac": 0.0,
1123
+ "batch_soft_sft_frac": 0.0,
1124
+ "batch_tf_frac": 0.0,
1125
+ "ce_loss": 0.32457938515717616,
1126
+ "epoch": 0.96,
1127
+ "grad_norm": 0.6953125,
1128
+ "kd_loss": 0.4011907008050457,
1129
+ "learning_rate": 3e-06,
1130
+ "loss": 0.5644,
1131
+ "masked_tokens": 103.4,
1132
+ "mean_t": 0.45781184462830427,
1133
+ "step": 450,
1134
+ "student_masked_tokens": 103.4
1135
+ },
1136
+ {
1137
+ "avg_mask_ratio": 0.47578654896933587,
1138
+ "avg_response_length": 214.6125,
1139
+ "avg_student_mask_ratio": 0.47578654896933587,
1140
+ "batch_ainp_frac": 0.0,
1141
+ "batch_inp_frac": 1.0,
1142
+ "batch_inp_oh_frac": 0.0,
1143
+ "batch_inp_par_frac": 0.0,
1144
+ "batch_inp_par_par_frac": 0.0,
1145
+ "batch_inp_par_reverse_frac": 0.0,
1146
+ "batch_rl_frac": 0.0,
1147
+ "batch_sft_frac": 0.0,
1148
+ "batch_soft_sft_frac": 0.0,
1149
+ "batch_tf_frac": 0.0,
1150
+ "ce_loss": 0.32885359905767475,
1151
+ "epoch": 0.9813333333333333,
1152
+ "grad_norm": 0.16015625,
1153
+ "kd_loss": 0.44463847501747294,
1154
+ "learning_rate": 3e-06,
1155
+ "loss": 0.635,
1156
+ "masked_tokens": 105.3125,
1157
+ "mean_t": 0.5075790266972036,
1158
+ "step": 460,
1159
+ "student_masked_tokens": 105.3125
1160
+ },
1161
+ {
1162
+ "avg_mask_ratio": 0.4782901787132557,
1163
+ "avg_response_length": 224.0952380952381,
1164
+ "avg_student_mask_ratio": 0.4782901787132557,
1165
+ "batch_ainp_frac": 0.0,
1166
+ "batch_inp_frac": 1.0,
1167
+ "batch_inp_oh_frac": 0.0,
1168
+ "batch_inp_par_frac": 0.0,
1169
+ "batch_inp_par_par_frac": 0.0,
1170
+ "batch_inp_par_reverse_frac": 0.0,
1171
+ "batch_rl_frac": 0.0,
1172
+ "batch_sft_frac": 0.0,
1173
+ "batch_soft_sft_frac": 0.0,
1174
+ "batch_tf_frac": 0.0,
1175
+ "ce_loss": 0.3393430382851702,
1176
+ "epoch": 1.0042666666666666,
1177
+ "grad_norm": 0.65625,
1178
+ "kd_loss": 0.5178591865708675,
1179
+ "learning_rate": 3e-06,
1180
+ "loss": 0.7769,
1181
+ "masked_tokens": 107.23809523809524,
1182
+ "mean_t": 0.5031429776822084,
1183
+ "step": 470,
1184
+ "student_masked_tokens": 107.23809523809524
1185
+ },
1186
+ {
1187
+ "avg_mask_ratio": 0.47575968883465974,
1188
+ "avg_response_length": 249.4125,
1189
+ "avg_student_mask_ratio": 0.47575968883465974,
1190
+ "batch_ainp_frac": 0.0,
1191
+ "batch_inp_frac": 1.0,
1192
+ "batch_inp_oh_frac": 0.0,
1193
+ "batch_inp_par_frac": 0.0,
1194
+ "batch_inp_par_par_frac": 0.0,
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": 0.44613247805159517,
1201
+ "epoch": 1.0256,
1202
+ "grad_norm": 0.498046875,
1203
+ "kd_loss": 0.5374264506522252,
1204
+ "learning_rate": 3e-06,
1205
+ "loss": 0.6772,
1206
+ "masked_tokens": 118.35,
1207
+ "mean_t": 0.504472183593316,
1208
+ "step": 480,
1209
+ "student_masked_tokens": 118.35
1210
+ },
1211
+ {
1212
+ "avg_mask_ratio": 0.4563717324635945,
1213
+ "avg_response_length": 232.0375,
1214
+ "avg_student_mask_ratio": 0.4563717324635945,
1215
+ "batch_ainp_frac": 0.0,
1216
+ "batch_inp_frac": 1.0,
1217
+ "batch_inp_oh_frac": 0.0,
1218
+ "batch_inp_par_frac": 0.0,
1219
+ "batch_inp_par_par_frac": 0.0,
1220
+ "batch_inp_par_reverse_frac": 0.0,
1221
+ "batch_rl_frac": 0.0,
1222
+ "batch_sft_frac": 0.0,
1223
+ "batch_soft_sft_frac": 0.0,
1224
+ "batch_tf_frac": 0.0,
1225
+ "ce_loss": 0.37626147485414096,
1226
+ "epoch": 1.0469333333333333,
1227
+ "grad_norm": 0.54296875,
1228
+ "kd_loss": 0.392788901903657,
1229
+ "learning_rate": 3e-06,
1230
+ "loss": 0.6047,
1231
+ "masked_tokens": 98.35,
1232
+ "mean_t": 0.4888980514719151,
1233
+ "step": 490,
1234
+ "student_masked_tokens": 98.35
1235
+ },
1236
+ {
1237
+ "avg_mask_ratio": 0.5079968665260821,
1238
+ "avg_response_length": 253.7875,
1239
+ "avg_student_mask_ratio": 0.5079968665260821,
1240
+ "batch_ainp_frac": 0.0,
1241
+ "batch_inp_frac": 1.0,
1242
+ "batch_inp_oh_frac": 0.0,
1243
+ "batch_inp_par_frac": 0.0,
1244
+ "batch_inp_par_par_frac": 0.0,
1245
+ "batch_inp_par_reverse_frac": 0.0,
1246
+ "batch_rl_frac": 0.0,
1247
+ "batch_sft_frac": 0.0,
1248
+ "batch_soft_sft_frac": 0.0,
1249
+ "batch_tf_frac": 0.0,
1250
+ "ce_loss": 0.30954629559880686,
1251
+ "epoch": 1.0682666666666667,
1252
+ "grad_norm": 0.291015625,
1253
+ "kd_loss": 0.4563873354276211,
1254
+ "learning_rate": 3e-06,
1255
+ "loss": 0.5996,
1256
+ "masked_tokens": 128.225,
1257
+ "mean_t": 0.5469163245841628,
1258
+ "step": 500,
1259
+ "student_masked_tokens": 128.225
1260
+ },
1261
+ {
1262
+ "avg_mask_ratio": 0.5109448074479588,
1263
+ "avg_response_length": 254.2,
1264
+ "avg_student_mask_ratio": 0.5109448074479588,
1265
+ "batch_ainp_frac": 0.0,
1266
+ "batch_inp_frac": 1.0,
1267
+ "batch_inp_oh_frac": 0.0,
1268
+ "batch_inp_par_frac": 0.0,
1269
+ "batch_inp_par_par_frac": 0.0,
1270
+ "batch_inp_par_reverse_frac": 0.0,
1271
+ "batch_rl_frac": 0.0,
1272
+ "batch_sft_frac": 0.0,
1273
+ "batch_soft_sft_frac": 0.0,
1274
+ "batch_tf_frac": 0.0,
1275
+ "ce_loss": 0.2868076219221166,
1276
+ "epoch": 1.0896,
1277
+ "grad_norm": 2.515625,
1278
+ "kd_loss": 0.5652106747879998,
1279
+ "learning_rate": 3e-06,
1280
+ "loss": 0.6398,
1281
+ "masked_tokens": 137.5875,
1282
+ "mean_t": 0.5275314710394013,
1283
+ "step": 510,
1284
+ "student_masked_tokens": 137.5875
1285
+ },
1286
+ {
1287
+ "avg_mask_ratio": 0.45396183808334173,
1288
+ "avg_response_length": 202.7625,
1289
+ "avg_student_mask_ratio": 0.45396183808334173,
1290
+ "batch_ainp_frac": 0.0,
1291
+ "batch_inp_frac": 1.0,
1292
+ "batch_inp_oh_frac": 0.0,
1293
+ "batch_inp_par_frac": 0.0,
1294
+ "batch_inp_par_par_frac": 0.0,
1295
+ "batch_inp_par_reverse_frac": 0.0,
1296
+ "batch_rl_frac": 0.0,
1297
+ "batch_sft_frac": 0.0,
1298
+ "batch_soft_sft_frac": 0.0,
1299
+ "batch_tf_frac": 0.0,
1300
+ "ce_loss": 0.38311037250946356,
1301
+ "epoch": 1.1109333333333333,
1302
+ "grad_norm": 0.6171875,
1303
+ "kd_loss": 0.423658079797778,
1304
+ "learning_rate": 3e-06,
1305
+ "loss": 0.6386,
1306
+ "masked_tokens": 87.0625,
1307
+ "mean_t": 0.49193521235138177,
1308
+ "step": 520,
1309
+ "student_masked_tokens": 87.0625
1310
+ },
1311
+ {
1312
+ "avg_mask_ratio": 0.47015948037151245,
1313
+ "avg_response_length": 214.275,
1314
+ "avg_student_mask_ratio": 0.47015948037151245,
1315
+ "batch_ainp_frac": 0.0,
1316
+ "batch_inp_frac": 1.0,
1317
+ "batch_inp_oh_frac": 0.0,
1318
+ "batch_inp_par_frac": 0.0,
1319
+ "batch_inp_par_par_frac": 0.0,
1320
+ "batch_inp_par_reverse_frac": 0.0,
1321
+ "batch_rl_frac": 0.0,
1322
+ "batch_sft_frac": 0.0,
1323
+ "batch_soft_sft_frac": 0.0,
1324
+ "batch_tf_frac": 0.0,
1325
+ "ce_loss": 0.47228433731506814,
1326
+ "epoch": 1.1322666666666668,
1327
+ "grad_norm": 0.609375,
1328
+ "kd_loss": 0.45688082203427316,
1329
+ "learning_rate": 3e-06,
1330
+ "loss": 0.737,
1331
+ "masked_tokens": 99.8625,
1332
+ "mean_t": 0.49621942077938,
1333
+ "step": 530,
1334
+ "student_masked_tokens": 99.8625
1335
+ },
1336
+ {
1337
+ "avg_mask_ratio": 0.4892866689246148,
1338
+ "avg_response_length": 231.3125,
1339
+ "avg_student_mask_ratio": 0.4892866689246148,
1340
+ "batch_ainp_frac": 0.0,
1341
+ "batch_inp_frac": 1.0,
1342
+ "batch_inp_oh_frac": 0.0,
1343
+ "batch_inp_par_frac": 0.0,
1344
+ "batch_inp_par_par_frac": 0.0,
1345
+ "batch_inp_par_reverse_frac": 0.0,
1346
+ "batch_rl_frac": 0.0,
1347
+ "batch_sft_frac": 0.0,
1348
+ "batch_soft_sft_frac": 0.0,
1349
+ "batch_tf_frac": 0.0,
1350
+ "ce_loss": 0.4080867745911064,
1351
+ "epoch": 1.1536,
1352
+ "grad_norm": 0.341796875,
1353
+ "kd_loss": 0.5618651450654625,
1354
+ "learning_rate": 3e-06,
1355
+ "loss": 0.6922,
1356
+ "masked_tokens": 107.375,
1357
+ "mean_t": 0.5208023569080978,
1358
+ "step": 540,
1359
+ "student_masked_tokens": 107.375
1360
+ },
1361
+ {
1362
+ "avg_mask_ratio": 0.4541942774085328,
1363
+ "avg_response_length": 213.525,
1364
+ "avg_student_mask_ratio": 0.4541942774085328,
1365
+ "batch_ainp_frac": 0.0,
1366
+ "batch_inp_frac": 1.0,
1367
+ "batch_inp_oh_frac": 0.0,
1368
+ "batch_inp_par_frac": 0.0,
1369
+ "batch_inp_par_par_frac": 0.0,
1370
+ "batch_inp_par_reverse_frac": 0.0,
1371
+ "batch_rl_frac": 0.0,
1372
+ "batch_sft_frac": 0.0,
1373
+ "batch_soft_sft_frac": 0.0,
1374
+ "batch_tf_frac": 0.0,
1375
+ "ce_loss": 0.22217674175137744,
1376
+ "epoch": 1.1749333333333334,
1377
+ "grad_norm": 0.2412109375,
1378
+ "kd_loss": 0.3673438885498399,
1379
+ "learning_rate": 3e-06,
1380
+ "loss": 0.5008,
1381
+ "masked_tokens": 97.8875,
1382
+ "mean_t": 0.4767197913257405,
1383
+ "step": 550,
1384
+ "student_masked_tokens": 97.8875
1385
+ },
1386
+ {
1387
+ "avg_mask_ratio": 0.39282396506750955,
1388
+ "avg_response_length": 231.4125,
1389
+ "avg_student_mask_ratio": 0.39282396506750955,
1390
+ "batch_ainp_frac": 0.0,
1391
+ "batch_inp_frac": 1.0,
1392
+ "batch_inp_oh_frac": 0.0,
1393
+ "batch_inp_par_frac": 0.0,
1394
+ "batch_inp_par_par_frac": 0.0,
1395
+ "batch_inp_par_reverse_frac": 0.0,
1396
+ "batch_rl_frac": 0.0,
1397
+ "batch_sft_frac": 0.0,
1398
+ "batch_soft_sft_frac": 0.0,
1399
+ "batch_tf_frac": 0.0,
1400
+ "ce_loss": 0.3512847523151777,
1401
+ "epoch": 1.1962666666666666,
1402
+ "grad_norm": 0.8828125,
1403
+ "kd_loss": 0.48686740984790616,
1404
+ "learning_rate": 3e-06,
1405
+ "loss": 0.5823,
1406
+ "masked_tokens": 99.2875,
1407
+ "mean_t": 0.4111072298779618,
1408
+ "step": 560,
1409
+ "student_masked_tokens": 99.2875
1410
+ },
1411
+ {
1412
+ "avg_mask_ratio": 0.4483634108910337,
1413
+ "avg_response_length": 230.1625,
1414
+ "avg_student_mask_ratio": 0.4483634108910337,
1415
+ "batch_ainp_frac": 0.0,
1416
+ "batch_inp_frac": 1.0,
1417
+ "batch_inp_oh_frac": 0.0,
1418
+ "batch_inp_par_frac": 0.0,
1419
+ "batch_inp_par_par_frac": 0.0,
1420
+ "batch_inp_par_reverse_frac": 0.0,
1421
+ "batch_rl_frac": 0.0,
1422
+ "batch_sft_frac": 0.0,
1423
+ "batch_soft_sft_frac": 0.0,
1424
+ "batch_tf_frac": 0.0,
1425
+ "ce_loss": 0.31345968546206676,
1426
+ "epoch": 1.2176,
1427
+ "grad_norm": 0.4453125,
1428
+ "kd_loss": 0.41564053312727084,
1429
+ "learning_rate": 3e-06,
1430
+ "loss": 0.5898,
1431
+ "masked_tokens": 108.9875,
1432
+ "mean_t": 0.48533305872697385,
1433
+ "step": 570,
1434
+ "student_masked_tokens": 108.9875
1435
+ },
1436
+ {
1437
+ "avg_mask_ratio": 0.465452536707744,
1438
+ "avg_response_length": 267.4375,
1439
+ "avg_student_mask_ratio": 0.465452536707744,
1440
+ "batch_ainp_frac": 0.0,
1441
+ "batch_inp_frac": 1.0,
1442
+ "batch_inp_oh_frac": 0.0,
1443
+ "batch_inp_par_frac": 0.0,
1444
+ "batch_inp_par_par_frac": 0.0,
1445
+ "batch_inp_par_reverse_frac": 0.0,
1446
+ "batch_rl_frac": 0.0,
1447
+ "batch_sft_frac": 0.0,
1448
+ "batch_soft_sft_frac": 0.0,
1449
+ "batch_tf_frac": 0.0,
1450
+ "ce_loss": 0.3618907347364768,
1451
+ "epoch": 1.2389333333333332,
1452
+ "grad_norm": 8.6875,
1453
+ "kd_loss": 0.4481006292516895,
1454
+ "learning_rate": 3e-06,
1455
+ "loss": 0.6314,
1456
+ "masked_tokens": 129.075,
1457
+ "mean_t": 0.49976949762785805,
1458
+ "step": 580,
1459
+ "student_masked_tokens": 129.075
1460
+ },
1461
+ {
1462
+ "avg_mask_ratio": 0.5225977989146486,
1463
+ "avg_response_length": 228.45,
1464
+ "avg_student_mask_ratio": 0.5225977989146486,
1465
+ "batch_ainp_frac": 0.0,
1466
+ "batch_inp_frac": 1.0,
1467
+ "batch_inp_oh_frac": 0.0,
1468
+ "batch_inp_par_frac": 0.0,
1469
+ "batch_inp_par_par_frac": 0.0,
1470
+ "batch_inp_par_reverse_frac": 0.0,
1471
+ "batch_rl_frac": 0.0,
1472
+ "batch_sft_frac": 0.0,
1473
+ "batch_soft_sft_frac": 0.0,
1474
+ "batch_tf_frac": 0.0,
1475
+ "ce_loss": 0.5639314363695348,
1476
+ "epoch": 1.2602666666666666,
1477
+ "grad_norm": 1.1328125,
1478
+ "kd_loss": 0.5351108588445992,
1479
+ "learning_rate": 3e-06,
1480
+ "loss": 0.8274,
1481
+ "masked_tokens": 121.675,
1482
+ "mean_t": 0.5521843038732186,
1483
+ "step": 590,
1484
+ "student_masked_tokens": 121.675
1485
+ },
1486
+ {
1487
+ "avg_mask_ratio": 0.44998724836623294,
1488
+ "avg_response_length": 236.7,
1489
+ "avg_student_mask_ratio": 0.44998724836623294,
1490
+ "batch_ainp_frac": 0.0,
1491
+ "batch_inp_frac": 1.0,
1492
+ "batch_inp_oh_frac": 0.0,
1493
+ "batch_inp_par_frac": 0.0,
1494
+ "batch_inp_par_par_frac": 0.0,
1495
+ "batch_inp_par_reverse_frac": 0.0,
1496
+ "batch_rl_frac": 0.0,
1497
+ "batch_sft_frac": 0.0,
1498
+ "batch_soft_sft_frac": 0.0,
1499
+ "batch_tf_frac": 0.0,
1500
+ "ce_loss": 0.3396833263838971,
1501
+ "epoch": 1.2816,
1502
+ "grad_norm": 0.365234375,
1503
+ "kd_loss": 0.41761890975592914,
1504
+ "learning_rate": 3e-06,
1505
+ "loss": 0.5752,
1506
+ "masked_tokens": 110.1625,
1507
+ "mean_t": 0.4788527532829903,
1508
+ "step": 600,
1509
+ "student_masked_tokens": 110.1625
1510
+ },
1511
+ {
1512
+ "avg_mask_ratio": 0.5042130865273066,
1513
+ "avg_response_length": 230.3375,
1514
+ "avg_student_mask_ratio": 0.5042130865273066,
1515
+ "batch_ainp_frac": 0.0,
1516
+ "batch_inp_frac": 1.0,
1517
+ "batch_inp_oh_frac": 0.0,
1518
+ "batch_inp_par_frac": 0.0,
1519
+ "batch_inp_par_par_frac": 0.0,
1520
+ "batch_inp_par_reverse_frac": 0.0,
1521
+ "batch_rl_frac": 0.0,
1522
+ "batch_sft_frac": 0.0,
1523
+ "batch_soft_sft_frac": 0.0,
1524
+ "batch_tf_frac": 0.0,
1525
+ "ce_loss": 0.35890077192343595,
1526
+ "epoch": 1.3029333333333333,
1527
+ "grad_norm": 0.28515625,
1528
+ "kd_loss": 0.5558427174539929,
1529
+ "learning_rate": 3e-06,
1530
+ "loss": 0.7657,
1531
+ "masked_tokens": 112.625,
1532
+ "mean_t": 0.5445419924799353,
1533
+ "step": 610,
1534
+ "student_masked_tokens": 112.625
1535
+ },
1536
+ {
1537
+ "avg_mask_ratio": 0.49637898594373836,
1538
+ "avg_response_length": 233.0625,
1539
+ "avg_student_mask_ratio": 0.49637898594373836,
1540
+ "batch_ainp_frac": 0.0,
1541
+ "batch_inp_frac": 1.0,
1542
+ "batch_inp_oh_frac": 0.0,
1543
+ "batch_inp_par_frac": 0.0,
1544
+ "batch_inp_par_par_frac": 0.0,
1545
+ "batch_inp_par_reverse_frac": 0.0,
1546
+ "batch_rl_frac": 0.0,
1547
+ "batch_sft_frac": 0.0,
1548
+ "batch_soft_sft_frac": 0.0,
1549
+ "batch_tf_frac": 0.0,
1550
+ "ce_loss": 0.32318839170733327,
1551
+ "epoch": 1.3242666666666667,
1552
+ "grad_norm": 0.515625,
1553
+ "kd_loss": 0.5518322235134179,
1554
+ "learning_rate": 3e-06,
1555
+ "loss": 0.6742,
1556
+ "masked_tokens": 111.25,
1557
+ "mean_t": 0.52490478400141,
1558
+ "step": 620,
1559
+ "student_masked_tokens": 111.25
1560
+ },
1561
+ {
1562
+ "avg_mask_ratio": 0.5177568581304512,
1563
+ "avg_response_length": 257.2125,
1564
+ "avg_student_mask_ratio": 0.5177568581304512,
1565
+ "batch_ainp_frac": 0.0,
1566
+ "batch_inp_frac": 1.0,
1567
+ "batch_inp_oh_frac": 0.0,
1568
+ "batch_inp_par_frac": 0.0,
1569
+ "batch_inp_par_par_frac": 0.0,
1570
+ "batch_inp_par_reverse_frac": 0.0,
1571
+ "batch_rl_frac": 0.0,
1572
+ "batch_sft_frac": 0.0,
1573
+ "batch_soft_sft_frac": 0.0,
1574
+ "batch_tf_frac": 0.0,
1575
+ "ce_loss": 0.5710563842050931,
1576
+ "epoch": 1.3456000000000001,
1577
+ "grad_norm": 1.3515625,
1578
+ "kd_loss": 0.5316411310721378,
1579
+ "learning_rate": 3e-06,
1580
+ "loss": 0.8598,
1581
+ "masked_tokens": 129.6125,
1582
+ "mean_t": 0.5564947265549562,
1583
+ "step": 630,
1584
+ "student_masked_tokens": 129.6125
1585
+ },
1586
+ {
1587
+ "avg_mask_ratio": 0.48226998368045315,
1588
+ "avg_response_length": 237.7125,
1589
+ "avg_student_mask_ratio": 0.48226998368045315,
1590
+ "batch_ainp_frac": 0.0,
1591
+ "batch_inp_frac": 1.0,
1592
+ "batch_inp_oh_frac": 0.0,
1593
+ "batch_inp_par_frac": 0.0,
1594
+ "batch_inp_par_par_frac": 0.0,
1595
+ "batch_inp_par_reverse_frac": 0.0,
1596
+ "batch_rl_frac": 0.0,
1597
+ "batch_sft_frac": 0.0,
1598
+ "batch_soft_sft_frac": 0.0,
1599
+ "batch_tf_frac": 0.0,
1600
+ "ce_loss": 0.2804489129174499,
1601
+ "epoch": 1.3669333333333333,
1602
+ "grad_norm": 0.2421875,
1603
+ "kd_loss": 0.3663112932188085,
1604
+ "learning_rate": 3e-06,
1605
+ "loss": 0.4584,
1606
+ "masked_tokens": 120.275,
1607
+ "mean_t": 0.5093393943971023,
1608
+ "step": 640,
1609
+ "student_masked_tokens": 120.275
1610
+ },
1611
+ {
1612
+ "avg_mask_ratio": 0.5306948523037136,
1613
+ "avg_response_length": 238.0125,
1614
+ "avg_student_mask_ratio": 0.5306948523037136,
1615
+ "batch_ainp_frac": 0.0,
1616
+ "batch_inp_frac": 1.0,
1617
+ "batch_inp_oh_frac": 0.0,
1618
+ "batch_inp_par_frac": 0.0,
1619
+ "batch_inp_par_par_frac": 0.0,
1620
+ "batch_inp_par_reverse_frac": 0.0,
1621
+ "batch_rl_frac": 0.0,
1622
+ "batch_sft_frac": 0.0,
1623
+ "batch_soft_sft_frac": 0.0,
1624
+ "batch_tf_frac": 0.0,
1625
+ "ce_loss": 0.475157093159612,
1626
+ "epoch": 1.3882666666666665,
1627
+ "grad_norm": 1.8125,
1628
+ "kd_loss": 0.5062341513834724,
1629
+ "learning_rate": 3e-06,
1630
+ "loss": 0.7115,
1631
+ "masked_tokens": 133.25,
1632
+ "mean_t": 0.5558586571365595,
1633
+ "step": 650,
1634
+ "student_masked_tokens": 133.25
1635
+ },
1636
+ {
1637
+ "avg_mask_ratio": 0.4821273953886703,
1638
+ "avg_response_length": 247.775,
1639
+ "avg_student_mask_ratio": 0.4821273953886703,
1640
+ "batch_ainp_frac": 0.0,
1641
+ "batch_inp_frac": 1.0,
1642
+ "batch_inp_oh_frac": 0.0,
1643
+ "batch_inp_par_frac": 0.0,
1644
+ "batch_inp_par_par_frac": 0.0,
1645
+ "batch_inp_par_reverse_frac": 0.0,
1646
+ "batch_rl_frac": 0.0,
1647
+ "batch_sft_frac": 0.0,
1648
+ "batch_soft_sft_frac": 0.0,
1649
+ "batch_tf_frac": 0.0,
1650
+ "ce_loss": 0.41770620119971225,
1651
+ "epoch": 1.4096,
1652
+ "grad_norm": 0.9375,
1653
+ "kd_loss": 0.425496905214095,
1654
+ "learning_rate": 3e-06,
1655
+ "loss": 0.6361,
1656
+ "masked_tokens": 128.875,
1657
+ "mean_t": 0.51307404555846,
1658
+ "step": 660,
1659
+ "student_masked_tokens": 128.875
1660
+ },
1661
+ {
1662
+ "avg_mask_ratio": 0.46056515555246735,
1663
+ "avg_response_length": 240.4375,
1664
+ "avg_student_mask_ratio": 0.46056515555246735,
1665
+ "batch_ainp_frac": 0.0,
1666
+ "batch_inp_frac": 1.0,
1667
+ "batch_inp_oh_frac": 0.0,
1668
+ "batch_inp_par_frac": 0.0,
1669
+ "batch_inp_par_par_frac": 0.0,
1670
+ "batch_inp_par_reverse_frac": 0.0,
1671
+ "batch_rl_frac": 0.0,
1672
+ "batch_sft_frac": 0.0,
1673
+ "batch_soft_sft_frac": 0.0,
1674
+ "batch_tf_frac": 0.0,
1675
+ "ce_loss": 0.24846992658117414,
1676
+ "epoch": 1.4309333333333334,
1677
+ "grad_norm": 0.60546875,
1678
+ "kd_loss": 0.34861083538812637,
1679
+ "learning_rate": 3e-06,
1680
+ "loss": 0.5112,
1681
+ "masked_tokens": 119.85,
1682
+ "mean_t": 0.4907285622088239,
1683
+ "step": 670,
1684
+ "student_masked_tokens": 119.85
1685
+ },
1686
+ {
1687
+ "avg_mask_ratio": 0.4666106043441687,
1688
+ "avg_response_length": 226.7375,
1689
+ "avg_student_mask_ratio": 0.4666106043441687,
1690
+ "batch_ainp_frac": 0.0,
1691
+ "batch_inp_frac": 1.0,
1692
+ "batch_inp_oh_frac": 0.0,
1693
+ "batch_inp_par_frac": 0.0,
1694
+ "batch_inp_par_par_frac": 0.0,
1695
+ "batch_inp_par_reverse_frac": 0.0,
1696
+ "batch_rl_frac": 0.0,
1697
+ "batch_sft_frac": 0.0,
1698
+ "batch_soft_sft_frac": 0.0,
1699
+ "batch_tf_frac": 0.0,
1700
+ "ce_loss": 0.4541423492493323,
1701
+ "epoch": 1.4522666666666666,
1702
+ "grad_norm": 0.51953125,
1703
+ "kd_loss": 0.4910934407485213,
1704
+ "learning_rate": 3e-06,
1705
+ "loss": 0.6946,
1706
+ "masked_tokens": 107.4625,
1707
+ "mean_t": 0.4913603452499956,
1708
+ "step": 680,
1709
+ "student_masked_tokens": 107.4625
1710
+ },
1711
+ {
1712
+ "avg_mask_ratio": 0.4790851596510038,
1713
+ "avg_response_length": 202.05,
1714
+ "avg_student_mask_ratio": 0.4790851596510038,
1715
+ "batch_ainp_frac": 0.0,
1716
+ "batch_inp_frac": 1.0,
1717
+ "batch_inp_oh_frac": 0.0,
1718
+ "batch_inp_par_frac": 0.0,
1719
+ "batch_inp_par_par_frac": 0.0,
1720
+ "batch_inp_par_reverse_frac": 0.0,
1721
+ "batch_rl_frac": 0.0,
1722
+ "batch_sft_frac": 0.0,
1723
+ "batch_soft_sft_frac": 0.0,
1724
+ "batch_tf_frac": 0.0,
1725
+ "ce_loss": 0.3711260147189023,
1726
+ "epoch": 1.4736,
1727
+ "grad_norm": 2.03125,
1728
+ "kd_loss": 0.41718243765291446,
1729
+ "learning_rate": 3e-06,
1730
+ "loss": 0.6313,
1731
+ "masked_tokens": 111.3125,
1732
+ "mean_t": 0.5133644798654131,
1733
+ "step": 690,
1734
+ "student_masked_tokens": 111.3125
1735
+ },
1736
+ {
1737
+ "avg_mask_ratio": 0.5250519359949977,
1738
+ "avg_response_length": 228.125,
1739
+ "avg_student_mask_ratio": 0.5250519359949977,
1740
+ "batch_ainp_frac": 0.0,
1741
+ "batch_inp_frac": 1.0,
1742
+ "batch_inp_oh_frac": 0.0,
1743
+ "batch_inp_par_frac": 0.0,
1744
+ "batch_inp_par_par_frac": 0.0,
1745
+ "batch_inp_par_reverse_frac": 0.0,
1746
+ "batch_rl_frac": 0.0,
1747
+ "batch_sft_frac": 0.0,
1748
+ "batch_soft_sft_frac": 0.0,
1749
+ "batch_tf_frac": 0.0,
1750
+ "ce_loss": 0.22230932631540554,
1751
+ "epoch": 1.4949333333333334,
1752
+ "grad_norm": 0.26171875,
1753
+ "kd_loss": 0.6619142963969352,
1754
+ "learning_rate": 3e-06,
1755
+ "loss": 0.7717,
1756
+ "masked_tokens": 132.55,
1757
+ "mean_t": 0.5625698395539075,
1758
+ "step": 700,
1759
+ "student_masked_tokens": 132.55
1760
+ },
1761
+ {
1762
+ "avg_mask_ratio": 0.4790433386107907,
1763
+ "avg_response_length": 212.5,
1764
+ "avg_student_mask_ratio": 0.4790433386107907,
1765
+ "batch_ainp_frac": 0.0,
1766
+ "batch_inp_frac": 1.0,
1767
+ "batch_inp_oh_frac": 0.0,
1768
+ "batch_inp_par_frac": 0.0,
1769
+ "batch_inp_par_par_frac": 0.0,
1770
+ "batch_inp_par_reverse_frac": 0.0,
1771
+ "batch_rl_frac": 0.0,
1772
+ "batch_sft_frac": 0.0,
1773
+ "batch_soft_sft_frac": 0.0,
1774
+ "batch_tf_frac": 0.0,
1775
+ "ce_loss": 0.24621229091012536,
1776
+ "epoch": 1.5162666666666667,
1777
+ "grad_norm": 0.2099609375,
1778
+ "kd_loss": 0.43454050603151584,
1779
+ "learning_rate": 3e-06,
1780
+ "loss": 0.5302,
1781
+ "masked_tokens": 108.7375,
1782
+ "mean_t": 0.5135623761918395,
1783
+ "step": 710,
1784
+ "student_masked_tokens": 108.7375
1785
+ },
1786
+ {
1787
+ "avg_mask_ratio": 0.47950589570682495,
1788
+ "avg_response_length": 227.075,
1789
+ "avg_student_mask_ratio": 0.47950589570682495,
1790
+ "batch_ainp_frac": 0.0,
1791
+ "batch_inp_frac": 1.0,
1792
+ "batch_inp_oh_frac": 0.0,
1793
+ "batch_inp_par_frac": 0.0,
1794
+ "batch_inp_par_par_frac": 0.0,
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": 0.36416104665024707,
1801
+ "epoch": 1.5375999999999999,
1802
+ "grad_norm": 0.75,
1803
+ "kd_loss": 0.5665610315164941,
1804
+ "learning_rate": 3e-06,
1805
+ "loss": 0.7121,
1806
+ "masked_tokens": 110.8,
1807
+ "mean_t": 0.5117021896177902,
1808
+ "step": 720,
1809
+ "student_masked_tokens": 110.8
1810
+ },
1811
+ {
1812
+ "avg_mask_ratio": 0.4604924251558259,
1813
+ "avg_response_length": 232.925,
1814
+ "avg_student_mask_ratio": 0.4604924251558259,
1815
+ "batch_ainp_frac": 0.0,
1816
+ "batch_inp_frac": 1.0,
1817
+ "batch_inp_oh_frac": 0.0,
1818
+ "batch_inp_par_frac": 0.0,
1819
+ "batch_inp_par_par_frac": 0.0,
1820
+ "batch_inp_par_reverse_frac": 0.0,
1821
+ "batch_rl_frac": 0.0,
1822
+ "batch_sft_frac": 0.0,
1823
+ "batch_soft_sft_frac": 0.0,
1824
+ "batch_tf_frac": 0.0,
1825
+ "ce_loss": 0.38923927966282007,
1826
+ "epoch": 1.5589333333333333,
1827
+ "grad_norm": 1.015625,
1828
+ "kd_loss": 0.4302867329986782,
1829
+ "learning_rate": 3e-06,
1830
+ "loss": 0.639,
1831
+ "masked_tokens": 104.9625,
1832
+ "mean_t": 0.49050743713742123,
1833
+ "step": 730,
1834
+ "student_masked_tokens": 104.9625
1835
+ },
1836
+ {
1837
+ "avg_mask_ratio": 0.5185885130194947,
1838
+ "avg_response_length": 183.325,
1839
+ "avg_student_mask_ratio": 0.5185885130194947,
1840
+ "batch_ainp_frac": 0.0,
1841
+ "batch_inp_frac": 1.0,
1842
+ "batch_inp_oh_frac": 0.0,
1843
+ "batch_inp_par_frac": 0.0,
1844
+ "batch_inp_par_par_frac": 0.0,
1845
+ "batch_inp_par_reverse_frac": 0.0,
1846
+ "batch_rl_frac": 0.0,
1847
+ "batch_sft_frac": 0.0,
1848
+ "batch_soft_sft_frac": 0.0,
1849
+ "batch_tf_frac": 0.0,
1850
+ "ce_loss": 0.3361817517367399,
1851
+ "epoch": 1.5802666666666667,
1852
+ "grad_norm": 0.40234375,
1853
+ "kd_loss": 0.5340734164818514,
1854
+ "learning_rate": 3e-06,
1855
+ "loss": 0.7461,
1856
+ "masked_tokens": 97.125,
1857
+ "mean_t": 0.5505168779753149,
1858
+ "step": 740,
1859
+ "student_masked_tokens": 97.125
1860
+ },
1861
+ {
1862
+ "avg_mask_ratio": 0.4191439319110941,
1863
+ "avg_response_length": 223.65,
1864
+ "avg_student_mask_ratio": 0.4191439319110941,
1865
+ "batch_ainp_frac": 0.0,
1866
+ "batch_inp_frac": 1.0,
1867
+ "batch_inp_oh_frac": 0.0,
1868
+ "batch_inp_par_frac": 0.0,
1869
+ "batch_inp_par_par_frac": 0.0,
1870
+ "batch_inp_par_reverse_frac": 0.0,
1871
+ "batch_rl_frac": 0.0,
1872
+ "batch_sft_frac": 0.0,
1873
+ "batch_soft_sft_frac": 0.0,
1874
+ "batch_tf_frac": 0.0,
1875
+ "ce_loss": 0.37429177601145514,
1876
+ "epoch": 1.6016,
1877
+ "grad_norm": 0.58203125,
1878
+ "kd_loss": 0.5036597276406856,
1879
+ "learning_rate": 3e-06,
1880
+ "loss": 0.6491,
1881
+ "masked_tokens": 95.3125,
1882
+ "mean_t": 0.4437690361432033,
1883
+ "step": 750,
1884
+ "student_masked_tokens": 95.3125
1885
+ },
1886
+ {
1887
+ "avg_mask_ratio": 0.46706983938929625,
1888
+ "avg_response_length": 216.0625,
1889
+ "avg_student_mask_ratio": 0.46706983938929625,
1890
+ "batch_ainp_frac": 0.0,
1891
+ "batch_inp_frac": 1.0,
1892
+ "batch_inp_oh_frac": 0.0,
1893
+ "batch_inp_par_frac": 0.0,
1894
+ "batch_inp_par_par_frac": 0.0,
1895
+ "batch_inp_par_reverse_frac": 0.0,
1896
+ "batch_rl_frac": 0.0,
1897
+ "batch_sft_frac": 0.0,
1898
+ "batch_soft_sft_frac": 0.0,
1899
+ "batch_tf_frac": 0.0,
1900
+ "ce_loss": 0.4449058656399984,
1901
+ "epoch": 1.6229333333333333,
1902
+ "grad_norm": 0.8203125,
1903
+ "kd_loss": 0.5661326096985168,
1904
+ "learning_rate": 3e-06,
1905
+ "loss": 0.7233,
1906
+ "masked_tokens": 107.7,
1907
+ "mean_t": 0.49132869170280175,
1908
+ "step": 760,
1909
+ "student_masked_tokens": 107.7
1910
+ },
1911
+ {
1912
+ "avg_mask_ratio": 0.44156218122225255,
1913
+ "avg_response_length": 259.675,
1914
+ "avg_student_mask_ratio": 0.44156218122225255,
1915
+ "batch_ainp_frac": 0.0,
1916
+ "batch_inp_frac": 1.0,
1917
+ "batch_inp_oh_frac": 0.0,
1918
+ "batch_inp_par_frac": 0.0,
1919
+ "batch_inp_par_par_frac": 0.0,
1920
+ "batch_inp_par_reverse_frac": 0.0,
1921
+ "batch_rl_frac": 0.0,
1922
+ "batch_sft_frac": 0.0,
1923
+ "batch_soft_sft_frac": 0.0,
1924
+ "batch_tf_frac": 0.0,
1925
+ "ce_loss": 0.25899335961771613,
1926
+ "epoch": 1.6442666666666668,
1927
+ "grad_norm": 0.396484375,
1928
+ "kd_loss": 0.4095979654902003,
1929
+ "learning_rate": 3e-06,
1930
+ "loss": 0.5099,
1931
+ "masked_tokens": 117.5,
1932
+ "mean_t": 0.4667695587326307,
1933
+ "step": 770,
1934
+ "student_masked_tokens": 117.5
1935
+ },
1936
+ {
1937
+ "avg_mask_ratio": 0.42836043585848527,
1938
+ "avg_response_length": 258.5125,
1939
+ "avg_student_mask_ratio": 0.42836043585848527,
1940
+ "batch_ainp_frac": 0.0,
1941
+ "batch_inp_frac": 1.0,
1942
+ "batch_inp_oh_frac": 0.0,
1943
+ "batch_inp_par_frac": 0.0,
1944
+ "batch_inp_par_par_frac": 0.0,
1945
+ "batch_inp_par_reverse_frac": 0.0,
1946
+ "batch_rl_frac": 0.0,
1947
+ "batch_sft_frac": 0.0,
1948
+ "batch_soft_sft_frac": 0.0,
1949
+ "batch_tf_frac": 0.0,
1950
+ "ce_loss": 0.2897560694203321,
1951
+ "epoch": 1.6656,
1952
+ "grad_norm": 0.2431640625,
1953
+ "kd_loss": 0.34635278815572546,
1954
+ "learning_rate": 3e-06,
1955
+ "loss": 0.4802,
1956
+ "masked_tokens": 119.0125,
1957
+ "mean_t": 0.44942845597106496,
1958
+ "step": 780,
1959
+ "student_masked_tokens": 119.0125
1960
+ },
1961
+ {
1962
+ "avg_mask_ratio": 0.46589430308085866,
1963
+ "avg_response_length": 222.3125,
1964
+ "avg_student_mask_ratio": 0.46589430308085866,
1965
+ "batch_ainp_frac": 0.0,
1966
+ "batch_inp_frac": 1.0,
1967
+ "batch_inp_oh_frac": 0.0,
1968
+ "batch_inp_par_frac": 0.0,
1969
+ "batch_inp_par_par_frac": 0.0,
1970
+ "batch_inp_par_reverse_frac": 0.0,
1971
+ "batch_rl_frac": 0.0,
1972
+ "batch_sft_frac": 0.0,
1973
+ "batch_soft_sft_frac": 0.0,
1974
+ "batch_tf_frac": 0.0,
1975
+ "ce_loss": 0.21603642557238345,
1976
+ "epoch": 1.6869333333333332,
1977
+ "grad_norm": 0.140625,
1978
+ "kd_loss": 0.33674514803767297,
1979
+ "learning_rate": 3e-06,
1980
+ "loss": 0.489,
1981
+ "masked_tokens": 103.25,
1982
+ "mean_t": 0.4993515375303105,
1983
+ "step": 790,
1984
+ "student_masked_tokens": 103.25
1985
+ },
1986
+ {
1987
+ "avg_mask_ratio": 0.46366424662992356,
1988
+ "avg_response_length": 219.6875,
1989
+ "avg_student_mask_ratio": 0.46366424662992356,
1990
+ "batch_ainp_frac": 0.0,
1991
+ "batch_inp_frac": 1.0,
1992
+ "batch_inp_oh_frac": 0.0,
1993
+ "batch_inp_par_frac": 0.0,
1994
+ "batch_inp_par_par_frac": 0.0,
1995
+ "batch_inp_par_reverse_frac": 0.0,
1996
+ "batch_rl_frac": 0.0,
1997
+ "batch_sft_frac": 0.0,
1998
+ "batch_soft_sft_frac": 0.0,
1999
+ "batch_tf_frac": 0.0,
2000
+ "ce_loss": 0.2663005536277069,
2001
+ "epoch": 1.7082666666666668,
2002
+ "grad_norm": 0.23828125,
2003
+ "kd_loss": 0.35138718315538425,
2004
+ "learning_rate": 3e-06,
2005
+ "loss": 0.5434,
2006
+ "masked_tokens": 104.5,
2007
+ "mean_t": 0.500370389316231,
2008
+ "step": 800,
2009
+ "student_masked_tokens": 104.5
2010
+ },
2011
+ {
2012
+ "avg_mask_ratio": 0.503375941584818,
2013
+ "avg_response_length": 237.85,
2014
+ "avg_student_mask_ratio": 0.503375941584818,
2015
+ "batch_ainp_frac": 0.0,
2016
+ "batch_inp_frac": 1.0,
2017
+ "batch_inp_oh_frac": 0.0,
2018
+ "batch_inp_par_frac": 0.0,
2019
+ "batch_inp_par_par_frac": 0.0,
2020
+ "batch_inp_par_reverse_frac": 0.0,
2021
+ "batch_rl_frac": 0.0,
2022
+ "batch_sft_frac": 0.0,
2023
+ "batch_soft_sft_frac": 0.0,
2024
+ "batch_tf_frac": 0.0,
2025
+ "ce_loss": 0.4813590554784753,
2026
+ "epoch": 1.7296,
2027
+ "grad_norm": 1.6015625,
2028
+ "kd_loss": 0.45312339970045057,
2029
+ "learning_rate": 3e-06,
2030
+ "loss": 0.706,
2031
+ "masked_tokens": 118.2125,
2032
+ "mean_t": 0.5317009104182944,
2033
+ "step": 810,
2034
+ "student_masked_tokens": 118.2125
2035
+ },
2036
+ {
2037
+ "avg_mask_ratio": 0.5110091455746442,
2038
+ "avg_response_length": 209.0875,
2039
+ "avg_student_mask_ratio": 0.5110091455746442,
2040
+ "batch_ainp_frac": 0.0,
2041
+ "batch_inp_frac": 1.0,
2042
+ "batch_inp_oh_frac": 0.0,
2043
+ "batch_inp_par_frac": 0.0,
2044
+ "batch_inp_par_par_frac": 0.0,
2045
+ "batch_inp_par_reverse_frac": 0.0,
2046
+ "batch_rl_frac": 0.0,
2047
+ "batch_sft_frac": 0.0,
2048
+ "batch_soft_sft_frac": 0.0,
2049
+ "batch_tf_frac": 0.0,
2050
+ "ce_loss": 0.4535834654417954,
2051
+ "epoch": 1.7509333333333332,
2052
+ "grad_norm": 0.70703125,
2053
+ "kd_loss": 0.5985253949772413,
2054
+ "learning_rate": 3e-06,
2055
+ "loss": 0.7794,
2056
+ "masked_tokens": 120.95,
2057
+ "mean_t": 0.5392061032878701,
2058
+ "step": 820,
2059
+ "student_masked_tokens": 120.95
2060
+ },
2061
+ {
2062
+ "avg_mask_ratio": 0.49899387182667854,
2063
+ "avg_response_length": 263.975,
2064
+ "avg_student_mask_ratio": 0.49899387182667854,
2065
+ "batch_ainp_frac": 0.0,
2066
+ "batch_inp_frac": 1.0,
2067
+ "batch_inp_oh_frac": 0.0,
2068
+ "batch_inp_par_frac": 0.0,
2069
+ "batch_inp_par_par_frac": 0.0,
2070
+ "batch_inp_par_reverse_frac": 0.0,
2071
+ "batch_rl_frac": 0.0,
2072
+ "batch_sft_frac": 0.0,
2073
+ "batch_soft_sft_frac": 0.0,
2074
+ "batch_tf_frac": 0.0,
2075
+ "ce_loss": 0.40083689643704473,
2076
+ "epoch": 1.7722666666666667,
2077
+ "grad_norm": 0.1708984375,
2078
+ "kd_loss": 0.5644028104892641,
2079
+ "learning_rate": 3e-06,
2080
+ "loss": 0.7632,
2081
+ "masked_tokens": 137.075,
2082
+ "mean_t": 0.5238314627087675,
2083
+ "step": 830,
2084
+ "student_masked_tokens": 137.075
2085
+ },
2086
+ {
2087
+ "avg_mask_ratio": 0.4997270987310912,
2088
+ "avg_response_length": 221.9,
2089
+ "avg_student_mask_ratio": 0.4997270987310912,
2090
+ "batch_ainp_frac": 0.0,
2091
+ "batch_inp_frac": 1.0,
2092
+ "batch_inp_oh_frac": 0.0,
2093
+ "batch_inp_par_frac": 0.0,
2094
+ "batch_inp_par_par_frac": 0.0,
2095
+ "batch_inp_par_reverse_frac": 0.0,
2096
+ "batch_rl_frac": 0.0,
2097
+ "batch_sft_frac": 0.0,
2098
+ "batch_soft_sft_frac": 0.0,
2099
+ "batch_tf_frac": 0.0,
2100
+ "ce_loss": 0.2870929398425915,
2101
+ "epoch": 1.7936,
2102
+ "grad_norm": 0.345703125,
2103
+ "kd_loss": 0.4698917509396324,
2104
+ "learning_rate": 3e-06,
2105
+ "loss": 0.6327,
2106
+ "masked_tokens": 114.525,
2107
+ "mean_t": 0.5301066277665086,
2108
+ "step": 840,
2109
+ "student_masked_tokens": 114.525
2110
+ },
2111
+ {
2112
+ "avg_mask_ratio": 0.4988076956477016,
2113
+ "avg_response_length": 225.5,
2114
+ "avg_student_mask_ratio": 0.4988076956477016,
2115
+ "batch_ainp_frac": 0.0,
2116
+ "batch_inp_frac": 1.0,
2117
+ "batch_inp_oh_frac": 0.0,
2118
+ "batch_inp_par_frac": 0.0,
2119
+ "batch_inp_par_par_frac": 0.0,
2120
+ "batch_inp_par_reverse_frac": 0.0,
2121
+ "batch_rl_frac": 0.0,
2122
+ "batch_sft_frac": 0.0,
2123
+ "batch_soft_sft_frac": 0.0,
2124
+ "batch_tf_frac": 0.0,
2125
+ "ce_loss": 0.3023421537889817,
2126
+ "epoch": 1.8149333333333333,
2127
+ "grad_norm": 0.443359375,
2128
+ "kd_loss": 0.3271854338312551,
2129
+ "learning_rate": 3e-06,
2130
+ "loss": 0.5634,
2131
+ "masked_tokens": 116.9125,
2132
+ "mean_t": 0.5343429344706238,
2133
+ "step": 850,
2134
+ "student_masked_tokens": 116.9125
2135
+ },
2136
+ {
2137
+ "avg_mask_ratio": 0.4635998342186213,
2138
+ "avg_response_length": 229.125,
2139
+ "avg_student_mask_ratio": 0.4635998342186213,
2140
+ "batch_ainp_frac": 0.0,
2141
+ "batch_inp_frac": 1.0,
2142
+ "batch_inp_oh_frac": 0.0,
2143
+ "batch_inp_par_frac": 0.0,
2144
+ "batch_inp_par_par_frac": 0.0,
2145
+ "batch_inp_par_reverse_frac": 0.0,
2146
+ "batch_rl_frac": 0.0,
2147
+ "batch_sft_frac": 0.0,
2148
+ "batch_soft_sft_frac": 0.0,
2149
+ "batch_tf_frac": 0.0,
2150
+ "ce_loss": 0.37467331880507115,
2151
+ "epoch": 1.8362666666666667,
2152
+ "grad_norm": 0.384765625,
2153
+ "kd_loss": 0.4431717619034316,
2154
+ "learning_rate": 3e-06,
2155
+ "loss": 0.5956,
2156
+ "masked_tokens": 109.675,
2157
+ "mean_t": 0.4791536889737472,
2158
+ "step": 860,
2159
+ "student_masked_tokens": 109.675
2160
+ },
2161
+ {
2162
+ "avg_mask_ratio": 0.49111039767740294,
2163
+ "avg_response_length": 229.1,
2164
+ "avg_student_mask_ratio": 0.49111039767740294,
2165
+ "batch_ainp_frac": 0.0,
2166
+ "batch_inp_frac": 1.0,
2167
+ "batch_inp_oh_frac": 0.0,
2168
+ "batch_inp_par_frac": 0.0,
2169
+ "batch_inp_par_par_frac": 0.0,
2170
+ "batch_inp_par_reverse_frac": 0.0,
2171
+ "batch_rl_frac": 0.0,
2172
+ "batch_sft_frac": 0.0,
2173
+ "batch_soft_sft_frac": 0.0,
2174
+ "batch_tf_frac": 0.0,
2175
+ "ce_loss": 0.3838037288314126,
2176
+ "epoch": 1.8576000000000001,
2177
+ "grad_norm": 0.333984375,
2178
+ "kd_loss": 0.47523635068355363,
2179
+ "learning_rate": 3e-06,
2180
+ "loss": 0.6859,
2181
+ "masked_tokens": 115.6625,
2182
+ "mean_t": 0.5203817339061061,
2183
+ "step": 870,
2184
+ "student_masked_tokens": 115.6625
2185
+ },
2186
+ {
2187
+ "avg_mask_ratio": 0.4427660425659269,
2188
+ "avg_response_length": 198.5625,
2189
+ "avg_student_mask_ratio": 0.4427660425659269,
2190
+ "batch_ainp_frac": 0.0,
2191
+ "batch_inp_frac": 1.0,
2192
+ "batch_inp_oh_frac": 0.0,
2193
+ "batch_inp_par_frac": 0.0,
2194
+ "batch_inp_par_par_frac": 0.0,
2195
+ "batch_inp_par_reverse_frac": 0.0,
2196
+ "batch_rl_frac": 0.0,
2197
+ "batch_sft_frac": 0.0,
2198
+ "batch_soft_sft_frac": 0.0,
2199
+ "batch_tf_frac": 0.0,
2200
+ "ce_loss": 0.33109274096627817,
2201
+ "epoch": 1.8789333333333333,
2202
+ "grad_norm": 1.0859375,
2203
+ "kd_loss": 0.46695662873548827,
2204
+ "learning_rate": 3e-06,
2205
+ "loss": 0.6284,
2206
+ "masked_tokens": 91.175,
2207
+ "mean_t": 0.4875184997683391,
2208
+ "step": 880,
2209
+ "student_masked_tokens": 91.175
2210
+ },
2211
+ {
2212
+ "avg_mask_ratio": 0.4464349385118112,
2213
+ "avg_response_length": 225.8375,
2214
+ "avg_student_mask_ratio": 0.4464349385118112,
2215
+ "batch_ainp_frac": 0.0,
2216
+ "batch_inp_frac": 1.0,
2217
+ "batch_inp_oh_frac": 0.0,
2218
+ "batch_inp_par_frac": 0.0,
2219
+ "batch_inp_par_par_frac": 0.0,
2220
+ "batch_inp_par_reverse_frac": 0.0,
2221
+ "batch_rl_frac": 0.0,
2222
+ "batch_sft_frac": 0.0,
2223
+ "batch_soft_sft_frac": 0.0,
2224
+ "batch_tf_frac": 0.0,
2225
+ "ce_loss": 0.22229116438190885,
2226
+ "epoch": 1.9002666666666665,
2227
+ "grad_norm": 0.12890625,
2228
+ "kd_loss": 0.4006316699657759,
2229
+ "learning_rate": 3e-06,
2230
+ "loss": 0.4934,
2231
+ "masked_tokens": 101.75,
2232
+ "mean_t": 0.4766692223958671,
2233
+ "step": 890,
2234
+ "student_masked_tokens": 101.75
2235
+ },
2236
+ {
2237
+ "avg_mask_ratio": 0.44976164362160487,
2238
+ "avg_response_length": 227.7875,
2239
+ "avg_student_mask_ratio": 0.44976164362160487,
2240
+ "batch_ainp_frac": 0.0,
2241
+ "batch_inp_frac": 1.0,
2242
+ "batch_inp_oh_frac": 0.0,
2243
+ "batch_inp_par_frac": 0.0,
2244
+ "batch_inp_par_par_frac": 0.0,
2245
+ "batch_inp_par_reverse_frac": 0.0,
2246
+ "batch_rl_frac": 0.0,
2247
+ "batch_sft_frac": 0.0,
2248
+ "batch_soft_sft_frac": 0.0,
2249
+ "batch_tf_frac": 0.0,
2250
+ "ce_loss": 0.38169105723031577,
2251
+ "epoch": 1.9216,
2252
+ "grad_norm": 1.765625,
2253
+ "kd_loss": 0.47280531010078086,
2254
+ "learning_rate": 3e-06,
2255
+ "loss": 0.6337,
2256
+ "masked_tokens": 103.475,
2257
+ "mean_t": 0.487134758150205,
2258
+ "step": 900,
2259
+ "student_masked_tokens": 103.475
2260
+ },
2261
+ {
2262
+ "avg_mask_ratio": 0.475579984736396,
2263
+ "avg_response_length": 245.1625,
2264
+ "avg_student_mask_ratio": 0.475579984736396,
2265
+ "batch_ainp_frac": 0.0,
2266
+ "batch_inp_frac": 1.0,
2267
+ "batch_inp_oh_frac": 0.0,
2268
+ "batch_inp_par_frac": 0.0,
2269
+ "batch_inp_par_par_frac": 0.0,
2270
+ "batch_inp_par_reverse_frac": 0.0,
2271
+ "batch_rl_frac": 0.0,
2272
+ "batch_sft_frac": 0.0,
2273
+ "batch_soft_sft_frac": 0.0,
2274
+ "batch_tf_frac": 0.0,
2275
+ "ce_loss": 0.27549623605577833,
2276
+ "epoch": 1.9429333333333334,
2277
+ "grad_norm": 0.451171875,
2278
+ "kd_loss": 0.4638562942510987,
2279
+ "learning_rate": 3e-06,
2280
+ "loss": 0.5387,
2281
+ "masked_tokens": 124.6375,
2282
+ "mean_t": 0.5027793228859082,
2283
+ "step": 910,
2284
+ "student_masked_tokens": 124.6375
2285
+ },
2286
+ {
2287
+ "avg_mask_ratio": 0.4688875659601763,
2288
+ "avg_response_length": 226.2875,
2289
+ "avg_student_mask_ratio": 0.4688875659601763,
2290
+ "batch_ainp_frac": 0.0,
2291
+ "batch_inp_frac": 1.0,
2292
+ "batch_inp_oh_frac": 0.0,
2293
+ "batch_inp_par_frac": 0.0,
2294
+ "batch_inp_par_par_frac": 0.0,
2295
+ "batch_inp_par_reverse_frac": 0.0,
2296
+ "batch_rl_frac": 0.0,
2297
+ "batch_sft_frac": 0.0,
2298
+ "batch_soft_sft_frac": 0.0,
2299
+ "batch_tf_frac": 0.0,
2300
+ "ce_loss": 0.2772836374151325,
2301
+ "epoch": 1.9642666666666666,
2302
+ "grad_norm": 0.416015625,
2303
+ "kd_loss": 0.44530672791033793,
2304
+ "learning_rate": 3e-06,
2305
+ "loss": 0.6177,
2306
+ "masked_tokens": 110.0125,
2307
+ "mean_t": 0.49417946098838,
2308
+ "step": 920,
2309
+ "student_masked_tokens": 110.0125
2310
+ },
2311
+ {
2312
+ "avg_mask_ratio": 0.47999348094454036,
2313
+ "avg_response_length": 237.05,
2314
+ "avg_student_mask_ratio": 0.47999348094454036,
2315
+ "batch_ainp_frac": 0.0,
2316
+ "batch_inp_frac": 1.0,
2317
+ "batch_inp_oh_frac": 0.0,
2318
+ "batch_inp_par_frac": 0.0,
2319
+ "batch_inp_par_par_frac": 0.0,
2320
+ "batch_inp_par_reverse_frac": 0.0,
2321
+ "batch_rl_frac": 0.0,
2322
+ "batch_sft_frac": 0.0,
2323
+ "batch_soft_sft_frac": 0.0,
2324
+ "batch_tf_frac": 0.0,
2325
+ "ce_loss": 0.2901802834984665,
2326
+ "epoch": 1.9856,
2327
+ "grad_norm": 0.37890625,
2328
+ "kd_loss": 0.4553093938939094,
2329
+ "learning_rate": 3e-06,
2330
+ "loss": 0.5905,
2331
+ "masked_tokens": 121.6,
2332
+ "mean_t": 0.5045580042526125,
2333
+ "step": 930,
2334
+ "student_masked_tokens": 121.6
2335
+ },
2336
+ {
2337
+ "avg_mask_ratio": 0.49413903727240505,
2338
+ "avg_response_length": 224.79761904761904,
2339
+ "avg_student_mask_ratio": 0.49413903727240505,
2340
+ "batch_ainp_frac": 0.0,
2341
+ "batch_inp_frac": 1.0,
2342
+ "batch_inp_oh_frac": 0.0,
2343
+ "batch_inp_par_frac": 0.0,
2344
+ "batch_inp_par_par_frac": 0.0,
2345
+ "batch_inp_par_reverse_frac": 0.0,
2346
+ "batch_rl_frac": 0.0,
2347
+ "batch_sft_frac": 0.0,
2348
+ "batch_soft_sft_frac": 0.0,
2349
+ "batch_tf_frac": 0.0,
2350
+ "ce_loss": 0.37941894131193166,
2351
+ "epoch": 2.0085333333333333,
2352
+ "grad_norm": 0.4921875,
2353
+ "kd_loss": 0.4946319753903075,
2354
+ "learning_rate": 3e-06,
2355
+ "loss": 0.6668,
2356
+ "masked_tokens": 120.5,
2357
+ "mean_t": 0.5321138524893849,
2358
+ "step": 940,
2359
+ "student_masked_tokens": 120.5
2360
+ },
2361
+ {
2362
+ "avg_mask_ratio": 0.4368605303927325,
2363
+ "avg_response_length": 240.9125,
2364
+ "avg_student_mask_ratio": 0.4368605303927325,
2365
+ "batch_ainp_frac": 0.0,
2366
+ "batch_inp_frac": 1.0,
2367
+ "batch_inp_oh_frac": 0.0,
2368
+ "batch_inp_par_frac": 0.0,
2369
+ "batch_inp_par_par_frac": 0.0,
2370
+ "batch_inp_par_reverse_frac": 0.0,
2371
+ "batch_rl_frac": 0.0,
2372
+ "batch_sft_frac": 0.0,
2373
+ "batch_soft_sft_frac": 0.0,
2374
+ "batch_tf_frac": 0.0,
2375
+ "ce_loss": 0.22575005246883392,
2376
+ "epoch": 2.0298666666666665,
2377
+ "grad_norm": 1.1875,
2378
+ "kd_loss": 0.4342805288508771,
2379
+ "learning_rate": 3e-06,
2380
+ "loss": 0.5248,
2381
+ "masked_tokens": 111.4125,
2382
+ "mean_t": 0.4632946296595037,
2383
+ "step": 950,
2384
+ "student_masked_tokens": 111.4125
2385
+ },
2386
+ {
2387
+ "avg_mask_ratio": 0.4988762516761199,
2388
+ "avg_response_length": 275.3,
2389
+ "avg_student_mask_ratio": 0.4988762516761199,
2390
+ "batch_ainp_frac": 0.0,
2391
+ "batch_inp_frac": 1.0,
2392
+ "batch_inp_oh_frac": 0.0,
2393
+ "batch_inp_par_frac": 0.0,
2394
+ "batch_inp_par_par_frac": 0.0,
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": 0.49722497609602667,
2401
+ "epoch": 2.0512,
2402
+ "grad_norm": 0.40625,
2403
+ "kd_loss": 0.5839257182941765,
2404
+ "learning_rate": 3e-06,
2405
+ "loss": 0.7523,
2406
+ "masked_tokens": 143.825,
2407
+ "mean_t": 0.5198000721400604,
2408
+ "step": 960,
2409
+ "student_masked_tokens": 143.825
2410
+ },
2411
+ {
2412
+ "avg_mask_ratio": 0.437801384011982,
2413
+ "avg_response_length": 236.2375,
2414
+ "avg_student_mask_ratio": 0.437801384011982,
2415
+ "batch_ainp_frac": 0.0,
2416
+ "batch_inp_frac": 1.0,
2417
+ "batch_inp_oh_frac": 0.0,
2418
+ "batch_inp_par_frac": 0.0,
2419
+ "batch_inp_par_par_frac": 0.0,
2420
+ "batch_inp_par_reverse_frac": 0.0,
2421
+ "batch_rl_frac": 0.0,
2422
+ "batch_sft_frac": 0.0,
2423
+ "batch_soft_sft_frac": 0.0,
2424
+ "batch_tf_frac": 0.0,
2425
+ "ce_loss": 0.2855980422358698,
2426
+ "epoch": 2.0725333333333333,
2427
+ "grad_norm": 0.4765625,
2428
+ "kd_loss": 0.35673561348757377,
2429
+ "learning_rate": 3e-06,
2430
+ "loss": 0.538,
2431
+ "masked_tokens": 107.025,
2432
+ "mean_t": 0.4703940597362816,
2433
+ "step": 970,
2434
+ "student_masked_tokens": 107.025
2435
+ },
2436
+ {
2437
+ "avg_mask_ratio": 0.42220073882490394,
2438
+ "avg_response_length": 230.8625,
2439
+ "avg_student_mask_ratio": 0.42220073882490394,
2440
+ "batch_ainp_frac": 0.0,
2441
+ "batch_inp_frac": 1.0,
2442
+ "batch_inp_oh_frac": 0.0,
2443
+ "batch_inp_par_frac": 0.0,
2444
+ "batch_inp_par_par_frac": 0.0,
2445
+ "batch_inp_par_reverse_frac": 0.0,
2446
+ "batch_rl_frac": 0.0,
2447
+ "batch_sft_frac": 0.0,
2448
+ "batch_soft_sft_frac": 0.0,
2449
+ "batch_tf_frac": 0.0,
2450
+ "ce_loss": 0.2906558813129777,
2451
+ "epoch": 2.0938666666666665,
2452
+ "grad_norm": 0.466796875,
2453
+ "kd_loss": 0.36284122784349504,
2454
+ "learning_rate": 3e-06,
2455
+ "loss": 0.4889,
2456
+ "masked_tokens": 97.7875,
2457
+ "mean_t": 0.4511947895749472,
2458
+ "step": 980,
2459
+ "student_masked_tokens": 97.7875
2460
+ },
2461
+ {
2462
+ "avg_mask_ratio": 0.4605769342277199,
2463
+ "avg_response_length": 262.0375,
2464
+ "avg_student_mask_ratio": 0.4605769342277199,
2465
+ "batch_ainp_frac": 0.0,
2466
+ "batch_inp_frac": 1.0,
2467
+ "batch_inp_oh_frac": 0.0,
2468
+ "batch_inp_par_frac": 0.0,
2469
+ "batch_inp_par_par_frac": 0.0,
2470
+ "batch_inp_par_reverse_frac": 0.0,
2471
+ "batch_rl_frac": 0.0,
2472
+ "batch_sft_frac": 0.0,
2473
+ "batch_soft_sft_frac": 0.0,
2474
+ "batch_tf_frac": 0.0,
2475
+ "ce_loss": 0.18629460762591635,
2476
+ "epoch": 2.1152,
2477
+ "grad_norm": 0.625,
2478
+ "kd_loss": 0.4187604939788798,
2479
+ "learning_rate": 3e-06,
2480
+ "loss": 0.5063,
2481
+ "masked_tokens": 122.0,
2482
+ "mean_t": 0.4923786667350214,
2483
+ "step": 990,
2484
+ "student_masked_tokens": 122.0
2485
+ },
2486
+ {
2487
+ "avg_mask_ratio": 0.4547682981239632,
2488
+ "avg_response_length": 215.3,
2489
+ "avg_student_mask_ratio": 0.4547682981239632,
2490
+ "batch_ainp_frac": 0.0,
2491
+ "batch_inp_frac": 1.0,
2492
+ "batch_inp_oh_frac": 0.0,
2493
+ "batch_inp_par_frac": 0.0,
2494
+ "batch_inp_par_par_frac": 0.0,
2495
+ "batch_inp_par_reverse_frac": 0.0,
2496
+ "batch_rl_frac": 0.0,
2497
+ "batch_sft_frac": 0.0,
2498
+ "batch_soft_sft_frac": 0.0,
2499
+ "batch_tf_frac": 0.0,
2500
+ "ce_loss": 0.26735156250199454,
2501
+ "epoch": 2.1365333333333334,
2502
+ "grad_norm": 0.26953125,
2503
+ "kd_loss": 0.3440752963605235,
2504
+ "learning_rate": 3e-06,
2505
+ "loss": 0.5169,
2506
+ "masked_tokens": 100.775,
2507
+ "mean_t": 0.4773523230338469,
2508
+ "step": 1000,
2509
+ "student_masked_tokens": 100.775
2510
+ }
2511
+ ],
2512
+ "logging_steps": 10,
2513
+ "max_steps": 1404,
2514
+ "num_input_tokens_seen": 0,
2515
+ "num_train_epochs": 3,
2516
+ "save_steps": 100,
2517
+ "stateful_callbacks": {
2518
+ "TrainerControl": {
2519
+ "args": {
2520
+ "should_epoch_stop": false,
2521
+ "should_evaluate": false,
2522
+ "should_log": false,
2523
+ "should_save": true,
2524
+ "should_training_stop": false
2525
+ },
2526
+ "attributes": {}
2527
+ }
2528
+ },
2529
+ "total_flos": 0.0,
2530
+ "train_batch_size": 1,
2531
+ "trial_name": null,
2532
+ "trial_params": null
2533
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89ae704e41a4a62f6ca56789c45ce45887326cd2f8d1e97e398e5ada4a93398c
3
+ size 8312
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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
+ "gate_proj",
28
+ "down_proj",
29
+ "up_proj",
30
+ "o_proj",
31
+ "q_proj",
32
+ "v_proj",
33
+ "k_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea0db96232a244b6cab173add792ca63b51cf8b25da144eaf8d1f4d360eb0131
3
+ size 2406624648
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4152fa051690004fa25524b0e8f1171a8278945d174cb036cbb0a1b9fde3d01c
3
+ size 671304442
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ebf0b85ec66092ce471ce68f6fcda64182d50a012837c2e4c9f9690fe3e5f5c3
3
+ size 14512
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/trainer_state.json ADDED
@@ -0,0 +1,2783 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.4931091487989761,
13
+ "avg_response_length": 225.975,
14
+ "avg_student_mask_ratio": 0.4931091487989761,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 1.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 0.0,
19
+ "batch_inp_par_par_frac": 0.0,
20
+ "batch_inp_par_reverse_frac": 0.0,
21
+ "batch_rl_frac": 0.0,
22
+ "batch_sft_frac": 0.0,
23
+ "batch_soft_sft_frac": 0.0,
24
+ "batch_tf_frac": 0.0,
25
+ "ce_loss": 0.5570551689027979,
26
+ "epoch": 0.021333333333333333,
27
+ "grad_norm": 0.404296875,
28
+ "kd_loss": 0.5375588692116253,
29
+ "learning_rate": 3e-06,
30
+ "loss": 0.8247,
31
+ "masked_tokens": 111.95,
32
+ "mean_t": 0.5145528071501758,
33
+ "step": 10,
34
+ "student_masked_tokens": 111.95
35
+ },
36
+ {
37
+ "avg_mask_ratio": 0.42058031369233506,
38
+ "avg_response_length": 255.2625,
39
+ "avg_student_mask_ratio": 0.42058031369233506,
40
+ "batch_ainp_frac": 0.0,
41
+ "batch_inp_frac": 1.0,
42
+ "batch_inp_oh_frac": 0.0,
43
+ "batch_inp_par_frac": 0.0,
44
+ "batch_inp_par_par_frac": 0.0,
45
+ "batch_inp_par_reverse_frac": 0.0,
46
+ "batch_rl_frac": 0.0,
47
+ "batch_sft_frac": 0.0,
48
+ "batch_soft_sft_frac": 0.0,
49
+ "batch_tf_frac": 0.0,
50
+ "ce_loss": 0.42826092825978324,
51
+ "epoch": 0.042666666666666665,
52
+ "grad_norm": 0.8046875,
53
+ "kd_loss": 0.4450965437417761,
54
+ "learning_rate": 3e-06,
55
+ "loss": 0.5724,
56
+ "masked_tokens": 97.35,
57
+ "mean_t": 0.43874448732240123,
58
+ "step": 20,
59
+ "student_masked_tokens": 97.35
60
+ },
61
+ {
62
+ "avg_mask_ratio": 0.4538542575784959,
63
+ "avg_response_length": 211.7625,
64
+ "avg_student_mask_ratio": 0.4538542575784959,
65
+ "batch_ainp_frac": 0.0,
66
+ "batch_inp_frac": 1.0,
67
+ "batch_inp_oh_frac": 0.0,
68
+ "batch_inp_par_frac": 0.0,
69
+ "batch_inp_par_par_frac": 0.0,
70
+ "batch_inp_par_reverse_frac": 0.0,
71
+ "batch_rl_frac": 0.0,
72
+ "batch_sft_frac": 0.0,
73
+ "batch_soft_sft_frac": 0.0,
74
+ "batch_tf_frac": 0.0,
75
+ "ce_loss": 0.4461815005188782,
76
+ "epoch": 0.064,
77
+ "grad_norm": 0.50390625,
78
+ "kd_loss": 0.5296064364436825,
79
+ "learning_rate": 3e-06,
80
+ "loss": 0.702,
81
+ "masked_tokens": 110.2,
82
+ "mean_t": 0.4803953981841914,
83
+ "step": 30,
84
+ "student_masked_tokens": 110.2
85
+ },
86
+ {
87
+ "avg_mask_ratio": 0.4207469140383182,
88
+ "avg_response_length": 224.125,
89
+ "avg_student_mask_ratio": 0.4207469140383182,
90
+ "batch_ainp_frac": 0.0,
91
+ "batch_inp_frac": 1.0,
92
+ "batch_inp_oh_frac": 0.0,
93
+ "batch_inp_par_frac": 0.0,
94
+ "batch_inp_par_par_frac": 0.0,
95
+ "batch_inp_par_reverse_frac": 0.0,
96
+ "batch_rl_frac": 0.0,
97
+ "batch_sft_frac": 0.0,
98
+ "batch_soft_sft_frac": 0.0,
99
+ "batch_tf_frac": 0.0,
100
+ "ce_loss": 0.38499989152683156,
101
+ "epoch": 0.08533333333333333,
102
+ "grad_norm": 1.671875,
103
+ "kd_loss": 0.33118802310931417,
104
+ "learning_rate": 3e-06,
105
+ "loss": 0.5529,
106
+ "masked_tokens": 98.1625,
107
+ "mean_t": 0.4569831106782658,
108
+ "step": 40,
109
+ "student_masked_tokens": 98.1625
110
+ },
111
+ {
112
+ "avg_mask_ratio": 0.43260439952719026,
113
+ "avg_response_length": 207.125,
114
+ "avg_student_mask_ratio": 0.43260439952719026,
115
+ "batch_ainp_frac": 0.0,
116
+ "batch_inp_frac": 1.0,
117
+ "batch_inp_oh_frac": 0.0,
118
+ "batch_inp_par_frac": 0.0,
119
+ "batch_inp_par_par_frac": 0.0,
120
+ "batch_inp_par_reverse_frac": 0.0,
121
+ "batch_rl_frac": 0.0,
122
+ "batch_sft_frac": 0.0,
123
+ "batch_soft_sft_frac": 0.0,
124
+ "batch_tf_frac": 0.0,
125
+ "ce_loss": 0.5240421466317912,
126
+ "epoch": 0.10666666666666667,
127
+ "grad_norm": 1.6875,
128
+ "kd_loss": 0.4270985169670894,
129
+ "learning_rate": 3e-06,
130
+ "loss": 0.671,
131
+ "masked_tokens": 85.05,
132
+ "mean_t": 0.4612453707959503,
133
+ "step": 50,
134
+ "student_masked_tokens": 85.05
135
+ },
136
+ {
137
+ "avg_mask_ratio": 0.46053453313652426,
138
+ "avg_response_length": 251.0875,
139
+ "avg_student_mask_ratio": 0.46053453313652426,
140
+ "batch_ainp_frac": 0.0,
141
+ "batch_inp_frac": 1.0,
142
+ "batch_inp_oh_frac": 0.0,
143
+ "batch_inp_par_frac": 0.0,
144
+ "batch_inp_par_par_frac": 0.0,
145
+ "batch_inp_par_reverse_frac": 0.0,
146
+ "batch_rl_frac": 0.0,
147
+ "batch_sft_frac": 0.0,
148
+ "batch_soft_sft_frac": 0.0,
149
+ "batch_tf_frac": 0.0,
150
+ "ce_loss": 0.5027546818272185,
151
+ "epoch": 0.128,
152
+ "grad_norm": 0.17578125,
153
+ "kd_loss": 0.3904111967755945,
154
+ "learning_rate": 3e-06,
155
+ "loss": 0.6672,
156
+ "masked_tokens": 120.9,
157
+ "mean_t": 0.48597636765334756,
158
+ "step": 60,
159
+ "student_masked_tokens": 120.9
160
+ },
161
+ {
162
+ "avg_mask_ratio": 0.5112146578729153,
163
+ "avg_response_length": 202.5875,
164
+ "avg_student_mask_ratio": 0.5112146578729153,
165
+ "batch_ainp_frac": 0.0,
166
+ "batch_inp_frac": 1.0,
167
+ "batch_inp_oh_frac": 0.0,
168
+ "batch_inp_par_frac": 0.0,
169
+ "batch_inp_par_par_frac": 0.0,
170
+ "batch_inp_par_reverse_frac": 0.0,
171
+ "batch_rl_frac": 0.0,
172
+ "batch_sft_frac": 0.0,
173
+ "batch_soft_sft_frac": 0.0,
174
+ "batch_tf_frac": 0.0,
175
+ "ce_loss": 0.7753003867959023,
176
+ "epoch": 0.14933333333333335,
177
+ "grad_norm": 0.953125,
178
+ "kd_loss": 0.4415664039527428,
179
+ "learning_rate": 3e-06,
180
+ "loss": 0.856,
181
+ "masked_tokens": 104.5875,
182
+ "mean_t": 0.5459650319069624,
183
+ "step": 70,
184
+ "student_masked_tokens": 104.5875
185
+ },
186
+ {
187
+ "avg_mask_ratio": 0.37548826879356056,
188
+ "avg_response_length": 225.85,
189
+ "avg_student_mask_ratio": 0.37548826879356056,
190
+ "batch_ainp_frac": 0.0,
191
+ "batch_inp_frac": 1.0,
192
+ "batch_inp_oh_frac": 0.0,
193
+ "batch_inp_par_frac": 0.0,
194
+ "batch_inp_par_par_frac": 0.0,
195
+ "batch_inp_par_reverse_frac": 0.0,
196
+ "batch_rl_frac": 0.0,
197
+ "batch_sft_frac": 0.0,
198
+ "batch_soft_sft_frac": 0.0,
199
+ "batch_tf_frac": 0.0,
200
+ "ce_loss": 0.3791731233859082,
201
+ "epoch": 0.17066666666666666,
202
+ "grad_norm": 0.1552734375,
203
+ "kd_loss": 0.31052538527774515,
204
+ "learning_rate": 3e-06,
205
+ "loss": 0.4843,
206
+ "masked_tokens": 85.0625,
207
+ "mean_t": 0.40758824030635876,
208
+ "step": 80,
209
+ "student_masked_tokens": 85.0625
210
+ },
211
+ {
212
+ "avg_mask_ratio": 0.5001560213277116,
213
+ "avg_response_length": 229.75,
214
+ "avg_student_mask_ratio": 0.5001560213277116,
215
+ "batch_ainp_frac": 0.0,
216
+ "batch_inp_frac": 1.0,
217
+ "batch_inp_oh_frac": 0.0,
218
+ "batch_inp_par_frac": 0.0,
219
+ "batch_inp_par_par_frac": 0.0,
220
+ "batch_inp_par_reverse_frac": 0.0,
221
+ "batch_rl_frac": 0.0,
222
+ "batch_sft_frac": 0.0,
223
+ "batch_soft_sft_frac": 0.0,
224
+ "batch_tf_frac": 0.0,
225
+ "ce_loss": 0.6899960007944174,
226
+ "epoch": 0.192,
227
+ "grad_norm": 1.25,
228
+ "kd_loss": 0.5995283465861896,
229
+ "learning_rate": 3e-06,
230
+ "loss": 0.9721,
231
+ "masked_tokens": 107.6625,
232
+ "mean_t": 0.5297661645396147,
233
+ "step": 90,
234
+ "student_masked_tokens": 107.6625
235
+ },
236
+ {
237
+ "avg_mask_ratio": 0.4576045103633078,
238
+ "avg_response_length": 208.0,
239
+ "avg_student_mask_ratio": 0.4576045103633078,
240
+ "batch_ainp_frac": 0.0,
241
+ "batch_inp_frac": 1.0,
242
+ "batch_inp_oh_frac": 0.0,
243
+ "batch_inp_par_frac": 0.0,
244
+ "batch_inp_par_par_frac": 0.0,
245
+ "batch_inp_par_reverse_frac": 0.0,
246
+ "batch_rl_frac": 0.0,
247
+ "batch_sft_frac": 0.0,
248
+ "batch_soft_sft_frac": 0.0,
249
+ "batch_tf_frac": 0.0,
250
+ "ce_loss": 0.41132245859021166,
251
+ "epoch": 0.21333333333333335,
252
+ "grad_norm": 0.64453125,
253
+ "kd_loss": 0.3813956479015957,
254
+ "learning_rate": 3e-06,
255
+ "loss": 0.6635,
256
+ "masked_tokens": 104.1625,
257
+ "mean_t": 0.4886587227345444,
258
+ "step": 100,
259
+ "student_masked_tokens": 104.1625
260
+ },
261
+ {
262
+ "avg_mask_ratio": 0.4877026333590038,
263
+ "avg_response_length": 213.0875,
264
+ "avg_student_mask_ratio": 0.4877026333590038,
265
+ "batch_ainp_frac": 0.0,
266
+ "batch_inp_frac": 1.0,
267
+ "batch_inp_oh_frac": 0.0,
268
+ "batch_inp_par_frac": 0.0,
269
+ "batch_inp_par_par_frac": 0.0,
270
+ "batch_inp_par_reverse_frac": 0.0,
271
+ "batch_rl_frac": 0.0,
272
+ "batch_sft_frac": 0.0,
273
+ "batch_soft_sft_frac": 0.0,
274
+ "batch_tf_frac": 0.0,
275
+ "ce_loss": 0.4612084587922368,
276
+ "epoch": 0.23466666666666666,
277
+ "grad_norm": 0.64453125,
278
+ "kd_loss": 0.5074845846289577,
279
+ "learning_rate": 3e-06,
280
+ "loss": 0.7993,
281
+ "masked_tokens": 102.075,
282
+ "mean_t": 0.5246987929102034,
283
+ "step": 110,
284
+ "student_masked_tokens": 102.075
285
+ },
286
+ {
287
+ "avg_mask_ratio": 0.45146879020612685,
288
+ "avg_response_length": 224.1875,
289
+ "avg_student_mask_ratio": 0.45146879020612685,
290
+ "batch_ainp_frac": 0.0,
291
+ "batch_inp_frac": 1.0,
292
+ "batch_inp_oh_frac": 0.0,
293
+ "batch_inp_par_frac": 0.0,
294
+ "batch_inp_par_par_frac": 0.0,
295
+ "batch_inp_par_reverse_frac": 0.0,
296
+ "batch_rl_frac": 0.0,
297
+ "batch_sft_frac": 0.0,
298
+ "batch_soft_sft_frac": 0.0,
299
+ "batch_tf_frac": 0.0,
300
+ "ce_loss": 0.3276976759495483,
301
+ "epoch": 0.256,
302
+ "grad_norm": 0.30078125,
303
+ "kd_loss": 0.41461311469229256,
304
+ "learning_rate": 3e-06,
305
+ "loss": 0.6088,
306
+ "masked_tokens": 100.525,
307
+ "mean_t": 0.4805434140143916,
308
+ "step": 120,
309
+ "student_masked_tokens": 100.525
310
+ },
311
+ {
312
+ "avg_mask_ratio": 0.4356566035945434,
313
+ "avg_response_length": 202.7,
314
+ "avg_student_mask_ratio": 0.4356566035945434,
315
+ "batch_ainp_frac": 0.0,
316
+ "batch_inp_frac": 1.0,
317
+ "batch_inp_oh_frac": 0.0,
318
+ "batch_inp_par_frac": 0.0,
319
+ "batch_inp_par_par_frac": 0.0,
320
+ "batch_inp_par_reverse_frac": 0.0,
321
+ "batch_rl_frac": 0.0,
322
+ "batch_sft_frac": 0.0,
323
+ "batch_soft_sft_frac": 0.0,
324
+ "batch_tf_frac": 0.0,
325
+ "ce_loss": 0.12710368948505674,
326
+ "epoch": 0.2773333333333333,
327
+ "grad_norm": 0.490234375,
328
+ "kd_loss": 0.23057804748218585,
329
+ "learning_rate": 3e-06,
330
+ "loss": 0.384,
331
+ "masked_tokens": 89.5625,
332
+ "mean_t": 0.47522516988683494,
333
+ "step": 130,
334
+ "student_masked_tokens": 89.5625
335
+ },
336
+ {
337
+ "avg_mask_ratio": 0.49419954856857656,
338
+ "avg_response_length": 255.625,
339
+ "avg_student_mask_ratio": 0.49419954856857656,
340
+ "batch_ainp_frac": 0.0,
341
+ "batch_inp_frac": 1.0,
342
+ "batch_inp_oh_frac": 0.0,
343
+ "batch_inp_par_frac": 0.0,
344
+ "batch_inp_par_par_frac": 0.0,
345
+ "batch_inp_par_reverse_frac": 0.0,
346
+ "batch_rl_frac": 0.0,
347
+ "batch_sft_frac": 0.0,
348
+ "batch_soft_sft_frac": 0.0,
349
+ "batch_tf_frac": 0.0,
350
+ "ce_loss": 0.48596099013025196,
351
+ "epoch": 0.2986666666666667,
352
+ "grad_norm": 0.859375,
353
+ "kd_loss": 0.5025483015746885,
354
+ "learning_rate": 3e-06,
355
+ "loss": 0.7892,
356
+ "masked_tokens": 136.575,
357
+ "mean_t": 0.5204090005659964,
358
+ "step": 140,
359
+ "student_masked_tokens": 136.575
360
+ },
361
+ {
362
+ "avg_mask_ratio": 0.4736677930341102,
363
+ "avg_response_length": 255.375,
364
+ "avg_student_mask_ratio": 0.4736677930341102,
365
+ "batch_ainp_frac": 0.0,
366
+ "batch_inp_frac": 1.0,
367
+ "batch_inp_oh_frac": 0.0,
368
+ "batch_inp_par_frac": 0.0,
369
+ "batch_inp_par_par_frac": 0.0,
370
+ "batch_inp_par_reverse_frac": 0.0,
371
+ "batch_rl_frac": 0.0,
372
+ "batch_sft_frac": 0.0,
373
+ "batch_soft_sft_frac": 0.0,
374
+ "batch_tf_frac": 0.0,
375
+ "ce_loss": 0.5493089448234059,
376
+ "epoch": 0.32,
377
+ "grad_norm": 0.076171875,
378
+ "kd_loss": 0.4892602212316547,
379
+ "learning_rate": 3e-06,
380
+ "loss": 0.7918,
381
+ "masked_tokens": 126.575,
382
+ "mean_t": 0.5012552456930279,
383
+ "step": 150,
384
+ "student_masked_tokens": 126.575
385
+ },
386
+ {
387
+ "avg_mask_ratio": 0.5032523009285796,
388
+ "avg_response_length": 209.325,
389
+ "avg_student_mask_ratio": 0.5032523009285796,
390
+ "batch_ainp_frac": 0.0,
391
+ "batch_inp_frac": 1.0,
392
+ "batch_inp_oh_frac": 0.0,
393
+ "batch_inp_par_frac": 0.0,
394
+ "batch_inp_par_par_frac": 0.0,
395
+ "batch_inp_par_reverse_frac": 0.0,
396
+ "batch_rl_frac": 0.0,
397
+ "batch_sft_frac": 0.0,
398
+ "batch_soft_sft_frac": 0.0,
399
+ "batch_tf_frac": 0.0,
400
+ "ce_loss": 0.5929547422666019,
401
+ "epoch": 0.3413333333333333,
402
+ "grad_norm": 1.171875,
403
+ "kd_loss": 0.44711892502580214,
404
+ "learning_rate": 3e-06,
405
+ "loss": 0.7472,
406
+ "masked_tokens": 99.425,
407
+ "mean_t": 0.5408745193795766,
408
+ "step": 160,
409
+ "student_masked_tokens": 99.425
410
+ },
411
+ {
412
+ "avg_mask_ratio": 0.4806730231270194,
413
+ "avg_response_length": 190.325,
414
+ "avg_student_mask_ratio": 0.4806730231270194,
415
+ "batch_ainp_frac": 0.0,
416
+ "batch_inp_frac": 1.0,
417
+ "batch_inp_oh_frac": 0.0,
418
+ "batch_inp_par_frac": 0.0,
419
+ "batch_inp_par_par_frac": 0.0,
420
+ "batch_inp_par_reverse_frac": 0.0,
421
+ "batch_rl_frac": 0.0,
422
+ "batch_sft_frac": 0.0,
423
+ "batch_soft_sft_frac": 0.0,
424
+ "batch_tf_frac": 0.0,
425
+ "ce_loss": 0.3583432949517601,
426
+ "epoch": 0.3626666666666667,
427
+ "grad_norm": 0.6171875,
428
+ "kd_loss": 0.4521343837219092,
429
+ "learning_rate": 3e-06,
430
+ "loss": 0.6358,
431
+ "masked_tokens": 89.825,
432
+ "mean_t": 0.5134547733236104,
433
+ "step": 170,
434
+ "student_masked_tokens": 89.825
435
+ },
436
+ {
437
+ "avg_mask_ratio": 0.45829249716189224,
438
+ "avg_response_length": 244.0,
439
+ "avg_student_mask_ratio": 0.45829249716189224,
440
+ "batch_ainp_frac": 0.0,
441
+ "batch_inp_frac": 1.0,
442
+ "batch_inp_oh_frac": 0.0,
443
+ "batch_inp_par_frac": 0.0,
444
+ "batch_inp_par_par_frac": 0.0,
445
+ "batch_inp_par_reverse_frac": 0.0,
446
+ "batch_rl_frac": 0.0,
447
+ "batch_sft_frac": 0.0,
448
+ "batch_soft_sft_frac": 0.0,
449
+ "batch_tf_frac": 0.0,
450
+ "ce_loss": 0.3746713957985094,
451
+ "epoch": 0.384,
452
+ "grad_norm": 0.49609375,
453
+ "kd_loss": 0.34934306121722897,
454
+ "learning_rate": 3e-06,
455
+ "loss": 0.5747,
456
+ "masked_tokens": 110.075,
457
+ "mean_t": 0.48226988823735156,
458
+ "step": 180,
459
+ "student_masked_tokens": 110.075
460
+ },
461
+ {
462
+ "avg_mask_ratio": 0.4842760307248682,
463
+ "avg_response_length": 233.675,
464
+ "avg_student_mask_ratio": 0.4842760307248682,
465
+ "batch_ainp_frac": 0.0,
466
+ "batch_inp_frac": 1.0,
467
+ "batch_inp_oh_frac": 0.0,
468
+ "batch_inp_par_frac": 0.0,
469
+ "batch_inp_par_par_frac": 0.0,
470
+ "batch_inp_par_reverse_frac": 0.0,
471
+ "batch_rl_frac": 0.0,
472
+ "batch_sft_frac": 0.0,
473
+ "batch_soft_sft_frac": 0.0,
474
+ "batch_tf_frac": 0.0,
475
+ "ce_loss": 0.5090123614077584,
476
+ "epoch": 0.4053333333333333,
477
+ "grad_norm": 1.6171875,
478
+ "kd_loss": 0.43204482231294605,
479
+ "learning_rate": 3e-06,
480
+ "loss": 0.7055,
481
+ "masked_tokens": 109.5875,
482
+ "mean_t": 0.5165087037021294,
483
+ "step": 190,
484
+ "student_masked_tokens": 109.5875
485
+ },
486
+ {
487
+ "avg_mask_ratio": 0.4665210062637925,
488
+ "avg_response_length": 197.8,
489
+ "avg_student_mask_ratio": 0.4665210062637925,
490
+ "batch_ainp_frac": 0.0,
491
+ "batch_inp_frac": 1.0,
492
+ "batch_inp_oh_frac": 0.0,
493
+ "batch_inp_par_frac": 0.0,
494
+ "batch_inp_par_par_frac": 0.0,
495
+ "batch_inp_par_reverse_frac": 0.0,
496
+ "batch_rl_frac": 0.0,
497
+ "batch_sft_frac": 0.0,
498
+ "batch_soft_sft_frac": 0.0,
499
+ "batch_tf_frac": 0.0,
500
+ "ce_loss": 0.27265903051802526,
501
+ "epoch": 0.4266666666666667,
502
+ "grad_norm": 0.318359375,
503
+ "kd_loss": 0.357759011555504,
504
+ "learning_rate": 3e-06,
505
+ "loss": 0.5013,
506
+ "masked_tokens": 97.0125,
507
+ "mean_t": 0.5073627714533359,
508
+ "step": 200,
509
+ "student_masked_tokens": 97.0125
510
+ },
511
+ {
512
+ "avg_mask_ratio": 0.45183838629163803,
513
+ "avg_response_length": 213.7875,
514
+ "avg_student_mask_ratio": 0.45183838629163803,
515
+ "batch_ainp_frac": 0.0,
516
+ "batch_inp_frac": 1.0,
517
+ "batch_inp_oh_frac": 0.0,
518
+ "batch_inp_par_frac": 0.0,
519
+ "batch_inp_par_par_frac": 0.0,
520
+ "batch_inp_par_reverse_frac": 0.0,
521
+ "batch_rl_frac": 0.0,
522
+ "batch_sft_frac": 0.0,
523
+ "batch_soft_sft_frac": 0.0,
524
+ "batch_tf_frac": 0.0,
525
+ "ce_loss": 0.21031314329709402,
526
+ "epoch": 0.448,
527
+ "grad_norm": 0.8359375,
528
+ "kd_loss": 0.38585986606940426,
529
+ "learning_rate": 3e-06,
530
+ "loss": 0.5103,
531
+ "masked_tokens": 96.0125,
532
+ "mean_t": 0.48430291628465055,
533
+ "step": 210,
534
+ "student_masked_tokens": 96.0125
535
+ },
536
+ {
537
+ "avg_mask_ratio": 0.5396152360364794,
538
+ "avg_response_length": 236.5125,
539
+ "avg_student_mask_ratio": 0.5396152360364794,
540
+ "batch_ainp_frac": 0.0,
541
+ "batch_inp_frac": 1.0,
542
+ "batch_inp_oh_frac": 0.0,
543
+ "batch_inp_par_frac": 0.0,
544
+ "batch_inp_par_par_frac": 0.0,
545
+ "batch_inp_par_reverse_frac": 0.0,
546
+ "batch_rl_frac": 0.0,
547
+ "batch_sft_frac": 0.0,
548
+ "batch_soft_sft_frac": 0.0,
549
+ "batch_tf_frac": 0.0,
550
+ "ce_loss": 0.6178499971098063,
551
+ "epoch": 0.4693333333333333,
552
+ "grad_norm": 0.96875,
553
+ "kd_loss": 0.46674597742967305,
554
+ "learning_rate": 3e-06,
555
+ "loss": 0.8495,
556
+ "masked_tokens": 125.35,
557
+ "mean_t": 0.570199209311977,
558
+ "step": 220,
559
+ "student_masked_tokens": 125.35
560
+ },
561
+ {
562
+ "avg_mask_ratio": 0.4403991688624956,
563
+ "avg_response_length": 252.7,
564
+ "avg_student_mask_ratio": 0.4403991688624956,
565
+ "batch_ainp_frac": 0.0,
566
+ "batch_inp_frac": 1.0,
567
+ "batch_inp_oh_frac": 0.0,
568
+ "batch_inp_par_frac": 0.0,
569
+ "batch_inp_par_par_frac": 0.0,
570
+ "batch_inp_par_reverse_frac": 0.0,
571
+ "batch_rl_frac": 0.0,
572
+ "batch_sft_frac": 0.0,
573
+ "batch_soft_sft_frac": 0.0,
574
+ "batch_tf_frac": 0.0,
575
+ "ce_loss": 0.25455478885055527,
576
+ "epoch": 0.49066666666666664,
577
+ "grad_norm": 0.5703125,
578
+ "kd_loss": 0.43305868929596725,
579
+ "learning_rate": 3e-06,
580
+ "loss": 0.6428,
581
+ "masked_tokens": 107.325,
582
+ "mean_t": 0.46891279935371133,
583
+ "step": 230,
584
+ "student_masked_tokens": 107.325
585
+ },
586
+ {
587
+ "avg_mask_ratio": 0.47419991258066146,
588
+ "avg_response_length": 212.85,
589
+ "avg_student_mask_ratio": 0.47419991258066146,
590
+ "batch_ainp_frac": 0.0,
591
+ "batch_inp_frac": 1.0,
592
+ "batch_inp_oh_frac": 0.0,
593
+ "batch_inp_par_frac": 0.0,
594
+ "batch_inp_par_par_frac": 0.0,
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": 0.32057130943685763,
601
+ "epoch": 0.512,
602
+ "grad_norm": 0.43359375,
603
+ "kd_loss": 0.5083060303753086,
604
+ "learning_rate": 3e-06,
605
+ "loss": 0.6986,
606
+ "masked_tokens": 106.9,
607
+ "mean_t": 0.502228345896583,
608
+ "step": 240,
609
+ "student_masked_tokens": 106.9
610
+ },
611
+ {
612
+ "avg_mask_ratio": 0.4464209079160355,
613
+ "avg_response_length": 243.475,
614
+ "avg_student_mask_ratio": 0.4464209079160355,
615
+ "batch_ainp_frac": 0.0,
616
+ "batch_inp_frac": 1.0,
617
+ "batch_inp_oh_frac": 0.0,
618
+ "batch_inp_par_frac": 0.0,
619
+ "batch_inp_par_par_frac": 0.0,
620
+ "batch_inp_par_reverse_frac": 0.0,
621
+ "batch_rl_frac": 0.0,
622
+ "batch_sft_frac": 0.0,
623
+ "batch_soft_sft_frac": 0.0,
624
+ "batch_tf_frac": 0.0,
625
+ "ce_loss": 0.33636454603331456,
626
+ "epoch": 0.5333333333333333,
627
+ "grad_norm": 0.1142578125,
628
+ "kd_loss": 0.41649795620701296,
629
+ "learning_rate": 3e-06,
630
+ "loss": 0.5666,
631
+ "masked_tokens": 112.7375,
632
+ "mean_t": 0.4733429416548461,
633
+ "step": 250,
634
+ "student_masked_tokens": 112.7375
635
+ },
636
+ {
637
+ "avg_mask_ratio": 0.4520751796895638,
638
+ "avg_response_length": 245.55,
639
+ "avg_student_mask_ratio": 0.4520751796895638,
640
+ "batch_ainp_frac": 0.0,
641
+ "batch_inp_frac": 1.0,
642
+ "batch_inp_oh_frac": 0.0,
643
+ "batch_inp_par_frac": 0.0,
644
+ "batch_inp_par_par_frac": 0.0,
645
+ "batch_inp_par_reverse_frac": 0.0,
646
+ "batch_rl_frac": 0.0,
647
+ "batch_sft_frac": 0.0,
648
+ "batch_soft_sft_frac": 0.0,
649
+ "batch_tf_frac": 0.0,
650
+ "ce_loss": 0.37478437887749577,
651
+ "epoch": 0.5546666666666666,
652
+ "grad_norm": 0.328125,
653
+ "kd_loss": 0.31532439299670545,
654
+ "learning_rate": 3e-06,
655
+ "loss": 0.5129,
656
+ "masked_tokens": 109.6375,
657
+ "mean_t": 0.4843149524240289,
658
+ "step": 260,
659
+ "student_masked_tokens": 109.6375
660
+ },
661
+ {
662
+ "avg_mask_ratio": 0.5305180630879477,
663
+ "avg_response_length": 224.45,
664
+ "avg_student_mask_ratio": 0.5305180630879477,
665
+ "batch_ainp_frac": 0.0,
666
+ "batch_inp_frac": 1.0,
667
+ "batch_inp_oh_frac": 0.0,
668
+ "batch_inp_par_frac": 0.0,
669
+ "batch_inp_par_par_frac": 0.0,
670
+ "batch_inp_par_reverse_frac": 0.0,
671
+ "batch_rl_frac": 0.0,
672
+ "batch_sft_frac": 0.0,
673
+ "batch_soft_sft_frac": 0.0,
674
+ "batch_tf_frac": 0.0,
675
+ "ce_loss": 0.42709534656005416,
676
+ "epoch": 0.576,
677
+ "grad_norm": 0.7578125,
678
+ "kd_loss": 0.5525495689224045,
679
+ "learning_rate": 3e-06,
680
+ "loss": 0.812,
681
+ "masked_tokens": 120.475,
682
+ "mean_t": 0.5643589949700981,
683
+ "step": 270,
684
+ "student_masked_tokens": 120.475
685
+ },
686
+ {
687
+ "avg_mask_ratio": 0.46451686368091033,
688
+ "avg_response_length": 254.825,
689
+ "avg_student_mask_ratio": 0.46451686368091033,
690
+ "batch_ainp_frac": 0.0,
691
+ "batch_inp_frac": 1.0,
692
+ "batch_inp_oh_frac": 0.0,
693
+ "batch_inp_par_frac": 0.0,
694
+ "batch_inp_par_par_frac": 0.0,
695
+ "batch_inp_par_reverse_frac": 0.0,
696
+ "batch_rl_frac": 0.0,
697
+ "batch_sft_frac": 0.0,
698
+ "batch_soft_sft_frac": 0.0,
699
+ "batch_tf_frac": 0.0,
700
+ "ce_loss": 0.31382316479499084,
701
+ "epoch": 0.5973333333333334,
702
+ "grad_norm": 0.90234375,
703
+ "kd_loss": 0.3957495673693458,
704
+ "learning_rate": 3e-06,
705
+ "loss": 0.6028,
706
+ "masked_tokens": 129.225,
707
+ "mean_t": 0.47818811538163575,
708
+ "step": 280,
709
+ "student_masked_tokens": 129.225
710
+ },
711
+ {
712
+ "avg_mask_ratio": 0.389662017847877,
713
+ "avg_response_length": 245.9125,
714
+ "avg_student_mask_ratio": 0.389662017847877,
715
+ "batch_ainp_frac": 0.0,
716
+ "batch_inp_frac": 1.0,
717
+ "batch_inp_oh_frac": 0.0,
718
+ "batch_inp_par_frac": 0.0,
719
+ "batch_inp_par_par_frac": 0.0,
720
+ "batch_inp_par_reverse_frac": 0.0,
721
+ "batch_rl_frac": 0.0,
722
+ "batch_sft_frac": 0.0,
723
+ "batch_soft_sft_frac": 0.0,
724
+ "batch_tf_frac": 0.0,
725
+ "ce_loss": 0.23645576389110373,
726
+ "epoch": 0.6186666666666667,
727
+ "grad_norm": 0.302734375,
728
+ "kd_loss": 0.27728830450374853,
729
+ "learning_rate": 3e-06,
730
+ "loss": 0.4314,
731
+ "masked_tokens": 99.8625,
732
+ "mean_t": 0.4088635521940887,
733
+ "step": 290,
734
+ "student_masked_tokens": 99.8625
735
+ },
736
+ {
737
+ "avg_mask_ratio": 0.44417000194080175,
738
+ "avg_response_length": 217.0375,
739
+ "avg_student_mask_ratio": 0.44417000194080175,
740
+ "batch_ainp_frac": 0.0,
741
+ "batch_inp_frac": 1.0,
742
+ "batch_inp_oh_frac": 0.0,
743
+ "batch_inp_par_frac": 0.0,
744
+ "batch_inp_par_par_frac": 0.0,
745
+ "batch_inp_par_reverse_frac": 0.0,
746
+ "batch_rl_frac": 0.0,
747
+ "batch_sft_frac": 0.0,
748
+ "batch_soft_sft_frac": 0.0,
749
+ "batch_tf_frac": 0.0,
750
+ "ce_loss": 0.3240562055096575,
751
+ "epoch": 0.64,
752
+ "grad_norm": 1.09375,
753
+ "kd_loss": 0.31930388437995133,
754
+ "learning_rate": 3e-06,
755
+ "loss": 0.5264,
756
+ "masked_tokens": 104.625,
757
+ "mean_t": 0.47984200695063917,
758
+ "step": 300,
759
+ "student_masked_tokens": 104.625
760
+ },
761
+ {
762
+ "avg_mask_ratio": 0.4706685543409549,
763
+ "avg_response_length": 175.45,
764
+ "avg_student_mask_ratio": 0.4706685543409549,
765
+ "batch_ainp_frac": 0.0,
766
+ "batch_inp_frac": 1.0,
767
+ "batch_inp_oh_frac": 0.0,
768
+ "batch_inp_par_frac": 0.0,
769
+ "batch_inp_par_par_frac": 0.0,
770
+ "batch_inp_par_reverse_frac": 0.0,
771
+ "batch_rl_frac": 0.0,
772
+ "batch_sft_frac": 0.0,
773
+ "batch_soft_sft_frac": 0.0,
774
+ "batch_tf_frac": 0.0,
775
+ "ce_loss": 0.34333510200582396,
776
+ "epoch": 0.6613333333333333,
777
+ "grad_norm": 1.234375,
778
+ "kd_loss": 0.5067149527083984,
779
+ "learning_rate": 3e-06,
780
+ "loss": 0.6534,
781
+ "masked_tokens": 84.875,
782
+ "mean_t": 0.5026606284547597,
783
+ "step": 310,
784
+ "student_masked_tokens": 84.875
785
+ },
786
+ {
787
+ "avg_mask_ratio": 0.4974605386145413,
788
+ "avg_response_length": 234.7875,
789
+ "avg_student_mask_ratio": 0.4974605386145413,
790
+ "batch_ainp_frac": 0.0,
791
+ "batch_inp_frac": 1.0,
792
+ "batch_inp_oh_frac": 0.0,
793
+ "batch_inp_par_frac": 0.0,
794
+ "batch_inp_par_par_frac": 0.0,
795
+ "batch_inp_par_reverse_frac": 0.0,
796
+ "batch_rl_frac": 0.0,
797
+ "batch_sft_frac": 0.0,
798
+ "batch_soft_sft_frac": 0.0,
799
+ "batch_tf_frac": 0.0,
800
+ "ce_loss": 0.34462752127872137,
801
+ "epoch": 0.6826666666666666,
802
+ "grad_norm": 0.333984375,
803
+ "kd_loss": 0.3942846609736307,
804
+ "learning_rate": 3e-06,
805
+ "loss": 0.7133,
806
+ "masked_tokens": 119.6,
807
+ "mean_t": 0.5293499688967132,
808
+ "step": 320,
809
+ "student_masked_tokens": 119.6
810
+ },
811
+ {
812
+ "avg_mask_ratio": 0.5112370474264025,
813
+ "avg_response_length": 236.0625,
814
+ "avg_student_mask_ratio": 0.5112370474264025,
815
+ "batch_ainp_frac": 0.0,
816
+ "batch_inp_frac": 1.0,
817
+ "batch_inp_oh_frac": 0.0,
818
+ "batch_inp_par_frac": 0.0,
819
+ "batch_inp_par_par_frac": 0.0,
820
+ "batch_inp_par_reverse_frac": 0.0,
821
+ "batch_rl_frac": 0.0,
822
+ "batch_sft_frac": 0.0,
823
+ "batch_soft_sft_frac": 0.0,
824
+ "batch_tf_frac": 0.0,
825
+ "ce_loss": 0.2974585796398969,
826
+ "epoch": 0.704,
827
+ "grad_norm": 0.44140625,
828
+ "kd_loss": 0.4301003347501496,
829
+ "learning_rate": 3e-06,
830
+ "loss": 0.6754,
831
+ "masked_tokens": 129.425,
832
+ "mean_t": 0.5426030711154454,
833
+ "step": 330,
834
+ "student_masked_tokens": 129.425
835
+ },
836
+ {
837
+ "avg_mask_ratio": 0.44370225080056114,
838
+ "avg_response_length": 241.4875,
839
+ "avg_student_mask_ratio": 0.44370225080056114,
840
+ "batch_ainp_frac": 0.0,
841
+ "batch_inp_frac": 1.0,
842
+ "batch_inp_oh_frac": 0.0,
843
+ "batch_inp_par_frac": 0.0,
844
+ "batch_inp_par_par_frac": 0.0,
845
+ "batch_inp_par_reverse_frac": 0.0,
846
+ "batch_rl_frac": 0.0,
847
+ "batch_sft_frac": 0.0,
848
+ "batch_soft_sft_frac": 0.0,
849
+ "batch_tf_frac": 0.0,
850
+ "ce_loss": 0.3732590021626493,
851
+ "epoch": 0.7253333333333334,
852
+ "grad_norm": 0.98046875,
853
+ "kd_loss": 0.4610515360019235,
854
+ "learning_rate": 3e-06,
855
+ "loss": 0.6627,
856
+ "masked_tokens": 108.775,
857
+ "mean_t": 0.47635243807453664,
858
+ "step": 340,
859
+ "student_masked_tokens": 108.775
860
+ },
861
+ {
862
+ "avg_mask_ratio": 0.49959173843380994,
863
+ "avg_response_length": 235.6375,
864
+ "avg_student_mask_ratio": 0.49959173843380994,
865
+ "batch_ainp_frac": 0.0,
866
+ "batch_inp_frac": 1.0,
867
+ "batch_inp_oh_frac": 0.0,
868
+ "batch_inp_par_frac": 0.0,
869
+ "batch_inp_par_par_frac": 0.0,
870
+ "batch_inp_par_reverse_frac": 0.0,
871
+ "batch_rl_frac": 0.0,
872
+ "batch_sft_frac": 0.0,
873
+ "batch_soft_sft_frac": 0.0,
874
+ "batch_tf_frac": 0.0,
875
+ "ce_loss": 0.48515336151417615,
876
+ "epoch": 0.7466666666666667,
877
+ "grad_norm": 0.92578125,
878
+ "kd_loss": 0.5031771080357654,
879
+ "learning_rate": 3e-06,
880
+ "loss": 0.7668,
881
+ "masked_tokens": 125.625,
882
+ "mean_t": 0.5268881446914747,
883
+ "step": 350,
884
+ "student_masked_tokens": 125.625
885
+ },
886
+ {
887
+ "avg_mask_ratio": 0.4744729608530179,
888
+ "avg_response_length": 246.1625,
889
+ "avg_student_mask_ratio": 0.4744729608530179,
890
+ "batch_ainp_frac": 0.0,
891
+ "batch_inp_frac": 1.0,
892
+ "batch_inp_oh_frac": 0.0,
893
+ "batch_inp_par_frac": 0.0,
894
+ "batch_inp_par_par_frac": 0.0,
895
+ "batch_inp_par_reverse_frac": 0.0,
896
+ "batch_rl_frac": 0.0,
897
+ "batch_sft_frac": 0.0,
898
+ "batch_soft_sft_frac": 0.0,
899
+ "batch_tf_frac": 0.0,
900
+ "ce_loss": 0.3005135279950082,
901
+ "epoch": 0.768,
902
+ "grad_norm": 0.169921875,
903
+ "kd_loss": 0.5216399239409879,
904
+ "learning_rate": 3e-06,
905
+ "loss": 0.6077,
906
+ "masked_tokens": 116.875,
907
+ "mean_t": 0.5040419134311378,
908
+ "step": 360,
909
+ "student_masked_tokens": 116.875
910
+ },
911
+ {
912
+ "avg_mask_ratio": 0.4738045462174341,
913
+ "avg_response_length": 257.575,
914
+ "avg_student_mask_ratio": 0.4738045462174341,
915
+ "batch_ainp_frac": 0.0,
916
+ "batch_inp_frac": 1.0,
917
+ "batch_inp_oh_frac": 0.0,
918
+ "batch_inp_par_frac": 0.0,
919
+ "batch_inp_par_par_frac": 0.0,
920
+ "batch_inp_par_reverse_frac": 0.0,
921
+ "batch_rl_frac": 0.0,
922
+ "batch_sft_frac": 0.0,
923
+ "batch_soft_sft_frac": 0.0,
924
+ "batch_tf_frac": 0.0,
925
+ "ce_loss": 0.5349442186782426,
926
+ "epoch": 0.7893333333333333,
927
+ "grad_norm": 0.201171875,
928
+ "kd_loss": 0.6039233199480805,
929
+ "learning_rate": 3e-06,
930
+ "loss": 0.7196,
931
+ "masked_tokens": 127.4625,
932
+ "mean_t": 0.5127181728370488,
933
+ "step": 370,
934
+ "student_masked_tokens": 127.4625
935
+ },
936
+ {
937
+ "avg_mask_ratio": 0.4512475330149755,
938
+ "avg_response_length": 209.8,
939
+ "avg_student_mask_ratio": 0.4512475330149755,
940
+ "batch_ainp_frac": 0.0,
941
+ "batch_inp_frac": 1.0,
942
+ "batch_inp_oh_frac": 0.0,
943
+ "batch_inp_par_frac": 0.0,
944
+ "batch_inp_par_par_frac": 0.0,
945
+ "batch_inp_par_reverse_frac": 0.0,
946
+ "batch_rl_frac": 0.0,
947
+ "batch_sft_frac": 0.0,
948
+ "batch_soft_sft_frac": 0.0,
949
+ "batch_tf_frac": 0.0,
950
+ "ce_loss": 0.19145508916275275,
951
+ "epoch": 0.8106666666666666,
952
+ "grad_norm": 0.6875,
953
+ "kd_loss": 0.4029755606519984,
954
+ "learning_rate": 3e-06,
955
+ "loss": 0.5055,
956
+ "masked_tokens": 100.8375,
957
+ "mean_t": 0.4825185665744357,
958
+ "step": 380,
959
+ "student_masked_tokens": 100.8375
960
+ },
961
+ {
962
+ "avg_mask_ratio": 0.4752940105390735,
963
+ "avg_response_length": 219.5625,
964
+ "avg_student_mask_ratio": 0.4752940105390735,
965
+ "batch_ainp_frac": 0.0,
966
+ "batch_inp_frac": 1.0,
967
+ "batch_inp_oh_frac": 0.0,
968
+ "batch_inp_par_frac": 0.0,
969
+ "batch_inp_par_par_frac": 0.0,
970
+ "batch_inp_par_reverse_frac": 0.0,
971
+ "batch_rl_frac": 0.0,
972
+ "batch_sft_frac": 0.0,
973
+ "batch_soft_sft_frac": 0.0,
974
+ "batch_tf_frac": 0.0,
975
+ "ce_loss": 0.4267096655552223,
976
+ "epoch": 0.832,
977
+ "grad_norm": 0.2578125,
978
+ "kd_loss": 0.4655849843487971,
979
+ "learning_rate": 3e-06,
980
+ "loss": 0.6749,
981
+ "masked_tokens": 112.375,
982
+ "mean_t": 0.5053101469413377,
983
+ "step": 390,
984
+ "student_masked_tokens": 112.375
985
+ },
986
+ {
987
+ "avg_mask_ratio": 0.47461870914557946,
988
+ "avg_response_length": 242.6125,
989
+ "avg_student_mask_ratio": 0.47461870914557946,
990
+ "batch_ainp_frac": 0.0,
991
+ "batch_inp_frac": 1.0,
992
+ "batch_inp_oh_frac": 0.0,
993
+ "batch_inp_par_frac": 0.0,
994
+ "batch_inp_par_par_frac": 0.0,
995
+ "batch_inp_par_reverse_frac": 0.0,
996
+ "batch_rl_frac": 0.0,
997
+ "batch_sft_frac": 0.0,
998
+ "batch_soft_sft_frac": 0.0,
999
+ "batch_tf_frac": 0.0,
1000
+ "ce_loss": 0.27868834779033025,
1001
+ "epoch": 0.8533333333333334,
1002
+ "grad_norm": 0.640625,
1003
+ "kd_loss": 0.5299579592951205,
1004
+ "learning_rate": 3e-06,
1005
+ "loss": 0.6538,
1006
+ "masked_tokens": 120.4125,
1007
+ "mean_t": 0.5052250675857067,
1008
+ "step": 400,
1009
+ "student_masked_tokens": 120.4125
1010
+ },
1011
+ {
1012
+ "avg_mask_ratio": 0.48321815438685006,
1013
+ "avg_response_length": 228.15,
1014
+ "avg_student_mask_ratio": 0.48321815438685006,
1015
+ "batch_ainp_frac": 0.0,
1016
+ "batch_inp_frac": 1.0,
1017
+ "batch_inp_oh_frac": 0.0,
1018
+ "batch_inp_par_frac": 0.0,
1019
+ "batch_inp_par_par_frac": 0.0,
1020
+ "batch_inp_par_reverse_frac": 0.0,
1021
+ "batch_rl_frac": 0.0,
1022
+ "batch_sft_frac": 0.0,
1023
+ "batch_soft_sft_frac": 0.0,
1024
+ "batch_tf_frac": 0.0,
1025
+ "ce_loss": 0.43057951200541994,
1026
+ "epoch": 0.8746666666666667,
1027
+ "grad_norm": 0.5390625,
1028
+ "kd_loss": 0.504674318619719,
1029
+ "learning_rate": 3e-06,
1030
+ "loss": 0.7381,
1031
+ "masked_tokens": 119.0,
1032
+ "mean_t": 0.5050956419203431,
1033
+ "step": 410,
1034
+ "student_masked_tokens": 119.0
1035
+ },
1036
+ {
1037
+ "avg_mask_ratio": 0.4379329536575824,
1038
+ "avg_response_length": 220.225,
1039
+ "avg_student_mask_ratio": 0.4379329536575824,
1040
+ "batch_ainp_frac": 0.0,
1041
+ "batch_inp_frac": 1.0,
1042
+ "batch_inp_oh_frac": 0.0,
1043
+ "batch_inp_par_frac": 0.0,
1044
+ "batch_inp_par_par_frac": 0.0,
1045
+ "batch_inp_par_reverse_frac": 0.0,
1046
+ "batch_rl_frac": 0.0,
1047
+ "batch_sft_frac": 0.0,
1048
+ "batch_soft_sft_frac": 0.0,
1049
+ "batch_tf_frac": 0.0,
1050
+ "ce_loss": 0.132674143492045,
1051
+ "epoch": 0.896,
1052
+ "grad_norm": 1.09375,
1053
+ "kd_loss": 0.27731474525324984,
1054
+ "learning_rate": 3e-06,
1055
+ "loss": 0.3953,
1056
+ "masked_tokens": 85.525,
1057
+ "mean_t": 0.4769687672611326,
1058
+ "step": 420,
1059
+ "student_masked_tokens": 85.525
1060
+ },
1061
+ {
1062
+ "avg_mask_ratio": 0.4674084897618741,
1063
+ "avg_response_length": 249.2125,
1064
+ "avg_student_mask_ratio": 0.4674084897618741,
1065
+ "batch_ainp_frac": 0.0,
1066
+ "batch_inp_frac": 1.0,
1067
+ "batch_inp_oh_frac": 0.0,
1068
+ "batch_inp_par_frac": 0.0,
1069
+ "batch_inp_par_par_frac": 0.0,
1070
+ "batch_inp_par_reverse_frac": 0.0,
1071
+ "batch_rl_frac": 0.0,
1072
+ "batch_sft_frac": 0.0,
1073
+ "batch_soft_sft_frac": 0.0,
1074
+ "batch_tf_frac": 0.0,
1075
+ "ce_loss": 0.37605725416574387,
1076
+ "epoch": 0.9173333333333333,
1077
+ "grad_norm": 0.43359375,
1078
+ "kd_loss": 0.49442086774362226,
1079
+ "learning_rate": 3e-06,
1080
+ "loss": 0.6699,
1081
+ "masked_tokens": 104.5625,
1082
+ "mean_t": 0.49262027950026094,
1083
+ "step": 430,
1084
+ "student_masked_tokens": 104.5625
1085
+ },
1086
+ {
1087
+ "avg_mask_ratio": 0.4415457699564286,
1088
+ "avg_response_length": 241.0875,
1089
+ "avg_student_mask_ratio": 0.4415457699564286,
1090
+ "batch_ainp_frac": 0.0,
1091
+ "batch_inp_frac": 1.0,
1092
+ "batch_inp_oh_frac": 0.0,
1093
+ "batch_inp_par_frac": 0.0,
1094
+ "batch_inp_par_par_frac": 0.0,
1095
+ "batch_inp_par_reverse_frac": 0.0,
1096
+ "batch_rl_frac": 0.0,
1097
+ "batch_sft_frac": 0.0,
1098
+ "batch_soft_sft_frac": 0.0,
1099
+ "batch_tf_frac": 0.0,
1100
+ "ce_loss": 0.3754083825901603,
1101
+ "epoch": 0.9386666666666666,
1102
+ "grad_norm": 0.6328125,
1103
+ "kd_loss": 0.45159815376919143,
1104
+ "learning_rate": 3e-06,
1105
+ "loss": 0.6585,
1106
+ "masked_tokens": 113.0875,
1107
+ "mean_t": 0.47046207524836064,
1108
+ "step": 440,
1109
+ "student_masked_tokens": 113.0875
1110
+ },
1111
+ {
1112
+ "avg_mask_ratio": 0.42486972180195154,
1113
+ "avg_response_length": 231.9875,
1114
+ "avg_student_mask_ratio": 0.42486972180195154,
1115
+ "batch_ainp_frac": 0.0,
1116
+ "batch_inp_frac": 1.0,
1117
+ "batch_inp_oh_frac": 0.0,
1118
+ "batch_inp_par_frac": 0.0,
1119
+ "batch_inp_par_par_frac": 0.0,
1120
+ "batch_inp_par_reverse_frac": 0.0,
1121
+ "batch_rl_frac": 0.0,
1122
+ "batch_sft_frac": 0.0,
1123
+ "batch_soft_sft_frac": 0.0,
1124
+ "batch_tf_frac": 0.0,
1125
+ "ce_loss": 0.32457938515717616,
1126
+ "epoch": 0.96,
1127
+ "grad_norm": 0.6953125,
1128
+ "kd_loss": 0.4011907008050457,
1129
+ "learning_rate": 3e-06,
1130
+ "loss": 0.5644,
1131
+ "masked_tokens": 103.4,
1132
+ "mean_t": 0.45781184462830427,
1133
+ "step": 450,
1134
+ "student_masked_tokens": 103.4
1135
+ },
1136
+ {
1137
+ "avg_mask_ratio": 0.47578654896933587,
1138
+ "avg_response_length": 214.6125,
1139
+ "avg_student_mask_ratio": 0.47578654896933587,
1140
+ "batch_ainp_frac": 0.0,
1141
+ "batch_inp_frac": 1.0,
1142
+ "batch_inp_oh_frac": 0.0,
1143
+ "batch_inp_par_frac": 0.0,
1144
+ "batch_inp_par_par_frac": 0.0,
1145
+ "batch_inp_par_reverse_frac": 0.0,
1146
+ "batch_rl_frac": 0.0,
1147
+ "batch_sft_frac": 0.0,
1148
+ "batch_soft_sft_frac": 0.0,
1149
+ "batch_tf_frac": 0.0,
1150
+ "ce_loss": 0.32885359905767475,
1151
+ "epoch": 0.9813333333333333,
1152
+ "grad_norm": 0.16015625,
1153
+ "kd_loss": 0.44463847501747294,
1154
+ "learning_rate": 3e-06,
1155
+ "loss": 0.635,
1156
+ "masked_tokens": 105.3125,
1157
+ "mean_t": 0.5075790266972036,
1158
+ "step": 460,
1159
+ "student_masked_tokens": 105.3125
1160
+ },
1161
+ {
1162
+ "avg_mask_ratio": 0.4782901787132557,
1163
+ "avg_response_length": 224.0952380952381,
1164
+ "avg_student_mask_ratio": 0.4782901787132557,
1165
+ "batch_ainp_frac": 0.0,
1166
+ "batch_inp_frac": 1.0,
1167
+ "batch_inp_oh_frac": 0.0,
1168
+ "batch_inp_par_frac": 0.0,
1169
+ "batch_inp_par_par_frac": 0.0,
1170
+ "batch_inp_par_reverse_frac": 0.0,
1171
+ "batch_rl_frac": 0.0,
1172
+ "batch_sft_frac": 0.0,
1173
+ "batch_soft_sft_frac": 0.0,
1174
+ "batch_tf_frac": 0.0,
1175
+ "ce_loss": 0.3393430382851702,
1176
+ "epoch": 1.0042666666666666,
1177
+ "grad_norm": 0.65625,
1178
+ "kd_loss": 0.5178591865708675,
1179
+ "learning_rate": 3e-06,
1180
+ "loss": 0.7769,
1181
+ "masked_tokens": 107.23809523809524,
1182
+ "mean_t": 0.5031429776822084,
1183
+ "step": 470,
1184
+ "student_masked_tokens": 107.23809523809524
1185
+ },
1186
+ {
1187
+ "avg_mask_ratio": 0.47575968883465974,
1188
+ "avg_response_length": 249.4125,
1189
+ "avg_student_mask_ratio": 0.47575968883465974,
1190
+ "batch_ainp_frac": 0.0,
1191
+ "batch_inp_frac": 1.0,
1192
+ "batch_inp_oh_frac": 0.0,
1193
+ "batch_inp_par_frac": 0.0,
1194
+ "batch_inp_par_par_frac": 0.0,
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": 0.44613247805159517,
1201
+ "epoch": 1.0256,
1202
+ "grad_norm": 0.498046875,
1203
+ "kd_loss": 0.5374264506522252,
1204
+ "learning_rate": 3e-06,
1205
+ "loss": 0.6772,
1206
+ "masked_tokens": 118.35,
1207
+ "mean_t": 0.504472183593316,
1208
+ "step": 480,
1209
+ "student_masked_tokens": 118.35
1210
+ },
1211
+ {
1212
+ "avg_mask_ratio": 0.4563717324635945,
1213
+ "avg_response_length": 232.0375,
1214
+ "avg_student_mask_ratio": 0.4563717324635945,
1215
+ "batch_ainp_frac": 0.0,
1216
+ "batch_inp_frac": 1.0,
1217
+ "batch_inp_oh_frac": 0.0,
1218
+ "batch_inp_par_frac": 0.0,
1219
+ "batch_inp_par_par_frac": 0.0,
1220
+ "batch_inp_par_reverse_frac": 0.0,
1221
+ "batch_rl_frac": 0.0,
1222
+ "batch_sft_frac": 0.0,
1223
+ "batch_soft_sft_frac": 0.0,
1224
+ "batch_tf_frac": 0.0,
1225
+ "ce_loss": 0.37626147485414096,
1226
+ "epoch": 1.0469333333333333,
1227
+ "grad_norm": 0.54296875,
1228
+ "kd_loss": 0.392788901903657,
1229
+ "learning_rate": 3e-06,
1230
+ "loss": 0.6047,
1231
+ "masked_tokens": 98.35,
1232
+ "mean_t": 0.4888980514719151,
1233
+ "step": 490,
1234
+ "student_masked_tokens": 98.35
1235
+ },
1236
+ {
1237
+ "avg_mask_ratio": 0.5079968665260821,
1238
+ "avg_response_length": 253.7875,
1239
+ "avg_student_mask_ratio": 0.5079968665260821,
1240
+ "batch_ainp_frac": 0.0,
1241
+ "batch_inp_frac": 1.0,
1242
+ "batch_inp_oh_frac": 0.0,
1243
+ "batch_inp_par_frac": 0.0,
1244
+ "batch_inp_par_par_frac": 0.0,
1245
+ "batch_inp_par_reverse_frac": 0.0,
1246
+ "batch_rl_frac": 0.0,
1247
+ "batch_sft_frac": 0.0,
1248
+ "batch_soft_sft_frac": 0.0,
1249
+ "batch_tf_frac": 0.0,
1250
+ "ce_loss": 0.30954629559880686,
1251
+ "epoch": 1.0682666666666667,
1252
+ "grad_norm": 0.291015625,
1253
+ "kd_loss": 0.4563873354276211,
1254
+ "learning_rate": 3e-06,
1255
+ "loss": 0.5996,
1256
+ "masked_tokens": 128.225,
1257
+ "mean_t": 0.5469163245841628,
1258
+ "step": 500,
1259
+ "student_masked_tokens": 128.225
1260
+ },
1261
+ {
1262
+ "avg_mask_ratio": 0.5109448074479588,
1263
+ "avg_response_length": 254.2,
1264
+ "avg_student_mask_ratio": 0.5109448074479588,
1265
+ "batch_ainp_frac": 0.0,
1266
+ "batch_inp_frac": 1.0,
1267
+ "batch_inp_oh_frac": 0.0,
1268
+ "batch_inp_par_frac": 0.0,
1269
+ "batch_inp_par_par_frac": 0.0,
1270
+ "batch_inp_par_reverse_frac": 0.0,
1271
+ "batch_rl_frac": 0.0,
1272
+ "batch_sft_frac": 0.0,
1273
+ "batch_soft_sft_frac": 0.0,
1274
+ "batch_tf_frac": 0.0,
1275
+ "ce_loss": 0.2868076219221166,
1276
+ "epoch": 1.0896,
1277
+ "grad_norm": 2.515625,
1278
+ "kd_loss": 0.5652106747879998,
1279
+ "learning_rate": 3e-06,
1280
+ "loss": 0.6398,
1281
+ "masked_tokens": 137.5875,
1282
+ "mean_t": 0.5275314710394013,
1283
+ "step": 510,
1284
+ "student_masked_tokens": 137.5875
1285
+ },
1286
+ {
1287
+ "avg_mask_ratio": 0.45396183808334173,
1288
+ "avg_response_length": 202.7625,
1289
+ "avg_student_mask_ratio": 0.45396183808334173,
1290
+ "batch_ainp_frac": 0.0,
1291
+ "batch_inp_frac": 1.0,
1292
+ "batch_inp_oh_frac": 0.0,
1293
+ "batch_inp_par_frac": 0.0,
1294
+ "batch_inp_par_par_frac": 0.0,
1295
+ "batch_inp_par_reverse_frac": 0.0,
1296
+ "batch_rl_frac": 0.0,
1297
+ "batch_sft_frac": 0.0,
1298
+ "batch_soft_sft_frac": 0.0,
1299
+ "batch_tf_frac": 0.0,
1300
+ "ce_loss": 0.38311037250946356,
1301
+ "epoch": 1.1109333333333333,
1302
+ "grad_norm": 0.6171875,
1303
+ "kd_loss": 0.423658079797778,
1304
+ "learning_rate": 3e-06,
1305
+ "loss": 0.6386,
1306
+ "masked_tokens": 87.0625,
1307
+ "mean_t": 0.49193521235138177,
1308
+ "step": 520,
1309
+ "student_masked_tokens": 87.0625
1310
+ },
1311
+ {
1312
+ "avg_mask_ratio": 0.47015948037151245,
1313
+ "avg_response_length": 214.275,
1314
+ "avg_student_mask_ratio": 0.47015948037151245,
1315
+ "batch_ainp_frac": 0.0,
1316
+ "batch_inp_frac": 1.0,
1317
+ "batch_inp_oh_frac": 0.0,
1318
+ "batch_inp_par_frac": 0.0,
1319
+ "batch_inp_par_par_frac": 0.0,
1320
+ "batch_inp_par_reverse_frac": 0.0,
1321
+ "batch_rl_frac": 0.0,
1322
+ "batch_sft_frac": 0.0,
1323
+ "batch_soft_sft_frac": 0.0,
1324
+ "batch_tf_frac": 0.0,
1325
+ "ce_loss": 0.47228433731506814,
1326
+ "epoch": 1.1322666666666668,
1327
+ "grad_norm": 0.609375,
1328
+ "kd_loss": 0.45688082203427316,
1329
+ "learning_rate": 3e-06,
1330
+ "loss": 0.737,
1331
+ "masked_tokens": 99.8625,
1332
+ "mean_t": 0.49621942077938,
1333
+ "step": 530,
1334
+ "student_masked_tokens": 99.8625
1335
+ },
1336
+ {
1337
+ "avg_mask_ratio": 0.4892866689246148,
1338
+ "avg_response_length": 231.3125,
1339
+ "avg_student_mask_ratio": 0.4892866689246148,
1340
+ "batch_ainp_frac": 0.0,
1341
+ "batch_inp_frac": 1.0,
1342
+ "batch_inp_oh_frac": 0.0,
1343
+ "batch_inp_par_frac": 0.0,
1344
+ "batch_inp_par_par_frac": 0.0,
1345
+ "batch_inp_par_reverse_frac": 0.0,
1346
+ "batch_rl_frac": 0.0,
1347
+ "batch_sft_frac": 0.0,
1348
+ "batch_soft_sft_frac": 0.0,
1349
+ "batch_tf_frac": 0.0,
1350
+ "ce_loss": 0.4080867745911064,
1351
+ "epoch": 1.1536,
1352
+ "grad_norm": 0.341796875,
1353
+ "kd_loss": 0.5618651450654625,
1354
+ "learning_rate": 3e-06,
1355
+ "loss": 0.6922,
1356
+ "masked_tokens": 107.375,
1357
+ "mean_t": 0.5208023569080978,
1358
+ "step": 540,
1359
+ "student_masked_tokens": 107.375
1360
+ },
1361
+ {
1362
+ "avg_mask_ratio": 0.4541942774085328,
1363
+ "avg_response_length": 213.525,
1364
+ "avg_student_mask_ratio": 0.4541942774085328,
1365
+ "batch_ainp_frac": 0.0,
1366
+ "batch_inp_frac": 1.0,
1367
+ "batch_inp_oh_frac": 0.0,
1368
+ "batch_inp_par_frac": 0.0,
1369
+ "batch_inp_par_par_frac": 0.0,
1370
+ "batch_inp_par_reverse_frac": 0.0,
1371
+ "batch_rl_frac": 0.0,
1372
+ "batch_sft_frac": 0.0,
1373
+ "batch_soft_sft_frac": 0.0,
1374
+ "batch_tf_frac": 0.0,
1375
+ "ce_loss": 0.22217674175137744,
1376
+ "epoch": 1.1749333333333334,
1377
+ "grad_norm": 0.2412109375,
1378
+ "kd_loss": 0.3673438885498399,
1379
+ "learning_rate": 3e-06,
1380
+ "loss": 0.5008,
1381
+ "masked_tokens": 97.8875,
1382
+ "mean_t": 0.4767197913257405,
1383
+ "step": 550,
1384
+ "student_masked_tokens": 97.8875
1385
+ },
1386
+ {
1387
+ "avg_mask_ratio": 0.39282396506750955,
1388
+ "avg_response_length": 231.4125,
1389
+ "avg_student_mask_ratio": 0.39282396506750955,
1390
+ "batch_ainp_frac": 0.0,
1391
+ "batch_inp_frac": 1.0,
1392
+ "batch_inp_oh_frac": 0.0,
1393
+ "batch_inp_par_frac": 0.0,
1394
+ "batch_inp_par_par_frac": 0.0,
1395
+ "batch_inp_par_reverse_frac": 0.0,
1396
+ "batch_rl_frac": 0.0,
1397
+ "batch_sft_frac": 0.0,
1398
+ "batch_soft_sft_frac": 0.0,
1399
+ "batch_tf_frac": 0.0,
1400
+ "ce_loss": 0.3512847523151777,
1401
+ "epoch": 1.1962666666666666,
1402
+ "grad_norm": 0.8828125,
1403
+ "kd_loss": 0.48686740984790616,
1404
+ "learning_rate": 3e-06,
1405
+ "loss": 0.5823,
1406
+ "masked_tokens": 99.2875,
1407
+ "mean_t": 0.4111072298779618,
1408
+ "step": 560,
1409
+ "student_masked_tokens": 99.2875
1410
+ },
1411
+ {
1412
+ "avg_mask_ratio": 0.4483634108910337,
1413
+ "avg_response_length": 230.1625,
1414
+ "avg_student_mask_ratio": 0.4483634108910337,
1415
+ "batch_ainp_frac": 0.0,
1416
+ "batch_inp_frac": 1.0,
1417
+ "batch_inp_oh_frac": 0.0,
1418
+ "batch_inp_par_frac": 0.0,
1419
+ "batch_inp_par_par_frac": 0.0,
1420
+ "batch_inp_par_reverse_frac": 0.0,
1421
+ "batch_rl_frac": 0.0,
1422
+ "batch_sft_frac": 0.0,
1423
+ "batch_soft_sft_frac": 0.0,
1424
+ "batch_tf_frac": 0.0,
1425
+ "ce_loss": 0.31345968546206676,
1426
+ "epoch": 1.2176,
1427
+ "grad_norm": 0.4453125,
1428
+ "kd_loss": 0.41564053312727084,
1429
+ "learning_rate": 3e-06,
1430
+ "loss": 0.5898,
1431
+ "masked_tokens": 108.9875,
1432
+ "mean_t": 0.48533305872697385,
1433
+ "step": 570,
1434
+ "student_masked_tokens": 108.9875
1435
+ },
1436
+ {
1437
+ "avg_mask_ratio": 0.465452536707744,
1438
+ "avg_response_length": 267.4375,
1439
+ "avg_student_mask_ratio": 0.465452536707744,
1440
+ "batch_ainp_frac": 0.0,
1441
+ "batch_inp_frac": 1.0,
1442
+ "batch_inp_oh_frac": 0.0,
1443
+ "batch_inp_par_frac": 0.0,
1444
+ "batch_inp_par_par_frac": 0.0,
1445
+ "batch_inp_par_reverse_frac": 0.0,
1446
+ "batch_rl_frac": 0.0,
1447
+ "batch_sft_frac": 0.0,
1448
+ "batch_soft_sft_frac": 0.0,
1449
+ "batch_tf_frac": 0.0,
1450
+ "ce_loss": 0.3618907347364768,
1451
+ "epoch": 1.2389333333333332,
1452
+ "grad_norm": 8.6875,
1453
+ "kd_loss": 0.4481006292516895,
1454
+ "learning_rate": 3e-06,
1455
+ "loss": 0.6314,
1456
+ "masked_tokens": 129.075,
1457
+ "mean_t": 0.49976949762785805,
1458
+ "step": 580,
1459
+ "student_masked_tokens": 129.075
1460
+ },
1461
+ {
1462
+ "avg_mask_ratio": 0.5225977989146486,
1463
+ "avg_response_length": 228.45,
1464
+ "avg_student_mask_ratio": 0.5225977989146486,
1465
+ "batch_ainp_frac": 0.0,
1466
+ "batch_inp_frac": 1.0,
1467
+ "batch_inp_oh_frac": 0.0,
1468
+ "batch_inp_par_frac": 0.0,
1469
+ "batch_inp_par_par_frac": 0.0,
1470
+ "batch_inp_par_reverse_frac": 0.0,
1471
+ "batch_rl_frac": 0.0,
1472
+ "batch_sft_frac": 0.0,
1473
+ "batch_soft_sft_frac": 0.0,
1474
+ "batch_tf_frac": 0.0,
1475
+ "ce_loss": 0.5639314363695348,
1476
+ "epoch": 1.2602666666666666,
1477
+ "grad_norm": 1.1328125,
1478
+ "kd_loss": 0.5351108588445992,
1479
+ "learning_rate": 3e-06,
1480
+ "loss": 0.8274,
1481
+ "masked_tokens": 121.675,
1482
+ "mean_t": 0.5521843038732186,
1483
+ "step": 590,
1484
+ "student_masked_tokens": 121.675
1485
+ },
1486
+ {
1487
+ "avg_mask_ratio": 0.44998724836623294,
1488
+ "avg_response_length": 236.7,
1489
+ "avg_student_mask_ratio": 0.44998724836623294,
1490
+ "batch_ainp_frac": 0.0,
1491
+ "batch_inp_frac": 1.0,
1492
+ "batch_inp_oh_frac": 0.0,
1493
+ "batch_inp_par_frac": 0.0,
1494
+ "batch_inp_par_par_frac": 0.0,
1495
+ "batch_inp_par_reverse_frac": 0.0,
1496
+ "batch_rl_frac": 0.0,
1497
+ "batch_sft_frac": 0.0,
1498
+ "batch_soft_sft_frac": 0.0,
1499
+ "batch_tf_frac": 0.0,
1500
+ "ce_loss": 0.3396833263838971,
1501
+ "epoch": 1.2816,
1502
+ "grad_norm": 0.365234375,
1503
+ "kd_loss": 0.41761890975592914,
1504
+ "learning_rate": 3e-06,
1505
+ "loss": 0.5752,
1506
+ "masked_tokens": 110.1625,
1507
+ "mean_t": 0.4788527532829903,
1508
+ "step": 600,
1509
+ "student_masked_tokens": 110.1625
1510
+ },
1511
+ {
1512
+ "avg_mask_ratio": 0.5042130865273066,
1513
+ "avg_response_length": 230.3375,
1514
+ "avg_student_mask_ratio": 0.5042130865273066,
1515
+ "batch_ainp_frac": 0.0,
1516
+ "batch_inp_frac": 1.0,
1517
+ "batch_inp_oh_frac": 0.0,
1518
+ "batch_inp_par_frac": 0.0,
1519
+ "batch_inp_par_par_frac": 0.0,
1520
+ "batch_inp_par_reverse_frac": 0.0,
1521
+ "batch_rl_frac": 0.0,
1522
+ "batch_sft_frac": 0.0,
1523
+ "batch_soft_sft_frac": 0.0,
1524
+ "batch_tf_frac": 0.0,
1525
+ "ce_loss": 0.35890077192343595,
1526
+ "epoch": 1.3029333333333333,
1527
+ "grad_norm": 0.28515625,
1528
+ "kd_loss": 0.5558427174539929,
1529
+ "learning_rate": 3e-06,
1530
+ "loss": 0.7657,
1531
+ "masked_tokens": 112.625,
1532
+ "mean_t": 0.5445419924799353,
1533
+ "step": 610,
1534
+ "student_masked_tokens": 112.625
1535
+ },
1536
+ {
1537
+ "avg_mask_ratio": 0.49637898594373836,
1538
+ "avg_response_length": 233.0625,
1539
+ "avg_student_mask_ratio": 0.49637898594373836,
1540
+ "batch_ainp_frac": 0.0,
1541
+ "batch_inp_frac": 1.0,
1542
+ "batch_inp_oh_frac": 0.0,
1543
+ "batch_inp_par_frac": 0.0,
1544
+ "batch_inp_par_par_frac": 0.0,
1545
+ "batch_inp_par_reverse_frac": 0.0,
1546
+ "batch_rl_frac": 0.0,
1547
+ "batch_sft_frac": 0.0,
1548
+ "batch_soft_sft_frac": 0.0,
1549
+ "batch_tf_frac": 0.0,
1550
+ "ce_loss": 0.32318839170733327,
1551
+ "epoch": 1.3242666666666667,
1552
+ "grad_norm": 0.515625,
1553
+ "kd_loss": 0.5518322235134179,
1554
+ "learning_rate": 3e-06,
1555
+ "loss": 0.6742,
1556
+ "masked_tokens": 111.25,
1557
+ "mean_t": 0.52490478400141,
1558
+ "step": 620,
1559
+ "student_masked_tokens": 111.25
1560
+ },
1561
+ {
1562
+ "avg_mask_ratio": 0.5177568581304512,
1563
+ "avg_response_length": 257.2125,
1564
+ "avg_student_mask_ratio": 0.5177568581304512,
1565
+ "batch_ainp_frac": 0.0,
1566
+ "batch_inp_frac": 1.0,
1567
+ "batch_inp_oh_frac": 0.0,
1568
+ "batch_inp_par_frac": 0.0,
1569
+ "batch_inp_par_par_frac": 0.0,
1570
+ "batch_inp_par_reverse_frac": 0.0,
1571
+ "batch_rl_frac": 0.0,
1572
+ "batch_sft_frac": 0.0,
1573
+ "batch_soft_sft_frac": 0.0,
1574
+ "batch_tf_frac": 0.0,
1575
+ "ce_loss": 0.5710563842050931,
1576
+ "epoch": 1.3456000000000001,
1577
+ "grad_norm": 1.3515625,
1578
+ "kd_loss": 0.5316411310721378,
1579
+ "learning_rate": 3e-06,
1580
+ "loss": 0.8598,
1581
+ "masked_tokens": 129.6125,
1582
+ "mean_t": 0.5564947265549562,
1583
+ "step": 630,
1584
+ "student_masked_tokens": 129.6125
1585
+ },
1586
+ {
1587
+ "avg_mask_ratio": 0.48226998368045315,
1588
+ "avg_response_length": 237.7125,
1589
+ "avg_student_mask_ratio": 0.48226998368045315,
1590
+ "batch_ainp_frac": 0.0,
1591
+ "batch_inp_frac": 1.0,
1592
+ "batch_inp_oh_frac": 0.0,
1593
+ "batch_inp_par_frac": 0.0,
1594
+ "batch_inp_par_par_frac": 0.0,
1595
+ "batch_inp_par_reverse_frac": 0.0,
1596
+ "batch_rl_frac": 0.0,
1597
+ "batch_sft_frac": 0.0,
1598
+ "batch_soft_sft_frac": 0.0,
1599
+ "batch_tf_frac": 0.0,
1600
+ "ce_loss": 0.2804489129174499,
1601
+ "epoch": 1.3669333333333333,
1602
+ "grad_norm": 0.2421875,
1603
+ "kd_loss": 0.3663112932188085,
1604
+ "learning_rate": 3e-06,
1605
+ "loss": 0.4584,
1606
+ "masked_tokens": 120.275,
1607
+ "mean_t": 0.5093393943971023,
1608
+ "step": 640,
1609
+ "student_masked_tokens": 120.275
1610
+ },
1611
+ {
1612
+ "avg_mask_ratio": 0.5306948523037136,
1613
+ "avg_response_length": 238.0125,
1614
+ "avg_student_mask_ratio": 0.5306948523037136,
1615
+ "batch_ainp_frac": 0.0,
1616
+ "batch_inp_frac": 1.0,
1617
+ "batch_inp_oh_frac": 0.0,
1618
+ "batch_inp_par_frac": 0.0,
1619
+ "batch_inp_par_par_frac": 0.0,
1620
+ "batch_inp_par_reverse_frac": 0.0,
1621
+ "batch_rl_frac": 0.0,
1622
+ "batch_sft_frac": 0.0,
1623
+ "batch_soft_sft_frac": 0.0,
1624
+ "batch_tf_frac": 0.0,
1625
+ "ce_loss": 0.475157093159612,
1626
+ "epoch": 1.3882666666666665,
1627
+ "grad_norm": 1.8125,
1628
+ "kd_loss": 0.5062341513834724,
1629
+ "learning_rate": 3e-06,
1630
+ "loss": 0.7115,
1631
+ "masked_tokens": 133.25,
1632
+ "mean_t": 0.5558586571365595,
1633
+ "step": 650,
1634
+ "student_masked_tokens": 133.25
1635
+ },
1636
+ {
1637
+ "avg_mask_ratio": 0.4821273953886703,
1638
+ "avg_response_length": 247.775,
1639
+ "avg_student_mask_ratio": 0.4821273953886703,
1640
+ "batch_ainp_frac": 0.0,
1641
+ "batch_inp_frac": 1.0,
1642
+ "batch_inp_oh_frac": 0.0,
1643
+ "batch_inp_par_frac": 0.0,
1644
+ "batch_inp_par_par_frac": 0.0,
1645
+ "batch_inp_par_reverse_frac": 0.0,
1646
+ "batch_rl_frac": 0.0,
1647
+ "batch_sft_frac": 0.0,
1648
+ "batch_soft_sft_frac": 0.0,
1649
+ "batch_tf_frac": 0.0,
1650
+ "ce_loss": 0.41770620119971225,
1651
+ "epoch": 1.4096,
1652
+ "grad_norm": 0.9375,
1653
+ "kd_loss": 0.425496905214095,
1654
+ "learning_rate": 3e-06,
1655
+ "loss": 0.6361,
1656
+ "masked_tokens": 128.875,
1657
+ "mean_t": 0.51307404555846,
1658
+ "step": 660,
1659
+ "student_masked_tokens": 128.875
1660
+ },
1661
+ {
1662
+ "avg_mask_ratio": 0.46056515555246735,
1663
+ "avg_response_length": 240.4375,
1664
+ "avg_student_mask_ratio": 0.46056515555246735,
1665
+ "batch_ainp_frac": 0.0,
1666
+ "batch_inp_frac": 1.0,
1667
+ "batch_inp_oh_frac": 0.0,
1668
+ "batch_inp_par_frac": 0.0,
1669
+ "batch_inp_par_par_frac": 0.0,
1670
+ "batch_inp_par_reverse_frac": 0.0,
1671
+ "batch_rl_frac": 0.0,
1672
+ "batch_sft_frac": 0.0,
1673
+ "batch_soft_sft_frac": 0.0,
1674
+ "batch_tf_frac": 0.0,
1675
+ "ce_loss": 0.24846992658117414,
1676
+ "epoch": 1.4309333333333334,
1677
+ "grad_norm": 0.60546875,
1678
+ "kd_loss": 0.34861083538812637,
1679
+ "learning_rate": 3e-06,
1680
+ "loss": 0.5112,
1681
+ "masked_tokens": 119.85,
1682
+ "mean_t": 0.4907285622088239,
1683
+ "step": 670,
1684
+ "student_masked_tokens": 119.85
1685
+ },
1686
+ {
1687
+ "avg_mask_ratio": 0.4666106043441687,
1688
+ "avg_response_length": 226.7375,
1689
+ "avg_student_mask_ratio": 0.4666106043441687,
1690
+ "batch_ainp_frac": 0.0,
1691
+ "batch_inp_frac": 1.0,
1692
+ "batch_inp_oh_frac": 0.0,
1693
+ "batch_inp_par_frac": 0.0,
1694
+ "batch_inp_par_par_frac": 0.0,
1695
+ "batch_inp_par_reverse_frac": 0.0,
1696
+ "batch_rl_frac": 0.0,
1697
+ "batch_sft_frac": 0.0,
1698
+ "batch_soft_sft_frac": 0.0,
1699
+ "batch_tf_frac": 0.0,
1700
+ "ce_loss": 0.4541423492493323,
1701
+ "epoch": 1.4522666666666666,
1702
+ "grad_norm": 0.51953125,
1703
+ "kd_loss": 0.4910934407485213,
1704
+ "learning_rate": 3e-06,
1705
+ "loss": 0.6946,
1706
+ "masked_tokens": 107.4625,
1707
+ "mean_t": 0.4913603452499956,
1708
+ "step": 680,
1709
+ "student_masked_tokens": 107.4625
1710
+ },
1711
+ {
1712
+ "avg_mask_ratio": 0.4790851596510038,
1713
+ "avg_response_length": 202.05,
1714
+ "avg_student_mask_ratio": 0.4790851596510038,
1715
+ "batch_ainp_frac": 0.0,
1716
+ "batch_inp_frac": 1.0,
1717
+ "batch_inp_oh_frac": 0.0,
1718
+ "batch_inp_par_frac": 0.0,
1719
+ "batch_inp_par_par_frac": 0.0,
1720
+ "batch_inp_par_reverse_frac": 0.0,
1721
+ "batch_rl_frac": 0.0,
1722
+ "batch_sft_frac": 0.0,
1723
+ "batch_soft_sft_frac": 0.0,
1724
+ "batch_tf_frac": 0.0,
1725
+ "ce_loss": 0.3711260147189023,
1726
+ "epoch": 1.4736,
1727
+ "grad_norm": 2.03125,
1728
+ "kd_loss": 0.41718243765291446,
1729
+ "learning_rate": 3e-06,
1730
+ "loss": 0.6313,
1731
+ "masked_tokens": 111.3125,
1732
+ "mean_t": 0.5133644798654131,
1733
+ "step": 690,
1734
+ "student_masked_tokens": 111.3125
1735
+ },
1736
+ {
1737
+ "avg_mask_ratio": 0.5250519359949977,
1738
+ "avg_response_length": 228.125,
1739
+ "avg_student_mask_ratio": 0.5250519359949977,
1740
+ "batch_ainp_frac": 0.0,
1741
+ "batch_inp_frac": 1.0,
1742
+ "batch_inp_oh_frac": 0.0,
1743
+ "batch_inp_par_frac": 0.0,
1744
+ "batch_inp_par_par_frac": 0.0,
1745
+ "batch_inp_par_reverse_frac": 0.0,
1746
+ "batch_rl_frac": 0.0,
1747
+ "batch_sft_frac": 0.0,
1748
+ "batch_soft_sft_frac": 0.0,
1749
+ "batch_tf_frac": 0.0,
1750
+ "ce_loss": 0.22230932631540554,
1751
+ "epoch": 1.4949333333333334,
1752
+ "grad_norm": 0.26171875,
1753
+ "kd_loss": 0.6619142963969352,
1754
+ "learning_rate": 3e-06,
1755
+ "loss": 0.7717,
1756
+ "masked_tokens": 132.55,
1757
+ "mean_t": 0.5625698395539075,
1758
+ "step": 700,
1759
+ "student_masked_tokens": 132.55
1760
+ },
1761
+ {
1762
+ "avg_mask_ratio": 0.4790433386107907,
1763
+ "avg_response_length": 212.5,
1764
+ "avg_student_mask_ratio": 0.4790433386107907,
1765
+ "batch_ainp_frac": 0.0,
1766
+ "batch_inp_frac": 1.0,
1767
+ "batch_inp_oh_frac": 0.0,
1768
+ "batch_inp_par_frac": 0.0,
1769
+ "batch_inp_par_par_frac": 0.0,
1770
+ "batch_inp_par_reverse_frac": 0.0,
1771
+ "batch_rl_frac": 0.0,
1772
+ "batch_sft_frac": 0.0,
1773
+ "batch_soft_sft_frac": 0.0,
1774
+ "batch_tf_frac": 0.0,
1775
+ "ce_loss": 0.24621229091012536,
1776
+ "epoch": 1.5162666666666667,
1777
+ "grad_norm": 0.2099609375,
1778
+ "kd_loss": 0.43454050603151584,
1779
+ "learning_rate": 3e-06,
1780
+ "loss": 0.5302,
1781
+ "masked_tokens": 108.7375,
1782
+ "mean_t": 0.5135623761918395,
1783
+ "step": 710,
1784
+ "student_masked_tokens": 108.7375
1785
+ },
1786
+ {
1787
+ "avg_mask_ratio": 0.47950589570682495,
1788
+ "avg_response_length": 227.075,
1789
+ "avg_student_mask_ratio": 0.47950589570682495,
1790
+ "batch_ainp_frac": 0.0,
1791
+ "batch_inp_frac": 1.0,
1792
+ "batch_inp_oh_frac": 0.0,
1793
+ "batch_inp_par_frac": 0.0,
1794
+ "batch_inp_par_par_frac": 0.0,
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": 0.36416104665024707,
1801
+ "epoch": 1.5375999999999999,
1802
+ "grad_norm": 0.75,
1803
+ "kd_loss": 0.5665610315164941,
1804
+ "learning_rate": 3e-06,
1805
+ "loss": 0.7121,
1806
+ "masked_tokens": 110.8,
1807
+ "mean_t": 0.5117021896177902,
1808
+ "step": 720,
1809
+ "student_masked_tokens": 110.8
1810
+ },
1811
+ {
1812
+ "avg_mask_ratio": 0.4604924251558259,
1813
+ "avg_response_length": 232.925,
1814
+ "avg_student_mask_ratio": 0.4604924251558259,
1815
+ "batch_ainp_frac": 0.0,
1816
+ "batch_inp_frac": 1.0,
1817
+ "batch_inp_oh_frac": 0.0,
1818
+ "batch_inp_par_frac": 0.0,
1819
+ "batch_inp_par_par_frac": 0.0,
1820
+ "batch_inp_par_reverse_frac": 0.0,
1821
+ "batch_rl_frac": 0.0,
1822
+ "batch_sft_frac": 0.0,
1823
+ "batch_soft_sft_frac": 0.0,
1824
+ "batch_tf_frac": 0.0,
1825
+ "ce_loss": 0.38923927966282007,
1826
+ "epoch": 1.5589333333333333,
1827
+ "grad_norm": 1.015625,
1828
+ "kd_loss": 0.4302867329986782,
1829
+ "learning_rate": 3e-06,
1830
+ "loss": 0.639,
1831
+ "masked_tokens": 104.9625,
1832
+ "mean_t": 0.49050743713742123,
1833
+ "step": 730,
1834
+ "student_masked_tokens": 104.9625
1835
+ },
1836
+ {
1837
+ "avg_mask_ratio": 0.5185885130194947,
1838
+ "avg_response_length": 183.325,
1839
+ "avg_student_mask_ratio": 0.5185885130194947,
1840
+ "batch_ainp_frac": 0.0,
1841
+ "batch_inp_frac": 1.0,
1842
+ "batch_inp_oh_frac": 0.0,
1843
+ "batch_inp_par_frac": 0.0,
1844
+ "batch_inp_par_par_frac": 0.0,
1845
+ "batch_inp_par_reverse_frac": 0.0,
1846
+ "batch_rl_frac": 0.0,
1847
+ "batch_sft_frac": 0.0,
1848
+ "batch_soft_sft_frac": 0.0,
1849
+ "batch_tf_frac": 0.0,
1850
+ "ce_loss": 0.3361817517367399,
1851
+ "epoch": 1.5802666666666667,
1852
+ "grad_norm": 0.40234375,
1853
+ "kd_loss": 0.5340734164818514,
1854
+ "learning_rate": 3e-06,
1855
+ "loss": 0.7461,
1856
+ "masked_tokens": 97.125,
1857
+ "mean_t": 0.5505168779753149,
1858
+ "step": 740,
1859
+ "student_masked_tokens": 97.125
1860
+ },
1861
+ {
1862
+ "avg_mask_ratio": 0.4191439319110941,
1863
+ "avg_response_length": 223.65,
1864
+ "avg_student_mask_ratio": 0.4191439319110941,
1865
+ "batch_ainp_frac": 0.0,
1866
+ "batch_inp_frac": 1.0,
1867
+ "batch_inp_oh_frac": 0.0,
1868
+ "batch_inp_par_frac": 0.0,
1869
+ "batch_inp_par_par_frac": 0.0,
1870
+ "batch_inp_par_reverse_frac": 0.0,
1871
+ "batch_rl_frac": 0.0,
1872
+ "batch_sft_frac": 0.0,
1873
+ "batch_soft_sft_frac": 0.0,
1874
+ "batch_tf_frac": 0.0,
1875
+ "ce_loss": 0.37429177601145514,
1876
+ "epoch": 1.6016,
1877
+ "grad_norm": 0.58203125,
1878
+ "kd_loss": 0.5036597276406856,
1879
+ "learning_rate": 3e-06,
1880
+ "loss": 0.6491,
1881
+ "masked_tokens": 95.3125,
1882
+ "mean_t": 0.4437690361432033,
1883
+ "step": 750,
1884
+ "student_masked_tokens": 95.3125
1885
+ },
1886
+ {
1887
+ "avg_mask_ratio": 0.46706983938929625,
1888
+ "avg_response_length": 216.0625,
1889
+ "avg_student_mask_ratio": 0.46706983938929625,
1890
+ "batch_ainp_frac": 0.0,
1891
+ "batch_inp_frac": 1.0,
1892
+ "batch_inp_oh_frac": 0.0,
1893
+ "batch_inp_par_frac": 0.0,
1894
+ "batch_inp_par_par_frac": 0.0,
1895
+ "batch_inp_par_reverse_frac": 0.0,
1896
+ "batch_rl_frac": 0.0,
1897
+ "batch_sft_frac": 0.0,
1898
+ "batch_soft_sft_frac": 0.0,
1899
+ "batch_tf_frac": 0.0,
1900
+ "ce_loss": 0.4449058656399984,
1901
+ "epoch": 1.6229333333333333,
1902
+ "grad_norm": 0.8203125,
1903
+ "kd_loss": 0.5661326096985168,
1904
+ "learning_rate": 3e-06,
1905
+ "loss": 0.7233,
1906
+ "masked_tokens": 107.7,
1907
+ "mean_t": 0.49132869170280175,
1908
+ "step": 760,
1909
+ "student_masked_tokens": 107.7
1910
+ },
1911
+ {
1912
+ "avg_mask_ratio": 0.44156218122225255,
1913
+ "avg_response_length": 259.675,
1914
+ "avg_student_mask_ratio": 0.44156218122225255,
1915
+ "batch_ainp_frac": 0.0,
1916
+ "batch_inp_frac": 1.0,
1917
+ "batch_inp_oh_frac": 0.0,
1918
+ "batch_inp_par_frac": 0.0,
1919
+ "batch_inp_par_par_frac": 0.0,
1920
+ "batch_inp_par_reverse_frac": 0.0,
1921
+ "batch_rl_frac": 0.0,
1922
+ "batch_sft_frac": 0.0,
1923
+ "batch_soft_sft_frac": 0.0,
1924
+ "batch_tf_frac": 0.0,
1925
+ "ce_loss": 0.25899335961771613,
1926
+ "epoch": 1.6442666666666668,
1927
+ "grad_norm": 0.396484375,
1928
+ "kd_loss": 0.4095979654902003,
1929
+ "learning_rate": 3e-06,
1930
+ "loss": 0.5099,
1931
+ "masked_tokens": 117.5,
1932
+ "mean_t": 0.4667695587326307,
1933
+ "step": 770,
1934
+ "student_masked_tokens": 117.5
1935
+ },
1936
+ {
1937
+ "avg_mask_ratio": 0.42836043585848527,
1938
+ "avg_response_length": 258.5125,
1939
+ "avg_student_mask_ratio": 0.42836043585848527,
1940
+ "batch_ainp_frac": 0.0,
1941
+ "batch_inp_frac": 1.0,
1942
+ "batch_inp_oh_frac": 0.0,
1943
+ "batch_inp_par_frac": 0.0,
1944
+ "batch_inp_par_par_frac": 0.0,
1945
+ "batch_inp_par_reverse_frac": 0.0,
1946
+ "batch_rl_frac": 0.0,
1947
+ "batch_sft_frac": 0.0,
1948
+ "batch_soft_sft_frac": 0.0,
1949
+ "batch_tf_frac": 0.0,
1950
+ "ce_loss": 0.2897560694203321,
1951
+ "epoch": 1.6656,
1952
+ "grad_norm": 0.2431640625,
1953
+ "kd_loss": 0.34635278815572546,
1954
+ "learning_rate": 3e-06,
1955
+ "loss": 0.4802,
1956
+ "masked_tokens": 119.0125,
1957
+ "mean_t": 0.44942845597106496,
1958
+ "step": 780,
1959
+ "student_masked_tokens": 119.0125
1960
+ },
1961
+ {
1962
+ "avg_mask_ratio": 0.46589430308085866,
1963
+ "avg_response_length": 222.3125,
1964
+ "avg_student_mask_ratio": 0.46589430308085866,
1965
+ "batch_ainp_frac": 0.0,
1966
+ "batch_inp_frac": 1.0,
1967
+ "batch_inp_oh_frac": 0.0,
1968
+ "batch_inp_par_frac": 0.0,
1969
+ "batch_inp_par_par_frac": 0.0,
1970
+ "batch_inp_par_reverse_frac": 0.0,
1971
+ "batch_rl_frac": 0.0,
1972
+ "batch_sft_frac": 0.0,
1973
+ "batch_soft_sft_frac": 0.0,
1974
+ "batch_tf_frac": 0.0,
1975
+ "ce_loss": 0.21603642557238345,
1976
+ "epoch": 1.6869333333333332,
1977
+ "grad_norm": 0.140625,
1978
+ "kd_loss": 0.33674514803767297,
1979
+ "learning_rate": 3e-06,
1980
+ "loss": 0.489,
1981
+ "masked_tokens": 103.25,
1982
+ "mean_t": 0.4993515375303105,
1983
+ "step": 790,
1984
+ "student_masked_tokens": 103.25
1985
+ },
1986
+ {
1987
+ "avg_mask_ratio": 0.46366424662992356,
1988
+ "avg_response_length": 219.6875,
1989
+ "avg_student_mask_ratio": 0.46366424662992356,
1990
+ "batch_ainp_frac": 0.0,
1991
+ "batch_inp_frac": 1.0,
1992
+ "batch_inp_oh_frac": 0.0,
1993
+ "batch_inp_par_frac": 0.0,
1994
+ "batch_inp_par_par_frac": 0.0,
1995
+ "batch_inp_par_reverse_frac": 0.0,
1996
+ "batch_rl_frac": 0.0,
1997
+ "batch_sft_frac": 0.0,
1998
+ "batch_soft_sft_frac": 0.0,
1999
+ "batch_tf_frac": 0.0,
2000
+ "ce_loss": 0.2663005536277069,
2001
+ "epoch": 1.7082666666666668,
2002
+ "grad_norm": 0.23828125,
2003
+ "kd_loss": 0.35138718315538425,
2004
+ "learning_rate": 3e-06,
2005
+ "loss": 0.5434,
2006
+ "masked_tokens": 104.5,
2007
+ "mean_t": 0.500370389316231,
2008
+ "step": 800,
2009
+ "student_masked_tokens": 104.5
2010
+ },
2011
+ {
2012
+ "avg_mask_ratio": 0.503375941584818,
2013
+ "avg_response_length": 237.85,
2014
+ "avg_student_mask_ratio": 0.503375941584818,
2015
+ "batch_ainp_frac": 0.0,
2016
+ "batch_inp_frac": 1.0,
2017
+ "batch_inp_oh_frac": 0.0,
2018
+ "batch_inp_par_frac": 0.0,
2019
+ "batch_inp_par_par_frac": 0.0,
2020
+ "batch_inp_par_reverse_frac": 0.0,
2021
+ "batch_rl_frac": 0.0,
2022
+ "batch_sft_frac": 0.0,
2023
+ "batch_soft_sft_frac": 0.0,
2024
+ "batch_tf_frac": 0.0,
2025
+ "ce_loss": 0.4813590554784753,
2026
+ "epoch": 1.7296,
2027
+ "grad_norm": 1.6015625,
2028
+ "kd_loss": 0.45312339970045057,
2029
+ "learning_rate": 3e-06,
2030
+ "loss": 0.706,
2031
+ "masked_tokens": 118.2125,
2032
+ "mean_t": 0.5317009104182944,
2033
+ "step": 810,
2034
+ "student_masked_tokens": 118.2125
2035
+ },
2036
+ {
2037
+ "avg_mask_ratio": 0.5110091455746442,
2038
+ "avg_response_length": 209.0875,
2039
+ "avg_student_mask_ratio": 0.5110091455746442,
2040
+ "batch_ainp_frac": 0.0,
2041
+ "batch_inp_frac": 1.0,
2042
+ "batch_inp_oh_frac": 0.0,
2043
+ "batch_inp_par_frac": 0.0,
2044
+ "batch_inp_par_par_frac": 0.0,
2045
+ "batch_inp_par_reverse_frac": 0.0,
2046
+ "batch_rl_frac": 0.0,
2047
+ "batch_sft_frac": 0.0,
2048
+ "batch_soft_sft_frac": 0.0,
2049
+ "batch_tf_frac": 0.0,
2050
+ "ce_loss": 0.4535834654417954,
2051
+ "epoch": 1.7509333333333332,
2052
+ "grad_norm": 0.70703125,
2053
+ "kd_loss": 0.5985253949772413,
2054
+ "learning_rate": 3e-06,
2055
+ "loss": 0.7794,
2056
+ "masked_tokens": 120.95,
2057
+ "mean_t": 0.5392061032878701,
2058
+ "step": 820,
2059
+ "student_masked_tokens": 120.95
2060
+ },
2061
+ {
2062
+ "avg_mask_ratio": 0.49899387182667854,
2063
+ "avg_response_length": 263.975,
2064
+ "avg_student_mask_ratio": 0.49899387182667854,
2065
+ "batch_ainp_frac": 0.0,
2066
+ "batch_inp_frac": 1.0,
2067
+ "batch_inp_oh_frac": 0.0,
2068
+ "batch_inp_par_frac": 0.0,
2069
+ "batch_inp_par_par_frac": 0.0,
2070
+ "batch_inp_par_reverse_frac": 0.0,
2071
+ "batch_rl_frac": 0.0,
2072
+ "batch_sft_frac": 0.0,
2073
+ "batch_soft_sft_frac": 0.0,
2074
+ "batch_tf_frac": 0.0,
2075
+ "ce_loss": 0.40083689643704473,
2076
+ "epoch": 1.7722666666666667,
2077
+ "grad_norm": 0.1708984375,
2078
+ "kd_loss": 0.5644028104892641,
2079
+ "learning_rate": 3e-06,
2080
+ "loss": 0.7632,
2081
+ "masked_tokens": 137.075,
2082
+ "mean_t": 0.5238314627087675,
2083
+ "step": 830,
2084
+ "student_masked_tokens": 137.075
2085
+ },
2086
+ {
2087
+ "avg_mask_ratio": 0.4997270987310912,
2088
+ "avg_response_length": 221.9,
2089
+ "avg_student_mask_ratio": 0.4997270987310912,
2090
+ "batch_ainp_frac": 0.0,
2091
+ "batch_inp_frac": 1.0,
2092
+ "batch_inp_oh_frac": 0.0,
2093
+ "batch_inp_par_frac": 0.0,
2094
+ "batch_inp_par_par_frac": 0.0,
2095
+ "batch_inp_par_reverse_frac": 0.0,
2096
+ "batch_rl_frac": 0.0,
2097
+ "batch_sft_frac": 0.0,
2098
+ "batch_soft_sft_frac": 0.0,
2099
+ "batch_tf_frac": 0.0,
2100
+ "ce_loss": 0.2870929398425915,
2101
+ "epoch": 1.7936,
2102
+ "grad_norm": 0.345703125,
2103
+ "kd_loss": 0.4698917509396324,
2104
+ "learning_rate": 3e-06,
2105
+ "loss": 0.6327,
2106
+ "masked_tokens": 114.525,
2107
+ "mean_t": 0.5301066277665086,
2108
+ "step": 840,
2109
+ "student_masked_tokens": 114.525
2110
+ },
2111
+ {
2112
+ "avg_mask_ratio": 0.4988076956477016,
2113
+ "avg_response_length": 225.5,
2114
+ "avg_student_mask_ratio": 0.4988076956477016,
2115
+ "batch_ainp_frac": 0.0,
2116
+ "batch_inp_frac": 1.0,
2117
+ "batch_inp_oh_frac": 0.0,
2118
+ "batch_inp_par_frac": 0.0,
2119
+ "batch_inp_par_par_frac": 0.0,
2120
+ "batch_inp_par_reverse_frac": 0.0,
2121
+ "batch_rl_frac": 0.0,
2122
+ "batch_sft_frac": 0.0,
2123
+ "batch_soft_sft_frac": 0.0,
2124
+ "batch_tf_frac": 0.0,
2125
+ "ce_loss": 0.3023421537889817,
2126
+ "epoch": 1.8149333333333333,
2127
+ "grad_norm": 0.443359375,
2128
+ "kd_loss": 0.3271854338312551,
2129
+ "learning_rate": 3e-06,
2130
+ "loss": 0.5634,
2131
+ "masked_tokens": 116.9125,
2132
+ "mean_t": 0.5343429344706238,
2133
+ "step": 850,
2134
+ "student_masked_tokens": 116.9125
2135
+ },
2136
+ {
2137
+ "avg_mask_ratio": 0.4635998342186213,
2138
+ "avg_response_length": 229.125,
2139
+ "avg_student_mask_ratio": 0.4635998342186213,
2140
+ "batch_ainp_frac": 0.0,
2141
+ "batch_inp_frac": 1.0,
2142
+ "batch_inp_oh_frac": 0.0,
2143
+ "batch_inp_par_frac": 0.0,
2144
+ "batch_inp_par_par_frac": 0.0,
2145
+ "batch_inp_par_reverse_frac": 0.0,
2146
+ "batch_rl_frac": 0.0,
2147
+ "batch_sft_frac": 0.0,
2148
+ "batch_soft_sft_frac": 0.0,
2149
+ "batch_tf_frac": 0.0,
2150
+ "ce_loss": 0.37467331880507115,
2151
+ "epoch": 1.8362666666666667,
2152
+ "grad_norm": 0.384765625,
2153
+ "kd_loss": 0.4431717619034316,
2154
+ "learning_rate": 3e-06,
2155
+ "loss": 0.5956,
2156
+ "masked_tokens": 109.675,
2157
+ "mean_t": 0.4791536889737472,
2158
+ "step": 860,
2159
+ "student_masked_tokens": 109.675
2160
+ },
2161
+ {
2162
+ "avg_mask_ratio": 0.49111039767740294,
2163
+ "avg_response_length": 229.1,
2164
+ "avg_student_mask_ratio": 0.49111039767740294,
2165
+ "batch_ainp_frac": 0.0,
2166
+ "batch_inp_frac": 1.0,
2167
+ "batch_inp_oh_frac": 0.0,
2168
+ "batch_inp_par_frac": 0.0,
2169
+ "batch_inp_par_par_frac": 0.0,
2170
+ "batch_inp_par_reverse_frac": 0.0,
2171
+ "batch_rl_frac": 0.0,
2172
+ "batch_sft_frac": 0.0,
2173
+ "batch_soft_sft_frac": 0.0,
2174
+ "batch_tf_frac": 0.0,
2175
+ "ce_loss": 0.3838037288314126,
2176
+ "epoch": 1.8576000000000001,
2177
+ "grad_norm": 0.333984375,
2178
+ "kd_loss": 0.47523635068355363,
2179
+ "learning_rate": 3e-06,
2180
+ "loss": 0.6859,
2181
+ "masked_tokens": 115.6625,
2182
+ "mean_t": 0.5203817339061061,
2183
+ "step": 870,
2184
+ "student_masked_tokens": 115.6625
2185
+ },
2186
+ {
2187
+ "avg_mask_ratio": 0.4427660425659269,
2188
+ "avg_response_length": 198.5625,
2189
+ "avg_student_mask_ratio": 0.4427660425659269,
2190
+ "batch_ainp_frac": 0.0,
2191
+ "batch_inp_frac": 1.0,
2192
+ "batch_inp_oh_frac": 0.0,
2193
+ "batch_inp_par_frac": 0.0,
2194
+ "batch_inp_par_par_frac": 0.0,
2195
+ "batch_inp_par_reverse_frac": 0.0,
2196
+ "batch_rl_frac": 0.0,
2197
+ "batch_sft_frac": 0.0,
2198
+ "batch_soft_sft_frac": 0.0,
2199
+ "batch_tf_frac": 0.0,
2200
+ "ce_loss": 0.33109274096627817,
2201
+ "epoch": 1.8789333333333333,
2202
+ "grad_norm": 1.0859375,
2203
+ "kd_loss": 0.46695662873548827,
2204
+ "learning_rate": 3e-06,
2205
+ "loss": 0.6284,
2206
+ "masked_tokens": 91.175,
2207
+ "mean_t": 0.4875184997683391,
2208
+ "step": 880,
2209
+ "student_masked_tokens": 91.175
2210
+ },
2211
+ {
2212
+ "avg_mask_ratio": 0.4464349385118112,
2213
+ "avg_response_length": 225.8375,
2214
+ "avg_student_mask_ratio": 0.4464349385118112,
2215
+ "batch_ainp_frac": 0.0,
2216
+ "batch_inp_frac": 1.0,
2217
+ "batch_inp_oh_frac": 0.0,
2218
+ "batch_inp_par_frac": 0.0,
2219
+ "batch_inp_par_par_frac": 0.0,
2220
+ "batch_inp_par_reverse_frac": 0.0,
2221
+ "batch_rl_frac": 0.0,
2222
+ "batch_sft_frac": 0.0,
2223
+ "batch_soft_sft_frac": 0.0,
2224
+ "batch_tf_frac": 0.0,
2225
+ "ce_loss": 0.22229116438190885,
2226
+ "epoch": 1.9002666666666665,
2227
+ "grad_norm": 0.12890625,
2228
+ "kd_loss": 0.4006316699657759,
2229
+ "learning_rate": 3e-06,
2230
+ "loss": 0.4934,
2231
+ "masked_tokens": 101.75,
2232
+ "mean_t": 0.4766692223958671,
2233
+ "step": 890,
2234
+ "student_masked_tokens": 101.75
2235
+ },
2236
+ {
2237
+ "avg_mask_ratio": 0.44976164362160487,
2238
+ "avg_response_length": 227.7875,
2239
+ "avg_student_mask_ratio": 0.44976164362160487,
2240
+ "batch_ainp_frac": 0.0,
2241
+ "batch_inp_frac": 1.0,
2242
+ "batch_inp_oh_frac": 0.0,
2243
+ "batch_inp_par_frac": 0.0,
2244
+ "batch_inp_par_par_frac": 0.0,
2245
+ "batch_inp_par_reverse_frac": 0.0,
2246
+ "batch_rl_frac": 0.0,
2247
+ "batch_sft_frac": 0.0,
2248
+ "batch_soft_sft_frac": 0.0,
2249
+ "batch_tf_frac": 0.0,
2250
+ "ce_loss": 0.38169105723031577,
2251
+ "epoch": 1.9216,
2252
+ "grad_norm": 1.765625,
2253
+ "kd_loss": 0.47280531010078086,
2254
+ "learning_rate": 3e-06,
2255
+ "loss": 0.6337,
2256
+ "masked_tokens": 103.475,
2257
+ "mean_t": 0.487134758150205,
2258
+ "step": 900,
2259
+ "student_masked_tokens": 103.475
2260
+ },
2261
+ {
2262
+ "avg_mask_ratio": 0.475579984736396,
2263
+ "avg_response_length": 245.1625,
2264
+ "avg_student_mask_ratio": 0.475579984736396,
2265
+ "batch_ainp_frac": 0.0,
2266
+ "batch_inp_frac": 1.0,
2267
+ "batch_inp_oh_frac": 0.0,
2268
+ "batch_inp_par_frac": 0.0,
2269
+ "batch_inp_par_par_frac": 0.0,
2270
+ "batch_inp_par_reverse_frac": 0.0,
2271
+ "batch_rl_frac": 0.0,
2272
+ "batch_sft_frac": 0.0,
2273
+ "batch_soft_sft_frac": 0.0,
2274
+ "batch_tf_frac": 0.0,
2275
+ "ce_loss": 0.27549623605577833,
2276
+ "epoch": 1.9429333333333334,
2277
+ "grad_norm": 0.451171875,
2278
+ "kd_loss": 0.4638562942510987,
2279
+ "learning_rate": 3e-06,
2280
+ "loss": 0.5387,
2281
+ "masked_tokens": 124.6375,
2282
+ "mean_t": 0.5027793228859082,
2283
+ "step": 910,
2284
+ "student_masked_tokens": 124.6375
2285
+ },
2286
+ {
2287
+ "avg_mask_ratio": 0.4688875659601763,
2288
+ "avg_response_length": 226.2875,
2289
+ "avg_student_mask_ratio": 0.4688875659601763,
2290
+ "batch_ainp_frac": 0.0,
2291
+ "batch_inp_frac": 1.0,
2292
+ "batch_inp_oh_frac": 0.0,
2293
+ "batch_inp_par_frac": 0.0,
2294
+ "batch_inp_par_par_frac": 0.0,
2295
+ "batch_inp_par_reverse_frac": 0.0,
2296
+ "batch_rl_frac": 0.0,
2297
+ "batch_sft_frac": 0.0,
2298
+ "batch_soft_sft_frac": 0.0,
2299
+ "batch_tf_frac": 0.0,
2300
+ "ce_loss": 0.2772836374151325,
2301
+ "epoch": 1.9642666666666666,
2302
+ "grad_norm": 0.416015625,
2303
+ "kd_loss": 0.44530672791033793,
2304
+ "learning_rate": 3e-06,
2305
+ "loss": 0.6177,
2306
+ "masked_tokens": 110.0125,
2307
+ "mean_t": 0.49417946098838,
2308
+ "step": 920,
2309
+ "student_masked_tokens": 110.0125
2310
+ },
2311
+ {
2312
+ "avg_mask_ratio": 0.47999348094454036,
2313
+ "avg_response_length": 237.05,
2314
+ "avg_student_mask_ratio": 0.47999348094454036,
2315
+ "batch_ainp_frac": 0.0,
2316
+ "batch_inp_frac": 1.0,
2317
+ "batch_inp_oh_frac": 0.0,
2318
+ "batch_inp_par_frac": 0.0,
2319
+ "batch_inp_par_par_frac": 0.0,
2320
+ "batch_inp_par_reverse_frac": 0.0,
2321
+ "batch_rl_frac": 0.0,
2322
+ "batch_sft_frac": 0.0,
2323
+ "batch_soft_sft_frac": 0.0,
2324
+ "batch_tf_frac": 0.0,
2325
+ "ce_loss": 0.2901802834984665,
2326
+ "epoch": 1.9856,
2327
+ "grad_norm": 0.37890625,
2328
+ "kd_loss": 0.4553093938939094,
2329
+ "learning_rate": 3e-06,
2330
+ "loss": 0.5905,
2331
+ "masked_tokens": 121.6,
2332
+ "mean_t": 0.5045580042526125,
2333
+ "step": 930,
2334
+ "student_masked_tokens": 121.6
2335
+ },
2336
+ {
2337
+ "avg_mask_ratio": 0.49413903727240505,
2338
+ "avg_response_length": 224.79761904761904,
2339
+ "avg_student_mask_ratio": 0.49413903727240505,
2340
+ "batch_ainp_frac": 0.0,
2341
+ "batch_inp_frac": 1.0,
2342
+ "batch_inp_oh_frac": 0.0,
2343
+ "batch_inp_par_frac": 0.0,
2344
+ "batch_inp_par_par_frac": 0.0,
2345
+ "batch_inp_par_reverse_frac": 0.0,
2346
+ "batch_rl_frac": 0.0,
2347
+ "batch_sft_frac": 0.0,
2348
+ "batch_soft_sft_frac": 0.0,
2349
+ "batch_tf_frac": 0.0,
2350
+ "ce_loss": 0.37941894131193166,
2351
+ "epoch": 2.0085333333333333,
2352
+ "grad_norm": 0.4921875,
2353
+ "kd_loss": 0.4946319753903075,
2354
+ "learning_rate": 3e-06,
2355
+ "loss": 0.6668,
2356
+ "masked_tokens": 120.5,
2357
+ "mean_t": 0.5321138524893849,
2358
+ "step": 940,
2359
+ "student_masked_tokens": 120.5
2360
+ },
2361
+ {
2362
+ "avg_mask_ratio": 0.4368605303927325,
2363
+ "avg_response_length": 240.9125,
2364
+ "avg_student_mask_ratio": 0.4368605303927325,
2365
+ "batch_ainp_frac": 0.0,
2366
+ "batch_inp_frac": 1.0,
2367
+ "batch_inp_oh_frac": 0.0,
2368
+ "batch_inp_par_frac": 0.0,
2369
+ "batch_inp_par_par_frac": 0.0,
2370
+ "batch_inp_par_reverse_frac": 0.0,
2371
+ "batch_rl_frac": 0.0,
2372
+ "batch_sft_frac": 0.0,
2373
+ "batch_soft_sft_frac": 0.0,
2374
+ "batch_tf_frac": 0.0,
2375
+ "ce_loss": 0.22575005246883392,
2376
+ "epoch": 2.0298666666666665,
2377
+ "grad_norm": 1.1875,
2378
+ "kd_loss": 0.4342805288508771,
2379
+ "learning_rate": 3e-06,
2380
+ "loss": 0.5248,
2381
+ "masked_tokens": 111.4125,
2382
+ "mean_t": 0.4632946296595037,
2383
+ "step": 950,
2384
+ "student_masked_tokens": 111.4125
2385
+ },
2386
+ {
2387
+ "avg_mask_ratio": 0.4988762516761199,
2388
+ "avg_response_length": 275.3,
2389
+ "avg_student_mask_ratio": 0.4988762516761199,
2390
+ "batch_ainp_frac": 0.0,
2391
+ "batch_inp_frac": 1.0,
2392
+ "batch_inp_oh_frac": 0.0,
2393
+ "batch_inp_par_frac": 0.0,
2394
+ "batch_inp_par_par_frac": 0.0,
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": 0.49722497609602667,
2401
+ "epoch": 2.0512,
2402
+ "grad_norm": 0.40625,
2403
+ "kd_loss": 0.5839257182941765,
2404
+ "learning_rate": 3e-06,
2405
+ "loss": 0.7523,
2406
+ "masked_tokens": 143.825,
2407
+ "mean_t": 0.5198000721400604,
2408
+ "step": 960,
2409
+ "student_masked_tokens": 143.825
2410
+ },
2411
+ {
2412
+ "avg_mask_ratio": 0.437801384011982,
2413
+ "avg_response_length": 236.2375,
2414
+ "avg_student_mask_ratio": 0.437801384011982,
2415
+ "batch_ainp_frac": 0.0,
2416
+ "batch_inp_frac": 1.0,
2417
+ "batch_inp_oh_frac": 0.0,
2418
+ "batch_inp_par_frac": 0.0,
2419
+ "batch_inp_par_par_frac": 0.0,
2420
+ "batch_inp_par_reverse_frac": 0.0,
2421
+ "batch_rl_frac": 0.0,
2422
+ "batch_sft_frac": 0.0,
2423
+ "batch_soft_sft_frac": 0.0,
2424
+ "batch_tf_frac": 0.0,
2425
+ "ce_loss": 0.2855980422358698,
2426
+ "epoch": 2.0725333333333333,
2427
+ "grad_norm": 0.4765625,
2428
+ "kd_loss": 0.35673561348757377,
2429
+ "learning_rate": 3e-06,
2430
+ "loss": 0.538,
2431
+ "masked_tokens": 107.025,
2432
+ "mean_t": 0.4703940597362816,
2433
+ "step": 970,
2434
+ "student_masked_tokens": 107.025
2435
+ },
2436
+ {
2437
+ "avg_mask_ratio": 0.42220073882490394,
2438
+ "avg_response_length": 230.8625,
2439
+ "avg_student_mask_ratio": 0.42220073882490394,
2440
+ "batch_ainp_frac": 0.0,
2441
+ "batch_inp_frac": 1.0,
2442
+ "batch_inp_oh_frac": 0.0,
2443
+ "batch_inp_par_frac": 0.0,
2444
+ "batch_inp_par_par_frac": 0.0,
2445
+ "batch_inp_par_reverse_frac": 0.0,
2446
+ "batch_rl_frac": 0.0,
2447
+ "batch_sft_frac": 0.0,
2448
+ "batch_soft_sft_frac": 0.0,
2449
+ "batch_tf_frac": 0.0,
2450
+ "ce_loss": 0.2906558813129777,
2451
+ "epoch": 2.0938666666666665,
2452
+ "grad_norm": 0.466796875,
2453
+ "kd_loss": 0.36284122784349504,
2454
+ "learning_rate": 3e-06,
2455
+ "loss": 0.4889,
2456
+ "masked_tokens": 97.7875,
2457
+ "mean_t": 0.4511947895749472,
2458
+ "step": 980,
2459
+ "student_masked_tokens": 97.7875
2460
+ },
2461
+ {
2462
+ "avg_mask_ratio": 0.4605769342277199,
2463
+ "avg_response_length": 262.0375,
2464
+ "avg_student_mask_ratio": 0.4605769342277199,
2465
+ "batch_ainp_frac": 0.0,
2466
+ "batch_inp_frac": 1.0,
2467
+ "batch_inp_oh_frac": 0.0,
2468
+ "batch_inp_par_frac": 0.0,
2469
+ "batch_inp_par_par_frac": 0.0,
2470
+ "batch_inp_par_reverse_frac": 0.0,
2471
+ "batch_rl_frac": 0.0,
2472
+ "batch_sft_frac": 0.0,
2473
+ "batch_soft_sft_frac": 0.0,
2474
+ "batch_tf_frac": 0.0,
2475
+ "ce_loss": 0.18629460762591635,
2476
+ "epoch": 2.1152,
2477
+ "grad_norm": 0.625,
2478
+ "kd_loss": 0.4187604939788798,
2479
+ "learning_rate": 3e-06,
2480
+ "loss": 0.5063,
2481
+ "masked_tokens": 122.0,
2482
+ "mean_t": 0.4923786667350214,
2483
+ "step": 990,
2484
+ "student_masked_tokens": 122.0
2485
+ },
2486
+ {
2487
+ "avg_mask_ratio": 0.4547682981239632,
2488
+ "avg_response_length": 215.3,
2489
+ "avg_student_mask_ratio": 0.4547682981239632,
2490
+ "batch_ainp_frac": 0.0,
2491
+ "batch_inp_frac": 1.0,
2492
+ "batch_inp_oh_frac": 0.0,
2493
+ "batch_inp_par_frac": 0.0,
2494
+ "batch_inp_par_par_frac": 0.0,
2495
+ "batch_inp_par_reverse_frac": 0.0,
2496
+ "batch_rl_frac": 0.0,
2497
+ "batch_sft_frac": 0.0,
2498
+ "batch_soft_sft_frac": 0.0,
2499
+ "batch_tf_frac": 0.0,
2500
+ "ce_loss": 0.26735156250199454,
2501
+ "epoch": 2.1365333333333334,
2502
+ "grad_norm": 0.26953125,
2503
+ "kd_loss": 0.3440752963605235,
2504
+ "learning_rate": 3e-06,
2505
+ "loss": 0.5169,
2506
+ "masked_tokens": 100.775,
2507
+ "mean_t": 0.4773523230338469,
2508
+ "step": 1000,
2509
+ "student_masked_tokens": 100.775
2510
+ },
2511
+ {
2512
+ "avg_mask_ratio": 0.43540415074676275,
2513
+ "avg_response_length": 215.0,
2514
+ "avg_student_mask_ratio": 0.43540415074676275,
2515
+ "batch_ainp_frac": 0.0,
2516
+ "batch_inp_frac": 1.0,
2517
+ "batch_inp_oh_frac": 0.0,
2518
+ "batch_inp_par_frac": 0.0,
2519
+ "batch_inp_par_par_frac": 0.0,
2520
+ "batch_inp_par_reverse_frac": 0.0,
2521
+ "batch_rl_frac": 0.0,
2522
+ "batch_sft_frac": 0.0,
2523
+ "batch_soft_sft_frac": 0.0,
2524
+ "batch_tf_frac": 0.0,
2525
+ "ce_loss": 0.28448459618934974,
2526
+ "epoch": 2.1578666666666666,
2527
+ "grad_norm": 0.2216796875,
2528
+ "kd_loss": 0.36393369872412384,
2529
+ "learning_rate": 3e-06,
2530
+ "loss": 0.503,
2531
+ "masked_tokens": 88.65,
2532
+ "mean_t": 0.4648138735938119,
2533
+ "step": 1010,
2534
+ "student_masked_tokens": 88.65
2535
+ },
2536
+ {
2537
+ "avg_mask_ratio": 0.5063220548443497,
2538
+ "avg_response_length": 206.9125,
2539
+ "avg_student_mask_ratio": 0.5063220548443497,
2540
+ "batch_ainp_frac": 0.0,
2541
+ "batch_inp_frac": 1.0,
2542
+ "batch_inp_oh_frac": 0.0,
2543
+ "batch_inp_par_frac": 0.0,
2544
+ "batch_inp_par_par_frac": 0.0,
2545
+ "batch_inp_par_reverse_frac": 0.0,
2546
+ "batch_rl_frac": 0.0,
2547
+ "batch_sft_frac": 0.0,
2548
+ "batch_soft_sft_frac": 0.0,
2549
+ "batch_tf_frac": 0.0,
2550
+ "ce_loss": 0.3141316814458378,
2551
+ "epoch": 2.1792,
2552
+ "grad_norm": 0.328125,
2553
+ "kd_loss": 0.49756694839059035,
2554
+ "learning_rate": 3e-06,
2555
+ "loss": 0.7133,
2556
+ "masked_tokens": 110.25,
2557
+ "mean_t": 0.5327763411332853,
2558
+ "step": 1020,
2559
+ "student_masked_tokens": 110.25
2560
+ },
2561
+ {
2562
+ "avg_mask_ratio": 0.46985941788880153,
2563
+ "avg_response_length": 220.05,
2564
+ "avg_student_mask_ratio": 0.46985941788880153,
2565
+ "batch_ainp_frac": 0.0,
2566
+ "batch_inp_frac": 1.0,
2567
+ "batch_inp_oh_frac": 0.0,
2568
+ "batch_inp_par_frac": 0.0,
2569
+ "batch_inp_par_par_frac": 0.0,
2570
+ "batch_inp_par_reverse_frac": 0.0,
2571
+ "batch_rl_frac": 0.0,
2572
+ "batch_sft_frac": 0.0,
2573
+ "batch_soft_sft_frac": 0.0,
2574
+ "batch_tf_frac": 0.0,
2575
+ "ce_loss": 0.4267214318231197,
2576
+ "epoch": 2.2005333333333335,
2577
+ "grad_norm": 0.423828125,
2578
+ "kd_loss": 0.4489077641891422,
2579
+ "learning_rate": 3e-06,
2580
+ "loss": 0.6384,
2581
+ "masked_tokens": 104.9,
2582
+ "mean_t": 0.5033508580760099,
2583
+ "step": 1030,
2584
+ "student_masked_tokens": 104.9
2585
+ },
2586
+ {
2587
+ "avg_mask_ratio": 0.49566771630197765,
2588
+ "avg_response_length": 213.7,
2589
+ "avg_student_mask_ratio": 0.49566771630197765,
2590
+ "batch_ainp_frac": 0.0,
2591
+ "batch_inp_frac": 1.0,
2592
+ "batch_inp_oh_frac": 0.0,
2593
+ "batch_inp_par_frac": 0.0,
2594
+ "batch_inp_par_par_frac": 0.0,
2595
+ "batch_inp_par_reverse_frac": 0.0,
2596
+ "batch_rl_frac": 0.0,
2597
+ "batch_sft_frac": 0.0,
2598
+ "batch_soft_sft_frac": 0.0,
2599
+ "batch_tf_frac": 0.0,
2600
+ "ce_loss": 0.2765686680849626,
2601
+ "epoch": 2.2218666666666667,
2602
+ "grad_norm": 0.74609375,
2603
+ "kd_loss": 0.5419906556950081,
2604
+ "learning_rate": 3e-06,
2605
+ "loss": 0.6686,
2606
+ "masked_tokens": 100.35,
2607
+ "mean_t": 0.5349024560535327,
2608
+ "step": 1040,
2609
+ "student_masked_tokens": 100.35
2610
+ },
2611
+ {
2612
+ "avg_mask_ratio": 0.5123252369463444,
2613
+ "avg_response_length": 239.1125,
2614
+ "avg_student_mask_ratio": 0.5123252369463444,
2615
+ "batch_ainp_frac": 0.0,
2616
+ "batch_inp_frac": 1.0,
2617
+ "batch_inp_oh_frac": 0.0,
2618
+ "batch_inp_par_frac": 0.0,
2619
+ "batch_inp_par_par_frac": 0.0,
2620
+ "batch_inp_par_reverse_frac": 0.0,
2621
+ "batch_rl_frac": 0.0,
2622
+ "batch_sft_frac": 0.0,
2623
+ "batch_soft_sft_frac": 0.0,
2624
+ "batch_tf_frac": 0.0,
2625
+ "ce_loss": 0.3092239250220928,
2626
+ "epoch": 2.2432,
2627
+ "grad_norm": 0.412109375,
2628
+ "kd_loss": 0.5601085751741266,
2629
+ "learning_rate": 3e-06,
2630
+ "loss": 0.6416,
2631
+ "masked_tokens": 123.0,
2632
+ "mean_t": 0.5457118917722255,
2633
+ "step": 1050,
2634
+ "student_masked_tokens": 123.0
2635
+ },
2636
+ {
2637
+ "avg_mask_ratio": 0.46218636581033934,
2638
+ "avg_response_length": 273.7875,
2639
+ "avg_student_mask_ratio": 0.46218636581033934,
2640
+ "batch_ainp_frac": 0.0,
2641
+ "batch_inp_frac": 1.0,
2642
+ "batch_inp_oh_frac": 0.0,
2643
+ "batch_inp_par_frac": 0.0,
2644
+ "batch_inp_par_par_frac": 0.0,
2645
+ "batch_inp_par_reverse_frac": 0.0,
2646
+ "batch_rl_frac": 0.0,
2647
+ "batch_sft_frac": 0.0,
2648
+ "batch_soft_sft_frac": 0.0,
2649
+ "batch_tf_frac": 0.0,
2650
+ "ce_loss": 0.33267747830594485,
2651
+ "epoch": 2.2645333333333335,
2652
+ "grad_norm": 0.3984375,
2653
+ "kd_loss": 0.49028674410892564,
2654
+ "learning_rate": 3e-06,
2655
+ "loss": 0.5827,
2656
+ "masked_tokens": 122.125,
2657
+ "mean_t": 0.48194136443780733,
2658
+ "step": 1060,
2659
+ "student_masked_tokens": 122.125
2660
+ },
2661
+ {
2662
+ "avg_mask_ratio": 0.48546303423354403,
2663
+ "avg_response_length": 260.8125,
2664
+ "avg_student_mask_ratio": 0.48546303423354403,
2665
+ "batch_ainp_frac": 0.0,
2666
+ "batch_inp_frac": 1.0,
2667
+ "batch_inp_oh_frac": 0.0,
2668
+ "batch_inp_par_frac": 0.0,
2669
+ "batch_inp_par_par_frac": 0.0,
2670
+ "batch_inp_par_reverse_frac": 0.0,
2671
+ "batch_rl_frac": 0.0,
2672
+ "batch_sft_frac": 0.0,
2673
+ "batch_soft_sft_frac": 0.0,
2674
+ "batch_tf_frac": 0.0,
2675
+ "ce_loss": 0.4495345233380249,
2676
+ "epoch": 2.2858666666666667,
2677
+ "grad_norm": 0.345703125,
2678
+ "kd_loss": 0.422707377332182,
2679
+ "learning_rate": 3e-06,
2680
+ "loss": 0.6038,
2681
+ "masked_tokens": 138.0375,
2682
+ "mean_t": 0.5015889146190602,
2683
+ "step": 1070,
2684
+ "student_masked_tokens": 138.0375
2685
+ },
2686
+ {
2687
+ "avg_mask_ratio": 0.4625250873621553,
2688
+ "avg_response_length": 219.8875,
2689
+ "avg_student_mask_ratio": 0.4625250873621553,
2690
+ "batch_ainp_frac": 0.0,
2691
+ "batch_inp_frac": 1.0,
2692
+ "batch_inp_oh_frac": 0.0,
2693
+ "batch_inp_par_frac": 0.0,
2694
+ "batch_inp_par_par_frac": 0.0,
2695
+ "batch_inp_par_reverse_frac": 0.0,
2696
+ "batch_rl_frac": 0.0,
2697
+ "batch_sft_frac": 0.0,
2698
+ "batch_soft_sft_frac": 0.0,
2699
+ "batch_tf_frac": 0.0,
2700
+ "ce_loss": 0.3075333785695989,
2701
+ "epoch": 2.3072,
2702
+ "grad_norm": 0.087890625,
2703
+ "kd_loss": 0.3667106795892323,
2704
+ "learning_rate": 3e-06,
2705
+ "loss": 0.4549,
2706
+ "masked_tokens": 100.5625,
2707
+ "mean_t": 0.4983203248586506,
2708
+ "step": 1080,
2709
+ "student_masked_tokens": 100.5625
2710
+ },
2711
+ {
2712
+ "avg_mask_ratio": 0.44248262273031286,
2713
+ "avg_response_length": 213.55,
2714
+ "avg_student_mask_ratio": 0.44248262273031286,
2715
+ "batch_ainp_frac": 0.0,
2716
+ "batch_inp_frac": 1.0,
2717
+ "batch_inp_oh_frac": 0.0,
2718
+ "batch_inp_par_frac": 0.0,
2719
+ "batch_inp_par_par_frac": 0.0,
2720
+ "batch_inp_par_reverse_frac": 0.0,
2721
+ "batch_rl_frac": 0.0,
2722
+ "batch_sft_frac": 0.0,
2723
+ "batch_soft_sft_frac": 0.0,
2724
+ "batch_tf_frac": 0.0,
2725
+ "ce_loss": 0.24968633314620092,
2726
+ "epoch": 2.3285333333333336,
2727
+ "grad_norm": 0.11572265625,
2728
+ "kd_loss": 0.3935246549681978,
2729
+ "learning_rate": 3e-06,
2730
+ "loss": 0.5053,
2731
+ "masked_tokens": 91.9125,
2732
+ "mean_t": 0.47094749807147307,
2733
+ "step": 1090,
2734
+ "student_masked_tokens": 91.9125
2735
+ },
2736
+ {
2737
+ "avg_mask_ratio": 0.5204601250356063,
2738
+ "avg_response_length": 246.1125,
2739
+ "avg_student_mask_ratio": 0.5204601250356063,
2740
+ "batch_ainp_frac": 0.0,
2741
+ "batch_inp_frac": 1.0,
2742
+ "batch_inp_oh_frac": 0.0,
2743
+ "batch_inp_par_frac": 0.0,
2744
+ "batch_inp_par_par_frac": 0.0,
2745
+ "batch_inp_par_reverse_frac": 0.0,
2746
+ "batch_rl_frac": 0.0,
2747
+ "batch_sft_frac": 0.0,
2748
+ "batch_soft_sft_frac": 0.0,
2749
+ "batch_tf_frac": 0.0,
2750
+ "ce_loss": 0.40365264619552477,
2751
+ "epoch": 2.3498666666666668,
2752
+ "grad_norm": 0.37109375,
2753
+ "kd_loss": 0.4355207666182196,
2754
+ "learning_rate": 3e-06,
2755
+ "loss": 0.6746,
2756
+ "masked_tokens": 133.1875,
2757
+ "mean_t": 0.5531192034482956,
2758
+ "step": 1100,
2759
+ "student_masked_tokens": 133.1875
2760
+ }
2761
+ ],
2762
+ "logging_steps": 10,
2763
+ "max_steps": 1404,
2764
+ "num_input_tokens_seen": 0,
2765
+ "num_train_epochs": 3,
2766
+ "save_steps": 100,
2767
+ "stateful_callbacks": {
2768
+ "TrainerControl": {
2769
+ "args": {
2770
+ "should_epoch_stop": false,
2771
+ "should_evaluate": false,
2772
+ "should_log": false,
2773
+ "should_save": true,
2774
+ "should_training_stop": false
2775
+ },
2776
+ "attributes": {}
2777
+ }
2778
+ },
2779
+ "total_flos": 0.0,
2780
+ "train_batch_size": 1,
2781
+ "trial_name": null,
2782
+ "trial_params": null
2783
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1100/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89ae704e41a4a62f6ca56789c45ce45887326cd2f8d1e97e398e5ada4a93398c
3
+ size 8312
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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
+ "gate_proj",
28
+ "down_proj",
29
+ "up_proj",
30
+ "o_proj",
31
+ "q_proj",
32
+ "v_proj",
33
+ "k_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb291f85216472941823a02c50c071b544197be99565c1a4cfeb03ab8292d425
3
+ size 2406624648
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2d2d46298a65b9c03cf05313711e1a42dabb2761fb50ad9c342ca544cbbbd5d6
3
+ size 671304442
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c8dcfaaa26518a9158534ae7671344dd035fcb11f3b40b193e3c3bd47123883
3
+ size 14512
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/trainer_state.json ADDED
@@ -0,0 +1,3033 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.4931091487989761,
13
+ "avg_response_length": 225.975,
14
+ "avg_student_mask_ratio": 0.4931091487989761,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 1.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 0.0,
19
+ "batch_inp_par_par_frac": 0.0,
20
+ "batch_inp_par_reverse_frac": 0.0,
21
+ "batch_rl_frac": 0.0,
22
+ "batch_sft_frac": 0.0,
23
+ "batch_soft_sft_frac": 0.0,
24
+ "batch_tf_frac": 0.0,
25
+ "ce_loss": 0.5570551689027979,
26
+ "epoch": 0.021333333333333333,
27
+ "grad_norm": 0.404296875,
28
+ "kd_loss": 0.5375588692116253,
29
+ "learning_rate": 3e-06,
30
+ "loss": 0.8247,
31
+ "masked_tokens": 111.95,
32
+ "mean_t": 0.5145528071501758,
33
+ "step": 10,
34
+ "student_masked_tokens": 111.95
35
+ },
36
+ {
37
+ "avg_mask_ratio": 0.42058031369233506,
38
+ "avg_response_length": 255.2625,
39
+ "avg_student_mask_ratio": 0.42058031369233506,
40
+ "batch_ainp_frac": 0.0,
41
+ "batch_inp_frac": 1.0,
42
+ "batch_inp_oh_frac": 0.0,
43
+ "batch_inp_par_frac": 0.0,
44
+ "batch_inp_par_par_frac": 0.0,
45
+ "batch_inp_par_reverse_frac": 0.0,
46
+ "batch_rl_frac": 0.0,
47
+ "batch_sft_frac": 0.0,
48
+ "batch_soft_sft_frac": 0.0,
49
+ "batch_tf_frac": 0.0,
50
+ "ce_loss": 0.42826092825978324,
51
+ "epoch": 0.042666666666666665,
52
+ "grad_norm": 0.8046875,
53
+ "kd_loss": 0.4450965437417761,
54
+ "learning_rate": 3e-06,
55
+ "loss": 0.5724,
56
+ "masked_tokens": 97.35,
57
+ "mean_t": 0.43874448732240123,
58
+ "step": 20,
59
+ "student_masked_tokens": 97.35
60
+ },
61
+ {
62
+ "avg_mask_ratio": 0.4538542575784959,
63
+ "avg_response_length": 211.7625,
64
+ "avg_student_mask_ratio": 0.4538542575784959,
65
+ "batch_ainp_frac": 0.0,
66
+ "batch_inp_frac": 1.0,
67
+ "batch_inp_oh_frac": 0.0,
68
+ "batch_inp_par_frac": 0.0,
69
+ "batch_inp_par_par_frac": 0.0,
70
+ "batch_inp_par_reverse_frac": 0.0,
71
+ "batch_rl_frac": 0.0,
72
+ "batch_sft_frac": 0.0,
73
+ "batch_soft_sft_frac": 0.0,
74
+ "batch_tf_frac": 0.0,
75
+ "ce_loss": 0.4461815005188782,
76
+ "epoch": 0.064,
77
+ "grad_norm": 0.50390625,
78
+ "kd_loss": 0.5296064364436825,
79
+ "learning_rate": 3e-06,
80
+ "loss": 0.702,
81
+ "masked_tokens": 110.2,
82
+ "mean_t": 0.4803953981841914,
83
+ "step": 30,
84
+ "student_masked_tokens": 110.2
85
+ },
86
+ {
87
+ "avg_mask_ratio": 0.4207469140383182,
88
+ "avg_response_length": 224.125,
89
+ "avg_student_mask_ratio": 0.4207469140383182,
90
+ "batch_ainp_frac": 0.0,
91
+ "batch_inp_frac": 1.0,
92
+ "batch_inp_oh_frac": 0.0,
93
+ "batch_inp_par_frac": 0.0,
94
+ "batch_inp_par_par_frac": 0.0,
95
+ "batch_inp_par_reverse_frac": 0.0,
96
+ "batch_rl_frac": 0.0,
97
+ "batch_sft_frac": 0.0,
98
+ "batch_soft_sft_frac": 0.0,
99
+ "batch_tf_frac": 0.0,
100
+ "ce_loss": 0.38499989152683156,
101
+ "epoch": 0.08533333333333333,
102
+ "grad_norm": 1.671875,
103
+ "kd_loss": 0.33118802310931417,
104
+ "learning_rate": 3e-06,
105
+ "loss": 0.5529,
106
+ "masked_tokens": 98.1625,
107
+ "mean_t": 0.4569831106782658,
108
+ "step": 40,
109
+ "student_masked_tokens": 98.1625
110
+ },
111
+ {
112
+ "avg_mask_ratio": 0.43260439952719026,
113
+ "avg_response_length": 207.125,
114
+ "avg_student_mask_ratio": 0.43260439952719026,
115
+ "batch_ainp_frac": 0.0,
116
+ "batch_inp_frac": 1.0,
117
+ "batch_inp_oh_frac": 0.0,
118
+ "batch_inp_par_frac": 0.0,
119
+ "batch_inp_par_par_frac": 0.0,
120
+ "batch_inp_par_reverse_frac": 0.0,
121
+ "batch_rl_frac": 0.0,
122
+ "batch_sft_frac": 0.0,
123
+ "batch_soft_sft_frac": 0.0,
124
+ "batch_tf_frac": 0.0,
125
+ "ce_loss": 0.5240421466317912,
126
+ "epoch": 0.10666666666666667,
127
+ "grad_norm": 1.6875,
128
+ "kd_loss": 0.4270985169670894,
129
+ "learning_rate": 3e-06,
130
+ "loss": 0.671,
131
+ "masked_tokens": 85.05,
132
+ "mean_t": 0.4612453707959503,
133
+ "step": 50,
134
+ "student_masked_tokens": 85.05
135
+ },
136
+ {
137
+ "avg_mask_ratio": 0.46053453313652426,
138
+ "avg_response_length": 251.0875,
139
+ "avg_student_mask_ratio": 0.46053453313652426,
140
+ "batch_ainp_frac": 0.0,
141
+ "batch_inp_frac": 1.0,
142
+ "batch_inp_oh_frac": 0.0,
143
+ "batch_inp_par_frac": 0.0,
144
+ "batch_inp_par_par_frac": 0.0,
145
+ "batch_inp_par_reverse_frac": 0.0,
146
+ "batch_rl_frac": 0.0,
147
+ "batch_sft_frac": 0.0,
148
+ "batch_soft_sft_frac": 0.0,
149
+ "batch_tf_frac": 0.0,
150
+ "ce_loss": 0.5027546818272185,
151
+ "epoch": 0.128,
152
+ "grad_norm": 0.17578125,
153
+ "kd_loss": 0.3904111967755945,
154
+ "learning_rate": 3e-06,
155
+ "loss": 0.6672,
156
+ "masked_tokens": 120.9,
157
+ "mean_t": 0.48597636765334756,
158
+ "step": 60,
159
+ "student_masked_tokens": 120.9
160
+ },
161
+ {
162
+ "avg_mask_ratio": 0.5112146578729153,
163
+ "avg_response_length": 202.5875,
164
+ "avg_student_mask_ratio": 0.5112146578729153,
165
+ "batch_ainp_frac": 0.0,
166
+ "batch_inp_frac": 1.0,
167
+ "batch_inp_oh_frac": 0.0,
168
+ "batch_inp_par_frac": 0.0,
169
+ "batch_inp_par_par_frac": 0.0,
170
+ "batch_inp_par_reverse_frac": 0.0,
171
+ "batch_rl_frac": 0.0,
172
+ "batch_sft_frac": 0.0,
173
+ "batch_soft_sft_frac": 0.0,
174
+ "batch_tf_frac": 0.0,
175
+ "ce_loss": 0.7753003867959023,
176
+ "epoch": 0.14933333333333335,
177
+ "grad_norm": 0.953125,
178
+ "kd_loss": 0.4415664039527428,
179
+ "learning_rate": 3e-06,
180
+ "loss": 0.856,
181
+ "masked_tokens": 104.5875,
182
+ "mean_t": 0.5459650319069624,
183
+ "step": 70,
184
+ "student_masked_tokens": 104.5875
185
+ },
186
+ {
187
+ "avg_mask_ratio": 0.37548826879356056,
188
+ "avg_response_length": 225.85,
189
+ "avg_student_mask_ratio": 0.37548826879356056,
190
+ "batch_ainp_frac": 0.0,
191
+ "batch_inp_frac": 1.0,
192
+ "batch_inp_oh_frac": 0.0,
193
+ "batch_inp_par_frac": 0.0,
194
+ "batch_inp_par_par_frac": 0.0,
195
+ "batch_inp_par_reverse_frac": 0.0,
196
+ "batch_rl_frac": 0.0,
197
+ "batch_sft_frac": 0.0,
198
+ "batch_soft_sft_frac": 0.0,
199
+ "batch_tf_frac": 0.0,
200
+ "ce_loss": 0.3791731233859082,
201
+ "epoch": 0.17066666666666666,
202
+ "grad_norm": 0.1552734375,
203
+ "kd_loss": 0.31052538527774515,
204
+ "learning_rate": 3e-06,
205
+ "loss": 0.4843,
206
+ "masked_tokens": 85.0625,
207
+ "mean_t": 0.40758824030635876,
208
+ "step": 80,
209
+ "student_masked_tokens": 85.0625
210
+ },
211
+ {
212
+ "avg_mask_ratio": 0.5001560213277116,
213
+ "avg_response_length": 229.75,
214
+ "avg_student_mask_ratio": 0.5001560213277116,
215
+ "batch_ainp_frac": 0.0,
216
+ "batch_inp_frac": 1.0,
217
+ "batch_inp_oh_frac": 0.0,
218
+ "batch_inp_par_frac": 0.0,
219
+ "batch_inp_par_par_frac": 0.0,
220
+ "batch_inp_par_reverse_frac": 0.0,
221
+ "batch_rl_frac": 0.0,
222
+ "batch_sft_frac": 0.0,
223
+ "batch_soft_sft_frac": 0.0,
224
+ "batch_tf_frac": 0.0,
225
+ "ce_loss": 0.6899960007944174,
226
+ "epoch": 0.192,
227
+ "grad_norm": 1.25,
228
+ "kd_loss": 0.5995283465861896,
229
+ "learning_rate": 3e-06,
230
+ "loss": 0.9721,
231
+ "masked_tokens": 107.6625,
232
+ "mean_t": 0.5297661645396147,
233
+ "step": 90,
234
+ "student_masked_tokens": 107.6625
235
+ },
236
+ {
237
+ "avg_mask_ratio": 0.4576045103633078,
238
+ "avg_response_length": 208.0,
239
+ "avg_student_mask_ratio": 0.4576045103633078,
240
+ "batch_ainp_frac": 0.0,
241
+ "batch_inp_frac": 1.0,
242
+ "batch_inp_oh_frac": 0.0,
243
+ "batch_inp_par_frac": 0.0,
244
+ "batch_inp_par_par_frac": 0.0,
245
+ "batch_inp_par_reverse_frac": 0.0,
246
+ "batch_rl_frac": 0.0,
247
+ "batch_sft_frac": 0.0,
248
+ "batch_soft_sft_frac": 0.0,
249
+ "batch_tf_frac": 0.0,
250
+ "ce_loss": 0.41132245859021166,
251
+ "epoch": 0.21333333333333335,
252
+ "grad_norm": 0.64453125,
253
+ "kd_loss": 0.3813956479015957,
254
+ "learning_rate": 3e-06,
255
+ "loss": 0.6635,
256
+ "masked_tokens": 104.1625,
257
+ "mean_t": 0.4886587227345444,
258
+ "step": 100,
259
+ "student_masked_tokens": 104.1625
260
+ },
261
+ {
262
+ "avg_mask_ratio": 0.4877026333590038,
263
+ "avg_response_length": 213.0875,
264
+ "avg_student_mask_ratio": 0.4877026333590038,
265
+ "batch_ainp_frac": 0.0,
266
+ "batch_inp_frac": 1.0,
267
+ "batch_inp_oh_frac": 0.0,
268
+ "batch_inp_par_frac": 0.0,
269
+ "batch_inp_par_par_frac": 0.0,
270
+ "batch_inp_par_reverse_frac": 0.0,
271
+ "batch_rl_frac": 0.0,
272
+ "batch_sft_frac": 0.0,
273
+ "batch_soft_sft_frac": 0.0,
274
+ "batch_tf_frac": 0.0,
275
+ "ce_loss": 0.4612084587922368,
276
+ "epoch": 0.23466666666666666,
277
+ "grad_norm": 0.64453125,
278
+ "kd_loss": 0.5074845846289577,
279
+ "learning_rate": 3e-06,
280
+ "loss": 0.7993,
281
+ "masked_tokens": 102.075,
282
+ "mean_t": 0.5246987929102034,
283
+ "step": 110,
284
+ "student_masked_tokens": 102.075
285
+ },
286
+ {
287
+ "avg_mask_ratio": 0.45146879020612685,
288
+ "avg_response_length": 224.1875,
289
+ "avg_student_mask_ratio": 0.45146879020612685,
290
+ "batch_ainp_frac": 0.0,
291
+ "batch_inp_frac": 1.0,
292
+ "batch_inp_oh_frac": 0.0,
293
+ "batch_inp_par_frac": 0.0,
294
+ "batch_inp_par_par_frac": 0.0,
295
+ "batch_inp_par_reverse_frac": 0.0,
296
+ "batch_rl_frac": 0.0,
297
+ "batch_sft_frac": 0.0,
298
+ "batch_soft_sft_frac": 0.0,
299
+ "batch_tf_frac": 0.0,
300
+ "ce_loss": 0.3276976759495483,
301
+ "epoch": 0.256,
302
+ "grad_norm": 0.30078125,
303
+ "kd_loss": 0.41461311469229256,
304
+ "learning_rate": 3e-06,
305
+ "loss": 0.6088,
306
+ "masked_tokens": 100.525,
307
+ "mean_t": 0.4805434140143916,
308
+ "step": 120,
309
+ "student_masked_tokens": 100.525
310
+ },
311
+ {
312
+ "avg_mask_ratio": 0.4356566035945434,
313
+ "avg_response_length": 202.7,
314
+ "avg_student_mask_ratio": 0.4356566035945434,
315
+ "batch_ainp_frac": 0.0,
316
+ "batch_inp_frac": 1.0,
317
+ "batch_inp_oh_frac": 0.0,
318
+ "batch_inp_par_frac": 0.0,
319
+ "batch_inp_par_par_frac": 0.0,
320
+ "batch_inp_par_reverse_frac": 0.0,
321
+ "batch_rl_frac": 0.0,
322
+ "batch_sft_frac": 0.0,
323
+ "batch_soft_sft_frac": 0.0,
324
+ "batch_tf_frac": 0.0,
325
+ "ce_loss": 0.12710368948505674,
326
+ "epoch": 0.2773333333333333,
327
+ "grad_norm": 0.490234375,
328
+ "kd_loss": 0.23057804748218585,
329
+ "learning_rate": 3e-06,
330
+ "loss": 0.384,
331
+ "masked_tokens": 89.5625,
332
+ "mean_t": 0.47522516988683494,
333
+ "step": 130,
334
+ "student_masked_tokens": 89.5625
335
+ },
336
+ {
337
+ "avg_mask_ratio": 0.49419954856857656,
338
+ "avg_response_length": 255.625,
339
+ "avg_student_mask_ratio": 0.49419954856857656,
340
+ "batch_ainp_frac": 0.0,
341
+ "batch_inp_frac": 1.0,
342
+ "batch_inp_oh_frac": 0.0,
343
+ "batch_inp_par_frac": 0.0,
344
+ "batch_inp_par_par_frac": 0.0,
345
+ "batch_inp_par_reverse_frac": 0.0,
346
+ "batch_rl_frac": 0.0,
347
+ "batch_sft_frac": 0.0,
348
+ "batch_soft_sft_frac": 0.0,
349
+ "batch_tf_frac": 0.0,
350
+ "ce_loss": 0.48596099013025196,
351
+ "epoch": 0.2986666666666667,
352
+ "grad_norm": 0.859375,
353
+ "kd_loss": 0.5025483015746885,
354
+ "learning_rate": 3e-06,
355
+ "loss": 0.7892,
356
+ "masked_tokens": 136.575,
357
+ "mean_t": 0.5204090005659964,
358
+ "step": 140,
359
+ "student_masked_tokens": 136.575
360
+ },
361
+ {
362
+ "avg_mask_ratio": 0.4736677930341102,
363
+ "avg_response_length": 255.375,
364
+ "avg_student_mask_ratio": 0.4736677930341102,
365
+ "batch_ainp_frac": 0.0,
366
+ "batch_inp_frac": 1.0,
367
+ "batch_inp_oh_frac": 0.0,
368
+ "batch_inp_par_frac": 0.0,
369
+ "batch_inp_par_par_frac": 0.0,
370
+ "batch_inp_par_reverse_frac": 0.0,
371
+ "batch_rl_frac": 0.0,
372
+ "batch_sft_frac": 0.0,
373
+ "batch_soft_sft_frac": 0.0,
374
+ "batch_tf_frac": 0.0,
375
+ "ce_loss": 0.5493089448234059,
376
+ "epoch": 0.32,
377
+ "grad_norm": 0.076171875,
378
+ "kd_loss": 0.4892602212316547,
379
+ "learning_rate": 3e-06,
380
+ "loss": 0.7918,
381
+ "masked_tokens": 126.575,
382
+ "mean_t": 0.5012552456930279,
383
+ "step": 150,
384
+ "student_masked_tokens": 126.575
385
+ },
386
+ {
387
+ "avg_mask_ratio": 0.5032523009285796,
388
+ "avg_response_length": 209.325,
389
+ "avg_student_mask_ratio": 0.5032523009285796,
390
+ "batch_ainp_frac": 0.0,
391
+ "batch_inp_frac": 1.0,
392
+ "batch_inp_oh_frac": 0.0,
393
+ "batch_inp_par_frac": 0.0,
394
+ "batch_inp_par_par_frac": 0.0,
395
+ "batch_inp_par_reverse_frac": 0.0,
396
+ "batch_rl_frac": 0.0,
397
+ "batch_sft_frac": 0.0,
398
+ "batch_soft_sft_frac": 0.0,
399
+ "batch_tf_frac": 0.0,
400
+ "ce_loss": 0.5929547422666019,
401
+ "epoch": 0.3413333333333333,
402
+ "grad_norm": 1.171875,
403
+ "kd_loss": 0.44711892502580214,
404
+ "learning_rate": 3e-06,
405
+ "loss": 0.7472,
406
+ "masked_tokens": 99.425,
407
+ "mean_t": 0.5408745193795766,
408
+ "step": 160,
409
+ "student_masked_tokens": 99.425
410
+ },
411
+ {
412
+ "avg_mask_ratio": 0.4806730231270194,
413
+ "avg_response_length": 190.325,
414
+ "avg_student_mask_ratio": 0.4806730231270194,
415
+ "batch_ainp_frac": 0.0,
416
+ "batch_inp_frac": 1.0,
417
+ "batch_inp_oh_frac": 0.0,
418
+ "batch_inp_par_frac": 0.0,
419
+ "batch_inp_par_par_frac": 0.0,
420
+ "batch_inp_par_reverse_frac": 0.0,
421
+ "batch_rl_frac": 0.0,
422
+ "batch_sft_frac": 0.0,
423
+ "batch_soft_sft_frac": 0.0,
424
+ "batch_tf_frac": 0.0,
425
+ "ce_loss": 0.3583432949517601,
426
+ "epoch": 0.3626666666666667,
427
+ "grad_norm": 0.6171875,
428
+ "kd_loss": 0.4521343837219092,
429
+ "learning_rate": 3e-06,
430
+ "loss": 0.6358,
431
+ "masked_tokens": 89.825,
432
+ "mean_t": 0.5134547733236104,
433
+ "step": 170,
434
+ "student_masked_tokens": 89.825
435
+ },
436
+ {
437
+ "avg_mask_ratio": 0.45829249716189224,
438
+ "avg_response_length": 244.0,
439
+ "avg_student_mask_ratio": 0.45829249716189224,
440
+ "batch_ainp_frac": 0.0,
441
+ "batch_inp_frac": 1.0,
442
+ "batch_inp_oh_frac": 0.0,
443
+ "batch_inp_par_frac": 0.0,
444
+ "batch_inp_par_par_frac": 0.0,
445
+ "batch_inp_par_reverse_frac": 0.0,
446
+ "batch_rl_frac": 0.0,
447
+ "batch_sft_frac": 0.0,
448
+ "batch_soft_sft_frac": 0.0,
449
+ "batch_tf_frac": 0.0,
450
+ "ce_loss": 0.3746713957985094,
451
+ "epoch": 0.384,
452
+ "grad_norm": 0.49609375,
453
+ "kd_loss": 0.34934306121722897,
454
+ "learning_rate": 3e-06,
455
+ "loss": 0.5747,
456
+ "masked_tokens": 110.075,
457
+ "mean_t": 0.48226988823735156,
458
+ "step": 180,
459
+ "student_masked_tokens": 110.075
460
+ },
461
+ {
462
+ "avg_mask_ratio": 0.4842760307248682,
463
+ "avg_response_length": 233.675,
464
+ "avg_student_mask_ratio": 0.4842760307248682,
465
+ "batch_ainp_frac": 0.0,
466
+ "batch_inp_frac": 1.0,
467
+ "batch_inp_oh_frac": 0.0,
468
+ "batch_inp_par_frac": 0.0,
469
+ "batch_inp_par_par_frac": 0.0,
470
+ "batch_inp_par_reverse_frac": 0.0,
471
+ "batch_rl_frac": 0.0,
472
+ "batch_sft_frac": 0.0,
473
+ "batch_soft_sft_frac": 0.0,
474
+ "batch_tf_frac": 0.0,
475
+ "ce_loss": 0.5090123614077584,
476
+ "epoch": 0.4053333333333333,
477
+ "grad_norm": 1.6171875,
478
+ "kd_loss": 0.43204482231294605,
479
+ "learning_rate": 3e-06,
480
+ "loss": 0.7055,
481
+ "masked_tokens": 109.5875,
482
+ "mean_t": 0.5165087037021294,
483
+ "step": 190,
484
+ "student_masked_tokens": 109.5875
485
+ },
486
+ {
487
+ "avg_mask_ratio": 0.4665210062637925,
488
+ "avg_response_length": 197.8,
489
+ "avg_student_mask_ratio": 0.4665210062637925,
490
+ "batch_ainp_frac": 0.0,
491
+ "batch_inp_frac": 1.0,
492
+ "batch_inp_oh_frac": 0.0,
493
+ "batch_inp_par_frac": 0.0,
494
+ "batch_inp_par_par_frac": 0.0,
495
+ "batch_inp_par_reverse_frac": 0.0,
496
+ "batch_rl_frac": 0.0,
497
+ "batch_sft_frac": 0.0,
498
+ "batch_soft_sft_frac": 0.0,
499
+ "batch_tf_frac": 0.0,
500
+ "ce_loss": 0.27265903051802526,
501
+ "epoch": 0.4266666666666667,
502
+ "grad_norm": 0.318359375,
503
+ "kd_loss": 0.357759011555504,
504
+ "learning_rate": 3e-06,
505
+ "loss": 0.5013,
506
+ "masked_tokens": 97.0125,
507
+ "mean_t": 0.5073627714533359,
508
+ "step": 200,
509
+ "student_masked_tokens": 97.0125
510
+ },
511
+ {
512
+ "avg_mask_ratio": 0.45183838629163803,
513
+ "avg_response_length": 213.7875,
514
+ "avg_student_mask_ratio": 0.45183838629163803,
515
+ "batch_ainp_frac": 0.0,
516
+ "batch_inp_frac": 1.0,
517
+ "batch_inp_oh_frac": 0.0,
518
+ "batch_inp_par_frac": 0.0,
519
+ "batch_inp_par_par_frac": 0.0,
520
+ "batch_inp_par_reverse_frac": 0.0,
521
+ "batch_rl_frac": 0.0,
522
+ "batch_sft_frac": 0.0,
523
+ "batch_soft_sft_frac": 0.0,
524
+ "batch_tf_frac": 0.0,
525
+ "ce_loss": 0.21031314329709402,
526
+ "epoch": 0.448,
527
+ "grad_norm": 0.8359375,
528
+ "kd_loss": 0.38585986606940426,
529
+ "learning_rate": 3e-06,
530
+ "loss": 0.5103,
531
+ "masked_tokens": 96.0125,
532
+ "mean_t": 0.48430291628465055,
533
+ "step": 210,
534
+ "student_masked_tokens": 96.0125
535
+ },
536
+ {
537
+ "avg_mask_ratio": 0.5396152360364794,
538
+ "avg_response_length": 236.5125,
539
+ "avg_student_mask_ratio": 0.5396152360364794,
540
+ "batch_ainp_frac": 0.0,
541
+ "batch_inp_frac": 1.0,
542
+ "batch_inp_oh_frac": 0.0,
543
+ "batch_inp_par_frac": 0.0,
544
+ "batch_inp_par_par_frac": 0.0,
545
+ "batch_inp_par_reverse_frac": 0.0,
546
+ "batch_rl_frac": 0.0,
547
+ "batch_sft_frac": 0.0,
548
+ "batch_soft_sft_frac": 0.0,
549
+ "batch_tf_frac": 0.0,
550
+ "ce_loss": 0.6178499971098063,
551
+ "epoch": 0.4693333333333333,
552
+ "grad_norm": 0.96875,
553
+ "kd_loss": 0.46674597742967305,
554
+ "learning_rate": 3e-06,
555
+ "loss": 0.8495,
556
+ "masked_tokens": 125.35,
557
+ "mean_t": 0.570199209311977,
558
+ "step": 220,
559
+ "student_masked_tokens": 125.35
560
+ },
561
+ {
562
+ "avg_mask_ratio": 0.4403991688624956,
563
+ "avg_response_length": 252.7,
564
+ "avg_student_mask_ratio": 0.4403991688624956,
565
+ "batch_ainp_frac": 0.0,
566
+ "batch_inp_frac": 1.0,
567
+ "batch_inp_oh_frac": 0.0,
568
+ "batch_inp_par_frac": 0.0,
569
+ "batch_inp_par_par_frac": 0.0,
570
+ "batch_inp_par_reverse_frac": 0.0,
571
+ "batch_rl_frac": 0.0,
572
+ "batch_sft_frac": 0.0,
573
+ "batch_soft_sft_frac": 0.0,
574
+ "batch_tf_frac": 0.0,
575
+ "ce_loss": 0.25455478885055527,
576
+ "epoch": 0.49066666666666664,
577
+ "grad_norm": 0.5703125,
578
+ "kd_loss": 0.43305868929596725,
579
+ "learning_rate": 3e-06,
580
+ "loss": 0.6428,
581
+ "masked_tokens": 107.325,
582
+ "mean_t": 0.46891279935371133,
583
+ "step": 230,
584
+ "student_masked_tokens": 107.325
585
+ },
586
+ {
587
+ "avg_mask_ratio": 0.47419991258066146,
588
+ "avg_response_length": 212.85,
589
+ "avg_student_mask_ratio": 0.47419991258066146,
590
+ "batch_ainp_frac": 0.0,
591
+ "batch_inp_frac": 1.0,
592
+ "batch_inp_oh_frac": 0.0,
593
+ "batch_inp_par_frac": 0.0,
594
+ "batch_inp_par_par_frac": 0.0,
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": 0.32057130943685763,
601
+ "epoch": 0.512,
602
+ "grad_norm": 0.43359375,
603
+ "kd_loss": 0.5083060303753086,
604
+ "learning_rate": 3e-06,
605
+ "loss": 0.6986,
606
+ "masked_tokens": 106.9,
607
+ "mean_t": 0.502228345896583,
608
+ "step": 240,
609
+ "student_masked_tokens": 106.9
610
+ },
611
+ {
612
+ "avg_mask_ratio": 0.4464209079160355,
613
+ "avg_response_length": 243.475,
614
+ "avg_student_mask_ratio": 0.4464209079160355,
615
+ "batch_ainp_frac": 0.0,
616
+ "batch_inp_frac": 1.0,
617
+ "batch_inp_oh_frac": 0.0,
618
+ "batch_inp_par_frac": 0.0,
619
+ "batch_inp_par_par_frac": 0.0,
620
+ "batch_inp_par_reverse_frac": 0.0,
621
+ "batch_rl_frac": 0.0,
622
+ "batch_sft_frac": 0.0,
623
+ "batch_soft_sft_frac": 0.0,
624
+ "batch_tf_frac": 0.0,
625
+ "ce_loss": 0.33636454603331456,
626
+ "epoch": 0.5333333333333333,
627
+ "grad_norm": 0.1142578125,
628
+ "kd_loss": 0.41649795620701296,
629
+ "learning_rate": 3e-06,
630
+ "loss": 0.5666,
631
+ "masked_tokens": 112.7375,
632
+ "mean_t": 0.4733429416548461,
633
+ "step": 250,
634
+ "student_masked_tokens": 112.7375
635
+ },
636
+ {
637
+ "avg_mask_ratio": 0.4520751796895638,
638
+ "avg_response_length": 245.55,
639
+ "avg_student_mask_ratio": 0.4520751796895638,
640
+ "batch_ainp_frac": 0.0,
641
+ "batch_inp_frac": 1.0,
642
+ "batch_inp_oh_frac": 0.0,
643
+ "batch_inp_par_frac": 0.0,
644
+ "batch_inp_par_par_frac": 0.0,
645
+ "batch_inp_par_reverse_frac": 0.0,
646
+ "batch_rl_frac": 0.0,
647
+ "batch_sft_frac": 0.0,
648
+ "batch_soft_sft_frac": 0.0,
649
+ "batch_tf_frac": 0.0,
650
+ "ce_loss": 0.37478437887749577,
651
+ "epoch": 0.5546666666666666,
652
+ "grad_norm": 0.328125,
653
+ "kd_loss": 0.31532439299670545,
654
+ "learning_rate": 3e-06,
655
+ "loss": 0.5129,
656
+ "masked_tokens": 109.6375,
657
+ "mean_t": 0.4843149524240289,
658
+ "step": 260,
659
+ "student_masked_tokens": 109.6375
660
+ },
661
+ {
662
+ "avg_mask_ratio": 0.5305180630879477,
663
+ "avg_response_length": 224.45,
664
+ "avg_student_mask_ratio": 0.5305180630879477,
665
+ "batch_ainp_frac": 0.0,
666
+ "batch_inp_frac": 1.0,
667
+ "batch_inp_oh_frac": 0.0,
668
+ "batch_inp_par_frac": 0.0,
669
+ "batch_inp_par_par_frac": 0.0,
670
+ "batch_inp_par_reverse_frac": 0.0,
671
+ "batch_rl_frac": 0.0,
672
+ "batch_sft_frac": 0.0,
673
+ "batch_soft_sft_frac": 0.0,
674
+ "batch_tf_frac": 0.0,
675
+ "ce_loss": 0.42709534656005416,
676
+ "epoch": 0.576,
677
+ "grad_norm": 0.7578125,
678
+ "kd_loss": 0.5525495689224045,
679
+ "learning_rate": 3e-06,
680
+ "loss": 0.812,
681
+ "masked_tokens": 120.475,
682
+ "mean_t": 0.5643589949700981,
683
+ "step": 270,
684
+ "student_masked_tokens": 120.475
685
+ },
686
+ {
687
+ "avg_mask_ratio": 0.46451686368091033,
688
+ "avg_response_length": 254.825,
689
+ "avg_student_mask_ratio": 0.46451686368091033,
690
+ "batch_ainp_frac": 0.0,
691
+ "batch_inp_frac": 1.0,
692
+ "batch_inp_oh_frac": 0.0,
693
+ "batch_inp_par_frac": 0.0,
694
+ "batch_inp_par_par_frac": 0.0,
695
+ "batch_inp_par_reverse_frac": 0.0,
696
+ "batch_rl_frac": 0.0,
697
+ "batch_sft_frac": 0.0,
698
+ "batch_soft_sft_frac": 0.0,
699
+ "batch_tf_frac": 0.0,
700
+ "ce_loss": 0.31382316479499084,
701
+ "epoch": 0.5973333333333334,
702
+ "grad_norm": 0.90234375,
703
+ "kd_loss": 0.3957495673693458,
704
+ "learning_rate": 3e-06,
705
+ "loss": 0.6028,
706
+ "masked_tokens": 129.225,
707
+ "mean_t": 0.47818811538163575,
708
+ "step": 280,
709
+ "student_masked_tokens": 129.225
710
+ },
711
+ {
712
+ "avg_mask_ratio": 0.389662017847877,
713
+ "avg_response_length": 245.9125,
714
+ "avg_student_mask_ratio": 0.389662017847877,
715
+ "batch_ainp_frac": 0.0,
716
+ "batch_inp_frac": 1.0,
717
+ "batch_inp_oh_frac": 0.0,
718
+ "batch_inp_par_frac": 0.0,
719
+ "batch_inp_par_par_frac": 0.0,
720
+ "batch_inp_par_reverse_frac": 0.0,
721
+ "batch_rl_frac": 0.0,
722
+ "batch_sft_frac": 0.0,
723
+ "batch_soft_sft_frac": 0.0,
724
+ "batch_tf_frac": 0.0,
725
+ "ce_loss": 0.23645576389110373,
726
+ "epoch": 0.6186666666666667,
727
+ "grad_norm": 0.302734375,
728
+ "kd_loss": 0.27728830450374853,
729
+ "learning_rate": 3e-06,
730
+ "loss": 0.4314,
731
+ "masked_tokens": 99.8625,
732
+ "mean_t": 0.4088635521940887,
733
+ "step": 290,
734
+ "student_masked_tokens": 99.8625
735
+ },
736
+ {
737
+ "avg_mask_ratio": 0.44417000194080175,
738
+ "avg_response_length": 217.0375,
739
+ "avg_student_mask_ratio": 0.44417000194080175,
740
+ "batch_ainp_frac": 0.0,
741
+ "batch_inp_frac": 1.0,
742
+ "batch_inp_oh_frac": 0.0,
743
+ "batch_inp_par_frac": 0.0,
744
+ "batch_inp_par_par_frac": 0.0,
745
+ "batch_inp_par_reverse_frac": 0.0,
746
+ "batch_rl_frac": 0.0,
747
+ "batch_sft_frac": 0.0,
748
+ "batch_soft_sft_frac": 0.0,
749
+ "batch_tf_frac": 0.0,
750
+ "ce_loss": 0.3240562055096575,
751
+ "epoch": 0.64,
752
+ "grad_norm": 1.09375,
753
+ "kd_loss": 0.31930388437995133,
754
+ "learning_rate": 3e-06,
755
+ "loss": 0.5264,
756
+ "masked_tokens": 104.625,
757
+ "mean_t": 0.47984200695063917,
758
+ "step": 300,
759
+ "student_masked_tokens": 104.625
760
+ },
761
+ {
762
+ "avg_mask_ratio": 0.4706685543409549,
763
+ "avg_response_length": 175.45,
764
+ "avg_student_mask_ratio": 0.4706685543409549,
765
+ "batch_ainp_frac": 0.0,
766
+ "batch_inp_frac": 1.0,
767
+ "batch_inp_oh_frac": 0.0,
768
+ "batch_inp_par_frac": 0.0,
769
+ "batch_inp_par_par_frac": 0.0,
770
+ "batch_inp_par_reverse_frac": 0.0,
771
+ "batch_rl_frac": 0.0,
772
+ "batch_sft_frac": 0.0,
773
+ "batch_soft_sft_frac": 0.0,
774
+ "batch_tf_frac": 0.0,
775
+ "ce_loss": 0.34333510200582396,
776
+ "epoch": 0.6613333333333333,
777
+ "grad_norm": 1.234375,
778
+ "kd_loss": 0.5067149527083984,
779
+ "learning_rate": 3e-06,
780
+ "loss": 0.6534,
781
+ "masked_tokens": 84.875,
782
+ "mean_t": 0.5026606284547597,
783
+ "step": 310,
784
+ "student_masked_tokens": 84.875
785
+ },
786
+ {
787
+ "avg_mask_ratio": 0.4974605386145413,
788
+ "avg_response_length": 234.7875,
789
+ "avg_student_mask_ratio": 0.4974605386145413,
790
+ "batch_ainp_frac": 0.0,
791
+ "batch_inp_frac": 1.0,
792
+ "batch_inp_oh_frac": 0.0,
793
+ "batch_inp_par_frac": 0.0,
794
+ "batch_inp_par_par_frac": 0.0,
795
+ "batch_inp_par_reverse_frac": 0.0,
796
+ "batch_rl_frac": 0.0,
797
+ "batch_sft_frac": 0.0,
798
+ "batch_soft_sft_frac": 0.0,
799
+ "batch_tf_frac": 0.0,
800
+ "ce_loss": 0.34462752127872137,
801
+ "epoch": 0.6826666666666666,
802
+ "grad_norm": 0.333984375,
803
+ "kd_loss": 0.3942846609736307,
804
+ "learning_rate": 3e-06,
805
+ "loss": 0.7133,
806
+ "masked_tokens": 119.6,
807
+ "mean_t": 0.5293499688967132,
808
+ "step": 320,
809
+ "student_masked_tokens": 119.6
810
+ },
811
+ {
812
+ "avg_mask_ratio": 0.5112370474264025,
813
+ "avg_response_length": 236.0625,
814
+ "avg_student_mask_ratio": 0.5112370474264025,
815
+ "batch_ainp_frac": 0.0,
816
+ "batch_inp_frac": 1.0,
817
+ "batch_inp_oh_frac": 0.0,
818
+ "batch_inp_par_frac": 0.0,
819
+ "batch_inp_par_par_frac": 0.0,
820
+ "batch_inp_par_reverse_frac": 0.0,
821
+ "batch_rl_frac": 0.0,
822
+ "batch_sft_frac": 0.0,
823
+ "batch_soft_sft_frac": 0.0,
824
+ "batch_tf_frac": 0.0,
825
+ "ce_loss": 0.2974585796398969,
826
+ "epoch": 0.704,
827
+ "grad_norm": 0.44140625,
828
+ "kd_loss": 0.4301003347501496,
829
+ "learning_rate": 3e-06,
830
+ "loss": 0.6754,
831
+ "masked_tokens": 129.425,
832
+ "mean_t": 0.5426030711154454,
833
+ "step": 330,
834
+ "student_masked_tokens": 129.425
835
+ },
836
+ {
837
+ "avg_mask_ratio": 0.44370225080056114,
838
+ "avg_response_length": 241.4875,
839
+ "avg_student_mask_ratio": 0.44370225080056114,
840
+ "batch_ainp_frac": 0.0,
841
+ "batch_inp_frac": 1.0,
842
+ "batch_inp_oh_frac": 0.0,
843
+ "batch_inp_par_frac": 0.0,
844
+ "batch_inp_par_par_frac": 0.0,
845
+ "batch_inp_par_reverse_frac": 0.0,
846
+ "batch_rl_frac": 0.0,
847
+ "batch_sft_frac": 0.0,
848
+ "batch_soft_sft_frac": 0.0,
849
+ "batch_tf_frac": 0.0,
850
+ "ce_loss": 0.3732590021626493,
851
+ "epoch": 0.7253333333333334,
852
+ "grad_norm": 0.98046875,
853
+ "kd_loss": 0.4610515360019235,
854
+ "learning_rate": 3e-06,
855
+ "loss": 0.6627,
856
+ "masked_tokens": 108.775,
857
+ "mean_t": 0.47635243807453664,
858
+ "step": 340,
859
+ "student_masked_tokens": 108.775
860
+ },
861
+ {
862
+ "avg_mask_ratio": 0.49959173843380994,
863
+ "avg_response_length": 235.6375,
864
+ "avg_student_mask_ratio": 0.49959173843380994,
865
+ "batch_ainp_frac": 0.0,
866
+ "batch_inp_frac": 1.0,
867
+ "batch_inp_oh_frac": 0.0,
868
+ "batch_inp_par_frac": 0.0,
869
+ "batch_inp_par_par_frac": 0.0,
870
+ "batch_inp_par_reverse_frac": 0.0,
871
+ "batch_rl_frac": 0.0,
872
+ "batch_sft_frac": 0.0,
873
+ "batch_soft_sft_frac": 0.0,
874
+ "batch_tf_frac": 0.0,
875
+ "ce_loss": 0.48515336151417615,
876
+ "epoch": 0.7466666666666667,
877
+ "grad_norm": 0.92578125,
878
+ "kd_loss": 0.5031771080357654,
879
+ "learning_rate": 3e-06,
880
+ "loss": 0.7668,
881
+ "masked_tokens": 125.625,
882
+ "mean_t": 0.5268881446914747,
883
+ "step": 350,
884
+ "student_masked_tokens": 125.625
885
+ },
886
+ {
887
+ "avg_mask_ratio": 0.4744729608530179,
888
+ "avg_response_length": 246.1625,
889
+ "avg_student_mask_ratio": 0.4744729608530179,
890
+ "batch_ainp_frac": 0.0,
891
+ "batch_inp_frac": 1.0,
892
+ "batch_inp_oh_frac": 0.0,
893
+ "batch_inp_par_frac": 0.0,
894
+ "batch_inp_par_par_frac": 0.0,
895
+ "batch_inp_par_reverse_frac": 0.0,
896
+ "batch_rl_frac": 0.0,
897
+ "batch_sft_frac": 0.0,
898
+ "batch_soft_sft_frac": 0.0,
899
+ "batch_tf_frac": 0.0,
900
+ "ce_loss": 0.3005135279950082,
901
+ "epoch": 0.768,
902
+ "grad_norm": 0.169921875,
903
+ "kd_loss": 0.5216399239409879,
904
+ "learning_rate": 3e-06,
905
+ "loss": 0.6077,
906
+ "masked_tokens": 116.875,
907
+ "mean_t": 0.5040419134311378,
908
+ "step": 360,
909
+ "student_masked_tokens": 116.875
910
+ },
911
+ {
912
+ "avg_mask_ratio": 0.4738045462174341,
913
+ "avg_response_length": 257.575,
914
+ "avg_student_mask_ratio": 0.4738045462174341,
915
+ "batch_ainp_frac": 0.0,
916
+ "batch_inp_frac": 1.0,
917
+ "batch_inp_oh_frac": 0.0,
918
+ "batch_inp_par_frac": 0.0,
919
+ "batch_inp_par_par_frac": 0.0,
920
+ "batch_inp_par_reverse_frac": 0.0,
921
+ "batch_rl_frac": 0.0,
922
+ "batch_sft_frac": 0.0,
923
+ "batch_soft_sft_frac": 0.0,
924
+ "batch_tf_frac": 0.0,
925
+ "ce_loss": 0.5349442186782426,
926
+ "epoch": 0.7893333333333333,
927
+ "grad_norm": 0.201171875,
928
+ "kd_loss": 0.6039233199480805,
929
+ "learning_rate": 3e-06,
930
+ "loss": 0.7196,
931
+ "masked_tokens": 127.4625,
932
+ "mean_t": 0.5127181728370488,
933
+ "step": 370,
934
+ "student_masked_tokens": 127.4625
935
+ },
936
+ {
937
+ "avg_mask_ratio": 0.4512475330149755,
938
+ "avg_response_length": 209.8,
939
+ "avg_student_mask_ratio": 0.4512475330149755,
940
+ "batch_ainp_frac": 0.0,
941
+ "batch_inp_frac": 1.0,
942
+ "batch_inp_oh_frac": 0.0,
943
+ "batch_inp_par_frac": 0.0,
944
+ "batch_inp_par_par_frac": 0.0,
945
+ "batch_inp_par_reverse_frac": 0.0,
946
+ "batch_rl_frac": 0.0,
947
+ "batch_sft_frac": 0.0,
948
+ "batch_soft_sft_frac": 0.0,
949
+ "batch_tf_frac": 0.0,
950
+ "ce_loss": 0.19145508916275275,
951
+ "epoch": 0.8106666666666666,
952
+ "grad_norm": 0.6875,
953
+ "kd_loss": 0.4029755606519984,
954
+ "learning_rate": 3e-06,
955
+ "loss": 0.5055,
956
+ "masked_tokens": 100.8375,
957
+ "mean_t": 0.4825185665744357,
958
+ "step": 380,
959
+ "student_masked_tokens": 100.8375
960
+ },
961
+ {
962
+ "avg_mask_ratio": 0.4752940105390735,
963
+ "avg_response_length": 219.5625,
964
+ "avg_student_mask_ratio": 0.4752940105390735,
965
+ "batch_ainp_frac": 0.0,
966
+ "batch_inp_frac": 1.0,
967
+ "batch_inp_oh_frac": 0.0,
968
+ "batch_inp_par_frac": 0.0,
969
+ "batch_inp_par_par_frac": 0.0,
970
+ "batch_inp_par_reverse_frac": 0.0,
971
+ "batch_rl_frac": 0.0,
972
+ "batch_sft_frac": 0.0,
973
+ "batch_soft_sft_frac": 0.0,
974
+ "batch_tf_frac": 0.0,
975
+ "ce_loss": 0.4267096655552223,
976
+ "epoch": 0.832,
977
+ "grad_norm": 0.2578125,
978
+ "kd_loss": 0.4655849843487971,
979
+ "learning_rate": 3e-06,
980
+ "loss": 0.6749,
981
+ "masked_tokens": 112.375,
982
+ "mean_t": 0.5053101469413377,
983
+ "step": 390,
984
+ "student_masked_tokens": 112.375
985
+ },
986
+ {
987
+ "avg_mask_ratio": 0.47461870914557946,
988
+ "avg_response_length": 242.6125,
989
+ "avg_student_mask_ratio": 0.47461870914557946,
990
+ "batch_ainp_frac": 0.0,
991
+ "batch_inp_frac": 1.0,
992
+ "batch_inp_oh_frac": 0.0,
993
+ "batch_inp_par_frac": 0.0,
994
+ "batch_inp_par_par_frac": 0.0,
995
+ "batch_inp_par_reverse_frac": 0.0,
996
+ "batch_rl_frac": 0.0,
997
+ "batch_sft_frac": 0.0,
998
+ "batch_soft_sft_frac": 0.0,
999
+ "batch_tf_frac": 0.0,
1000
+ "ce_loss": 0.27868834779033025,
1001
+ "epoch": 0.8533333333333334,
1002
+ "grad_norm": 0.640625,
1003
+ "kd_loss": 0.5299579592951205,
1004
+ "learning_rate": 3e-06,
1005
+ "loss": 0.6538,
1006
+ "masked_tokens": 120.4125,
1007
+ "mean_t": 0.5052250675857067,
1008
+ "step": 400,
1009
+ "student_masked_tokens": 120.4125
1010
+ },
1011
+ {
1012
+ "avg_mask_ratio": 0.48321815438685006,
1013
+ "avg_response_length": 228.15,
1014
+ "avg_student_mask_ratio": 0.48321815438685006,
1015
+ "batch_ainp_frac": 0.0,
1016
+ "batch_inp_frac": 1.0,
1017
+ "batch_inp_oh_frac": 0.0,
1018
+ "batch_inp_par_frac": 0.0,
1019
+ "batch_inp_par_par_frac": 0.0,
1020
+ "batch_inp_par_reverse_frac": 0.0,
1021
+ "batch_rl_frac": 0.0,
1022
+ "batch_sft_frac": 0.0,
1023
+ "batch_soft_sft_frac": 0.0,
1024
+ "batch_tf_frac": 0.0,
1025
+ "ce_loss": 0.43057951200541994,
1026
+ "epoch": 0.8746666666666667,
1027
+ "grad_norm": 0.5390625,
1028
+ "kd_loss": 0.504674318619719,
1029
+ "learning_rate": 3e-06,
1030
+ "loss": 0.7381,
1031
+ "masked_tokens": 119.0,
1032
+ "mean_t": 0.5050956419203431,
1033
+ "step": 410,
1034
+ "student_masked_tokens": 119.0
1035
+ },
1036
+ {
1037
+ "avg_mask_ratio": 0.4379329536575824,
1038
+ "avg_response_length": 220.225,
1039
+ "avg_student_mask_ratio": 0.4379329536575824,
1040
+ "batch_ainp_frac": 0.0,
1041
+ "batch_inp_frac": 1.0,
1042
+ "batch_inp_oh_frac": 0.0,
1043
+ "batch_inp_par_frac": 0.0,
1044
+ "batch_inp_par_par_frac": 0.0,
1045
+ "batch_inp_par_reverse_frac": 0.0,
1046
+ "batch_rl_frac": 0.0,
1047
+ "batch_sft_frac": 0.0,
1048
+ "batch_soft_sft_frac": 0.0,
1049
+ "batch_tf_frac": 0.0,
1050
+ "ce_loss": 0.132674143492045,
1051
+ "epoch": 0.896,
1052
+ "grad_norm": 1.09375,
1053
+ "kd_loss": 0.27731474525324984,
1054
+ "learning_rate": 3e-06,
1055
+ "loss": 0.3953,
1056
+ "masked_tokens": 85.525,
1057
+ "mean_t": 0.4769687672611326,
1058
+ "step": 420,
1059
+ "student_masked_tokens": 85.525
1060
+ },
1061
+ {
1062
+ "avg_mask_ratio": 0.4674084897618741,
1063
+ "avg_response_length": 249.2125,
1064
+ "avg_student_mask_ratio": 0.4674084897618741,
1065
+ "batch_ainp_frac": 0.0,
1066
+ "batch_inp_frac": 1.0,
1067
+ "batch_inp_oh_frac": 0.0,
1068
+ "batch_inp_par_frac": 0.0,
1069
+ "batch_inp_par_par_frac": 0.0,
1070
+ "batch_inp_par_reverse_frac": 0.0,
1071
+ "batch_rl_frac": 0.0,
1072
+ "batch_sft_frac": 0.0,
1073
+ "batch_soft_sft_frac": 0.0,
1074
+ "batch_tf_frac": 0.0,
1075
+ "ce_loss": 0.37605725416574387,
1076
+ "epoch": 0.9173333333333333,
1077
+ "grad_norm": 0.43359375,
1078
+ "kd_loss": 0.49442086774362226,
1079
+ "learning_rate": 3e-06,
1080
+ "loss": 0.6699,
1081
+ "masked_tokens": 104.5625,
1082
+ "mean_t": 0.49262027950026094,
1083
+ "step": 430,
1084
+ "student_masked_tokens": 104.5625
1085
+ },
1086
+ {
1087
+ "avg_mask_ratio": 0.4415457699564286,
1088
+ "avg_response_length": 241.0875,
1089
+ "avg_student_mask_ratio": 0.4415457699564286,
1090
+ "batch_ainp_frac": 0.0,
1091
+ "batch_inp_frac": 1.0,
1092
+ "batch_inp_oh_frac": 0.0,
1093
+ "batch_inp_par_frac": 0.0,
1094
+ "batch_inp_par_par_frac": 0.0,
1095
+ "batch_inp_par_reverse_frac": 0.0,
1096
+ "batch_rl_frac": 0.0,
1097
+ "batch_sft_frac": 0.0,
1098
+ "batch_soft_sft_frac": 0.0,
1099
+ "batch_tf_frac": 0.0,
1100
+ "ce_loss": 0.3754083825901603,
1101
+ "epoch": 0.9386666666666666,
1102
+ "grad_norm": 0.6328125,
1103
+ "kd_loss": 0.45159815376919143,
1104
+ "learning_rate": 3e-06,
1105
+ "loss": 0.6585,
1106
+ "masked_tokens": 113.0875,
1107
+ "mean_t": 0.47046207524836064,
1108
+ "step": 440,
1109
+ "student_masked_tokens": 113.0875
1110
+ },
1111
+ {
1112
+ "avg_mask_ratio": 0.42486972180195154,
1113
+ "avg_response_length": 231.9875,
1114
+ "avg_student_mask_ratio": 0.42486972180195154,
1115
+ "batch_ainp_frac": 0.0,
1116
+ "batch_inp_frac": 1.0,
1117
+ "batch_inp_oh_frac": 0.0,
1118
+ "batch_inp_par_frac": 0.0,
1119
+ "batch_inp_par_par_frac": 0.0,
1120
+ "batch_inp_par_reverse_frac": 0.0,
1121
+ "batch_rl_frac": 0.0,
1122
+ "batch_sft_frac": 0.0,
1123
+ "batch_soft_sft_frac": 0.0,
1124
+ "batch_tf_frac": 0.0,
1125
+ "ce_loss": 0.32457938515717616,
1126
+ "epoch": 0.96,
1127
+ "grad_norm": 0.6953125,
1128
+ "kd_loss": 0.4011907008050457,
1129
+ "learning_rate": 3e-06,
1130
+ "loss": 0.5644,
1131
+ "masked_tokens": 103.4,
1132
+ "mean_t": 0.45781184462830427,
1133
+ "step": 450,
1134
+ "student_masked_tokens": 103.4
1135
+ },
1136
+ {
1137
+ "avg_mask_ratio": 0.47578654896933587,
1138
+ "avg_response_length": 214.6125,
1139
+ "avg_student_mask_ratio": 0.47578654896933587,
1140
+ "batch_ainp_frac": 0.0,
1141
+ "batch_inp_frac": 1.0,
1142
+ "batch_inp_oh_frac": 0.0,
1143
+ "batch_inp_par_frac": 0.0,
1144
+ "batch_inp_par_par_frac": 0.0,
1145
+ "batch_inp_par_reverse_frac": 0.0,
1146
+ "batch_rl_frac": 0.0,
1147
+ "batch_sft_frac": 0.0,
1148
+ "batch_soft_sft_frac": 0.0,
1149
+ "batch_tf_frac": 0.0,
1150
+ "ce_loss": 0.32885359905767475,
1151
+ "epoch": 0.9813333333333333,
1152
+ "grad_norm": 0.16015625,
1153
+ "kd_loss": 0.44463847501747294,
1154
+ "learning_rate": 3e-06,
1155
+ "loss": 0.635,
1156
+ "masked_tokens": 105.3125,
1157
+ "mean_t": 0.5075790266972036,
1158
+ "step": 460,
1159
+ "student_masked_tokens": 105.3125
1160
+ },
1161
+ {
1162
+ "avg_mask_ratio": 0.4782901787132557,
1163
+ "avg_response_length": 224.0952380952381,
1164
+ "avg_student_mask_ratio": 0.4782901787132557,
1165
+ "batch_ainp_frac": 0.0,
1166
+ "batch_inp_frac": 1.0,
1167
+ "batch_inp_oh_frac": 0.0,
1168
+ "batch_inp_par_frac": 0.0,
1169
+ "batch_inp_par_par_frac": 0.0,
1170
+ "batch_inp_par_reverse_frac": 0.0,
1171
+ "batch_rl_frac": 0.0,
1172
+ "batch_sft_frac": 0.0,
1173
+ "batch_soft_sft_frac": 0.0,
1174
+ "batch_tf_frac": 0.0,
1175
+ "ce_loss": 0.3393430382851702,
1176
+ "epoch": 1.0042666666666666,
1177
+ "grad_norm": 0.65625,
1178
+ "kd_loss": 0.5178591865708675,
1179
+ "learning_rate": 3e-06,
1180
+ "loss": 0.7769,
1181
+ "masked_tokens": 107.23809523809524,
1182
+ "mean_t": 0.5031429776822084,
1183
+ "step": 470,
1184
+ "student_masked_tokens": 107.23809523809524
1185
+ },
1186
+ {
1187
+ "avg_mask_ratio": 0.47575968883465974,
1188
+ "avg_response_length": 249.4125,
1189
+ "avg_student_mask_ratio": 0.47575968883465974,
1190
+ "batch_ainp_frac": 0.0,
1191
+ "batch_inp_frac": 1.0,
1192
+ "batch_inp_oh_frac": 0.0,
1193
+ "batch_inp_par_frac": 0.0,
1194
+ "batch_inp_par_par_frac": 0.0,
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": 0.44613247805159517,
1201
+ "epoch": 1.0256,
1202
+ "grad_norm": 0.498046875,
1203
+ "kd_loss": 0.5374264506522252,
1204
+ "learning_rate": 3e-06,
1205
+ "loss": 0.6772,
1206
+ "masked_tokens": 118.35,
1207
+ "mean_t": 0.504472183593316,
1208
+ "step": 480,
1209
+ "student_masked_tokens": 118.35
1210
+ },
1211
+ {
1212
+ "avg_mask_ratio": 0.4563717324635945,
1213
+ "avg_response_length": 232.0375,
1214
+ "avg_student_mask_ratio": 0.4563717324635945,
1215
+ "batch_ainp_frac": 0.0,
1216
+ "batch_inp_frac": 1.0,
1217
+ "batch_inp_oh_frac": 0.0,
1218
+ "batch_inp_par_frac": 0.0,
1219
+ "batch_inp_par_par_frac": 0.0,
1220
+ "batch_inp_par_reverse_frac": 0.0,
1221
+ "batch_rl_frac": 0.0,
1222
+ "batch_sft_frac": 0.0,
1223
+ "batch_soft_sft_frac": 0.0,
1224
+ "batch_tf_frac": 0.0,
1225
+ "ce_loss": 0.37626147485414096,
1226
+ "epoch": 1.0469333333333333,
1227
+ "grad_norm": 0.54296875,
1228
+ "kd_loss": 0.392788901903657,
1229
+ "learning_rate": 3e-06,
1230
+ "loss": 0.6047,
1231
+ "masked_tokens": 98.35,
1232
+ "mean_t": 0.4888980514719151,
1233
+ "step": 490,
1234
+ "student_masked_tokens": 98.35
1235
+ },
1236
+ {
1237
+ "avg_mask_ratio": 0.5079968665260821,
1238
+ "avg_response_length": 253.7875,
1239
+ "avg_student_mask_ratio": 0.5079968665260821,
1240
+ "batch_ainp_frac": 0.0,
1241
+ "batch_inp_frac": 1.0,
1242
+ "batch_inp_oh_frac": 0.0,
1243
+ "batch_inp_par_frac": 0.0,
1244
+ "batch_inp_par_par_frac": 0.0,
1245
+ "batch_inp_par_reverse_frac": 0.0,
1246
+ "batch_rl_frac": 0.0,
1247
+ "batch_sft_frac": 0.0,
1248
+ "batch_soft_sft_frac": 0.0,
1249
+ "batch_tf_frac": 0.0,
1250
+ "ce_loss": 0.30954629559880686,
1251
+ "epoch": 1.0682666666666667,
1252
+ "grad_norm": 0.291015625,
1253
+ "kd_loss": 0.4563873354276211,
1254
+ "learning_rate": 3e-06,
1255
+ "loss": 0.5996,
1256
+ "masked_tokens": 128.225,
1257
+ "mean_t": 0.5469163245841628,
1258
+ "step": 500,
1259
+ "student_masked_tokens": 128.225
1260
+ },
1261
+ {
1262
+ "avg_mask_ratio": 0.5109448074479588,
1263
+ "avg_response_length": 254.2,
1264
+ "avg_student_mask_ratio": 0.5109448074479588,
1265
+ "batch_ainp_frac": 0.0,
1266
+ "batch_inp_frac": 1.0,
1267
+ "batch_inp_oh_frac": 0.0,
1268
+ "batch_inp_par_frac": 0.0,
1269
+ "batch_inp_par_par_frac": 0.0,
1270
+ "batch_inp_par_reverse_frac": 0.0,
1271
+ "batch_rl_frac": 0.0,
1272
+ "batch_sft_frac": 0.0,
1273
+ "batch_soft_sft_frac": 0.0,
1274
+ "batch_tf_frac": 0.0,
1275
+ "ce_loss": 0.2868076219221166,
1276
+ "epoch": 1.0896,
1277
+ "grad_norm": 2.515625,
1278
+ "kd_loss": 0.5652106747879998,
1279
+ "learning_rate": 3e-06,
1280
+ "loss": 0.6398,
1281
+ "masked_tokens": 137.5875,
1282
+ "mean_t": 0.5275314710394013,
1283
+ "step": 510,
1284
+ "student_masked_tokens": 137.5875
1285
+ },
1286
+ {
1287
+ "avg_mask_ratio": 0.45396183808334173,
1288
+ "avg_response_length": 202.7625,
1289
+ "avg_student_mask_ratio": 0.45396183808334173,
1290
+ "batch_ainp_frac": 0.0,
1291
+ "batch_inp_frac": 1.0,
1292
+ "batch_inp_oh_frac": 0.0,
1293
+ "batch_inp_par_frac": 0.0,
1294
+ "batch_inp_par_par_frac": 0.0,
1295
+ "batch_inp_par_reverse_frac": 0.0,
1296
+ "batch_rl_frac": 0.0,
1297
+ "batch_sft_frac": 0.0,
1298
+ "batch_soft_sft_frac": 0.0,
1299
+ "batch_tf_frac": 0.0,
1300
+ "ce_loss": 0.38311037250946356,
1301
+ "epoch": 1.1109333333333333,
1302
+ "grad_norm": 0.6171875,
1303
+ "kd_loss": 0.423658079797778,
1304
+ "learning_rate": 3e-06,
1305
+ "loss": 0.6386,
1306
+ "masked_tokens": 87.0625,
1307
+ "mean_t": 0.49193521235138177,
1308
+ "step": 520,
1309
+ "student_masked_tokens": 87.0625
1310
+ },
1311
+ {
1312
+ "avg_mask_ratio": 0.47015948037151245,
1313
+ "avg_response_length": 214.275,
1314
+ "avg_student_mask_ratio": 0.47015948037151245,
1315
+ "batch_ainp_frac": 0.0,
1316
+ "batch_inp_frac": 1.0,
1317
+ "batch_inp_oh_frac": 0.0,
1318
+ "batch_inp_par_frac": 0.0,
1319
+ "batch_inp_par_par_frac": 0.0,
1320
+ "batch_inp_par_reverse_frac": 0.0,
1321
+ "batch_rl_frac": 0.0,
1322
+ "batch_sft_frac": 0.0,
1323
+ "batch_soft_sft_frac": 0.0,
1324
+ "batch_tf_frac": 0.0,
1325
+ "ce_loss": 0.47228433731506814,
1326
+ "epoch": 1.1322666666666668,
1327
+ "grad_norm": 0.609375,
1328
+ "kd_loss": 0.45688082203427316,
1329
+ "learning_rate": 3e-06,
1330
+ "loss": 0.737,
1331
+ "masked_tokens": 99.8625,
1332
+ "mean_t": 0.49621942077938,
1333
+ "step": 530,
1334
+ "student_masked_tokens": 99.8625
1335
+ },
1336
+ {
1337
+ "avg_mask_ratio": 0.4892866689246148,
1338
+ "avg_response_length": 231.3125,
1339
+ "avg_student_mask_ratio": 0.4892866689246148,
1340
+ "batch_ainp_frac": 0.0,
1341
+ "batch_inp_frac": 1.0,
1342
+ "batch_inp_oh_frac": 0.0,
1343
+ "batch_inp_par_frac": 0.0,
1344
+ "batch_inp_par_par_frac": 0.0,
1345
+ "batch_inp_par_reverse_frac": 0.0,
1346
+ "batch_rl_frac": 0.0,
1347
+ "batch_sft_frac": 0.0,
1348
+ "batch_soft_sft_frac": 0.0,
1349
+ "batch_tf_frac": 0.0,
1350
+ "ce_loss": 0.4080867745911064,
1351
+ "epoch": 1.1536,
1352
+ "grad_norm": 0.341796875,
1353
+ "kd_loss": 0.5618651450654625,
1354
+ "learning_rate": 3e-06,
1355
+ "loss": 0.6922,
1356
+ "masked_tokens": 107.375,
1357
+ "mean_t": 0.5208023569080978,
1358
+ "step": 540,
1359
+ "student_masked_tokens": 107.375
1360
+ },
1361
+ {
1362
+ "avg_mask_ratio": 0.4541942774085328,
1363
+ "avg_response_length": 213.525,
1364
+ "avg_student_mask_ratio": 0.4541942774085328,
1365
+ "batch_ainp_frac": 0.0,
1366
+ "batch_inp_frac": 1.0,
1367
+ "batch_inp_oh_frac": 0.0,
1368
+ "batch_inp_par_frac": 0.0,
1369
+ "batch_inp_par_par_frac": 0.0,
1370
+ "batch_inp_par_reverse_frac": 0.0,
1371
+ "batch_rl_frac": 0.0,
1372
+ "batch_sft_frac": 0.0,
1373
+ "batch_soft_sft_frac": 0.0,
1374
+ "batch_tf_frac": 0.0,
1375
+ "ce_loss": 0.22217674175137744,
1376
+ "epoch": 1.1749333333333334,
1377
+ "grad_norm": 0.2412109375,
1378
+ "kd_loss": 0.3673438885498399,
1379
+ "learning_rate": 3e-06,
1380
+ "loss": 0.5008,
1381
+ "masked_tokens": 97.8875,
1382
+ "mean_t": 0.4767197913257405,
1383
+ "step": 550,
1384
+ "student_masked_tokens": 97.8875
1385
+ },
1386
+ {
1387
+ "avg_mask_ratio": 0.39282396506750955,
1388
+ "avg_response_length": 231.4125,
1389
+ "avg_student_mask_ratio": 0.39282396506750955,
1390
+ "batch_ainp_frac": 0.0,
1391
+ "batch_inp_frac": 1.0,
1392
+ "batch_inp_oh_frac": 0.0,
1393
+ "batch_inp_par_frac": 0.0,
1394
+ "batch_inp_par_par_frac": 0.0,
1395
+ "batch_inp_par_reverse_frac": 0.0,
1396
+ "batch_rl_frac": 0.0,
1397
+ "batch_sft_frac": 0.0,
1398
+ "batch_soft_sft_frac": 0.0,
1399
+ "batch_tf_frac": 0.0,
1400
+ "ce_loss": 0.3512847523151777,
1401
+ "epoch": 1.1962666666666666,
1402
+ "grad_norm": 0.8828125,
1403
+ "kd_loss": 0.48686740984790616,
1404
+ "learning_rate": 3e-06,
1405
+ "loss": 0.5823,
1406
+ "masked_tokens": 99.2875,
1407
+ "mean_t": 0.4111072298779618,
1408
+ "step": 560,
1409
+ "student_masked_tokens": 99.2875
1410
+ },
1411
+ {
1412
+ "avg_mask_ratio": 0.4483634108910337,
1413
+ "avg_response_length": 230.1625,
1414
+ "avg_student_mask_ratio": 0.4483634108910337,
1415
+ "batch_ainp_frac": 0.0,
1416
+ "batch_inp_frac": 1.0,
1417
+ "batch_inp_oh_frac": 0.0,
1418
+ "batch_inp_par_frac": 0.0,
1419
+ "batch_inp_par_par_frac": 0.0,
1420
+ "batch_inp_par_reverse_frac": 0.0,
1421
+ "batch_rl_frac": 0.0,
1422
+ "batch_sft_frac": 0.0,
1423
+ "batch_soft_sft_frac": 0.0,
1424
+ "batch_tf_frac": 0.0,
1425
+ "ce_loss": 0.31345968546206676,
1426
+ "epoch": 1.2176,
1427
+ "grad_norm": 0.4453125,
1428
+ "kd_loss": 0.41564053312727084,
1429
+ "learning_rate": 3e-06,
1430
+ "loss": 0.5898,
1431
+ "masked_tokens": 108.9875,
1432
+ "mean_t": 0.48533305872697385,
1433
+ "step": 570,
1434
+ "student_masked_tokens": 108.9875
1435
+ },
1436
+ {
1437
+ "avg_mask_ratio": 0.465452536707744,
1438
+ "avg_response_length": 267.4375,
1439
+ "avg_student_mask_ratio": 0.465452536707744,
1440
+ "batch_ainp_frac": 0.0,
1441
+ "batch_inp_frac": 1.0,
1442
+ "batch_inp_oh_frac": 0.0,
1443
+ "batch_inp_par_frac": 0.0,
1444
+ "batch_inp_par_par_frac": 0.0,
1445
+ "batch_inp_par_reverse_frac": 0.0,
1446
+ "batch_rl_frac": 0.0,
1447
+ "batch_sft_frac": 0.0,
1448
+ "batch_soft_sft_frac": 0.0,
1449
+ "batch_tf_frac": 0.0,
1450
+ "ce_loss": 0.3618907347364768,
1451
+ "epoch": 1.2389333333333332,
1452
+ "grad_norm": 8.6875,
1453
+ "kd_loss": 0.4481006292516895,
1454
+ "learning_rate": 3e-06,
1455
+ "loss": 0.6314,
1456
+ "masked_tokens": 129.075,
1457
+ "mean_t": 0.49976949762785805,
1458
+ "step": 580,
1459
+ "student_masked_tokens": 129.075
1460
+ },
1461
+ {
1462
+ "avg_mask_ratio": 0.5225977989146486,
1463
+ "avg_response_length": 228.45,
1464
+ "avg_student_mask_ratio": 0.5225977989146486,
1465
+ "batch_ainp_frac": 0.0,
1466
+ "batch_inp_frac": 1.0,
1467
+ "batch_inp_oh_frac": 0.0,
1468
+ "batch_inp_par_frac": 0.0,
1469
+ "batch_inp_par_par_frac": 0.0,
1470
+ "batch_inp_par_reverse_frac": 0.0,
1471
+ "batch_rl_frac": 0.0,
1472
+ "batch_sft_frac": 0.0,
1473
+ "batch_soft_sft_frac": 0.0,
1474
+ "batch_tf_frac": 0.0,
1475
+ "ce_loss": 0.5639314363695348,
1476
+ "epoch": 1.2602666666666666,
1477
+ "grad_norm": 1.1328125,
1478
+ "kd_loss": 0.5351108588445992,
1479
+ "learning_rate": 3e-06,
1480
+ "loss": 0.8274,
1481
+ "masked_tokens": 121.675,
1482
+ "mean_t": 0.5521843038732186,
1483
+ "step": 590,
1484
+ "student_masked_tokens": 121.675
1485
+ },
1486
+ {
1487
+ "avg_mask_ratio": 0.44998724836623294,
1488
+ "avg_response_length": 236.7,
1489
+ "avg_student_mask_ratio": 0.44998724836623294,
1490
+ "batch_ainp_frac": 0.0,
1491
+ "batch_inp_frac": 1.0,
1492
+ "batch_inp_oh_frac": 0.0,
1493
+ "batch_inp_par_frac": 0.0,
1494
+ "batch_inp_par_par_frac": 0.0,
1495
+ "batch_inp_par_reverse_frac": 0.0,
1496
+ "batch_rl_frac": 0.0,
1497
+ "batch_sft_frac": 0.0,
1498
+ "batch_soft_sft_frac": 0.0,
1499
+ "batch_tf_frac": 0.0,
1500
+ "ce_loss": 0.3396833263838971,
1501
+ "epoch": 1.2816,
1502
+ "grad_norm": 0.365234375,
1503
+ "kd_loss": 0.41761890975592914,
1504
+ "learning_rate": 3e-06,
1505
+ "loss": 0.5752,
1506
+ "masked_tokens": 110.1625,
1507
+ "mean_t": 0.4788527532829903,
1508
+ "step": 600,
1509
+ "student_masked_tokens": 110.1625
1510
+ },
1511
+ {
1512
+ "avg_mask_ratio": 0.5042130865273066,
1513
+ "avg_response_length": 230.3375,
1514
+ "avg_student_mask_ratio": 0.5042130865273066,
1515
+ "batch_ainp_frac": 0.0,
1516
+ "batch_inp_frac": 1.0,
1517
+ "batch_inp_oh_frac": 0.0,
1518
+ "batch_inp_par_frac": 0.0,
1519
+ "batch_inp_par_par_frac": 0.0,
1520
+ "batch_inp_par_reverse_frac": 0.0,
1521
+ "batch_rl_frac": 0.0,
1522
+ "batch_sft_frac": 0.0,
1523
+ "batch_soft_sft_frac": 0.0,
1524
+ "batch_tf_frac": 0.0,
1525
+ "ce_loss": 0.35890077192343595,
1526
+ "epoch": 1.3029333333333333,
1527
+ "grad_norm": 0.28515625,
1528
+ "kd_loss": 0.5558427174539929,
1529
+ "learning_rate": 3e-06,
1530
+ "loss": 0.7657,
1531
+ "masked_tokens": 112.625,
1532
+ "mean_t": 0.5445419924799353,
1533
+ "step": 610,
1534
+ "student_masked_tokens": 112.625
1535
+ },
1536
+ {
1537
+ "avg_mask_ratio": 0.49637898594373836,
1538
+ "avg_response_length": 233.0625,
1539
+ "avg_student_mask_ratio": 0.49637898594373836,
1540
+ "batch_ainp_frac": 0.0,
1541
+ "batch_inp_frac": 1.0,
1542
+ "batch_inp_oh_frac": 0.0,
1543
+ "batch_inp_par_frac": 0.0,
1544
+ "batch_inp_par_par_frac": 0.0,
1545
+ "batch_inp_par_reverse_frac": 0.0,
1546
+ "batch_rl_frac": 0.0,
1547
+ "batch_sft_frac": 0.0,
1548
+ "batch_soft_sft_frac": 0.0,
1549
+ "batch_tf_frac": 0.0,
1550
+ "ce_loss": 0.32318839170733327,
1551
+ "epoch": 1.3242666666666667,
1552
+ "grad_norm": 0.515625,
1553
+ "kd_loss": 0.5518322235134179,
1554
+ "learning_rate": 3e-06,
1555
+ "loss": 0.6742,
1556
+ "masked_tokens": 111.25,
1557
+ "mean_t": 0.52490478400141,
1558
+ "step": 620,
1559
+ "student_masked_tokens": 111.25
1560
+ },
1561
+ {
1562
+ "avg_mask_ratio": 0.5177568581304512,
1563
+ "avg_response_length": 257.2125,
1564
+ "avg_student_mask_ratio": 0.5177568581304512,
1565
+ "batch_ainp_frac": 0.0,
1566
+ "batch_inp_frac": 1.0,
1567
+ "batch_inp_oh_frac": 0.0,
1568
+ "batch_inp_par_frac": 0.0,
1569
+ "batch_inp_par_par_frac": 0.0,
1570
+ "batch_inp_par_reverse_frac": 0.0,
1571
+ "batch_rl_frac": 0.0,
1572
+ "batch_sft_frac": 0.0,
1573
+ "batch_soft_sft_frac": 0.0,
1574
+ "batch_tf_frac": 0.0,
1575
+ "ce_loss": 0.5710563842050931,
1576
+ "epoch": 1.3456000000000001,
1577
+ "grad_norm": 1.3515625,
1578
+ "kd_loss": 0.5316411310721378,
1579
+ "learning_rate": 3e-06,
1580
+ "loss": 0.8598,
1581
+ "masked_tokens": 129.6125,
1582
+ "mean_t": 0.5564947265549562,
1583
+ "step": 630,
1584
+ "student_masked_tokens": 129.6125
1585
+ },
1586
+ {
1587
+ "avg_mask_ratio": 0.48226998368045315,
1588
+ "avg_response_length": 237.7125,
1589
+ "avg_student_mask_ratio": 0.48226998368045315,
1590
+ "batch_ainp_frac": 0.0,
1591
+ "batch_inp_frac": 1.0,
1592
+ "batch_inp_oh_frac": 0.0,
1593
+ "batch_inp_par_frac": 0.0,
1594
+ "batch_inp_par_par_frac": 0.0,
1595
+ "batch_inp_par_reverse_frac": 0.0,
1596
+ "batch_rl_frac": 0.0,
1597
+ "batch_sft_frac": 0.0,
1598
+ "batch_soft_sft_frac": 0.0,
1599
+ "batch_tf_frac": 0.0,
1600
+ "ce_loss": 0.2804489129174499,
1601
+ "epoch": 1.3669333333333333,
1602
+ "grad_norm": 0.2421875,
1603
+ "kd_loss": 0.3663112932188085,
1604
+ "learning_rate": 3e-06,
1605
+ "loss": 0.4584,
1606
+ "masked_tokens": 120.275,
1607
+ "mean_t": 0.5093393943971023,
1608
+ "step": 640,
1609
+ "student_masked_tokens": 120.275
1610
+ },
1611
+ {
1612
+ "avg_mask_ratio": 0.5306948523037136,
1613
+ "avg_response_length": 238.0125,
1614
+ "avg_student_mask_ratio": 0.5306948523037136,
1615
+ "batch_ainp_frac": 0.0,
1616
+ "batch_inp_frac": 1.0,
1617
+ "batch_inp_oh_frac": 0.0,
1618
+ "batch_inp_par_frac": 0.0,
1619
+ "batch_inp_par_par_frac": 0.0,
1620
+ "batch_inp_par_reverse_frac": 0.0,
1621
+ "batch_rl_frac": 0.0,
1622
+ "batch_sft_frac": 0.0,
1623
+ "batch_soft_sft_frac": 0.0,
1624
+ "batch_tf_frac": 0.0,
1625
+ "ce_loss": 0.475157093159612,
1626
+ "epoch": 1.3882666666666665,
1627
+ "grad_norm": 1.8125,
1628
+ "kd_loss": 0.5062341513834724,
1629
+ "learning_rate": 3e-06,
1630
+ "loss": 0.7115,
1631
+ "masked_tokens": 133.25,
1632
+ "mean_t": 0.5558586571365595,
1633
+ "step": 650,
1634
+ "student_masked_tokens": 133.25
1635
+ },
1636
+ {
1637
+ "avg_mask_ratio": 0.4821273953886703,
1638
+ "avg_response_length": 247.775,
1639
+ "avg_student_mask_ratio": 0.4821273953886703,
1640
+ "batch_ainp_frac": 0.0,
1641
+ "batch_inp_frac": 1.0,
1642
+ "batch_inp_oh_frac": 0.0,
1643
+ "batch_inp_par_frac": 0.0,
1644
+ "batch_inp_par_par_frac": 0.0,
1645
+ "batch_inp_par_reverse_frac": 0.0,
1646
+ "batch_rl_frac": 0.0,
1647
+ "batch_sft_frac": 0.0,
1648
+ "batch_soft_sft_frac": 0.0,
1649
+ "batch_tf_frac": 0.0,
1650
+ "ce_loss": 0.41770620119971225,
1651
+ "epoch": 1.4096,
1652
+ "grad_norm": 0.9375,
1653
+ "kd_loss": 0.425496905214095,
1654
+ "learning_rate": 3e-06,
1655
+ "loss": 0.6361,
1656
+ "masked_tokens": 128.875,
1657
+ "mean_t": 0.51307404555846,
1658
+ "step": 660,
1659
+ "student_masked_tokens": 128.875
1660
+ },
1661
+ {
1662
+ "avg_mask_ratio": 0.46056515555246735,
1663
+ "avg_response_length": 240.4375,
1664
+ "avg_student_mask_ratio": 0.46056515555246735,
1665
+ "batch_ainp_frac": 0.0,
1666
+ "batch_inp_frac": 1.0,
1667
+ "batch_inp_oh_frac": 0.0,
1668
+ "batch_inp_par_frac": 0.0,
1669
+ "batch_inp_par_par_frac": 0.0,
1670
+ "batch_inp_par_reverse_frac": 0.0,
1671
+ "batch_rl_frac": 0.0,
1672
+ "batch_sft_frac": 0.0,
1673
+ "batch_soft_sft_frac": 0.0,
1674
+ "batch_tf_frac": 0.0,
1675
+ "ce_loss": 0.24846992658117414,
1676
+ "epoch": 1.4309333333333334,
1677
+ "grad_norm": 0.60546875,
1678
+ "kd_loss": 0.34861083538812637,
1679
+ "learning_rate": 3e-06,
1680
+ "loss": 0.5112,
1681
+ "masked_tokens": 119.85,
1682
+ "mean_t": 0.4907285622088239,
1683
+ "step": 670,
1684
+ "student_masked_tokens": 119.85
1685
+ },
1686
+ {
1687
+ "avg_mask_ratio": 0.4666106043441687,
1688
+ "avg_response_length": 226.7375,
1689
+ "avg_student_mask_ratio": 0.4666106043441687,
1690
+ "batch_ainp_frac": 0.0,
1691
+ "batch_inp_frac": 1.0,
1692
+ "batch_inp_oh_frac": 0.0,
1693
+ "batch_inp_par_frac": 0.0,
1694
+ "batch_inp_par_par_frac": 0.0,
1695
+ "batch_inp_par_reverse_frac": 0.0,
1696
+ "batch_rl_frac": 0.0,
1697
+ "batch_sft_frac": 0.0,
1698
+ "batch_soft_sft_frac": 0.0,
1699
+ "batch_tf_frac": 0.0,
1700
+ "ce_loss": 0.4541423492493323,
1701
+ "epoch": 1.4522666666666666,
1702
+ "grad_norm": 0.51953125,
1703
+ "kd_loss": 0.4910934407485213,
1704
+ "learning_rate": 3e-06,
1705
+ "loss": 0.6946,
1706
+ "masked_tokens": 107.4625,
1707
+ "mean_t": 0.4913603452499956,
1708
+ "step": 680,
1709
+ "student_masked_tokens": 107.4625
1710
+ },
1711
+ {
1712
+ "avg_mask_ratio": 0.4790851596510038,
1713
+ "avg_response_length": 202.05,
1714
+ "avg_student_mask_ratio": 0.4790851596510038,
1715
+ "batch_ainp_frac": 0.0,
1716
+ "batch_inp_frac": 1.0,
1717
+ "batch_inp_oh_frac": 0.0,
1718
+ "batch_inp_par_frac": 0.0,
1719
+ "batch_inp_par_par_frac": 0.0,
1720
+ "batch_inp_par_reverse_frac": 0.0,
1721
+ "batch_rl_frac": 0.0,
1722
+ "batch_sft_frac": 0.0,
1723
+ "batch_soft_sft_frac": 0.0,
1724
+ "batch_tf_frac": 0.0,
1725
+ "ce_loss": 0.3711260147189023,
1726
+ "epoch": 1.4736,
1727
+ "grad_norm": 2.03125,
1728
+ "kd_loss": 0.41718243765291446,
1729
+ "learning_rate": 3e-06,
1730
+ "loss": 0.6313,
1731
+ "masked_tokens": 111.3125,
1732
+ "mean_t": 0.5133644798654131,
1733
+ "step": 690,
1734
+ "student_masked_tokens": 111.3125
1735
+ },
1736
+ {
1737
+ "avg_mask_ratio": 0.5250519359949977,
1738
+ "avg_response_length": 228.125,
1739
+ "avg_student_mask_ratio": 0.5250519359949977,
1740
+ "batch_ainp_frac": 0.0,
1741
+ "batch_inp_frac": 1.0,
1742
+ "batch_inp_oh_frac": 0.0,
1743
+ "batch_inp_par_frac": 0.0,
1744
+ "batch_inp_par_par_frac": 0.0,
1745
+ "batch_inp_par_reverse_frac": 0.0,
1746
+ "batch_rl_frac": 0.0,
1747
+ "batch_sft_frac": 0.0,
1748
+ "batch_soft_sft_frac": 0.0,
1749
+ "batch_tf_frac": 0.0,
1750
+ "ce_loss": 0.22230932631540554,
1751
+ "epoch": 1.4949333333333334,
1752
+ "grad_norm": 0.26171875,
1753
+ "kd_loss": 0.6619142963969352,
1754
+ "learning_rate": 3e-06,
1755
+ "loss": 0.7717,
1756
+ "masked_tokens": 132.55,
1757
+ "mean_t": 0.5625698395539075,
1758
+ "step": 700,
1759
+ "student_masked_tokens": 132.55
1760
+ },
1761
+ {
1762
+ "avg_mask_ratio": 0.4790433386107907,
1763
+ "avg_response_length": 212.5,
1764
+ "avg_student_mask_ratio": 0.4790433386107907,
1765
+ "batch_ainp_frac": 0.0,
1766
+ "batch_inp_frac": 1.0,
1767
+ "batch_inp_oh_frac": 0.0,
1768
+ "batch_inp_par_frac": 0.0,
1769
+ "batch_inp_par_par_frac": 0.0,
1770
+ "batch_inp_par_reverse_frac": 0.0,
1771
+ "batch_rl_frac": 0.0,
1772
+ "batch_sft_frac": 0.0,
1773
+ "batch_soft_sft_frac": 0.0,
1774
+ "batch_tf_frac": 0.0,
1775
+ "ce_loss": 0.24621229091012536,
1776
+ "epoch": 1.5162666666666667,
1777
+ "grad_norm": 0.2099609375,
1778
+ "kd_loss": 0.43454050603151584,
1779
+ "learning_rate": 3e-06,
1780
+ "loss": 0.5302,
1781
+ "masked_tokens": 108.7375,
1782
+ "mean_t": 0.5135623761918395,
1783
+ "step": 710,
1784
+ "student_masked_tokens": 108.7375
1785
+ },
1786
+ {
1787
+ "avg_mask_ratio": 0.47950589570682495,
1788
+ "avg_response_length": 227.075,
1789
+ "avg_student_mask_ratio": 0.47950589570682495,
1790
+ "batch_ainp_frac": 0.0,
1791
+ "batch_inp_frac": 1.0,
1792
+ "batch_inp_oh_frac": 0.0,
1793
+ "batch_inp_par_frac": 0.0,
1794
+ "batch_inp_par_par_frac": 0.0,
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": 0.36416104665024707,
1801
+ "epoch": 1.5375999999999999,
1802
+ "grad_norm": 0.75,
1803
+ "kd_loss": 0.5665610315164941,
1804
+ "learning_rate": 3e-06,
1805
+ "loss": 0.7121,
1806
+ "masked_tokens": 110.8,
1807
+ "mean_t": 0.5117021896177902,
1808
+ "step": 720,
1809
+ "student_masked_tokens": 110.8
1810
+ },
1811
+ {
1812
+ "avg_mask_ratio": 0.4604924251558259,
1813
+ "avg_response_length": 232.925,
1814
+ "avg_student_mask_ratio": 0.4604924251558259,
1815
+ "batch_ainp_frac": 0.0,
1816
+ "batch_inp_frac": 1.0,
1817
+ "batch_inp_oh_frac": 0.0,
1818
+ "batch_inp_par_frac": 0.0,
1819
+ "batch_inp_par_par_frac": 0.0,
1820
+ "batch_inp_par_reverse_frac": 0.0,
1821
+ "batch_rl_frac": 0.0,
1822
+ "batch_sft_frac": 0.0,
1823
+ "batch_soft_sft_frac": 0.0,
1824
+ "batch_tf_frac": 0.0,
1825
+ "ce_loss": 0.38923927966282007,
1826
+ "epoch": 1.5589333333333333,
1827
+ "grad_norm": 1.015625,
1828
+ "kd_loss": 0.4302867329986782,
1829
+ "learning_rate": 3e-06,
1830
+ "loss": 0.639,
1831
+ "masked_tokens": 104.9625,
1832
+ "mean_t": 0.49050743713742123,
1833
+ "step": 730,
1834
+ "student_masked_tokens": 104.9625
1835
+ },
1836
+ {
1837
+ "avg_mask_ratio": 0.5185885130194947,
1838
+ "avg_response_length": 183.325,
1839
+ "avg_student_mask_ratio": 0.5185885130194947,
1840
+ "batch_ainp_frac": 0.0,
1841
+ "batch_inp_frac": 1.0,
1842
+ "batch_inp_oh_frac": 0.0,
1843
+ "batch_inp_par_frac": 0.0,
1844
+ "batch_inp_par_par_frac": 0.0,
1845
+ "batch_inp_par_reverse_frac": 0.0,
1846
+ "batch_rl_frac": 0.0,
1847
+ "batch_sft_frac": 0.0,
1848
+ "batch_soft_sft_frac": 0.0,
1849
+ "batch_tf_frac": 0.0,
1850
+ "ce_loss": 0.3361817517367399,
1851
+ "epoch": 1.5802666666666667,
1852
+ "grad_norm": 0.40234375,
1853
+ "kd_loss": 0.5340734164818514,
1854
+ "learning_rate": 3e-06,
1855
+ "loss": 0.7461,
1856
+ "masked_tokens": 97.125,
1857
+ "mean_t": 0.5505168779753149,
1858
+ "step": 740,
1859
+ "student_masked_tokens": 97.125
1860
+ },
1861
+ {
1862
+ "avg_mask_ratio": 0.4191439319110941,
1863
+ "avg_response_length": 223.65,
1864
+ "avg_student_mask_ratio": 0.4191439319110941,
1865
+ "batch_ainp_frac": 0.0,
1866
+ "batch_inp_frac": 1.0,
1867
+ "batch_inp_oh_frac": 0.0,
1868
+ "batch_inp_par_frac": 0.0,
1869
+ "batch_inp_par_par_frac": 0.0,
1870
+ "batch_inp_par_reverse_frac": 0.0,
1871
+ "batch_rl_frac": 0.0,
1872
+ "batch_sft_frac": 0.0,
1873
+ "batch_soft_sft_frac": 0.0,
1874
+ "batch_tf_frac": 0.0,
1875
+ "ce_loss": 0.37429177601145514,
1876
+ "epoch": 1.6016,
1877
+ "grad_norm": 0.58203125,
1878
+ "kd_loss": 0.5036597276406856,
1879
+ "learning_rate": 3e-06,
1880
+ "loss": 0.6491,
1881
+ "masked_tokens": 95.3125,
1882
+ "mean_t": 0.4437690361432033,
1883
+ "step": 750,
1884
+ "student_masked_tokens": 95.3125
1885
+ },
1886
+ {
1887
+ "avg_mask_ratio": 0.46706983938929625,
1888
+ "avg_response_length": 216.0625,
1889
+ "avg_student_mask_ratio": 0.46706983938929625,
1890
+ "batch_ainp_frac": 0.0,
1891
+ "batch_inp_frac": 1.0,
1892
+ "batch_inp_oh_frac": 0.0,
1893
+ "batch_inp_par_frac": 0.0,
1894
+ "batch_inp_par_par_frac": 0.0,
1895
+ "batch_inp_par_reverse_frac": 0.0,
1896
+ "batch_rl_frac": 0.0,
1897
+ "batch_sft_frac": 0.0,
1898
+ "batch_soft_sft_frac": 0.0,
1899
+ "batch_tf_frac": 0.0,
1900
+ "ce_loss": 0.4449058656399984,
1901
+ "epoch": 1.6229333333333333,
1902
+ "grad_norm": 0.8203125,
1903
+ "kd_loss": 0.5661326096985168,
1904
+ "learning_rate": 3e-06,
1905
+ "loss": 0.7233,
1906
+ "masked_tokens": 107.7,
1907
+ "mean_t": 0.49132869170280175,
1908
+ "step": 760,
1909
+ "student_masked_tokens": 107.7
1910
+ },
1911
+ {
1912
+ "avg_mask_ratio": 0.44156218122225255,
1913
+ "avg_response_length": 259.675,
1914
+ "avg_student_mask_ratio": 0.44156218122225255,
1915
+ "batch_ainp_frac": 0.0,
1916
+ "batch_inp_frac": 1.0,
1917
+ "batch_inp_oh_frac": 0.0,
1918
+ "batch_inp_par_frac": 0.0,
1919
+ "batch_inp_par_par_frac": 0.0,
1920
+ "batch_inp_par_reverse_frac": 0.0,
1921
+ "batch_rl_frac": 0.0,
1922
+ "batch_sft_frac": 0.0,
1923
+ "batch_soft_sft_frac": 0.0,
1924
+ "batch_tf_frac": 0.0,
1925
+ "ce_loss": 0.25899335961771613,
1926
+ "epoch": 1.6442666666666668,
1927
+ "grad_norm": 0.396484375,
1928
+ "kd_loss": 0.4095979654902003,
1929
+ "learning_rate": 3e-06,
1930
+ "loss": 0.5099,
1931
+ "masked_tokens": 117.5,
1932
+ "mean_t": 0.4667695587326307,
1933
+ "step": 770,
1934
+ "student_masked_tokens": 117.5
1935
+ },
1936
+ {
1937
+ "avg_mask_ratio": 0.42836043585848527,
1938
+ "avg_response_length": 258.5125,
1939
+ "avg_student_mask_ratio": 0.42836043585848527,
1940
+ "batch_ainp_frac": 0.0,
1941
+ "batch_inp_frac": 1.0,
1942
+ "batch_inp_oh_frac": 0.0,
1943
+ "batch_inp_par_frac": 0.0,
1944
+ "batch_inp_par_par_frac": 0.0,
1945
+ "batch_inp_par_reverse_frac": 0.0,
1946
+ "batch_rl_frac": 0.0,
1947
+ "batch_sft_frac": 0.0,
1948
+ "batch_soft_sft_frac": 0.0,
1949
+ "batch_tf_frac": 0.0,
1950
+ "ce_loss": 0.2897560694203321,
1951
+ "epoch": 1.6656,
1952
+ "grad_norm": 0.2431640625,
1953
+ "kd_loss": 0.34635278815572546,
1954
+ "learning_rate": 3e-06,
1955
+ "loss": 0.4802,
1956
+ "masked_tokens": 119.0125,
1957
+ "mean_t": 0.44942845597106496,
1958
+ "step": 780,
1959
+ "student_masked_tokens": 119.0125
1960
+ },
1961
+ {
1962
+ "avg_mask_ratio": 0.46589430308085866,
1963
+ "avg_response_length": 222.3125,
1964
+ "avg_student_mask_ratio": 0.46589430308085866,
1965
+ "batch_ainp_frac": 0.0,
1966
+ "batch_inp_frac": 1.0,
1967
+ "batch_inp_oh_frac": 0.0,
1968
+ "batch_inp_par_frac": 0.0,
1969
+ "batch_inp_par_par_frac": 0.0,
1970
+ "batch_inp_par_reverse_frac": 0.0,
1971
+ "batch_rl_frac": 0.0,
1972
+ "batch_sft_frac": 0.0,
1973
+ "batch_soft_sft_frac": 0.0,
1974
+ "batch_tf_frac": 0.0,
1975
+ "ce_loss": 0.21603642557238345,
1976
+ "epoch": 1.6869333333333332,
1977
+ "grad_norm": 0.140625,
1978
+ "kd_loss": 0.33674514803767297,
1979
+ "learning_rate": 3e-06,
1980
+ "loss": 0.489,
1981
+ "masked_tokens": 103.25,
1982
+ "mean_t": 0.4993515375303105,
1983
+ "step": 790,
1984
+ "student_masked_tokens": 103.25
1985
+ },
1986
+ {
1987
+ "avg_mask_ratio": 0.46366424662992356,
1988
+ "avg_response_length": 219.6875,
1989
+ "avg_student_mask_ratio": 0.46366424662992356,
1990
+ "batch_ainp_frac": 0.0,
1991
+ "batch_inp_frac": 1.0,
1992
+ "batch_inp_oh_frac": 0.0,
1993
+ "batch_inp_par_frac": 0.0,
1994
+ "batch_inp_par_par_frac": 0.0,
1995
+ "batch_inp_par_reverse_frac": 0.0,
1996
+ "batch_rl_frac": 0.0,
1997
+ "batch_sft_frac": 0.0,
1998
+ "batch_soft_sft_frac": 0.0,
1999
+ "batch_tf_frac": 0.0,
2000
+ "ce_loss": 0.2663005536277069,
2001
+ "epoch": 1.7082666666666668,
2002
+ "grad_norm": 0.23828125,
2003
+ "kd_loss": 0.35138718315538425,
2004
+ "learning_rate": 3e-06,
2005
+ "loss": 0.5434,
2006
+ "masked_tokens": 104.5,
2007
+ "mean_t": 0.500370389316231,
2008
+ "step": 800,
2009
+ "student_masked_tokens": 104.5
2010
+ },
2011
+ {
2012
+ "avg_mask_ratio": 0.503375941584818,
2013
+ "avg_response_length": 237.85,
2014
+ "avg_student_mask_ratio": 0.503375941584818,
2015
+ "batch_ainp_frac": 0.0,
2016
+ "batch_inp_frac": 1.0,
2017
+ "batch_inp_oh_frac": 0.0,
2018
+ "batch_inp_par_frac": 0.0,
2019
+ "batch_inp_par_par_frac": 0.0,
2020
+ "batch_inp_par_reverse_frac": 0.0,
2021
+ "batch_rl_frac": 0.0,
2022
+ "batch_sft_frac": 0.0,
2023
+ "batch_soft_sft_frac": 0.0,
2024
+ "batch_tf_frac": 0.0,
2025
+ "ce_loss": 0.4813590554784753,
2026
+ "epoch": 1.7296,
2027
+ "grad_norm": 1.6015625,
2028
+ "kd_loss": 0.45312339970045057,
2029
+ "learning_rate": 3e-06,
2030
+ "loss": 0.706,
2031
+ "masked_tokens": 118.2125,
2032
+ "mean_t": 0.5317009104182944,
2033
+ "step": 810,
2034
+ "student_masked_tokens": 118.2125
2035
+ },
2036
+ {
2037
+ "avg_mask_ratio": 0.5110091455746442,
2038
+ "avg_response_length": 209.0875,
2039
+ "avg_student_mask_ratio": 0.5110091455746442,
2040
+ "batch_ainp_frac": 0.0,
2041
+ "batch_inp_frac": 1.0,
2042
+ "batch_inp_oh_frac": 0.0,
2043
+ "batch_inp_par_frac": 0.0,
2044
+ "batch_inp_par_par_frac": 0.0,
2045
+ "batch_inp_par_reverse_frac": 0.0,
2046
+ "batch_rl_frac": 0.0,
2047
+ "batch_sft_frac": 0.0,
2048
+ "batch_soft_sft_frac": 0.0,
2049
+ "batch_tf_frac": 0.0,
2050
+ "ce_loss": 0.4535834654417954,
2051
+ "epoch": 1.7509333333333332,
2052
+ "grad_norm": 0.70703125,
2053
+ "kd_loss": 0.5985253949772413,
2054
+ "learning_rate": 3e-06,
2055
+ "loss": 0.7794,
2056
+ "masked_tokens": 120.95,
2057
+ "mean_t": 0.5392061032878701,
2058
+ "step": 820,
2059
+ "student_masked_tokens": 120.95
2060
+ },
2061
+ {
2062
+ "avg_mask_ratio": 0.49899387182667854,
2063
+ "avg_response_length": 263.975,
2064
+ "avg_student_mask_ratio": 0.49899387182667854,
2065
+ "batch_ainp_frac": 0.0,
2066
+ "batch_inp_frac": 1.0,
2067
+ "batch_inp_oh_frac": 0.0,
2068
+ "batch_inp_par_frac": 0.0,
2069
+ "batch_inp_par_par_frac": 0.0,
2070
+ "batch_inp_par_reverse_frac": 0.0,
2071
+ "batch_rl_frac": 0.0,
2072
+ "batch_sft_frac": 0.0,
2073
+ "batch_soft_sft_frac": 0.0,
2074
+ "batch_tf_frac": 0.0,
2075
+ "ce_loss": 0.40083689643704473,
2076
+ "epoch": 1.7722666666666667,
2077
+ "grad_norm": 0.1708984375,
2078
+ "kd_loss": 0.5644028104892641,
2079
+ "learning_rate": 3e-06,
2080
+ "loss": 0.7632,
2081
+ "masked_tokens": 137.075,
2082
+ "mean_t": 0.5238314627087675,
2083
+ "step": 830,
2084
+ "student_masked_tokens": 137.075
2085
+ },
2086
+ {
2087
+ "avg_mask_ratio": 0.4997270987310912,
2088
+ "avg_response_length": 221.9,
2089
+ "avg_student_mask_ratio": 0.4997270987310912,
2090
+ "batch_ainp_frac": 0.0,
2091
+ "batch_inp_frac": 1.0,
2092
+ "batch_inp_oh_frac": 0.0,
2093
+ "batch_inp_par_frac": 0.0,
2094
+ "batch_inp_par_par_frac": 0.0,
2095
+ "batch_inp_par_reverse_frac": 0.0,
2096
+ "batch_rl_frac": 0.0,
2097
+ "batch_sft_frac": 0.0,
2098
+ "batch_soft_sft_frac": 0.0,
2099
+ "batch_tf_frac": 0.0,
2100
+ "ce_loss": 0.2870929398425915,
2101
+ "epoch": 1.7936,
2102
+ "grad_norm": 0.345703125,
2103
+ "kd_loss": 0.4698917509396324,
2104
+ "learning_rate": 3e-06,
2105
+ "loss": 0.6327,
2106
+ "masked_tokens": 114.525,
2107
+ "mean_t": 0.5301066277665086,
2108
+ "step": 840,
2109
+ "student_masked_tokens": 114.525
2110
+ },
2111
+ {
2112
+ "avg_mask_ratio": 0.4988076956477016,
2113
+ "avg_response_length": 225.5,
2114
+ "avg_student_mask_ratio": 0.4988076956477016,
2115
+ "batch_ainp_frac": 0.0,
2116
+ "batch_inp_frac": 1.0,
2117
+ "batch_inp_oh_frac": 0.0,
2118
+ "batch_inp_par_frac": 0.0,
2119
+ "batch_inp_par_par_frac": 0.0,
2120
+ "batch_inp_par_reverse_frac": 0.0,
2121
+ "batch_rl_frac": 0.0,
2122
+ "batch_sft_frac": 0.0,
2123
+ "batch_soft_sft_frac": 0.0,
2124
+ "batch_tf_frac": 0.0,
2125
+ "ce_loss": 0.3023421537889817,
2126
+ "epoch": 1.8149333333333333,
2127
+ "grad_norm": 0.443359375,
2128
+ "kd_loss": 0.3271854338312551,
2129
+ "learning_rate": 3e-06,
2130
+ "loss": 0.5634,
2131
+ "masked_tokens": 116.9125,
2132
+ "mean_t": 0.5343429344706238,
2133
+ "step": 850,
2134
+ "student_masked_tokens": 116.9125
2135
+ },
2136
+ {
2137
+ "avg_mask_ratio": 0.4635998342186213,
2138
+ "avg_response_length": 229.125,
2139
+ "avg_student_mask_ratio": 0.4635998342186213,
2140
+ "batch_ainp_frac": 0.0,
2141
+ "batch_inp_frac": 1.0,
2142
+ "batch_inp_oh_frac": 0.0,
2143
+ "batch_inp_par_frac": 0.0,
2144
+ "batch_inp_par_par_frac": 0.0,
2145
+ "batch_inp_par_reverse_frac": 0.0,
2146
+ "batch_rl_frac": 0.0,
2147
+ "batch_sft_frac": 0.0,
2148
+ "batch_soft_sft_frac": 0.0,
2149
+ "batch_tf_frac": 0.0,
2150
+ "ce_loss": 0.37467331880507115,
2151
+ "epoch": 1.8362666666666667,
2152
+ "grad_norm": 0.384765625,
2153
+ "kd_loss": 0.4431717619034316,
2154
+ "learning_rate": 3e-06,
2155
+ "loss": 0.5956,
2156
+ "masked_tokens": 109.675,
2157
+ "mean_t": 0.4791536889737472,
2158
+ "step": 860,
2159
+ "student_masked_tokens": 109.675
2160
+ },
2161
+ {
2162
+ "avg_mask_ratio": 0.49111039767740294,
2163
+ "avg_response_length": 229.1,
2164
+ "avg_student_mask_ratio": 0.49111039767740294,
2165
+ "batch_ainp_frac": 0.0,
2166
+ "batch_inp_frac": 1.0,
2167
+ "batch_inp_oh_frac": 0.0,
2168
+ "batch_inp_par_frac": 0.0,
2169
+ "batch_inp_par_par_frac": 0.0,
2170
+ "batch_inp_par_reverse_frac": 0.0,
2171
+ "batch_rl_frac": 0.0,
2172
+ "batch_sft_frac": 0.0,
2173
+ "batch_soft_sft_frac": 0.0,
2174
+ "batch_tf_frac": 0.0,
2175
+ "ce_loss": 0.3838037288314126,
2176
+ "epoch": 1.8576000000000001,
2177
+ "grad_norm": 0.333984375,
2178
+ "kd_loss": 0.47523635068355363,
2179
+ "learning_rate": 3e-06,
2180
+ "loss": 0.6859,
2181
+ "masked_tokens": 115.6625,
2182
+ "mean_t": 0.5203817339061061,
2183
+ "step": 870,
2184
+ "student_masked_tokens": 115.6625
2185
+ },
2186
+ {
2187
+ "avg_mask_ratio": 0.4427660425659269,
2188
+ "avg_response_length": 198.5625,
2189
+ "avg_student_mask_ratio": 0.4427660425659269,
2190
+ "batch_ainp_frac": 0.0,
2191
+ "batch_inp_frac": 1.0,
2192
+ "batch_inp_oh_frac": 0.0,
2193
+ "batch_inp_par_frac": 0.0,
2194
+ "batch_inp_par_par_frac": 0.0,
2195
+ "batch_inp_par_reverse_frac": 0.0,
2196
+ "batch_rl_frac": 0.0,
2197
+ "batch_sft_frac": 0.0,
2198
+ "batch_soft_sft_frac": 0.0,
2199
+ "batch_tf_frac": 0.0,
2200
+ "ce_loss": 0.33109274096627817,
2201
+ "epoch": 1.8789333333333333,
2202
+ "grad_norm": 1.0859375,
2203
+ "kd_loss": 0.46695662873548827,
2204
+ "learning_rate": 3e-06,
2205
+ "loss": 0.6284,
2206
+ "masked_tokens": 91.175,
2207
+ "mean_t": 0.4875184997683391,
2208
+ "step": 880,
2209
+ "student_masked_tokens": 91.175
2210
+ },
2211
+ {
2212
+ "avg_mask_ratio": 0.4464349385118112,
2213
+ "avg_response_length": 225.8375,
2214
+ "avg_student_mask_ratio": 0.4464349385118112,
2215
+ "batch_ainp_frac": 0.0,
2216
+ "batch_inp_frac": 1.0,
2217
+ "batch_inp_oh_frac": 0.0,
2218
+ "batch_inp_par_frac": 0.0,
2219
+ "batch_inp_par_par_frac": 0.0,
2220
+ "batch_inp_par_reverse_frac": 0.0,
2221
+ "batch_rl_frac": 0.0,
2222
+ "batch_sft_frac": 0.0,
2223
+ "batch_soft_sft_frac": 0.0,
2224
+ "batch_tf_frac": 0.0,
2225
+ "ce_loss": 0.22229116438190885,
2226
+ "epoch": 1.9002666666666665,
2227
+ "grad_norm": 0.12890625,
2228
+ "kd_loss": 0.4006316699657759,
2229
+ "learning_rate": 3e-06,
2230
+ "loss": 0.4934,
2231
+ "masked_tokens": 101.75,
2232
+ "mean_t": 0.4766692223958671,
2233
+ "step": 890,
2234
+ "student_masked_tokens": 101.75
2235
+ },
2236
+ {
2237
+ "avg_mask_ratio": 0.44976164362160487,
2238
+ "avg_response_length": 227.7875,
2239
+ "avg_student_mask_ratio": 0.44976164362160487,
2240
+ "batch_ainp_frac": 0.0,
2241
+ "batch_inp_frac": 1.0,
2242
+ "batch_inp_oh_frac": 0.0,
2243
+ "batch_inp_par_frac": 0.0,
2244
+ "batch_inp_par_par_frac": 0.0,
2245
+ "batch_inp_par_reverse_frac": 0.0,
2246
+ "batch_rl_frac": 0.0,
2247
+ "batch_sft_frac": 0.0,
2248
+ "batch_soft_sft_frac": 0.0,
2249
+ "batch_tf_frac": 0.0,
2250
+ "ce_loss": 0.38169105723031577,
2251
+ "epoch": 1.9216,
2252
+ "grad_norm": 1.765625,
2253
+ "kd_loss": 0.47280531010078086,
2254
+ "learning_rate": 3e-06,
2255
+ "loss": 0.6337,
2256
+ "masked_tokens": 103.475,
2257
+ "mean_t": 0.487134758150205,
2258
+ "step": 900,
2259
+ "student_masked_tokens": 103.475
2260
+ },
2261
+ {
2262
+ "avg_mask_ratio": 0.475579984736396,
2263
+ "avg_response_length": 245.1625,
2264
+ "avg_student_mask_ratio": 0.475579984736396,
2265
+ "batch_ainp_frac": 0.0,
2266
+ "batch_inp_frac": 1.0,
2267
+ "batch_inp_oh_frac": 0.0,
2268
+ "batch_inp_par_frac": 0.0,
2269
+ "batch_inp_par_par_frac": 0.0,
2270
+ "batch_inp_par_reverse_frac": 0.0,
2271
+ "batch_rl_frac": 0.0,
2272
+ "batch_sft_frac": 0.0,
2273
+ "batch_soft_sft_frac": 0.0,
2274
+ "batch_tf_frac": 0.0,
2275
+ "ce_loss": 0.27549623605577833,
2276
+ "epoch": 1.9429333333333334,
2277
+ "grad_norm": 0.451171875,
2278
+ "kd_loss": 0.4638562942510987,
2279
+ "learning_rate": 3e-06,
2280
+ "loss": 0.5387,
2281
+ "masked_tokens": 124.6375,
2282
+ "mean_t": 0.5027793228859082,
2283
+ "step": 910,
2284
+ "student_masked_tokens": 124.6375
2285
+ },
2286
+ {
2287
+ "avg_mask_ratio": 0.4688875659601763,
2288
+ "avg_response_length": 226.2875,
2289
+ "avg_student_mask_ratio": 0.4688875659601763,
2290
+ "batch_ainp_frac": 0.0,
2291
+ "batch_inp_frac": 1.0,
2292
+ "batch_inp_oh_frac": 0.0,
2293
+ "batch_inp_par_frac": 0.0,
2294
+ "batch_inp_par_par_frac": 0.0,
2295
+ "batch_inp_par_reverse_frac": 0.0,
2296
+ "batch_rl_frac": 0.0,
2297
+ "batch_sft_frac": 0.0,
2298
+ "batch_soft_sft_frac": 0.0,
2299
+ "batch_tf_frac": 0.0,
2300
+ "ce_loss": 0.2772836374151325,
2301
+ "epoch": 1.9642666666666666,
2302
+ "grad_norm": 0.416015625,
2303
+ "kd_loss": 0.44530672791033793,
2304
+ "learning_rate": 3e-06,
2305
+ "loss": 0.6177,
2306
+ "masked_tokens": 110.0125,
2307
+ "mean_t": 0.49417946098838,
2308
+ "step": 920,
2309
+ "student_masked_tokens": 110.0125
2310
+ },
2311
+ {
2312
+ "avg_mask_ratio": 0.47999348094454036,
2313
+ "avg_response_length": 237.05,
2314
+ "avg_student_mask_ratio": 0.47999348094454036,
2315
+ "batch_ainp_frac": 0.0,
2316
+ "batch_inp_frac": 1.0,
2317
+ "batch_inp_oh_frac": 0.0,
2318
+ "batch_inp_par_frac": 0.0,
2319
+ "batch_inp_par_par_frac": 0.0,
2320
+ "batch_inp_par_reverse_frac": 0.0,
2321
+ "batch_rl_frac": 0.0,
2322
+ "batch_sft_frac": 0.0,
2323
+ "batch_soft_sft_frac": 0.0,
2324
+ "batch_tf_frac": 0.0,
2325
+ "ce_loss": 0.2901802834984665,
2326
+ "epoch": 1.9856,
2327
+ "grad_norm": 0.37890625,
2328
+ "kd_loss": 0.4553093938939094,
2329
+ "learning_rate": 3e-06,
2330
+ "loss": 0.5905,
2331
+ "masked_tokens": 121.6,
2332
+ "mean_t": 0.5045580042526125,
2333
+ "step": 930,
2334
+ "student_masked_tokens": 121.6
2335
+ },
2336
+ {
2337
+ "avg_mask_ratio": 0.49413903727240505,
2338
+ "avg_response_length": 224.79761904761904,
2339
+ "avg_student_mask_ratio": 0.49413903727240505,
2340
+ "batch_ainp_frac": 0.0,
2341
+ "batch_inp_frac": 1.0,
2342
+ "batch_inp_oh_frac": 0.0,
2343
+ "batch_inp_par_frac": 0.0,
2344
+ "batch_inp_par_par_frac": 0.0,
2345
+ "batch_inp_par_reverse_frac": 0.0,
2346
+ "batch_rl_frac": 0.0,
2347
+ "batch_sft_frac": 0.0,
2348
+ "batch_soft_sft_frac": 0.0,
2349
+ "batch_tf_frac": 0.0,
2350
+ "ce_loss": 0.37941894131193166,
2351
+ "epoch": 2.0085333333333333,
2352
+ "grad_norm": 0.4921875,
2353
+ "kd_loss": 0.4946319753903075,
2354
+ "learning_rate": 3e-06,
2355
+ "loss": 0.6668,
2356
+ "masked_tokens": 120.5,
2357
+ "mean_t": 0.5321138524893849,
2358
+ "step": 940,
2359
+ "student_masked_tokens": 120.5
2360
+ },
2361
+ {
2362
+ "avg_mask_ratio": 0.4368605303927325,
2363
+ "avg_response_length": 240.9125,
2364
+ "avg_student_mask_ratio": 0.4368605303927325,
2365
+ "batch_ainp_frac": 0.0,
2366
+ "batch_inp_frac": 1.0,
2367
+ "batch_inp_oh_frac": 0.0,
2368
+ "batch_inp_par_frac": 0.0,
2369
+ "batch_inp_par_par_frac": 0.0,
2370
+ "batch_inp_par_reverse_frac": 0.0,
2371
+ "batch_rl_frac": 0.0,
2372
+ "batch_sft_frac": 0.0,
2373
+ "batch_soft_sft_frac": 0.0,
2374
+ "batch_tf_frac": 0.0,
2375
+ "ce_loss": 0.22575005246883392,
2376
+ "epoch": 2.0298666666666665,
2377
+ "grad_norm": 1.1875,
2378
+ "kd_loss": 0.4342805288508771,
2379
+ "learning_rate": 3e-06,
2380
+ "loss": 0.5248,
2381
+ "masked_tokens": 111.4125,
2382
+ "mean_t": 0.4632946296595037,
2383
+ "step": 950,
2384
+ "student_masked_tokens": 111.4125
2385
+ },
2386
+ {
2387
+ "avg_mask_ratio": 0.4988762516761199,
2388
+ "avg_response_length": 275.3,
2389
+ "avg_student_mask_ratio": 0.4988762516761199,
2390
+ "batch_ainp_frac": 0.0,
2391
+ "batch_inp_frac": 1.0,
2392
+ "batch_inp_oh_frac": 0.0,
2393
+ "batch_inp_par_frac": 0.0,
2394
+ "batch_inp_par_par_frac": 0.0,
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": 0.49722497609602667,
2401
+ "epoch": 2.0512,
2402
+ "grad_norm": 0.40625,
2403
+ "kd_loss": 0.5839257182941765,
2404
+ "learning_rate": 3e-06,
2405
+ "loss": 0.7523,
2406
+ "masked_tokens": 143.825,
2407
+ "mean_t": 0.5198000721400604,
2408
+ "step": 960,
2409
+ "student_masked_tokens": 143.825
2410
+ },
2411
+ {
2412
+ "avg_mask_ratio": 0.437801384011982,
2413
+ "avg_response_length": 236.2375,
2414
+ "avg_student_mask_ratio": 0.437801384011982,
2415
+ "batch_ainp_frac": 0.0,
2416
+ "batch_inp_frac": 1.0,
2417
+ "batch_inp_oh_frac": 0.0,
2418
+ "batch_inp_par_frac": 0.0,
2419
+ "batch_inp_par_par_frac": 0.0,
2420
+ "batch_inp_par_reverse_frac": 0.0,
2421
+ "batch_rl_frac": 0.0,
2422
+ "batch_sft_frac": 0.0,
2423
+ "batch_soft_sft_frac": 0.0,
2424
+ "batch_tf_frac": 0.0,
2425
+ "ce_loss": 0.2855980422358698,
2426
+ "epoch": 2.0725333333333333,
2427
+ "grad_norm": 0.4765625,
2428
+ "kd_loss": 0.35673561348757377,
2429
+ "learning_rate": 3e-06,
2430
+ "loss": 0.538,
2431
+ "masked_tokens": 107.025,
2432
+ "mean_t": 0.4703940597362816,
2433
+ "step": 970,
2434
+ "student_masked_tokens": 107.025
2435
+ },
2436
+ {
2437
+ "avg_mask_ratio": 0.42220073882490394,
2438
+ "avg_response_length": 230.8625,
2439
+ "avg_student_mask_ratio": 0.42220073882490394,
2440
+ "batch_ainp_frac": 0.0,
2441
+ "batch_inp_frac": 1.0,
2442
+ "batch_inp_oh_frac": 0.0,
2443
+ "batch_inp_par_frac": 0.0,
2444
+ "batch_inp_par_par_frac": 0.0,
2445
+ "batch_inp_par_reverse_frac": 0.0,
2446
+ "batch_rl_frac": 0.0,
2447
+ "batch_sft_frac": 0.0,
2448
+ "batch_soft_sft_frac": 0.0,
2449
+ "batch_tf_frac": 0.0,
2450
+ "ce_loss": 0.2906558813129777,
2451
+ "epoch": 2.0938666666666665,
2452
+ "grad_norm": 0.466796875,
2453
+ "kd_loss": 0.36284122784349504,
2454
+ "learning_rate": 3e-06,
2455
+ "loss": 0.4889,
2456
+ "masked_tokens": 97.7875,
2457
+ "mean_t": 0.4511947895749472,
2458
+ "step": 980,
2459
+ "student_masked_tokens": 97.7875
2460
+ },
2461
+ {
2462
+ "avg_mask_ratio": 0.4605769342277199,
2463
+ "avg_response_length": 262.0375,
2464
+ "avg_student_mask_ratio": 0.4605769342277199,
2465
+ "batch_ainp_frac": 0.0,
2466
+ "batch_inp_frac": 1.0,
2467
+ "batch_inp_oh_frac": 0.0,
2468
+ "batch_inp_par_frac": 0.0,
2469
+ "batch_inp_par_par_frac": 0.0,
2470
+ "batch_inp_par_reverse_frac": 0.0,
2471
+ "batch_rl_frac": 0.0,
2472
+ "batch_sft_frac": 0.0,
2473
+ "batch_soft_sft_frac": 0.0,
2474
+ "batch_tf_frac": 0.0,
2475
+ "ce_loss": 0.18629460762591635,
2476
+ "epoch": 2.1152,
2477
+ "grad_norm": 0.625,
2478
+ "kd_loss": 0.4187604939788798,
2479
+ "learning_rate": 3e-06,
2480
+ "loss": 0.5063,
2481
+ "masked_tokens": 122.0,
2482
+ "mean_t": 0.4923786667350214,
2483
+ "step": 990,
2484
+ "student_masked_tokens": 122.0
2485
+ },
2486
+ {
2487
+ "avg_mask_ratio": 0.4547682981239632,
2488
+ "avg_response_length": 215.3,
2489
+ "avg_student_mask_ratio": 0.4547682981239632,
2490
+ "batch_ainp_frac": 0.0,
2491
+ "batch_inp_frac": 1.0,
2492
+ "batch_inp_oh_frac": 0.0,
2493
+ "batch_inp_par_frac": 0.0,
2494
+ "batch_inp_par_par_frac": 0.0,
2495
+ "batch_inp_par_reverse_frac": 0.0,
2496
+ "batch_rl_frac": 0.0,
2497
+ "batch_sft_frac": 0.0,
2498
+ "batch_soft_sft_frac": 0.0,
2499
+ "batch_tf_frac": 0.0,
2500
+ "ce_loss": 0.26735156250199454,
2501
+ "epoch": 2.1365333333333334,
2502
+ "grad_norm": 0.26953125,
2503
+ "kd_loss": 0.3440752963605235,
2504
+ "learning_rate": 3e-06,
2505
+ "loss": 0.5169,
2506
+ "masked_tokens": 100.775,
2507
+ "mean_t": 0.4773523230338469,
2508
+ "step": 1000,
2509
+ "student_masked_tokens": 100.775
2510
+ },
2511
+ {
2512
+ "avg_mask_ratio": 0.43540415074676275,
2513
+ "avg_response_length": 215.0,
2514
+ "avg_student_mask_ratio": 0.43540415074676275,
2515
+ "batch_ainp_frac": 0.0,
2516
+ "batch_inp_frac": 1.0,
2517
+ "batch_inp_oh_frac": 0.0,
2518
+ "batch_inp_par_frac": 0.0,
2519
+ "batch_inp_par_par_frac": 0.0,
2520
+ "batch_inp_par_reverse_frac": 0.0,
2521
+ "batch_rl_frac": 0.0,
2522
+ "batch_sft_frac": 0.0,
2523
+ "batch_soft_sft_frac": 0.0,
2524
+ "batch_tf_frac": 0.0,
2525
+ "ce_loss": 0.28448459618934974,
2526
+ "epoch": 2.1578666666666666,
2527
+ "grad_norm": 0.2216796875,
2528
+ "kd_loss": 0.36393369872412384,
2529
+ "learning_rate": 3e-06,
2530
+ "loss": 0.503,
2531
+ "masked_tokens": 88.65,
2532
+ "mean_t": 0.4648138735938119,
2533
+ "step": 1010,
2534
+ "student_masked_tokens": 88.65
2535
+ },
2536
+ {
2537
+ "avg_mask_ratio": 0.5063220548443497,
2538
+ "avg_response_length": 206.9125,
2539
+ "avg_student_mask_ratio": 0.5063220548443497,
2540
+ "batch_ainp_frac": 0.0,
2541
+ "batch_inp_frac": 1.0,
2542
+ "batch_inp_oh_frac": 0.0,
2543
+ "batch_inp_par_frac": 0.0,
2544
+ "batch_inp_par_par_frac": 0.0,
2545
+ "batch_inp_par_reverse_frac": 0.0,
2546
+ "batch_rl_frac": 0.0,
2547
+ "batch_sft_frac": 0.0,
2548
+ "batch_soft_sft_frac": 0.0,
2549
+ "batch_tf_frac": 0.0,
2550
+ "ce_loss": 0.3141316814458378,
2551
+ "epoch": 2.1792,
2552
+ "grad_norm": 0.328125,
2553
+ "kd_loss": 0.49756694839059035,
2554
+ "learning_rate": 3e-06,
2555
+ "loss": 0.7133,
2556
+ "masked_tokens": 110.25,
2557
+ "mean_t": 0.5327763411332853,
2558
+ "step": 1020,
2559
+ "student_masked_tokens": 110.25
2560
+ },
2561
+ {
2562
+ "avg_mask_ratio": 0.46985941788880153,
2563
+ "avg_response_length": 220.05,
2564
+ "avg_student_mask_ratio": 0.46985941788880153,
2565
+ "batch_ainp_frac": 0.0,
2566
+ "batch_inp_frac": 1.0,
2567
+ "batch_inp_oh_frac": 0.0,
2568
+ "batch_inp_par_frac": 0.0,
2569
+ "batch_inp_par_par_frac": 0.0,
2570
+ "batch_inp_par_reverse_frac": 0.0,
2571
+ "batch_rl_frac": 0.0,
2572
+ "batch_sft_frac": 0.0,
2573
+ "batch_soft_sft_frac": 0.0,
2574
+ "batch_tf_frac": 0.0,
2575
+ "ce_loss": 0.4267214318231197,
2576
+ "epoch": 2.2005333333333335,
2577
+ "grad_norm": 0.423828125,
2578
+ "kd_loss": 0.4489077641891422,
2579
+ "learning_rate": 3e-06,
2580
+ "loss": 0.6384,
2581
+ "masked_tokens": 104.9,
2582
+ "mean_t": 0.5033508580760099,
2583
+ "step": 1030,
2584
+ "student_masked_tokens": 104.9
2585
+ },
2586
+ {
2587
+ "avg_mask_ratio": 0.49566771630197765,
2588
+ "avg_response_length": 213.7,
2589
+ "avg_student_mask_ratio": 0.49566771630197765,
2590
+ "batch_ainp_frac": 0.0,
2591
+ "batch_inp_frac": 1.0,
2592
+ "batch_inp_oh_frac": 0.0,
2593
+ "batch_inp_par_frac": 0.0,
2594
+ "batch_inp_par_par_frac": 0.0,
2595
+ "batch_inp_par_reverse_frac": 0.0,
2596
+ "batch_rl_frac": 0.0,
2597
+ "batch_sft_frac": 0.0,
2598
+ "batch_soft_sft_frac": 0.0,
2599
+ "batch_tf_frac": 0.0,
2600
+ "ce_loss": 0.2765686680849626,
2601
+ "epoch": 2.2218666666666667,
2602
+ "grad_norm": 0.74609375,
2603
+ "kd_loss": 0.5419906556950081,
2604
+ "learning_rate": 3e-06,
2605
+ "loss": 0.6686,
2606
+ "masked_tokens": 100.35,
2607
+ "mean_t": 0.5349024560535327,
2608
+ "step": 1040,
2609
+ "student_masked_tokens": 100.35
2610
+ },
2611
+ {
2612
+ "avg_mask_ratio": 0.5123252369463444,
2613
+ "avg_response_length": 239.1125,
2614
+ "avg_student_mask_ratio": 0.5123252369463444,
2615
+ "batch_ainp_frac": 0.0,
2616
+ "batch_inp_frac": 1.0,
2617
+ "batch_inp_oh_frac": 0.0,
2618
+ "batch_inp_par_frac": 0.0,
2619
+ "batch_inp_par_par_frac": 0.0,
2620
+ "batch_inp_par_reverse_frac": 0.0,
2621
+ "batch_rl_frac": 0.0,
2622
+ "batch_sft_frac": 0.0,
2623
+ "batch_soft_sft_frac": 0.0,
2624
+ "batch_tf_frac": 0.0,
2625
+ "ce_loss": 0.3092239250220928,
2626
+ "epoch": 2.2432,
2627
+ "grad_norm": 0.412109375,
2628
+ "kd_loss": 0.5601085751741266,
2629
+ "learning_rate": 3e-06,
2630
+ "loss": 0.6416,
2631
+ "masked_tokens": 123.0,
2632
+ "mean_t": 0.5457118917722255,
2633
+ "step": 1050,
2634
+ "student_masked_tokens": 123.0
2635
+ },
2636
+ {
2637
+ "avg_mask_ratio": 0.46218636581033934,
2638
+ "avg_response_length": 273.7875,
2639
+ "avg_student_mask_ratio": 0.46218636581033934,
2640
+ "batch_ainp_frac": 0.0,
2641
+ "batch_inp_frac": 1.0,
2642
+ "batch_inp_oh_frac": 0.0,
2643
+ "batch_inp_par_frac": 0.0,
2644
+ "batch_inp_par_par_frac": 0.0,
2645
+ "batch_inp_par_reverse_frac": 0.0,
2646
+ "batch_rl_frac": 0.0,
2647
+ "batch_sft_frac": 0.0,
2648
+ "batch_soft_sft_frac": 0.0,
2649
+ "batch_tf_frac": 0.0,
2650
+ "ce_loss": 0.33267747830594485,
2651
+ "epoch": 2.2645333333333335,
2652
+ "grad_norm": 0.3984375,
2653
+ "kd_loss": 0.49028674410892564,
2654
+ "learning_rate": 3e-06,
2655
+ "loss": 0.5827,
2656
+ "masked_tokens": 122.125,
2657
+ "mean_t": 0.48194136443780733,
2658
+ "step": 1060,
2659
+ "student_masked_tokens": 122.125
2660
+ },
2661
+ {
2662
+ "avg_mask_ratio": 0.48546303423354403,
2663
+ "avg_response_length": 260.8125,
2664
+ "avg_student_mask_ratio": 0.48546303423354403,
2665
+ "batch_ainp_frac": 0.0,
2666
+ "batch_inp_frac": 1.0,
2667
+ "batch_inp_oh_frac": 0.0,
2668
+ "batch_inp_par_frac": 0.0,
2669
+ "batch_inp_par_par_frac": 0.0,
2670
+ "batch_inp_par_reverse_frac": 0.0,
2671
+ "batch_rl_frac": 0.0,
2672
+ "batch_sft_frac": 0.0,
2673
+ "batch_soft_sft_frac": 0.0,
2674
+ "batch_tf_frac": 0.0,
2675
+ "ce_loss": 0.4495345233380249,
2676
+ "epoch": 2.2858666666666667,
2677
+ "grad_norm": 0.345703125,
2678
+ "kd_loss": 0.422707377332182,
2679
+ "learning_rate": 3e-06,
2680
+ "loss": 0.6038,
2681
+ "masked_tokens": 138.0375,
2682
+ "mean_t": 0.5015889146190602,
2683
+ "step": 1070,
2684
+ "student_masked_tokens": 138.0375
2685
+ },
2686
+ {
2687
+ "avg_mask_ratio": 0.4625250873621553,
2688
+ "avg_response_length": 219.8875,
2689
+ "avg_student_mask_ratio": 0.4625250873621553,
2690
+ "batch_ainp_frac": 0.0,
2691
+ "batch_inp_frac": 1.0,
2692
+ "batch_inp_oh_frac": 0.0,
2693
+ "batch_inp_par_frac": 0.0,
2694
+ "batch_inp_par_par_frac": 0.0,
2695
+ "batch_inp_par_reverse_frac": 0.0,
2696
+ "batch_rl_frac": 0.0,
2697
+ "batch_sft_frac": 0.0,
2698
+ "batch_soft_sft_frac": 0.0,
2699
+ "batch_tf_frac": 0.0,
2700
+ "ce_loss": 0.3075333785695989,
2701
+ "epoch": 2.3072,
2702
+ "grad_norm": 0.087890625,
2703
+ "kd_loss": 0.3667106795892323,
2704
+ "learning_rate": 3e-06,
2705
+ "loss": 0.4549,
2706
+ "masked_tokens": 100.5625,
2707
+ "mean_t": 0.4983203248586506,
2708
+ "step": 1080,
2709
+ "student_masked_tokens": 100.5625
2710
+ },
2711
+ {
2712
+ "avg_mask_ratio": 0.44248262273031286,
2713
+ "avg_response_length": 213.55,
2714
+ "avg_student_mask_ratio": 0.44248262273031286,
2715
+ "batch_ainp_frac": 0.0,
2716
+ "batch_inp_frac": 1.0,
2717
+ "batch_inp_oh_frac": 0.0,
2718
+ "batch_inp_par_frac": 0.0,
2719
+ "batch_inp_par_par_frac": 0.0,
2720
+ "batch_inp_par_reverse_frac": 0.0,
2721
+ "batch_rl_frac": 0.0,
2722
+ "batch_sft_frac": 0.0,
2723
+ "batch_soft_sft_frac": 0.0,
2724
+ "batch_tf_frac": 0.0,
2725
+ "ce_loss": 0.24968633314620092,
2726
+ "epoch": 2.3285333333333336,
2727
+ "grad_norm": 0.11572265625,
2728
+ "kd_loss": 0.3935246549681978,
2729
+ "learning_rate": 3e-06,
2730
+ "loss": 0.5053,
2731
+ "masked_tokens": 91.9125,
2732
+ "mean_t": 0.47094749807147307,
2733
+ "step": 1090,
2734
+ "student_masked_tokens": 91.9125
2735
+ },
2736
+ {
2737
+ "avg_mask_ratio": 0.5204601250356063,
2738
+ "avg_response_length": 246.1125,
2739
+ "avg_student_mask_ratio": 0.5204601250356063,
2740
+ "batch_ainp_frac": 0.0,
2741
+ "batch_inp_frac": 1.0,
2742
+ "batch_inp_oh_frac": 0.0,
2743
+ "batch_inp_par_frac": 0.0,
2744
+ "batch_inp_par_par_frac": 0.0,
2745
+ "batch_inp_par_reverse_frac": 0.0,
2746
+ "batch_rl_frac": 0.0,
2747
+ "batch_sft_frac": 0.0,
2748
+ "batch_soft_sft_frac": 0.0,
2749
+ "batch_tf_frac": 0.0,
2750
+ "ce_loss": 0.40365264619552477,
2751
+ "epoch": 2.3498666666666668,
2752
+ "grad_norm": 0.37109375,
2753
+ "kd_loss": 0.4355207666182196,
2754
+ "learning_rate": 3e-06,
2755
+ "loss": 0.6746,
2756
+ "masked_tokens": 133.1875,
2757
+ "mean_t": 0.5531192034482956,
2758
+ "step": 1100,
2759
+ "student_masked_tokens": 133.1875
2760
+ },
2761
+ {
2762
+ "avg_mask_ratio": 0.447697223268915,
2763
+ "avg_response_length": 226.6375,
2764
+ "avg_student_mask_ratio": 0.447697223268915,
2765
+ "batch_ainp_frac": 0.0,
2766
+ "batch_inp_frac": 1.0,
2767
+ "batch_inp_oh_frac": 0.0,
2768
+ "batch_inp_par_frac": 0.0,
2769
+ "batch_inp_par_par_frac": 0.0,
2770
+ "batch_inp_par_reverse_frac": 0.0,
2771
+ "batch_rl_frac": 0.0,
2772
+ "batch_sft_frac": 0.0,
2773
+ "batch_soft_sft_frac": 0.0,
2774
+ "batch_tf_frac": 0.0,
2775
+ "ce_loss": 0.323583398951439,
2776
+ "epoch": 2.3712,
2777
+ "grad_norm": 1.0625,
2778
+ "kd_loss": 0.3599243894114693,
2779
+ "learning_rate": 3e-06,
2780
+ "loss": 0.5035,
2781
+ "masked_tokens": 103.475,
2782
+ "mean_t": 0.4757364276825683,
2783
+ "step": 1110,
2784
+ "student_masked_tokens": 103.475
2785
+ },
2786
+ {
2787
+ "avg_mask_ratio": 0.4670829998096451,
2788
+ "avg_response_length": 245.0375,
2789
+ "avg_student_mask_ratio": 0.4670829998096451,
2790
+ "batch_ainp_frac": 0.0,
2791
+ "batch_inp_frac": 1.0,
2792
+ "batch_inp_oh_frac": 0.0,
2793
+ "batch_inp_par_frac": 0.0,
2794
+ "batch_inp_par_par_frac": 0.0,
2795
+ "batch_inp_par_reverse_frac": 0.0,
2796
+ "batch_rl_frac": 0.0,
2797
+ "batch_sft_frac": 0.0,
2798
+ "batch_soft_sft_frac": 0.0,
2799
+ "batch_tf_frac": 0.0,
2800
+ "ce_loss": 0.2362255901227968,
2801
+ "epoch": 2.392533333333333,
2802
+ "grad_norm": 0.91015625,
2803
+ "kd_loss": 0.350646685710654,
2804
+ "learning_rate": 3e-06,
2805
+ "loss": 0.5669,
2806
+ "masked_tokens": 114.175,
2807
+ "mean_t": 0.5013068238971755,
2808
+ "step": 1120,
2809
+ "student_masked_tokens": 114.175
2810
+ },
2811
+ {
2812
+ "avg_mask_ratio": 0.5008096542558633,
2813
+ "avg_response_length": 273.125,
2814
+ "avg_student_mask_ratio": 0.5008096542558633,
2815
+ "batch_ainp_frac": 0.0,
2816
+ "batch_inp_frac": 1.0,
2817
+ "batch_inp_oh_frac": 0.0,
2818
+ "batch_inp_par_frac": 0.0,
2819
+ "batch_inp_par_par_frac": 0.0,
2820
+ "batch_inp_par_reverse_frac": 0.0,
2821
+ "batch_rl_frac": 0.0,
2822
+ "batch_sft_frac": 0.0,
2823
+ "batch_soft_sft_frac": 0.0,
2824
+ "batch_tf_frac": 0.0,
2825
+ "ce_loss": 0.45858247512928757,
2826
+ "epoch": 2.413866666666667,
2827
+ "grad_norm": 0.8359375,
2828
+ "kd_loss": 0.5058724594353151,
2829
+ "learning_rate": 3e-06,
2830
+ "loss": 0.6836,
2831
+ "masked_tokens": 144.2625,
2832
+ "mean_t": 0.5303254407714121,
2833
+ "step": 1130,
2834
+ "student_masked_tokens": 144.2625
2835
+ },
2836
+ {
2837
+ "avg_mask_ratio": 0.4456572526367381,
2838
+ "avg_response_length": 217.9125,
2839
+ "avg_student_mask_ratio": 0.4456572526367381,
2840
+ "batch_ainp_frac": 0.0,
2841
+ "batch_inp_frac": 1.0,
2842
+ "batch_inp_oh_frac": 0.0,
2843
+ "batch_inp_par_frac": 0.0,
2844
+ "batch_inp_par_par_frac": 0.0,
2845
+ "batch_inp_par_reverse_frac": 0.0,
2846
+ "batch_rl_frac": 0.0,
2847
+ "batch_sft_frac": 0.0,
2848
+ "batch_soft_sft_frac": 0.0,
2849
+ "batch_tf_frac": 0.0,
2850
+ "ce_loss": 0.20887070082535503,
2851
+ "epoch": 2.4352,
2852
+ "grad_norm": 7.21875,
2853
+ "kd_loss": 0.44113304006314136,
2854
+ "learning_rate": 3e-06,
2855
+ "loss": 0.5599,
2856
+ "masked_tokens": 103.125,
2857
+ "mean_t": 0.4845335395424627,
2858
+ "step": 1140,
2859
+ "student_masked_tokens": 103.125
2860
+ },
2861
+ {
2862
+ "avg_mask_ratio": 0.5291026248247362,
2863
+ "avg_response_length": 212.5875,
2864
+ "avg_student_mask_ratio": 0.5291026248247362,
2865
+ "batch_ainp_frac": 0.0,
2866
+ "batch_inp_frac": 1.0,
2867
+ "batch_inp_oh_frac": 0.0,
2868
+ "batch_inp_par_frac": 0.0,
2869
+ "batch_inp_par_par_frac": 0.0,
2870
+ "batch_inp_par_reverse_frac": 0.0,
2871
+ "batch_rl_frac": 0.0,
2872
+ "batch_sft_frac": 0.0,
2873
+ "batch_soft_sft_frac": 0.0,
2874
+ "batch_tf_frac": 0.0,
2875
+ "ce_loss": 0.3553736373823654,
2876
+ "epoch": 2.4565333333333332,
2877
+ "grad_norm": 0.365234375,
2878
+ "kd_loss": 0.4455350318000953,
2879
+ "learning_rate": 3e-06,
2880
+ "loss": 0.7113,
2881
+ "masked_tokens": 110.025,
2882
+ "mean_t": 0.5690932425903157,
2883
+ "step": 1150,
2884
+ "student_masked_tokens": 110.025
2885
+ },
2886
+ {
2887
+ "avg_mask_ratio": 0.46949602509848776,
2888
+ "avg_response_length": 229.475,
2889
+ "avg_student_mask_ratio": 0.46949602509848776,
2890
+ "batch_ainp_frac": 0.0,
2891
+ "batch_inp_frac": 1.0,
2892
+ "batch_inp_oh_frac": 0.0,
2893
+ "batch_inp_par_frac": 0.0,
2894
+ "batch_inp_par_par_frac": 0.0,
2895
+ "batch_inp_par_reverse_frac": 0.0,
2896
+ "batch_rl_frac": 0.0,
2897
+ "batch_sft_frac": 0.0,
2898
+ "batch_soft_sft_frac": 0.0,
2899
+ "batch_tf_frac": 0.0,
2900
+ "ce_loss": 0.3682716753066188,
2901
+ "epoch": 2.4778666666666664,
2902
+ "grad_norm": 0.98828125,
2903
+ "kd_loss": 0.4137534074947894,
2904
+ "learning_rate": 3e-06,
2905
+ "loss": 0.5178,
2906
+ "masked_tokens": 104.4375,
2907
+ "mean_t": 0.5040684466948733,
2908
+ "step": 1160,
2909
+ "student_masked_tokens": 104.4375
2910
+ },
2911
+ {
2912
+ "avg_mask_ratio": 0.480710746452678,
2913
+ "avg_response_length": 247.7625,
2914
+ "avg_student_mask_ratio": 0.480710746452678,
2915
+ "batch_ainp_frac": 0.0,
2916
+ "batch_inp_frac": 1.0,
2917
+ "batch_inp_oh_frac": 0.0,
2918
+ "batch_inp_par_frac": 0.0,
2919
+ "batch_inp_par_par_frac": 0.0,
2920
+ "batch_inp_par_reverse_frac": 0.0,
2921
+ "batch_rl_frac": 0.0,
2922
+ "batch_sft_frac": 0.0,
2923
+ "batch_soft_sft_frac": 0.0,
2924
+ "batch_tf_frac": 0.0,
2925
+ "ce_loss": 0.3276487763844216,
2926
+ "epoch": 2.4992,
2927
+ "grad_norm": 0.67578125,
2928
+ "kd_loss": 0.39728453117754725,
2929
+ "learning_rate": 3e-06,
2930
+ "loss": 0.5996,
2931
+ "masked_tokens": 123.9125,
2932
+ "mean_t": 0.5114516971167177,
2933
+ "step": 1170,
2934
+ "student_masked_tokens": 123.9125
2935
+ },
2936
+ {
2937
+ "avg_mask_ratio": 0.41960311922593974,
2938
+ "avg_response_length": 215.025,
2939
+ "avg_student_mask_ratio": 0.41960311922593974,
2940
+ "batch_ainp_frac": 0.0,
2941
+ "batch_inp_frac": 1.0,
2942
+ "batch_inp_oh_frac": 0.0,
2943
+ "batch_inp_par_frac": 0.0,
2944
+ "batch_inp_par_par_frac": 0.0,
2945
+ "batch_inp_par_reverse_frac": 0.0,
2946
+ "batch_rl_frac": 0.0,
2947
+ "batch_sft_frac": 0.0,
2948
+ "batch_soft_sft_frac": 0.0,
2949
+ "batch_tf_frac": 0.0,
2950
+ "ce_loss": 0.3255401685638958,
2951
+ "epoch": 2.5205333333333333,
2952
+ "grad_norm": 0.69140625,
2953
+ "kd_loss": 0.31952214344050844,
2954
+ "learning_rate": 3e-06,
2955
+ "loss": 0.5148,
2956
+ "masked_tokens": 85.3625,
2957
+ "mean_t": 0.4491677140351385,
2958
+ "step": 1180,
2959
+ "student_masked_tokens": 85.3625
2960
+ },
2961
+ {
2962
+ "avg_mask_ratio": 0.522994744987227,
2963
+ "avg_response_length": 220.3375,
2964
+ "avg_student_mask_ratio": 0.522994744987227,
2965
+ "batch_ainp_frac": 0.0,
2966
+ "batch_inp_frac": 1.0,
2967
+ "batch_inp_oh_frac": 0.0,
2968
+ "batch_inp_par_frac": 0.0,
2969
+ "batch_inp_par_par_frac": 0.0,
2970
+ "batch_inp_par_reverse_frac": 0.0,
2971
+ "batch_rl_frac": 0.0,
2972
+ "batch_sft_frac": 0.0,
2973
+ "batch_soft_sft_frac": 0.0,
2974
+ "batch_tf_frac": 0.0,
2975
+ "ce_loss": 0.46823025978692384,
2976
+ "epoch": 2.5418666666666665,
2977
+ "grad_norm": 0.87890625,
2978
+ "kd_loss": 0.48877327183990926,
2979
+ "learning_rate": 3e-06,
2980
+ "loss": 0.7125,
2981
+ "masked_tokens": 124.2875,
2982
+ "mean_t": 0.5590635397238657,
2983
+ "step": 1190,
2984
+ "student_masked_tokens": 124.2875
2985
+ },
2986
+ {
2987
+ "avg_mask_ratio": 0.4730891800048994,
2988
+ "avg_response_length": 215.675,
2989
+ "avg_student_mask_ratio": 0.4730891800048994,
2990
+ "batch_ainp_frac": 0.0,
2991
+ "batch_inp_frac": 1.0,
2992
+ "batch_inp_oh_frac": 0.0,
2993
+ "batch_inp_par_frac": 0.0,
2994
+ "batch_inp_par_par_frac": 0.0,
2995
+ "batch_inp_par_reverse_frac": 0.0,
2996
+ "batch_rl_frac": 0.0,
2997
+ "batch_sft_frac": 0.0,
2998
+ "batch_soft_sft_frac": 0.0,
2999
+ "batch_tf_frac": 0.0,
3000
+ "ce_loss": 0.25984036189172455,
3001
+ "epoch": 2.5632,
3002
+ "grad_norm": 0.34765625,
3003
+ "kd_loss": 0.3999250433967063,
3004
+ "learning_rate": 3e-06,
3005
+ "loss": 0.5566,
3006
+ "masked_tokens": 98.775,
3007
+ "mean_t": 0.506370971655997,
3008
+ "step": 1200,
3009
+ "student_masked_tokens": 98.775
3010
+ }
3011
+ ],
3012
+ "logging_steps": 10,
3013
+ "max_steps": 1404,
3014
+ "num_input_tokens_seen": 0,
3015
+ "num_train_epochs": 3,
3016
+ "save_steps": 100,
3017
+ "stateful_callbacks": {
3018
+ "TrainerControl": {
3019
+ "args": {
3020
+ "should_epoch_stop": false,
3021
+ "should_evaluate": false,
3022
+ "should_log": false,
3023
+ "should_save": true,
3024
+ "should_training_stop": false
3025
+ },
3026
+ "attributes": {}
3027
+ }
3028
+ },
3029
+ "total_flos": 0.0,
3030
+ "train_batch_size": 1,
3031
+ "trial_name": null,
3032
+ "trial_params": null
3033
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-1200/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89ae704e41a4a62f6ca56789c45ce45887326cd2f8d1e97e398e5ada4a93398c
3
+ size 8312
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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
+ "gate_proj",
28
+ "down_proj",
29
+ "up_proj",
30
+ "o_proj",
31
+ "q_proj",
32
+ "v_proj",
33
+ "k_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:373a5e8685ef586dfb53f3d0e5d25bce673968b92f9aa3aa9a0b664f70d1aad9
3
+ size 2406624648
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5dad61a855c3db0b23d7549282c2eb83eae465eeae3ad5b3c593adc89daba8b5
3
+ size 671304442
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e3d1d06bd17db661594f307eb1a293c09413ad06c18d1facb6bda1bbe2f3940a
3
+ size 14512
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e69b6548f92233a5d3cb22aa7c60d5d4d1f37d04ac969eb521f7a7c36271ae54
3
+ size 14512
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/trainer_state.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 0.4931091487989761,
13
+ "avg_response_length": 225.975,
14
+ "avg_student_mask_ratio": 0.4931091487989761,
15
+ "batch_ainp_frac": 0.0,
16
+ "batch_inp_frac": 1.0,
17
+ "batch_inp_oh_frac": 0.0,
18
+ "batch_inp_par_frac": 0.0,
19
+ "batch_inp_par_par_frac": 0.0,
20
+ "batch_inp_par_reverse_frac": 0.0,
21
+ "batch_rl_frac": 0.0,
22
+ "batch_sft_frac": 0.0,
23
+ "batch_soft_sft_frac": 0.0,
24
+ "batch_tf_frac": 0.0,
25
+ "ce_loss": 0.5570551689027979,
26
+ "epoch": 0.021333333333333333,
27
+ "grad_norm": 0.404296875,
28
+ "kd_loss": 0.5375588692116253,
29
+ "learning_rate": 3e-06,
30
+ "loss": 0.8247,
31
+ "masked_tokens": 111.95,
32
+ "mean_t": 0.5145528071501758,
33
+ "step": 10,
34
+ "student_masked_tokens": 111.95
35
+ },
36
+ {
37
+ "avg_mask_ratio": 0.42058031369233506,
38
+ "avg_response_length": 255.2625,
39
+ "avg_student_mask_ratio": 0.42058031369233506,
40
+ "batch_ainp_frac": 0.0,
41
+ "batch_inp_frac": 1.0,
42
+ "batch_inp_oh_frac": 0.0,
43
+ "batch_inp_par_frac": 0.0,
44
+ "batch_inp_par_par_frac": 0.0,
45
+ "batch_inp_par_reverse_frac": 0.0,
46
+ "batch_rl_frac": 0.0,
47
+ "batch_sft_frac": 0.0,
48
+ "batch_soft_sft_frac": 0.0,
49
+ "batch_tf_frac": 0.0,
50
+ "ce_loss": 0.42826092825978324,
51
+ "epoch": 0.042666666666666665,
52
+ "grad_norm": 0.8046875,
53
+ "kd_loss": 0.4450965437417761,
54
+ "learning_rate": 3e-06,
55
+ "loss": 0.5724,
56
+ "masked_tokens": 97.35,
57
+ "mean_t": 0.43874448732240123,
58
+ "step": 20,
59
+ "student_masked_tokens": 97.35
60
+ },
61
+ {
62
+ "avg_mask_ratio": 0.4538542575784959,
63
+ "avg_response_length": 211.7625,
64
+ "avg_student_mask_ratio": 0.4538542575784959,
65
+ "batch_ainp_frac": 0.0,
66
+ "batch_inp_frac": 1.0,
67
+ "batch_inp_oh_frac": 0.0,
68
+ "batch_inp_par_frac": 0.0,
69
+ "batch_inp_par_par_frac": 0.0,
70
+ "batch_inp_par_reverse_frac": 0.0,
71
+ "batch_rl_frac": 0.0,
72
+ "batch_sft_frac": 0.0,
73
+ "batch_soft_sft_frac": 0.0,
74
+ "batch_tf_frac": 0.0,
75
+ "ce_loss": 0.4461815005188782,
76
+ "epoch": 0.064,
77
+ "grad_norm": 0.50390625,
78
+ "kd_loss": 0.5296064364436825,
79
+ "learning_rate": 3e-06,
80
+ "loss": 0.702,
81
+ "masked_tokens": 110.2,
82
+ "mean_t": 0.4803953981841914,
83
+ "step": 30,
84
+ "student_masked_tokens": 110.2
85
+ },
86
+ {
87
+ "avg_mask_ratio": 0.4207469140383182,
88
+ "avg_response_length": 224.125,
89
+ "avg_student_mask_ratio": 0.4207469140383182,
90
+ "batch_ainp_frac": 0.0,
91
+ "batch_inp_frac": 1.0,
92
+ "batch_inp_oh_frac": 0.0,
93
+ "batch_inp_par_frac": 0.0,
94
+ "batch_inp_par_par_frac": 0.0,
95
+ "batch_inp_par_reverse_frac": 0.0,
96
+ "batch_rl_frac": 0.0,
97
+ "batch_sft_frac": 0.0,
98
+ "batch_soft_sft_frac": 0.0,
99
+ "batch_tf_frac": 0.0,
100
+ "ce_loss": 0.38499989152683156,
101
+ "epoch": 0.08533333333333333,
102
+ "grad_norm": 1.671875,
103
+ "kd_loss": 0.33118802310931417,
104
+ "learning_rate": 3e-06,
105
+ "loss": 0.5529,
106
+ "masked_tokens": 98.1625,
107
+ "mean_t": 0.4569831106782658,
108
+ "step": 40,
109
+ "student_masked_tokens": 98.1625
110
+ },
111
+ {
112
+ "avg_mask_ratio": 0.43260439952719026,
113
+ "avg_response_length": 207.125,
114
+ "avg_student_mask_ratio": 0.43260439952719026,
115
+ "batch_ainp_frac": 0.0,
116
+ "batch_inp_frac": 1.0,
117
+ "batch_inp_oh_frac": 0.0,
118
+ "batch_inp_par_frac": 0.0,
119
+ "batch_inp_par_par_frac": 0.0,
120
+ "batch_inp_par_reverse_frac": 0.0,
121
+ "batch_rl_frac": 0.0,
122
+ "batch_sft_frac": 0.0,
123
+ "batch_soft_sft_frac": 0.0,
124
+ "batch_tf_frac": 0.0,
125
+ "ce_loss": 0.5240421466317912,
126
+ "epoch": 0.10666666666666667,
127
+ "grad_norm": 1.6875,
128
+ "kd_loss": 0.4270985169670894,
129
+ "learning_rate": 3e-06,
130
+ "loss": 0.671,
131
+ "masked_tokens": 85.05,
132
+ "mean_t": 0.4612453707959503,
133
+ "step": 50,
134
+ "student_masked_tokens": 85.05
135
+ },
136
+ {
137
+ "avg_mask_ratio": 0.46053453313652426,
138
+ "avg_response_length": 251.0875,
139
+ "avg_student_mask_ratio": 0.46053453313652426,
140
+ "batch_ainp_frac": 0.0,
141
+ "batch_inp_frac": 1.0,
142
+ "batch_inp_oh_frac": 0.0,
143
+ "batch_inp_par_frac": 0.0,
144
+ "batch_inp_par_par_frac": 0.0,
145
+ "batch_inp_par_reverse_frac": 0.0,
146
+ "batch_rl_frac": 0.0,
147
+ "batch_sft_frac": 0.0,
148
+ "batch_soft_sft_frac": 0.0,
149
+ "batch_tf_frac": 0.0,
150
+ "ce_loss": 0.5027546818272185,
151
+ "epoch": 0.128,
152
+ "grad_norm": 0.17578125,
153
+ "kd_loss": 0.3904111967755945,
154
+ "learning_rate": 3e-06,
155
+ "loss": 0.6672,
156
+ "masked_tokens": 120.9,
157
+ "mean_t": 0.48597636765334756,
158
+ "step": 60,
159
+ "student_masked_tokens": 120.9
160
+ },
161
+ {
162
+ "avg_mask_ratio": 0.5112146578729153,
163
+ "avg_response_length": 202.5875,
164
+ "avg_student_mask_ratio": 0.5112146578729153,
165
+ "batch_ainp_frac": 0.0,
166
+ "batch_inp_frac": 1.0,
167
+ "batch_inp_oh_frac": 0.0,
168
+ "batch_inp_par_frac": 0.0,
169
+ "batch_inp_par_par_frac": 0.0,
170
+ "batch_inp_par_reverse_frac": 0.0,
171
+ "batch_rl_frac": 0.0,
172
+ "batch_sft_frac": 0.0,
173
+ "batch_soft_sft_frac": 0.0,
174
+ "batch_tf_frac": 0.0,
175
+ "ce_loss": 0.7753003867959023,
176
+ "epoch": 0.14933333333333335,
177
+ "grad_norm": 0.953125,
178
+ "kd_loss": 0.4415664039527428,
179
+ "learning_rate": 3e-06,
180
+ "loss": 0.856,
181
+ "masked_tokens": 104.5875,
182
+ "mean_t": 0.5459650319069624,
183
+ "step": 70,
184
+ "student_masked_tokens": 104.5875
185
+ },
186
+ {
187
+ "avg_mask_ratio": 0.37548826879356056,
188
+ "avg_response_length": 225.85,
189
+ "avg_student_mask_ratio": 0.37548826879356056,
190
+ "batch_ainp_frac": 0.0,
191
+ "batch_inp_frac": 1.0,
192
+ "batch_inp_oh_frac": 0.0,
193
+ "batch_inp_par_frac": 0.0,
194
+ "batch_inp_par_par_frac": 0.0,
195
+ "batch_inp_par_reverse_frac": 0.0,
196
+ "batch_rl_frac": 0.0,
197
+ "batch_sft_frac": 0.0,
198
+ "batch_soft_sft_frac": 0.0,
199
+ "batch_tf_frac": 0.0,
200
+ "ce_loss": 0.3791731233859082,
201
+ "epoch": 0.17066666666666666,
202
+ "grad_norm": 0.1552734375,
203
+ "kd_loss": 0.31052538527774515,
204
+ "learning_rate": 3e-06,
205
+ "loss": 0.4843,
206
+ "masked_tokens": 85.0625,
207
+ "mean_t": 0.40758824030635876,
208
+ "step": 80,
209
+ "student_masked_tokens": 85.0625
210
+ },
211
+ {
212
+ "avg_mask_ratio": 0.5001560213277116,
213
+ "avg_response_length": 229.75,
214
+ "avg_student_mask_ratio": 0.5001560213277116,
215
+ "batch_ainp_frac": 0.0,
216
+ "batch_inp_frac": 1.0,
217
+ "batch_inp_oh_frac": 0.0,
218
+ "batch_inp_par_frac": 0.0,
219
+ "batch_inp_par_par_frac": 0.0,
220
+ "batch_inp_par_reverse_frac": 0.0,
221
+ "batch_rl_frac": 0.0,
222
+ "batch_sft_frac": 0.0,
223
+ "batch_soft_sft_frac": 0.0,
224
+ "batch_tf_frac": 0.0,
225
+ "ce_loss": 0.6899960007944174,
226
+ "epoch": 0.192,
227
+ "grad_norm": 1.25,
228
+ "kd_loss": 0.5995283465861896,
229
+ "learning_rate": 3e-06,
230
+ "loss": 0.9721,
231
+ "masked_tokens": 107.6625,
232
+ "mean_t": 0.5297661645396147,
233
+ "step": 90,
234
+ "student_masked_tokens": 107.6625
235
+ },
236
+ {
237
+ "avg_mask_ratio": 0.4576045103633078,
238
+ "avg_response_length": 208.0,
239
+ "avg_student_mask_ratio": 0.4576045103633078,
240
+ "batch_ainp_frac": 0.0,
241
+ "batch_inp_frac": 1.0,
242
+ "batch_inp_oh_frac": 0.0,
243
+ "batch_inp_par_frac": 0.0,
244
+ "batch_inp_par_par_frac": 0.0,
245
+ "batch_inp_par_reverse_frac": 0.0,
246
+ "batch_rl_frac": 0.0,
247
+ "batch_sft_frac": 0.0,
248
+ "batch_soft_sft_frac": 0.0,
249
+ "batch_tf_frac": 0.0,
250
+ "ce_loss": 0.41132245859021166,
251
+ "epoch": 0.21333333333333335,
252
+ "grad_norm": 0.64453125,
253
+ "kd_loss": 0.3813956479015957,
254
+ "learning_rate": 3e-06,
255
+ "loss": 0.6635,
256
+ "masked_tokens": 104.1625,
257
+ "mean_t": 0.4886587227345444,
258
+ "step": 100,
259
+ "student_masked_tokens": 104.1625
260
+ },
261
+ {
262
+ "avg_mask_ratio": 0.4877026333590038,
263
+ "avg_response_length": 213.0875,
264
+ "avg_student_mask_ratio": 0.4877026333590038,
265
+ "batch_ainp_frac": 0.0,
266
+ "batch_inp_frac": 1.0,
267
+ "batch_inp_oh_frac": 0.0,
268
+ "batch_inp_par_frac": 0.0,
269
+ "batch_inp_par_par_frac": 0.0,
270
+ "batch_inp_par_reverse_frac": 0.0,
271
+ "batch_rl_frac": 0.0,
272
+ "batch_sft_frac": 0.0,
273
+ "batch_soft_sft_frac": 0.0,
274
+ "batch_tf_frac": 0.0,
275
+ "ce_loss": 0.4612084587922368,
276
+ "epoch": 0.23466666666666666,
277
+ "grad_norm": 0.64453125,
278
+ "kd_loss": 0.5074845846289577,
279
+ "learning_rate": 3e-06,
280
+ "loss": 0.7993,
281
+ "masked_tokens": 102.075,
282
+ "mean_t": 0.5246987929102034,
283
+ "step": 110,
284
+ "student_masked_tokens": 102.075
285
+ },
286
+ {
287
+ "avg_mask_ratio": 0.45146879020612685,
288
+ "avg_response_length": 224.1875,
289
+ "avg_student_mask_ratio": 0.45146879020612685,
290
+ "batch_ainp_frac": 0.0,
291
+ "batch_inp_frac": 1.0,
292
+ "batch_inp_oh_frac": 0.0,
293
+ "batch_inp_par_frac": 0.0,
294
+ "batch_inp_par_par_frac": 0.0,
295
+ "batch_inp_par_reverse_frac": 0.0,
296
+ "batch_rl_frac": 0.0,
297
+ "batch_sft_frac": 0.0,
298
+ "batch_soft_sft_frac": 0.0,
299
+ "batch_tf_frac": 0.0,
300
+ "ce_loss": 0.3276976759495483,
301
+ "epoch": 0.256,
302
+ "grad_norm": 0.30078125,
303
+ "kd_loss": 0.41461311469229256,
304
+ "learning_rate": 3e-06,
305
+ "loss": 0.6088,
306
+ "masked_tokens": 100.525,
307
+ "mean_t": 0.4805434140143916,
308
+ "step": 120,
309
+ "student_masked_tokens": 100.525
310
+ },
311
+ {
312
+ "avg_mask_ratio": 0.4356566035945434,
313
+ "avg_response_length": 202.7,
314
+ "avg_student_mask_ratio": 0.4356566035945434,
315
+ "batch_ainp_frac": 0.0,
316
+ "batch_inp_frac": 1.0,
317
+ "batch_inp_oh_frac": 0.0,
318
+ "batch_inp_par_frac": 0.0,
319
+ "batch_inp_par_par_frac": 0.0,
320
+ "batch_inp_par_reverse_frac": 0.0,
321
+ "batch_rl_frac": 0.0,
322
+ "batch_sft_frac": 0.0,
323
+ "batch_soft_sft_frac": 0.0,
324
+ "batch_tf_frac": 0.0,
325
+ "ce_loss": 0.12710368948505674,
326
+ "epoch": 0.2773333333333333,
327
+ "grad_norm": 0.490234375,
328
+ "kd_loss": 0.23057804748218585,
329
+ "learning_rate": 3e-06,
330
+ "loss": 0.384,
331
+ "masked_tokens": 89.5625,
332
+ "mean_t": 0.47522516988683494,
333
+ "step": 130,
334
+ "student_masked_tokens": 89.5625
335
+ },
336
+ {
337
+ "avg_mask_ratio": 0.49419954856857656,
338
+ "avg_response_length": 255.625,
339
+ "avg_student_mask_ratio": 0.49419954856857656,
340
+ "batch_ainp_frac": 0.0,
341
+ "batch_inp_frac": 1.0,
342
+ "batch_inp_oh_frac": 0.0,
343
+ "batch_inp_par_frac": 0.0,
344
+ "batch_inp_par_par_frac": 0.0,
345
+ "batch_inp_par_reverse_frac": 0.0,
346
+ "batch_rl_frac": 0.0,
347
+ "batch_sft_frac": 0.0,
348
+ "batch_soft_sft_frac": 0.0,
349
+ "batch_tf_frac": 0.0,
350
+ "ce_loss": 0.48596099013025196,
351
+ "epoch": 0.2986666666666667,
352
+ "grad_norm": 0.859375,
353
+ "kd_loss": 0.5025483015746885,
354
+ "learning_rate": 3e-06,
355
+ "loss": 0.7892,
356
+ "masked_tokens": 136.575,
357
+ "mean_t": 0.5204090005659964,
358
+ "step": 140,
359
+ "student_masked_tokens": 136.575
360
+ },
361
+ {
362
+ "avg_mask_ratio": 0.4736677930341102,
363
+ "avg_response_length": 255.375,
364
+ "avg_student_mask_ratio": 0.4736677930341102,
365
+ "batch_ainp_frac": 0.0,
366
+ "batch_inp_frac": 1.0,
367
+ "batch_inp_oh_frac": 0.0,
368
+ "batch_inp_par_frac": 0.0,
369
+ "batch_inp_par_par_frac": 0.0,
370
+ "batch_inp_par_reverse_frac": 0.0,
371
+ "batch_rl_frac": 0.0,
372
+ "batch_sft_frac": 0.0,
373
+ "batch_soft_sft_frac": 0.0,
374
+ "batch_tf_frac": 0.0,
375
+ "ce_loss": 0.5493089448234059,
376
+ "epoch": 0.32,
377
+ "grad_norm": 0.076171875,
378
+ "kd_loss": 0.4892602212316547,
379
+ "learning_rate": 3e-06,
380
+ "loss": 0.7918,
381
+ "masked_tokens": 126.575,
382
+ "mean_t": 0.5012552456930279,
383
+ "step": 150,
384
+ "student_masked_tokens": 126.575
385
+ },
386
+ {
387
+ "avg_mask_ratio": 0.5032523009285796,
388
+ "avg_response_length": 209.325,
389
+ "avg_student_mask_ratio": 0.5032523009285796,
390
+ "batch_ainp_frac": 0.0,
391
+ "batch_inp_frac": 1.0,
392
+ "batch_inp_oh_frac": 0.0,
393
+ "batch_inp_par_frac": 0.0,
394
+ "batch_inp_par_par_frac": 0.0,
395
+ "batch_inp_par_reverse_frac": 0.0,
396
+ "batch_rl_frac": 0.0,
397
+ "batch_sft_frac": 0.0,
398
+ "batch_soft_sft_frac": 0.0,
399
+ "batch_tf_frac": 0.0,
400
+ "ce_loss": 0.5929547422666019,
401
+ "epoch": 0.3413333333333333,
402
+ "grad_norm": 1.171875,
403
+ "kd_loss": 0.44711892502580214,
404
+ "learning_rate": 3e-06,
405
+ "loss": 0.7472,
406
+ "masked_tokens": 99.425,
407
+ "mean_t": 0.5408745193795766,
408
+ "step": 160,
409
+ "student_masked_tokens": 99.425
410
+ },
411
+ {
412
+ "avg_mask_ratio": 0.4806730231270194,
413
+ "avg_response_length": 190.325,
414
+ "avg_student_mask_ratio": 0.4806730231270194,
415
+ "batch_ainp_frac": 0.0,
416
+ "batch_inp_frac": 1.0,
417
+ "batch_inp_oh_frac": 0.0,
418
+ "batch_inp_par_frac": 0.0,
419
+ "batch_inp_par_par_frac": 0.0,
420
+ "batch_inp_par_reverse_frac": 0.0,
421
+ "batch_rl_frac": 0.0,
422
+ "batch_sft_frac": 0.0,
423
+ "batch_soft_sft_frac": 0.0,
424
+ "batch_tf_frac": 0.0,
425
+ "ce_loss": 0.3583432949517601,
426
+ "epoch": 0.3626666666666667,
427
+ "grad_norm": 0.6171875,
428
+ "kd_loss": 0.4521343837219092,
429
+ "learning_rate": 3e-06,
430
+ "loss": 0.6358,
431
+ "masked_tokens": 89.825,
432
+ "mean_t": 0.5134547733236104,
433
+ "step": 170,
434
+ "student_masked_tokens": 89.825
435
+ },
436
+ {
437
+ "avg_mask_ratio": 0.45829249716189224,
438
+ "avg_response_length": 244.0,
439
+ "avg_student_mask_ratio": 0.45829249716189224,
440
+ "batch_ainp_frac": 0.0,
441
+ "batch_inp_frac": 1.0,
442
+ "batch_inp_oh_frac": 0.0,
443
+ "batch_inp_par_frac": 0.0,
444
+ "batch_inp_par_par_frac": 0.0,
445
+ "batch_inp_par_reverse_frac": 0.0,
446
+ "batch_rl_frac": 0.0,
447
+ "batch_sft_frac": 0.0,
448
+ "batch_soft_sft_frac": 0.0,
449
+ "batch_tf_frac": 0.0,
450
+ "ce_loss": 0.3746713957985094,
451
+ "epoch": 0.384,
452
+ "grad_norm": 0.49609375,
453
+ "kd_loss": 0.34934306121722897,
454
+ "learning_rate": 3e-06,
455
+ "loss": 0.5747,
456
+ "masked_tokens": 110.075,
457
+ "mean_t": 0.48226988823735156,
458
+ "step": 180,
459
+ "student_masked_tokens": 110.075
460
+ },
461
+ {
462
+ "avg_mask_ratio": 0.4842760307248682,
463
+ "avg_response_length": 233.675,
464
+ "avg_student_mask_ratio": 0.4842760307248682,
465
+ "batch_ainp_frac": 0.0,
466
+ "batch_inp_frac": 1.0,
467
+ "batch_inp_oh_frac": 0.0,
468
+ "batch_inp_par_frac": 0.0,
469
+ "batch_inp_par_par_frac": 0.0,
470
+ "batch_inp_par_reverse_frac": 0.0,
471
+ "batch_rl_frac": 0.0,
472
+ "batch_sft_frac": 0.0,
473
+ "batch_soft_sft_frac": 0.0,
474
+ "batch_tf_frac": 0.0,
475
+ "ce_loss": 0.5090123614077584,
476
+ "epoch": 0.4053333333333333,
477
+ "grad_norm": 1.6171875,
478
+ "kd_loss": 0.43204482231294605,
479
+ "learning_rate": 3e-06,
480
+ "loss": 0.7055,
481
+ "masked_tokens": 109.5875,
482
+ "mean_t": 0.5165087037021294,
483
+ "step": 190,
484
+ "student_masked_tokens": 109.5875
485
+ },
486
+ {
487
+ "avg_mask_ratio": 0.4665210062637925,
488
+ "avg_response_length": 197.8,
489
+ "avg_student_mask_ratio": 0.4665210062637925,
490
+ "batch_ainp_frac": 0.0,
491
+ "batch_inp_frac": 1.0,
492
+ "batch_inp_oh_frac": 0.0,
493
+ "batch_inp_par_frac": 0.0,
494
+ "batch_inp_par_par_frac": 0.0,
495
+ "batch_inp_par_reverse_frac": 0.0,
496
+ "batch_rl_frac": 0.0,
497
+ "batch_sft_frac": 0.0,
498
+ "batch_soft_sft_frac": 0.0,
499
+ "batch_tf_frac": 0.0,
500
+ "ce_loss": 0.27265903051802526,
501
+ "epoch": 0.4266666666666667,
502
+ "grad_norm": 0.318359375,
503
+ "kd_loss": 0.357759011555504,
504
+ "learning_rate": 3e-06,
505
+ "loss": 0.5013,
506
+ "masked_tokens": 97.0125,
507
+ "mean_t": 0.5073627714533359,
508
+ "step": 200,
509
+ "student_masked_tokens": 97.0125
510
+ }
511
+ ],
512
+ "logging_steps": 10,
513
+ "max_steps": 1404,
514
+ "num_input_tokens_seen": 0,
515
+ "num_train_epochs": 3,
516
+ "save_steps": 100,
517
+ "stateful_callbacks": {
518
+ "TrainerControl": {
519
+ "args": {
520
+ "should_epoch_stop": false,
521
+ "should_evaluate": false,
522
+ "should_log": false,
523
+ "should_save": true,
524
+ "should_training_stop": false
525
+ },
526
+ "attributes": {}
527
+ }
528
+ },
529
+ "total_flos": 0.0,
530
+ "train_batch_size": 1,
531
+ "trial_name": null,
532
+ "trial_params": null
533
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-200/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89ae704e41a4a62f6ca56789c45ce45887326cd2f8d1e97e398e5ada4a93398c
3
+ size 8312
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/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/unmask_tags_leave_last_step_gold1_target1_ce0.5/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
+ "gate_proj",
28
+ "down_proj",
29
+ "up_proj",
30
+ "o_proj",
31
+ "q_proj",
32
+ "v_proj",
33
+ "k_proj"
34
+ ],
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_dora": false,
38
+ "use_rslora": false
39
+ }
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-300/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0628a26dc3698e8b26ee3e8c5839bcdddc4a15d0501673a3faedd536e06be7ef
3
+ size 2406624648
math/INP/unmask_tags_leave_last_step_gold1_target1_ce0.5/checkpoint-300/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bd2918a717be1688c778cbe8e58099e66551411a02a7d7a8ed6319e7827c1aca
3
+ size 671304442