brthor commited on
Commit
3925ca9
·
verified ·
1 Parent(s): c8777c2

Upload full trainer checkpoint step=2026 (profile=llama3_1_8b_base_grpo, run_type=grpo)

Browse files
.gitattributes CHANGED
@@ -147,3 +147,4 @@ meta-llama__meta-llama-3.1-8b/sft/checkpoints/checkpoint-23456/tokenizer.json fi
147
  meta-llama__meta-llama-3.1-8b/sft/checkpoints/checkpoint-23460/tokenizer.json filter=lfs diff=lfs merge=lfs -text
148
  meta-llama__meta-llama-3.1-8b/sft/epoch_0004/tokenizer.json filter=lfs diff=lfs merge=lfs -text
149
  meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-1013/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
147
  meta-llama__meta-llama-3.1-8b/sft/checkpoints/checkpoint-23460/tokenizer.json filter=lfs diff=lfs merge=lfs -text
148
  meta-llama__meta-llama-3.1-8b/sft/epoch_0004/tokenizer.json filter=lfs diff=lfs merge=lfs -text
149
  meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-1013/tokenizer.json filter=lfs diff=lfs merge=lfs -text
150
+ meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/tokenizer.json filter=lfs diff=lfs merge=lfs -text
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/README.md ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: unsloth/Meta-Llama-3.1-8B
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:unsloth/Meta-Llama-3.1-8B
7
+ - grpo
8
+ - lora
9
+ - transformers
10
+ - trl
11
+ - unsloth
12
+ ---
13
+
14
+ # Model Card for Model ID
15
+
16
+ <!-- Provide a quick summary of what the model is/does. -->
17
+
18
+
19
+
20
+ ## Model Details
21
+
22
+ ### Model Description
23
+
24
+ <!-- Provide a longer summary of what this model is. -->
25
+
26
+
27
+
28
+ - **Developed by:** [More Information Needed]
29
+ - **Funded by [optional]:** [More Information Needed]
30
+ - **Shared by [optional]:** [More Information Needed]
31
+ - **Model type:** [More Information Needed]
32
+ - **Language(s) (NLP):** [More Information Needed]
33
+ - **License:** [More Information Needed]
34
+ - **Finetuned from model [optional]:** [More Information Needed]
35
+
36
+ ### Model Sources [optional]
37
+
38
+ <!-- Provide the basic links for the model. -->
39
+
40
+ - **Repository:** [More Information Needed]
41
+ - **Paper [optional]:** [More Information Needed]
42
+ - **Demo [optional]:** [More Information Needed]
43
+
44
+ ## Uses
45
+
46
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
47
+
48
+ ### Direct Use
49
+
50
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
51
+
52
+ [More Information Needed]
53
+
54
+ ### Downstream Use [optional]
55
+
56
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
57
+
58
+ [More Information Needed]
59
+
60
+ ### Out-of-Scope Use
61
+
62
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
63
+
64
+ [More Information Needed]
65
+
66
+ ## Bias, Risks, and Limitations
67
+
68
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
69
+
70
+ [More Information Needed]
71
+
72
+ ### Recommendations
73
+
74
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
75
+
76
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
77
+
78
+ ## How to Get Started with the Model
79
+
80
+ Use the code below to get started with the model.
81
+
82
+ [More Information Needed]
83
+
84
+ ## Training Details
85
+
86
+ ### Training Data
87
+
88
+ <!-- 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. -->
89
+
90
+ [More Information Needed]
91
+
92
+ ### Training Procedure
93
+
94
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
95
+
96
+ #### Preprocessing [optional]
97
+
98
+ [More Information Needed]
99
+
100
+
101
+ #### Training Hyperparameters
102
+
103
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
104
+
105
+ #### Speeds, Sizes, Times [optional]
106
+
107
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
108
+
109
+ [More Information Needed]
110
+
111
+ ## Evaluation
112
+
113
+ <!-- This section describes the evaluation protocols and provides the results. -->
114
+
115
+ ### Testing Data, Factors & Metrics
116
+
117
+ #### Testing Data
118
+
119
+ <!-- This should link to a Dataset Card if possible. -->
120
+
121
+ [More Information Needed]
122
+
123
+ #### Factors
124
+
125
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
126
+
127
+ [More Information Needed]
128
+
129
+ #### Metrics
130
+
131
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
132
+
133
+ [More Information Needed]
134
+
135
+ ### Results
136
+
137
+ [More Information Needed]
138
+
139
+ #### Summary
140
+
141
+
142
+
143
+ ## Model Examination [optional]
144
+
145
+ <!-- Relevant interpretability work for the model goes here -->
146
+
147
+ [More Information Needed]
148
+
149
+ ## Environmental Impact
150
+
151
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
152
+
153
+ 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).
154
+
155
+ - **Hardware Type:** [More Information Needed]
156
+ - **Hours used:** [More Information Needed]
157
+ - **Cloud Provider:** [More Information Needed]
158
+ - **Compute Region:** [More Information Needed]
159
+ - **Carbon Emitted:** [More Information Needed]
160
+
161
+ ## Technical Specifications [optional]
162
+
163
+ ### Model Architecture and Objective
164
+
165
+ [More Information Needed]
166
+
167
+ ### Compute Infrastructure
168
+
169
+ [More Information Needed]
170
+
171
+ #### Hardware
172
+
173
+ [More Information Needed]
174
+
175
+ #### Software
176
+
177
+ [More Information Needed]
178
+
179
+ ## Citation [optional]
180
+
181
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
182
+
183
+ **BibTeX:**
184
+
185
+ [More Information Needed]
186
+
187
+ **APA:**
188
+
189
+ [More Information Needed]
190
+
191
+ ## Glossary [optional]
192
+
193
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
194
+
195
+ [More Information Needed]
196
+
197
+ ## More Information [optional]
198
+
199
+ [More Information Needed]
200
+
201
+ ## Model Card Authors [optional]
202
+
203
+ [More Information Needed]
204
+
205
+ ## Model Card Contact
206
+
207
+ [More Information Needed]
208
+ ### Framework versions
209
+
210
+ - PEFT 0.18.1
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/adapter_config.json ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": {
6
+ "base_model_class": "LlamaForCausalLM",
7
+ "parent_library": "transformers.models.llama.modeling_llama",
8
+ "unsloth_fixed": true
9
+ },
10
+ "base_model_name_or_path": "unsloth/Meta-Llama-3.1-8B",
11
+ "bias": "none",
12
+ "corda_config": null,
13
+ "ensure_weight_tying": false,
14
+ "eva_config": null,
15
+ "exclude_modules": null,
16
+ "fan_in_fan_out": false,
17
+ "inference_mode": true,
18
+ "init_lora_weights": true,
19
+ "layer_replication": null,
20
+ "layers_pattern": null,
21
+ "layers_to_transform": null,
22
+ "loftq_config": {},
23
+ "lora_alpha": 16,
24
+ "lora_bias": false,
25
+ "lora_dropout": 0.0,
26
+ "megatron_config": null,
27
+ "megatron_core": "megatron.core",
28
+ "modules_to_save": null,
29
+ "peft_type": "LORA",
30
+ "peft_version": "0.18.1",
31
+ "qalora_group_size": 16,
32
+ "r": 8,
33
+ "rank_pattern": {},
34
+ "revision": null,
35
+ "target_modules": [
36
+ "o_proj",
37
+ "down_proj",
38
+ "up_proj",
39
+ "gate_proj",
40
+ "k_proj",
41
+ "v_proj",
42
+ "q_proj"
43
+ ],
44
+ "target_parameters": null,
45
+ "task_type": "CAUSAL_LM",
46
+ "trainable_token_indices": null,
47
+ "use_dora": false,
48
+ "use_qalora": false,
49
+ "use_rslora": false
50
+ }
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:50d4423335b38d92fd7ebe7e63ac271dd88b77d67b2913c813c54f1a8bc2c8f2
3
+ size 83945296
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/chat_template.jinja ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ messages[0]['content'] + '
2
+
3
+ ' }}{% set loop_messages = messages[1:] %}{% else %}{{ '' + '
4
+
5
+ ' }}{% set loop_messages = messages %}{% endif %}{% for message in loop_messages %}{% if message['role'] == 'user' %}{{ '### Instruction:
6
+ ' + message['content'] + '
7
+
8
+ ' }}{% elif message['role'] == 'assistant' %}{{ '### Response:
9
+ ' + message['content'] + eos_token + '
10
+
11
+ ' }}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '### Response:
12
+ ' }}{% endif %}
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8eff1430f0a5904d595dcef1e17dc9c50e2f044e25f78e5ece8fcfab7ae4bb5f
3
+ size 43127973
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:368c188f4019dee48f24d30cffeef109d0d367163b73096063e69c443f7ebbdf
3
+ size 14645
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5e8eb9979eb54b15304ceed82df37a0d182460ea9f12d3e89c602d4ebdec5445
3
+ size 1465
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a65c6c5f9764771aa485e6a1f5e63d7d9af8477fe0777148c17476ecb2e09a05
3
+ size 17210099
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/tokenizer_config.json ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<|begin_of_text|>",
4
+ "clean_up_tokenization_spaces": true,
5
+ "eos_token": "<|end_of_text|>",
6
+ "from_slow": true,
7
+ "is_local": false,
8
+ "legacy": false,
9
+ "model_input_names": [
10
+ "input_ids",
11
+ "attention_mask"
12
+ ],
13
+ "model_max_length": 131072,
14
+ "pad_token": "<|finetune_right_pad_id|>",
15
+ "padding_side": "right",
16
+ "tokenizer_class": "TokenizersBackend",
17
+ "unk_token": null
18
+ }
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/trainer_state.json ADDED
@@ -0,0 +1,2239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.003598515124065292,
6
+ "eval_steps": 500,
7
+ "global_step": 2026,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "clip_ratio/high_max": 0.0,
14
+ "clip_ratio/high_mean": 0.0,
15
+ "clip_ratio/low_mean": 0.0,
16
+ "clip_ratio/low_min": 0.0,
17
+ "clip_ratio/region_mean": 0.0,
18
+ "completion_length": 65.81,
19
+ "completions/clipped_ratio": 0.0,
20
+ "completions/max_length": 66.76923076923077,
21
+ "completions/max_terminated_length": 66.76923076923077,
22
+ "completions/mean_length": 66.54807692307692,
23
+ "completions/mean_terminated_length": 66.54807692307692,
24
+ "completions/min_length": 66.3076923076923,
25
+ "completions/min_terminated_length": 66.3076923076923,
26
+ "epoch": 4.440418465036145e-05,
27
+ "frac_reward_zero_std": 0.5384615384615384,
28
+ "grad_norm": 0.025798887014389038,
29
+ "kl": 0.18038831524550913,
30
+ "learning_rate": 1.0656956994738129e-09,
31
+ "loss": -0.0065249383449554445,
32
+ "num_tokens": 24649.0,
33
+ "reward": 5.7312158437875604e-09,
34
+ "reward_std": 0.4934053604419415,
35
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.7312158437875604e-09,
36
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
37
+ "step": 25
38
+ },
39
+ {
40
+ "clip_ratio/high_max": 0.0,
41
+ "clip_ratio/high_mean": 0.0,
42
+ "clip_ratio/low_mean": 0.0,
43
+ "clip_ratio/low_min": 0.0,
44
+ "clip_ratio/region_mean": 0.0,
45
+ "completion_length": 65.5,
46
+ "completions/clipped_ratio": 0.0,
47
+ "completions/max_length": 64.75,
48
+ "completions/max_terminated_length": 64.75,
49
+ "completions/mean_length": 64.6875,
50
+ "completions/mean_terminated_length": 64.6875,
51
+ "completions/min_length": 64.5,
52
+ "completions/min_terminated_length": 64.5,
53
+ "epoch": 8.88083693007229e-05,
54
+ "frac_reward_zero_std": 0.8333333333333334,
55
+ "grad_norm": 0.27390509843826294,
56
+ "kl": 0.16198864802718163,
57
+ "learning_rate": 2.175795386425701e-09,
58
+ "loss": 0.008163155317306518,
59
+ "num_tokens": 46547.0,
60
+ "reward": 2.483526865641276e-09,
61
+ "reward_std": 0.17817416787147522,
62
+ "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
63
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.17817415793736777,
64
+ "step": 50
65
+ },
66
+ {
67
+ "clip_ratio/high_max": 0.0,
68
+ "clip_ratio/high_mean": 0.0,
69
+ "clip_ratio/low_mean": 0.0,
70
+ "clip_ratio/low_min": 0.0,
71
+ "clip_ratio/region_mean": 0.0,
72
+ "completion_length": 52.52,
73
+ "completions/clipped_ratio": 0.0,
74
+ "completions/max_length": 52.61538461538461,
75
+ "completions/max_terminated_length": 52.61538461538461,
76
+ "completions/mean_length": 52.42307692307692,
77
+ "completions/mean_terminated_length": 52.42307692307692,
78
+ "completions/min_length": 52.30769230769231,
79
+ "completions/min_terminated_length": 52.30769230769231,
80
+ "epoch": 0.00013321255395108434,
81
+ "frac_reward_zero_std": 0.8461538461538461,
82
+ "grad_norm": 1.6365996771128266e-06,
83
+ "kl": 0.20123317912220956,
84
+ "learning_rate": 3.285895073377589e-09,
85
+ "loss": 0.0007231433689594269,
86
+ "num_tokens": 66431.0,
87
+ "reward": 0.0,
88
+ "reward_std": 0.16446845348064715,
89
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
90
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.16446845348064715,
91
+ "step": 75
92
+ },
93
+ {
94
+ "clip_ratio/high_max": 0.0,
95
+ "clip_ratio/high_mean": 0.0,
96
+ "clip_ratio/low_mean": 0.0,
97
+ "clip_ratio/low_min": 0.0,
98
+ "clip_ratio/region_mean": 0.0,
99
+ "completion_length": 54.32,
100
+ "completions/clipped_ratio": 0.0,
101
+ "completions/max_length": 54.5,
102
+ "completions/max_terminated_length": 54.5,
103
+ "completions/mean_length": 54.5,
104
+ "completions/mean_terminated_length": 54.5,
105
+ "completions/min_length": 54.5,
106
+ "completions/min_terminated_length": 54.5,
107
+ "epoch": 0.0001776167386014458,
108
+ "frac_reward_zero_std": 0.5833333333333334,
109
+ "grad_norm": 0.00011817843187600374,
110
+ "kl": 0.253861972540617,
111
+ "learning_rate": 4.3959947603294776e-09,
112
+ "loss": 0.0025739344954490663,
113
+ "num_tokens": 85543.0,
114
+ "reward": 3.725290298461914e-09,
115
+ "reward_std": 0.4454353948434194,
116
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.725290298461914e-09,
117
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
118
+ "step": 100
119
+ },
120
+ {
121
+ "clip_ratio/high_max": 0.0,
122
+ "clip_ratio/high_mean": 0.0,
123
+ "clip_ratio/low_mean": 0.0,
124
+ "clip_ratio/low_min": 0.0,
125
+ "clip_ratio/region_mean": 0.0,
126
+ "completion_length": 51.7,
127
+ "completions/clipped_ratio": 0.0,
128
+ "completions/max_length": 51.0,
129
+ "completions/max_terminated_length": 51.0,
130
+ "completions/mean_length": 50.86538461538461,
131
+ "completions/mean_terminated_length": 50.86538461538461,
132
+ "completions/min_length": 50.38461538461539,
133
+ "completions/min_terminated_length": 50.38461538461539,
134
+ "epoch": 0.00022202092325180725,
135
+ "frac_reward_zero_std": 0.5384615384615384,
136
+ "grad_norm": 0.3782321810722351,
137
+ "kl": 0.24278869599103928,
138
+ "learning_rate": 5.506094447281366e-09,
139
+ "loss": 0.0006042173504829406,
140
+ "num_tokens": 105273.0,
141
+ "reward": 2.292486337515024e-09,
142
+ "reward_std": 0.49340536961188686,
143
+ "rewards/TranscriptCorrectionGrpoReward/mean": 2.292486337515024e-09,
144
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
145
+ "step": 125
146
+ },
147
+ {
148
+ "clip_ratio/high_max": 0.0,
149
+ "clip_ratio/high_mean": 0.0,
150
+ "clip_ratio/low_mean": 0.0,
151
+ "clip_ratio/low_min": 0.0,
152
+ "clip_ratio/region_mean": 0.0,
153
+ "completion_length": 73.29,
154
+ "completions/clipped_ratio": 0.0,
155
+ "completions/max_length": 75.33333333333333,
156
+ "completions/max_terminated_length": 75.33333333333333,
157
+ "completions/mean_length": 75.09375,
158
+ "completions/mean_terminated_length": 75.09375,
159
+ "completions/min_length": 74.91666666666667,
160
+ "completions/min_terminated_length": 74.91666666666667,
161
+ "epoch": 0.0002664251079021687,
162
+ "frac_reward_zero_std": 0.6666666666666666,
163
+ "grad_norm": 0.1665542721748352,
164
+ "kl": 0.20266253776848317,
165
+ "learning_rate": 6.6161941342332544e-09,
166
+ "loss": 0.001955866813659668,
167
+ "num_tokens": 130370.0,
168
+ "reward": 4.967053731282552e-09,
169
+ "reward_std": 0.35634831587473553,
170
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
171
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
172
+ "step": 150
173
+ },
174
+ {
175
+ "clip_ratio/high_max": 0.0,
176
+ "clip_ratio/high_mean": 0.0,
177
+ "clip_ratio/low_mean": 0.0,
178
+ "clip_ratio/low_min": 0.0,
179
+ "clip_ratio/region_mean": 0.0,
180
+ "completion_length": 57.74,
181
+ "completions/clipped_ratio": 0.0,
182
+ "completions/max_length": 58.0,
183
+ "completions/max_terminated_length": 58.0,
184
+ "completions/mean_length": 57.94230769230769,
185
+ "completions/mean_terminated_length": 57.94230769230769,
186
+ "completions/min_length": 57.84615384615385,
187
+ "completions/min_terminated_length": 57.84615384615385,
188
+ "epoch": 0.00031082929255253015,
189
+ "frac_reward_zero_std": 0.6153846153846154,
190
+ "grad_norm": 0.22071869671344757,
191
+ "kl": 0.20556429855525493,
192
+ "learning_rate": 7.726293821185143e-09,
193
+ "loss": 0.0019431695342063903,
194
+ "num_tokens": 152108.0,
195
+ "reward": 5.7312158437875604e-09,
196
+ "reward_std": 0.4111711337016179,
197
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.7312158437875604e-09,
198
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
199
+ "step": 175
200
+ },
201
+ {
202
+ "clip_ratio/high_max": 0.0,
203
+ "clip_ratio/high_mean": 0.0,
204
+ "clip_ratio/low_mean": 0.0,
205
+ "clip_ratio/low_min": 0.0,
206
+ "clip_ratio/region_mean": 0.0,
207
+ "completion_length": 66.59,
208
+ "completions/clipped_ratio": 0.0,
209
+ "completions/max_length": 66.91666666666667,
210
+ "completions/max_terminated_length": 66.91666666666667,
211
+ "completions/mean_length": 66.73958333333333,
212
+ "completions/mean_terminated_length": 66.73958333333333,
213
+ "completions/min_length": 66.33333333333333,
214
+ "completions/min_terminated_length": 66.33333333333333,
215
+ "epoch": 0.0003552334772028916,
216
+ "frac_reward_zero_std": 0.5,
217
+ "grad_norm": 6.216925248736516e-05,
218
+ "kl": 0.17584349907934665,
219
+ "learning_rate": 8.83639350813703e-09,
220
+ "loss": 0.0011808258295059205,
221
+ "num_tokens": 174947.0,
222
+ "reward": 4.346172014872233e-09,
223
+ "reward_std": 0.5345224738121033,
224
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.346172014872233e-09,
225
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.5345224738121033,
226
+ "step": 200
227
+ },
228
+ {
229
+ "clip_ratio/high_max": 0.0,
230
+ "clip_ratio/high_mean": 0.0,
231
+ "clip_ratio/low_mean": 0.0,
232
+ "clip_ratio/low_min": 0.0,
233
+ "clip_ratio/region_mean": 0.0,
234
+ "completion_length": 65.44,
235
+ "completions/clipped_ratio": 0.0,
236
+ "completions/max_length": 65.92307692307692,
237
+ "completions/max_terminated_length": 65.92307692307692,
238
+ "completions/mean_length": 65.61538461538461,
239
+ "completions/mean_terminated_length": 65.61538461538461,
240
+ "completions/min_length": 65.38461538461539,
241
+ "completions/min_terminated_length": 65.38461538461539,
242
+ "epoch": 0.000399637661853253,
243
+ "frac_reward_zero_std": 0.6153846153846154,
244
+ "grad_norm": 0.12366796284914017,
245
+ "kl": 0.19227153606712818,
246
+ "learning_rate": 9.94649319508892e-09,
247
+ "loss": 0.0005045509338378906,
248
+ "num_tokens": 199651.0,
249
+ "reward": 1.7193647531362681e-09,
250
+ "reward_std": 0.4111711337016179,
251
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.7193647531362681e-09,
252
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
253
+ "step": 225
254
+ },
255
+ {
256
+ "clip_ratio/high_max": 0.0,
257
+ "clip_ratio/high_mean": 0.0,
258
+ "clip_ratio/low_mean": 0.0,
259
+ "clip_ratio/low_min": 0.0,
260
+ "clip_ratio/region_mean": 0.0,
261
+ "completion_length": 66.15,
262
+ "completions/clipped_ratio": 0.0,
263
+ "completions/max_length": 66.25,
264
+ "completions/max_terminated_length": 66.25,
265
+ "completions/mean_length": 65.98958333333333,
266
+ "completions/mean_terminated_length": 65.98958333333333,
267
+ "completions/min_length": 65.83333333333333,
268
+ "completions/min_terminated_length": 65.83333333333333,
269
+ "epoch": 0.0004440418465036145,
270
+ "frac_reward_zero_std": 0.5833333333333334,
271
+ "grad_norm": 5.099513509776443e-05,
272
+ "kl": 0.19273025602102278,
273
+ "learning_rate": 1.1056592882040807e-08,
274
+ "loss": 0.002573988139629364,
275
+ "num_tokens": 222250.0,
276
+ "reward": 4.346172014872233e-09,
277
+ "reward_std": 0.4454353948434194,
278
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.346172014872233e-09,
279
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
280
+ "step": 250
281
+ },
282
+ {
283
+ "clip_ratio/high_max": 0.0,
284
+ "clip_ratio/high_mean": 0.0,
285
+ "clip_ratio/low_mean": 0.0,
286
+ "clip_ratio/low_min": 0.0,
287
+ "clip_ratio/region_mean": 0.0,
288
+ "completion_length": 64.71,
289
+ "completions/clipped_ratio": 0.0,
290
+ "completions/max_length": 66.92307692307692,
291
+ "completions/max_terminated_length": 66.92307692307692,
292
+ "completions/mean_length": 66.6826923076923,
293
+ "completions/mean_terminated_length": 66.6826923076923,
294
+ "completions/min_length": 66.38461538461539,
295
+ "completions/min_terminated_length": 66.38461538461539,
296
+ "epoch": 0.0004884460311539759,
297
+ "frac_reward_zero_std": 0.5384615384615384,
298
+ "grad_norm": 0.04118318483233452,
299
+ "kl": 0.17756690874695777,
300
+ "learning_rate": 1.2166692568992695e-08,
301
+ "loss": -0.0014105629920959472,
302
+ "num_tokens": 247777.0,
303
+ "reward": 5.73121584378756e-10,
304
+ "reward_std": 0.49340535127199614,
305
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.73121584378756e-10,
306
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
307
+ "step": 275
308
+ },
309
+ {
310
+ "clip_ratio/high_max": 0.0,
311
+ "clip_ratio/high_mean": 0.0,
312
+ "clip_ratio/low_mean": 0.0,
313
+ "clip_ratio/low_min": 0.0,
314
+ "clip_ratio/region_mean": 0.0,
315
+ "completion_length": 72.83,
316
+ "completions/clipped_ratio": 0.0,
317
+ "completions/max_length": 71.25,
318
+ "completions/max_terminated_length": 71.25,
319
+ "completions/mean_length": 71.03125,
320
+ "completions/mean_terminated_length": 71.03125,
321
+ "completions/min_length": 70.91666666666667,
322
+ "completions/min_terminated_length": 70.91666666666667,
323
+ "epoch": 0.0005328502158043374,
324
+ "frac_reward_zero_std": 0.75,
325
+ "grad_norm": 7.932442713354249e-06,
326
+ "kl": 0.16507803082466124,
327
+ "learning_rate": 1.3276792255944584e-08,
328
+ "loss": 0.005102531909942627,
329
+ "num_tokens": 271724.0,
330
+ "reward": 3.725290298461914e-09,
331
+ "reward_std": 0.2672612468401591,
332
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.725290298461914e-09,
333
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.26726123690605164,
334
+ "step": 300
335
+ },
336
+ {
337
+ "clip_ratio/high_max": 0.0,
338
+ "clip_ratio/high_mean": 0.0,
339
+ "clip_ratio/low_mean": 0.0,
340
+ "clip_ratio/low_min": 0.0,
341
+ "clip_ratio/region_mean": 0.0,
342
+ "completion_length": 68.11,
343
+ "completions/clipped_ratio": 0.0,
344
+ "completions/max_length": 67.92307692307692,
345
+ "completions/max_terminated_length": 67.92307692307692,
346
+ "completions/mean_length": 67.74038461538461,
347
+ "completions/mean_terminated_length": 67.74038461538461,
348
+ "completions/min_length": 67.61538461538461,
349
+ "completions/min_terminated_length": 67.61538461538461,
350
+ "epoch": 0.0005772544004546988,
351
+ "frac_reward_zero_std": 0.6923076923076923,
352
+ "grad_norm": 0.2366252988576889,
353
+ "kl": 0.18056891307234765,
354
+ "learning_rate": 1.438689194289647e-08,
355
+ "loss": 0.0016832691431045532,
356
+ "num_tokens": 297265.0,
357
+ "reward": 2.8656079218937802e-09,
358
+ "reward_std": 0.3289369069612943,
359
+ "rewards/TranscriptCorrectionGrpoReward/mean": 2.8656079218937802e-09,
360
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.3289369069612943,
361
+ "step": 325
362
+ },
363
+ {
364
+ "clip_ratio/high_max": 0.0,
365
+ "clip_ratio/high_mean": 0.0,
366
+ "clip_ratio/low_mean": 0.0,
367
+ "clip_ratio/low_min": 0.0,
368
+ "clip_ratio/region_mean": 0.0,
369
+ "completion_length": 54.2,
370
+ "completions/clipped_ratio": 0.0,
371
+ "completions/max_length": 54.083333333333336,
372
+ "completions/max_terminated_length": 54.083333333333336,
373
+ "completions/mean_length": 54.020833333333336,
374
+ "completions/mean_terminated_length": 54.020833333333336,
375
+ "completions/min_length": 53.916666666666664,
376
+ "completions/min_terminated_length": 53.916666666666664,
377
+ "epoch": 0.0006216585851050603,
378
+ "frac_reward_zero_std": 0.75,
379
+ "grad_norm": 2.0182937987556215e-06,
380
+ "kl": 0.22566444158554078,
381
+ "learning_rate": 1.549699162984836e-08,
382
+ "loss": 0.003195708990097046,
383
+ "num_tokens": 316355.0,
384
+ "reward": 0.0,
385
+ "reward_std": 0.26726123690605164,
386
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
387
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.26726123690605164,
388
+ "step": 350
389
+ },
390
+ {
391
+ "clip_ratio/high_max": 0.0,
392
+ "clip_ratio/high_mean": 0.0,
393
+ "clip_ratio/low_mean": 0.000317460335791111,
394
+ "clip_ratio/low_min": 0.000317460335791111,
395
+ "clip_ratio/region_mean": 0.000317460335791111,
396
+ "completion_length": 67.73,
397
+ "completions/clipped_ratio": 0.0,
398
+ "completions/max_length": 67.76923076923077,
399
+ "completions/max_terminated_length": 67.76923076923077,
400
+ "completions/mean_length": 67.25,
401
+ "completions/mean_terminated_length": 67.25,
402
+ "completions/min_length": 66.6923076923077,
403
+ "completions/min_terminated_length": 66.6923076923077,
404
+ "epoch": 0.0006660627697554218,
405
+ "frac_reward_zero_std": 0.3076923076923077,
406
+ "grad_norm": 0.16900327801704407,
407
+ "kl": 0.1710970998555422,
408
+ "learning_rate": 1.660709131680025e-08,
409
+ "loss": -0.004613960087299347,
410
+ "num_tokens": 341645.0,
411
+ "reward": 5.158094259408804e-09,
412
+ "reward_std": 0.7401080314929669,
413
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.158094259408804e-09,
414
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.7401080406629122,
415
+ "step": 375
416
+ },
417
+ {
418
+ "clip_ratio/high_max": 0.0,
419
+ "clip_ratio/high_mean": 0.0,
420
+ "clip_ratio/low_mean": 0.0,
421
+ "clip_ratio/low_min": 0.0,
422
+ "clip_ratio/region_mean": 0.0,
423
+ "completion_length": 61.17,
424
+ "completions/clipped_ratio": 0.0,
425
+ "completions/max_length": 61.833333333333336,
426
+ "completions/max_terminated_length": 61.833333333333336,
427
+ "completions/mean_length": 61.416666666666664,
428
+ "completions/mean_terminated_length": 61.416666666666664,
429
+ "completions/min_length": 61.25,
430
+ "completions/min_terminated_length": 61.25,
431
+ "epoch": 0.0007104669544057832,
432
+ "frac_reward_zero_std": 0.6666666666666666,
433
+ "grad_norm": 8.416860168836138e-07,
434
+ "kl": 0.21405472174286844,
435
+ "learning_rate": 1.7717191003752138e-08,
436
+ "loss": 0.00976666271686554,
437
+ "num_tokens": 362597.0,
438
+ "reward": 0.0,
439
+ "reward_std": 0.35634831587473553,
440
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
441
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634832580884296,
442
+ "step": 400
443
+ },
444
+ {
445
+ "clip_ratio/high_max": 0.0,
446
+ "clip_ratio/high_mean": 0.0,
447
+ "clip_ratio/low_mean": 0.0,
448
+ "clip_ratio/low_min": 0.0,
449
+ "clip_ratio/region_mean": 0.0,
450
+ "completion_length": 70.06,
451
+ "completions/clipped_ratio": 0.0,
452
+ "completions/max_length": 70.07692307692308,
453
+ "completions/max_terminated_length": 70.07692307692308,
454
+ "completions/mean_length": 69.78846153846153,
455
+ "completions/mean_terminated_length": 69.78846153846153,
456
+ "completions/min_length": 69.3076923076923,
457
+ "completions/min_terminated_length": 69.3076923076923,
458
+ "epoch": 0.0007548711390561447,
459
+ "frac_reward_zero_std": 0.3076923076923077,
460
+ "grad_norm": 0.15129399299621582,
461
+ "kl": 0.18921885184943676,
462
+ "learning_rate": 1.8827290690704024e-08,
463
+ "loss": -0.001335275173187256,
464
+ "num_tokens": 388727.0,
465
+ "reward": 6.8774590125450725e-09,
466
+ "reward_std": 0.7401080314929669,
467
+ "rewards/TranscriptCorrectionGrpoReward/mean": 6.8774590125450725e-09,
468
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.7401080406629122,
469
+ "step": 425
470
+ },
471
+ {
472
+ "clip_ratio/high_max": 0.0,
473
+ "clip_ratio/high_mean": 0.0,
474
+ "clip_ratio/low_mean": 0.0,
475
+ "clip_ratio/low_min": 0.0,
476
+ "clip_ratio/region_mean": 0.0,
477
+ "completion_length": 60.39,
478
+ "completions/clipped_ratio": 0.0,
479
+ "completions/max_length": 60.5,
480
+ "completions/max_terminated_length": 60.5,
481
+ "completions/mean_length": 60.28125,
482
+ "completions/mean_terminated_length": 60.28125,
483
+ "completions/min_length": 60.25,
484
+ "completions/min_terminated_length": 60.25,
485
+ "epoch": 0.000799275323706506,
486
+ "frac_reward_zero_std": 0.6666666666666666,
487
+ "grad_norm": 0.0004283797461539507,
488
+ "kl": 0.19731507159769535,
489
+ "learning_rate": 1.9937390377655913e-08,
490
+ "loss": 0.002477347552776337,
491
+ "num_tokens": 409458.0,
492
+ "reward": 0.0,
493
+ "reward_std": 0.35634832580884296,
494
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
495
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
496
+ "step": 450
497
+ },
498
+ {
499
+ "clip_ratio/high_max": 0.0,
500
+ "clip_ratio/high_mean": 0.0,
501
+ "clip_ratio/low_mean": 0.0,
502
+ "clip_ratio/low_min": 0.0,
503
+ "clip_ratio/region_mean": 0.0,
504
+ "completion_length": 70.39,
505
+ "completions/clipped_ratio": 0.0,
506
+ "completions/max_length": 69.46153846153847,
507
+ "completions/max_terminated_length": 69.46153846153847,
508
+ "completions/mean_length": 69.45192307692308,
509
+ "completions/mean_terminated_length": 69.45192307692308,
510
+ "completions/min_length": 69.38461538461539,
511
+ "completions/min_terminated_length": 69.38461538461539,
512
+ "epoch": 0.0008436795083568675,
513
+ "frac_reward_zero_std": 0.8461538461538461,
514
+ "grad_norm": 6.861191195639549e-06,
515
+ "kl": 0.17384251914918422,
516
+ "learning_rate": 2.10474900646078e-08,
517
+ "loss": 0.0019164717197418214,
518
+ "num_tokens": 435369.0,
519
+ "reward": 0.0,
520
+ "reward_std": 0.16446845348064715,
521
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
522
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.16446845348064715,
523
+ "step": 475
524
+ },
525
+ {
526
+ "clip_ratio/high_max": 0.0,
527
+ "clip_ratio/high_mean": 0.0,
528
+ "clip_ratio/low_mean": 0.0,
529
+ "clip_ratio/low_min": 0.0,
530
+ "clip_ratio/region_mean": 0.0,
531
+ "completion_length": 61.43,
532
+ "completions/clipped_ratio": 0.0,
533
+ "completions/max_length": 62.25,
534
+ "completions/max_terminated_length": 62.25,
535
+ "completions/mean_length": 62.072916666666664,
536
+ "completions/mean_terminated_length": 62.072916666666664,
537
+ "completions/min_length": 61.666666666666664,
538
+ "completions/min_terminated_length": 61.666666666666664,
539
+ "epoch": 0.000888083693007229,
540
+ "frac_reward_zero_std": 0.6666666666666666,
541
+ "grad_norm": 0.1936454176902771,
542
+ "kl": 0.18575097121298312,
543
+ "learning_rate": 2.2157589751559692e-08,
544
+ "loss": 0.00039461255073547363,
545
+ "num_tokens": 456096.0,
546
+ "reward": -6.20881716410319e-09,
547
+ "reward_std": 0.35634831587473553,
548
+ "rewards/TranscriptCorrectionGrpoReward/mean": -6.20881716410319e-09,
549
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
550
+ "step": 500
551
+ },
552
+ {
553
+ "clip_ratio/high_max": 0.0,
554
+ "clip_ratio/high_mean": 0.0,
555
+ "clip_ratio/low_mean": 0.0,
556
+ "clip_ratio/low_min": 0.0,
557
+ "clip_ratio/region_mean": 0.0,
558
+ "completion_length": 53.47,
559
+ "completions/clipped_ratio": 0.0,
560
+ "completions/max_length": 53.92307692307692,
561
+ "completions/max_terminated_length": 53.92307692307692,
562
+ "completions/mean_length": 53.64423076923077,
563
+ "completions/mean_terminated_length": 53.64423076923077,
564
+ "completions/min_length": 53.15384615384615,
565
+ "completions/min_terminated_length": 53.15384615384615,
566
+ "epoch": 0.0009324878776575904,
567
+ "frac_reward_zero_std": 0.7692307692307693,
568
+ "grad_norm": 2.9146491215215065e-05,
569
+ "kl": 0.21476560801267625,
570
+ "learning_rate": 2.3267689438511578e-08,
571
+ "loss": 0.0002883000299334526,
572
+ "num_tokens": 476723.0,
573
+ "reward": 0.0,
574
+ "reward_std": 0.24670268022097075,
575
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
576
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.24670268022097075,
577
+ "step": 525
578
+ },
579
+ {
580
+ "clip_ratio/high_max": 0.0,
581
+ "clip_ratio/high_mean": 0.0,
582
+ "clip_ratio/low_mean": 0.0,
583
+ "clip_ratio/low_min": 0.0,
584
+ "clip_ratio/region_mean": 0.0,
585
+ "completion_length": 78.73,
586
+ "completions/clipped_ratio": 0.0,
587
+ "completions/max_length": 79.91666666666667,
588
+ "completions/max_terminated_length": 79.91666666666667,
589
+ "completions/mean_length": 79.59375,
590
+ "completions/mean_terminated_length": 79.59375,
591
+ "completions/min_length": 78.91666666666667,
592
+ "completions/min_terminated_length": 78.91666666666667,
593
+ "epoch": 0.0009768920623079518,
594
+ "frac_reward_zero_std": 0.5833333333333334,
595
+ "grad_norm": 1.4861758245388046e-05,
596
+ "kl": 0.15855858221650124,
597
+ "learning_rate": 2.4377789125463464e-08,
598
+ "loss": 0.0011431390047073363,
599
+ "num_tokens": 503564.0,
600
+ "reward": 4.346172014872233e-09,
601
+ "reward_std": 0.4454353948434194,
602
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.346172014872233e-09,
603
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
604
+ "step": 550
605
+ },
606
+ {
607
+ "clip_ratio/high_max": 0.0,
608
+ "clip_ratio/high_mean": 0.0,
609
+ "clip_ratio/low_mean": 0.0,
610
+ "clip_ratio/low_min": 0.0,
611
+ "clip_ratio/region_mean": 0.0,
612
+ "completion_length": 70.19,
613
+ "completions/clipped_ratio": 0.0,
614
+ "completions/max_length": 72.07692307692308,
615
+ "completions/max_terminated_length": 72.07692307692308,
616
+ "completions/mean_length": 71.83653846153847,
617
+ "completions/mean_terminated_length": 71.83653846153847,
618
+ "completions/min_length": 71.53846153846153,
619
+ "completions/min_terminated_length": 71.53846153846153,
620
+ "epoch": 0.0010212962469583133,
621
+ "frac_reward_zero_std": 0.6153846153846154,
622
+ "grad_norm": 0.19313402473926544,
623
+ "kl": 0.18091401405632496,
624
+ "learning_rate": 2.5487888812415353e-08,
625
+ "loss": 0.010396603345870972,
626
+ "num_tokens": 530171.0,
627
+ "reward": -5.7312158437875604e-09,
628
+ "reward_std": 0.4111711337016179,
629
+ "rewards/TranscriptCorrectionGrpoReward/mean": -5.7312158437875604e-09,
630
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
631
+ "step": 575
632
+ },
633
+ {
634
+ "clip_ratio/high_max": 0.0,
635
+ "clip_ratio/high_mean": 0.0,
636
+ "clip_ratio/low_mean": 0.0,
637
+ "clip_ratio/low_min": 0.0,
638
+ "clip_ratio/region_mean": 0.0,
639
+ "completion_length": 69.74,
640
+ "completions/clipped_ratio": 0.0,
641
+ "completions/max_length": 68.0,
642
+ "completions/max_terminated_length": 68.0,
643
+ "completions/mean_length": 67.9375,
644
+ "completions/mean_terminated_length": 67.9375,
645
+ "completions/min_length": 67.66666666666667,
646
+ "completions/min_terminated_length": 67.66666666666667,
647
+ "epoch": 0.0010657004316086747,
648
+ "frac_reward_zero_std": 0.5833333333333334,
649
+ "grad_norm": 0.1186411902308464,
650
+ "kl": 0.16490939274430275,
651
+ "learning_rate": 2.659798849936724e-08,
652
+ "loss": -0.006589226126670837,
653
+ "num_tokens": 553341.0,
654
+ "reward": -2.949188152949015e-09,
655
+ "reward_std": 0.44543540477752686,
656
+ "rewards/TranscriptCorrectionGrpoReward/mean": -2.949188152949015e-09,
657
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
658
+ "step": 600
659
+ },
660
+ {
661
+ "clip_ratio/high_max": 0.0,
662
+ "clip_ratio/high_mean": 0.0,
663
+ "clip_ratio/low_mean": 0.0,
664
+ "clip_ratio/low_min": 0.0,
665
+ "clip_ratio/region_mean": 0.0,
666
+ "completion_length": 69.76,
667
+ "completions/clipped_ratio": 0.0,
668
+ "completions/max_length": 70.61538461538461,
669
+ "completions/max_terminated_length": 70.61538461538461,
670
+ "completions/mean_length": 70.23076923076923,
671
+ "completions/mean_terminated_length": 70.23076923076923,
672
+ "completions/min_length": 69.76923076923077,
673
+ "completions/min_terminated_length": 69.76923076923077,
674
+ "epoch": 0.0011101046162590362,
675
+ "frac_reward_zero_std": 0.38461538461538464,
676
+ "grad_norm": 0.00016267380851786584,
677
+ "kl": 0.1838107569515705,
678
+ "learning_rate": 2.770808818631913e-08,
679
+ "loss": 0.0022651535272598267,
680
+ "num_tokens": 579229.0,
681
+ "reward": -3.4387295062725362e-09,
682
+ "reward_std": 0.6578738139225886,
683
+ "rewards/TranscriptCorrectionGrpoReward/mean": -3.4387295062725362e-09,
684
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.6578738322624793,
685
+ "step": 625
686
+ },
687
+ {
688
+ "clip_ratio/high_max": 0.0,
689
+ "clip_ratio/high_mean": 0.0,
690
+ "clip_ratio/low_mean": 0.0,
691
+ "clip_ratio/low_min": 0.0,
692
+ "clip_ratio/region_mean": 0.0,
693
+ "completion_length": 73.21,
694
+ "completions/clipped_ratio": 0.0,
695
+ "completions/max_length": 73.33333333333333,
696
+ "completions/max_terminated_length": 73.33333333333333,
697
+ "completions/mean_length": 72.84375,
698
+ "completions/mean_terminated_length": 72.84375,
699
+ "completions/min_length": 71.66666666666667,
700
+ "completions/min_terminated_length": 71.66666666666667,
701
+ "epoch": 0.0011545088009093977,
702
+ "frac_reward_zero_std": 0.3333333333333333,
703
+ "grad_norm": 2.573253914306406e-05,
704
+ "kl": 0.17695149526000023,
705
+ "learning_rate": 2.8818187873271017e-08,
706
+ "loss": -9.701214730739594e-05,
707
+ "num_tokens": 603254.0,
708
+ "reward": 4.035731156667073e-09,
709
+ "reward_std": 0.7126966516176859,
710
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.035731156667073e-09,
711
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.7126966317494711,
712
+ "step": 650
713
+ },
714
+ {
715
+ "clip_ratio/high_max": 0.0,
716
+ "clip_ratio/high_mean": 0.0,
717
+ "clip_ratio/low_mean": 0.0,
718
+ "clip_ratio/low_min": 0.0,
719
+ "clip_ratio/region_mean": 0.0,
720
+ "completion_length": 82.16,
721
+ "completions/clipped_ratio": 0.0,
722
+ "completions/max_length": 81.46153846153847,
723
+ "completions/max_terminated_length": 81.46153846153847,
724
+ "completions/mean_length": 81.1923076923077,
725
+ "completions/mean_terminated_length": 81.1923076923077,
726
+ "completions/min_length": 81.15384615384616,
727
+ "completions/min_terminated_length": 81.15384615384616,
728
+ "epoch": 0.0011989129855597591,
729
+ "frac_reward_zero_std": 0.6923076923076923,
730
+ "grad_norm": 1.8145580042983056e-06,
731
+ "kl": 0.15570140190422535,
732
+ "learning_rate": 2.9928287560222903e-08,
733
+ "loss": 0.0013257008790969848,
734
+ "num_tokens": 632778.0,
735
+ "reward": 0.0,
736
+ "reward_std": 0.3289369069612943,
737
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
738
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.3289369069612943,
739
+ "step": 675
740
+ },
741
+ {
742
+ "clip_ratio/high_max": 0.0,
743
+ "clip_ratio/high_mean": 0.0,
744
+ "clip_ratio/low_mean": 8.849557489156722e-05,
745
+ "clip_ratio/low_min": 8.849557489156722e-05,
746
+ "clip_ratio/region_mean": 8.849557489156722e-05,
747
+ "completion_length": 61.01,
748
+ "completions/clipped_ratio": 0.0,
749
+ "completions/max_length": 61.5,
750
+ "completions/max_terminated_length": 61.5,
751
+ "completions/mean_length": 61.177083333333336,
752
+ "completions/mean_terminated_length": 61.177083333333336,
753
+ "completions/min_length": 60.25,
754
+ "completions/min_terminated_length": 60.25,
755
+ "epoch": 0.0012433171702101206,
756
+ "frac_reward_zero_std": 0.5,
757
+ "grad_norm": 0.0008123472798615694,
758
+ "kl": 0.20670814238488674,
759
+ "learning_rate": 3.103838724717479e-08,
760
+ "loss": -0.002791093587875366,
761
+ "num_tokens": 654243.0,
762
+ "reward": 6.20881716410319e-09,
763
+ "reward_std": 0.5345224738121033,
764
+ "rewards/TranscriptCorrectionGrpoReward/mean": 6.20881716410319e-09,
765
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.5345224738121033,
766
+ "step": 700
767
+ },
768
+ {
769
+ "clip_ratio/high_max": 0.0,
770
+ "clip_ratio/high_mean": 0.0,
771
+ "clip_ratio/low_mean": 0.0,
772
+ "clip_ratio/low_min": 0.0,
773
+ "clip_ratio/region_mean": 0.0,
774
+ "completion_length": 66.55,
775
+ "completions/clipped_ratio": 0.0,
776
+ "completions/max_length": 66.61538461538461,
777
+ "completions/max_terminated_length": 66.61538461538461,
778
+ "completions/mean_length": 66.29807692307692,
779
+ "completions/mean_terminated_length": 66.29807692307692,
780
+ "completions/min_length": 66.07692307692308,
781
+ "completions/min_terminated_length": 66.07692307692308,
782
+ "epoch": 0.001287721354860482,
783
+ "frac_reward_zero_std": 0.6153846153846154,
784
+ "grad_norm": 1.722803517623106e-06,
785
+ "kl": 0.18176497109234332,
786
+ "learning_rate": 3.214848693412668e-08,
787
+ "loss": 0.0018749050796031952,
788
+ "num_tokens": 678762.0,
789
+ "reward": -2.292486337515024e-09,
790
+ "reward_std": 0.4111711337016179,
791
+ "rewards/TranscriptCorrectionGrpoReward/mean": -2.292486337515024e-09,
792
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
793
+ "step": 725
794
+ },
795
+ {
796
+ "clip_ratio/high_max": 0.0,
797
+ "clip_ratio/high_mean": 0.0,
798
+ "clip_ratio/low_mean": 0.0,
799
+ "clip_ratio/low_min": 0.0,
800
+ "clip_ratio/region_mean": 0.0,
801
+ "completion_length": 64.79,
802
+ "completions/clipped_ratio": 0.0,
803
+ "completions/max_length": 65.16666666666667,
804
+ "completions/max_terminated_length": 65.16666666666667,
805
+ "completions/mean_length": 64.98958333333333,
806
+ "completions/mean_terminated_length": 64.98958333333333,
807
+ "completions/min_length": 64.91666666666667,
808
+ "completions/min_terminated_length": 64.91666666666667,
809
+ "epoch": 0.0013321255395108435,
810
+ "frac_reward_zero_std": 0.6666666666666666,
811
+ "grad_norm": 0.0003243849496357143,
812
+ "kl": 0.1709983678907156,
813
+ "learning_rate": 3.325858662107857e-08,
814
+ "loss": 0.0017569869756698608,
815
+ "num_tokens": 700945.0,
816
+ "reward": 7.450580596923828e-09,
817
+ "reward_std": 0.35634832580884296,
818
+ "rewards/TranscriptCorrectionGrpoReward/mean": 7.450580596923828e-09,
819
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
820
+ "step": 750
821
+ },
822
+ {
823
+ "clip_ratio/high_max": 0.0,
824
+ "clip_ratio/high_mean": 0.0,
825
+ "clip_ratio/low_mean": 0.0,
826
+ "clip_ratio/low_min": 0.0,
827
+ "clip_ratio/region_mean": 0.0,
828
+ "completion_length": 72.83,
829
+ "completions/clipped_ratio": 0.0,
830
+ "completions/max_length": 73.46153846153847,
831
+ "completions/max_terminated_length": 73.46153846153847,
832
+ "completions/mean_length": 73.29807692307692,
833
+ "completions/mean_terminated_length": 73.29807692307692,
834
+ "completions/min_length": 73.15384615384616,
835
+ "completions/min_terminated_length": 73.15384615384616,
836
+ "epoch": 0.001376529724161205,
837
+ "frac_reward_zero_std": 0.38461538461538464,
838
+ "grad_norm": 0.2545297145843506,
839
+ "kl": 0.18227736078202725,
840
+ "learning_rate": 3.4368686308030454e-08,
841
+ "loss": -0.009244667887687684,
842
+ "num_tokens": 728240.0,
843
+ "reward": 0.0,
844
+ "reward_std": 0.6578738139225886,
845
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
846
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.6578738139225886,
847
+ "step": 775
848
+ },
849
+ {
850
+ "clip_ratio/high_max": 0.0,
851
+ "clip_ratio/high_mean": 0.0,
852
+ "clip_ratio/low_mean": 0.0,
853
+ "clip_ratio/low_min": 0.0,
854
+ "clip_ratio/region_mean": 0.0,
855
+ "completion_length": 62.07,
856
+ "completions/clipped_ratio": 0.0,
857
+ "completions/max_length": 61.416666666666664,
858
+ "completions/max_terminated_length": 61.416666666666664,
859
+ "completions/mean_length": 61.114583333333336,
860
+ "completions/mean_terminated_length": 61.114583333333336,
861
+ "completions/min_length": 60.916666666666664,
862
+ "completions/min_terminated_length": 60.916666666666664,
863
+ "epoch": 0.0014209339088115665,
864
+ "frac_reward_zero_std": 0.6666666666666666,
865
+ "grad_norm": 0.0001137235522037372,
866
+ "kl": 0.20930900558829307,
867
+ "learning_rate": 3.547878599498235e-08,
868
+ "loss": 0.012507272958755493,
869
+ "num_tokens": 749099.0,
870
+ "reward": 0.0,
871
+ "reward_std": 0.35634830594062805,
872
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
873
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
874
+ "step": 800
875
+ },
876
+ {
877
+ "clip_ratio/high_max": 0.0,
878
+ "clip_ratio/high_mean": 0.0,
879
+ "clip_ratio/low_mean": 0.0,
880
+ "clip_ratio/low_min": 0.0,
881
+ "clip_ratio/region_mean": 0.0,
882
+ "completion_length": 79.05,
883
+ "completions/clipped_ratio": 0.0,
884
+ "completions/max_length": 79.38461538461539,
885
+ "completions/max_terminated_length": 79.38461538461539,
886
+ "completions/mean_length": 79.08653846153847,
887
+ "completions/mean_terminated_length": 79.08653846153847,
888
+ "completions/min_length": 78.92307692307692,
889
+ "completions/min_terminated_length": 78.92307692307692,
890
+ "epoch": 0.001465338093461928,
891
+ "frac_reward_zero_std": 0.46153846153846156,
892
+ "grad_norm": 0.00025333481607958674,
893
+ "kl": 0.1384541442245245,
894
+ "learning_rate": 3.658888568193424e-08,
895
+ "loss": 0.0011119496822357177,
896
+ "num_tokens": 778348.0,
897
+ "reward": 3.4387295062725362e-09,
898
+ "reward_std": 0.5756396055221558,
899
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.4387295062725362e-09,
900
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.575639587182265,
901
+ "step": 825
902
+ },
903
+ {
904
+ "clip_ratio/high_max": 0.0,
905
+ "clip_ratio/high_mean": 0.0,
906
+ "clip_ratio/low_mean": 0.0,
907
+ "clip_ratio/low_min": 0.0,
908
+ "clip_ratio/region_mean": 0.0,
909
+ "completion_length": 71.36,
910
+ "completions/clipped_ratio": 0.0,
911
+ "completions/max_length": 71.16666666666667,
912
+ "completions/max_terminated_length": 71.16666666666667,
913
+ "completions/mean_length": 71.0,
914
+ "completions/mean_terminated_length": 71.0,
915
+ "completions/min_length": 70.91666666666667,
916
+ "completions/min_terminated_length": 70.91666666666667,
917
+ "epoch": 0.0015097422781122894,
918
+ "frac_reward_zero_std": 0.5833333333333334,
919
+ "grad_norm": 0.013177081011235714,
920
+ "kl": 0.17892150908708573,
921
+ "learning_rate": 3.769898536888613e-08,
922
+ "loss": 0.002372233867645264,
923
+ "num_tokens": 802668.0,
924
+ "reward": 3.725290298461914e-09,
925
+ "reward_std": 0.4454353948434194,
926
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.725290298461914e-09,
927
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
928
+ "step": 850
929
+ },
930
+ {
931
+ "clip_ratio/high_max": 0.0,
932
+ "clip_ratio/high_mean": 0.0,
933
+ "clip_ratio/low_mean": 0.00011363636702299118,
934
+ "clip_ratio/low_min": 0.00011363636702299118,
935
+ "clip_ratio/region_mean": 0.00011363636702299118,
936
+ "completion_length": 78.5,
937
+ "completions/clipped_ratio": 0.0,
938
+ "completions/max_length": 76.84615384615384,
939
+ "completions/max_terminated_length": 76.84615384615384,
940
+ "completions/mean_length": 76.71153846153847,
941
+ "completions/mean_terminated_length": 76.71153846153847,
942
+ "completions/min_length": 76.38461538461539,
943
+ "completions/min_terminated_length": 76.38461538461539,
944
+ "epoch": 0.0015541464627626509,
945
+ "frac_reward_zero_std": 0.6153846153846154,
946
+ "grad_norm": 0.38749536871910095,
947
+ "kl": 0.17304533801972866,
948
+ "learning_rate": 3.880908505583801e-08,
949
+ "loss": -0.004066205024719239,
950
+ "num_tokens": 831014.0,
951
+ "reward": 2.292486337515024e-09,
952
+ "reward_std": 0.4111711337016179,
953
+ "rewards/TranscriptCorrectionGrpoReward/mean": 2.292486337515024e-09,
954
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
955
+ "step": 875
956
+ },
957
+ {
958
+ "clip_ratio/high_max": 0.0,
959
+ "clip_ratio/high_mean": 0.0,
960
+ "clip_ratio/low_mean": 0.0,
961
+ "clip_ratio/low_min": 0.0,
962
+ "clip_ratio/region_mean": 0.0,
963
+ "completion_length": 69.88,
964
+ "completions/clipped_ratio": 0.0,
965
+ "completions/max_length": 71.66666666666667,
966
+ "completions/max_terminated_length": 71.66666666666667,
967
+ "completions/mean_length": 71.45833333333333,
968
+ "completions/mean_terminated_length": 71.45833333333333,
969
+ "completions/min_length": 71.25,
970
+ "completions/min_terminated_length": 71.25,
971
+ "epoch": 0.001598550647413012,
972
+ "frac_reward_zero_std": 0.5,
973
+ "grad_norm": 0.2461424618959427,
974
+ "kl": 0.2068350876867771,
975
+ "learning_rate": 3.99191847427899e-08,
976
+ "loss": 0.007263134121894836,
977
+ "num_tokens": 855450.0,
978
+ "reward": 6.20881716410319e-09,
979
+ "reward_std": 0.5345224638779958,
980
+ "rewards/TranscriptCorrectionGrpoReward/mean": 6.20881716410319e-09,
981
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.5345224738121033,
982
+ "step": 900
983
+ },
984
+ {
985
+ "clip_ratio/high_max": 0.0,
986
+ "clip_ratio/high_mean": 0.0,
987
+ "clip_ratio/low_mean": 0.0,
988
+ "clip_ratio/low_min": 0.0,
989
+ "clip_ratio/region_mean": 0.0,
990
+ "completion_length": 72.65,
991
+ "completions/clipped_ratio": 0.0,
992
+ "completions/max_length": 74.76923076923077,
993
+ "completions/max_terminated_length": 74.76923076923077,
994
+ "completions/mean_length": 74.50961538461539,
995
+ "completions/mean_terminated_length": 74.50961538461539,
996
+ "completions/min_length": 74.23076923076923,
997
+ "completions/min_terminated_length": 74.23076923076923,
998
+ "epoch": 0.0016429548320633736,
999
+ "frac_reward_zero_std": 0.6923076923076923,
1000
+ "grad_norm": 3.1216541174217127e-07,
1001
+ "kl": 0.1654582280665636,
1002
+ "learning_rate": 4.102928442974179e-08,
1003
+ "loss": 0.0022979089617729187,
1004
+ "num_tokens": 883151.0,
1005
+ "reward": -1.7193647531362681e-09,
1006
+ "reward_std": 0.3289369069612943,
1007
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.7193647531362681e-09,
1008
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.3289369069612943,
1009
+ "step": 925
1010
+ },
1011
+ {
1012
+ "clip_ratio/high_max": 0.0,
1013
+ "clip_ratio/high_mean": 0.0,
1014
+ "clip_ratio/low_mean": 0.0,
1015
+ "clip_ratio/low_min": 0.0,
1016
+ "clip_ratio/region_mean": 0.0,
1017
+ "completion_length": 66.25,
1018
+ "completions/clipped_ratio": 0.0,
1019
+ "completions/max_length": 64.33333333333333,
1020
+ "completions/max_terminated_length": 64.33333333333333,
1021
+ "completions/mean_length": 63.96875,
1022
+ "completions/mean_terminated_length": 63.96875,
1023
+ "completions/min_length": 63.833333333333336,
1024
+ "completions/min_terminated_length": 63.833333333333336,
1025
+ "epoch": 0.001687359016713735,
1026
+ "frac_reward_zero_std": 0.5,
1027
+ "grad_norm": 0.3891129791736603,
1028
+ "kl": 0.17953816015273333,
1029
+ "learning_rate": 4.213938411669368e-08,
1030
+ "loss": 0.0029295945167541505,
1031
+ "num_tokens": 905076.0,
1032
+ "reward": 3.725290298461914e-09,
1033
+ "reward_std": 0.5345224738121033,
1034
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.725290298461914e-09,
1035
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.5345224738121033,
1036
+ "step": 950
1037
+ },
1038
+ {
1039
+ "clip_ratio/high_max": 0.0,
1040
+ "clip_ratio/high_mean": 0.0,
1041
+ "clip_ratio/low_mean": 0.0,
1042
+ "clip_ratio/low_min": 0.0,
1043
+ "clip_ratio/region_mean": 0.0,
1044
+ "completion_length": 64.85,
1045
+ "completions/clipped_ratio": 0.0,
1046
+ "completions/max_length": 65.23076923076923,
1047
+ "completions/max_terminated_length": 65.23076923076923,
1048
+ "completions/mean_length": 64.9326923076923,
1049
+ "completions/mean_terminated_length": 64.9326923076923,
1050
+ "completions/min_length": 64.38461538461539,
1051
+ "completions/min_terminated_length": 64.38461538461539,
1052
+ "epoch": 0.0017317632013640965,
1053
+ "frac_reward_zero_std": 0.5384615384615384,
1054
+ "grad_norm": 2.965283556477516e-06,
1055
+ "kl": 0.19248799994587898,
1056
+ "learning_rate": 4.324948380364556e-08,
1057
+ "loss": 0.0011707821488380432,
1058
+ "num_tokens": 929237.0,
1059
+ "reward": 1.7193647531362681e-09,
1060
+ "reward_std": 0.4934053604419415,
1061
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.7193647531362681e-09,
1062
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
1063
+ "step": 975
1064
+ },
1065
+ {
1066
+ "clip_ratio/high_max": 0.0,
1067
+ "clip_ratio/high_mean": 0.0,
1068
+ "clip_ratio/low_mean": 0.0,
1069
+ "clip_ratio/low_min": 0.0,
1070
+ "clip_ratio/region_mean": 0.0,
1071
+ "completion_length": 58.09,
1072
+ "completions/clipped_ratio": 0.0,
1073
+ "completions/max_length": 58.0,
1074
+ "completions/max_terminated_length": 58.0,
1075
+ "completions/mean_length": 57.71875,
1076
+ "completions/mean_terminated_length": 57.71875,
1077
+ "completions/min_length": 57.416666666666664,
1078
+ "completions/min_terminated_length": 57.416666666666664,
1079
+ "epoch": 0.001776167386014458,
1080
+ "frac_reward_zero_std": 0.6666666666666666,
1081
+ "grad_norm": 0.0015717418864369392,
1082
+ "kl": 0.26020569302141666,
1083
+ "learning_rate": 4.435958349059745e-08,
1084
+ "loss": 0.00038701888173818587,
1085
+ "num_tokens": 949458.0,
1086
+ "reward": -3.725290298461914e-09,
1087
+ "reward_std": 0.35634831587473553,
1088
+ "rewards/TranscriptCorrectionGrpoReward/mean": -3.725290298461914e-09,
1089
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
1090
+ "step": 1000
1091
+ },
1092
+ {
1093
+ "eval_cer_subset": 0.01423290504339896,
1094
+ "eval_cer_subset_edit_distance": 874,
1095
+ "eval_cer_subset_groups": 250,
1096
+ "eval_cer_subset_items": 250,
1097
+ "eval_cer_subset_ref_chars": 61407,
1098
+ "eval_loss": 0.03394044979410833,
1099
+ "step": 1013
1100
+ },
1101
+ {
1102
+ "clip_ratio/high_max": 0.0,
1103
+ "clip_ratio/high_mean": 0.0,
1104
+ "clip_ratio/low_mean": 0.0,
1105
+ "clip_ratio/low_min": 0.0,
1106
+ "clip_ratio/region_mean": 0.0,
1107
+ "completion_length": 75.01,
1108
+ "completions/clipped_ratio": 0.0,
1109
+ "completions/max_length": 75.15384615384616,
1110
+ "completions/max_terminated_length": 75.15384615384616,
1111
+ "completions/mean_length": 74.70192307692308,
1112
+ "completions/mean_terminated_length": 74.70192307692308,
1113
+ "completions/min_length": 74.46153846153847,
1114
+ "completions/min_terminated_length": 74.46153846153847,
1115
+ "epoch": 0.0018205715706648194,
1116
+ "frac_reward_zero_std": 0.38461538461538464,
1117
+ "grad_norm": 0.00012233731104061007,
1118
+ "kl": 0.16456884417682885,
1119
+ "learning_rate": 4.5469683177549347e-08,
1120
+ "loss": 0.002440948486328125,
1121
+ "num_tokens": 976371.0,
1122
+ "reward": 5.158094259408804e-09,
1123
+ "reward_std": 0.6578738139225886,
1124
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.158094259408804e-09,
1125
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.6578738139225886,
1126
+ "step": 1025
1127
+ },
1128
+ {
1129
+ "clip_ratio/high_max": 0.0,
1130
+ "clip_ratio/high_mean": 0.0,
1131
+ "clip_ratio/low_mean": 0.0,
1132
+ "clip_ratio/low_min": 0.0,
1133
+ "clip_ratio/region_mean": 0.0,
1134
+ "completion_length": 48.74,
1135
+ "completions/clipped_ratio": 0.0,
1136
+ "completions/max_length": 48.083333333333336,
1137
+ "completions/max_terminated_length": 48.083333333333336,
1138
+ "completions/mean_length": 47.979166666666664,
1139
+ "completions/mean_terminated_length": 47.979166666666664,
1140
+ "completions/min_length": 47.833333333333336,
1141
+ "completions/min_terminated_length": 47.833333333333336,
1142
+ "epoch": 0.001864975755315181,
1143
+ "frac_reward_zero_std": 0.6666666666666666,
1144
+ "grad_norm": 1.2304620213399176e-05,
1145
+ "kl": 0.2251258111000061,
1146
+ "learning_rate": 4.6579782864501236e-08,
1147
+ "loss": 0.0019197005033493042,
1148
+ "num_tokens": 993665.0,
1149
+ "reward": 1.241763432820638e-09,
1150
+ "reward_std": 0.35634832580884296,
1151
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-09,
1152
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
1153
+ "step": 1050
1154
+ },
1155
+ {
1156
+ "clip_ratio/high_max": 0.0,
1157
+ "clip_ratio/high_mean": 0.0,
1158
+ "clip_ratio/low_mean": 0.0,
1159
+ "clip_ratio/low_min": 0.0,
1160
+ "clip_ratio/region_mean": 0.0,
1161
+ "completion_length": 64.11,
1162
+ "completions/clipped_ratio": 0.0,
1163
+ "completions/max_length": 64.15384615384616,
1164
+ "completions/max_terminated_length": 64.15384615384616,
1165
+ "completions/mean_length": 63.875,
1166
+ "completions/mean_terminated_length": 63.875,
1167
+ "completions/min_length": 63.69230769230769,
1168
+ "completions/min_terminated_length": 63.69230769230769,
1169
+ "epoch": 0.0019093799399655424,
1170
+ "frac_reward_zero_std": 0.6923076923076923,
1171
+ "grad_norm": 1.1637333727776422e-06,
1172
+ "kl": 0.1967310430854559,
1173
+ "learning_rate": 4.768988255145312e-08,
1174
+ "loss": 0.0027909329533576965,
1175
+ "num_tokens": 1017476.0,
1176
+ "reward": 1.146243168757512e-09,
1177
+ "reward_std": 0.328936897791349,
1178
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.146243168757512e-09,
1179
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.3289369069612943,
1180
+ "step": 1075
1181
+ },
1182
+ {
1183
+ "clip_ratio/high_max": 0.0,
1184
+ "clip_ratio/high_mean": 0.0,
1185
+ "clip_ratio/low_mean": 0.0,
1186
+ "clip_ratio/low_min": 0.0,
1187
+ "clip_ratio/region_mean": 0.0,
1188
+ "completion_length": 60.78,
1189
+ "completions/clipped_ratio": 0.0,
1190
+ "completions/max_length": 61.0,
1191
+ "completions/max_terminated_length": 61.0,
1192
+ "completions/mean_length": 60.895833333333336,
1193
+ "completions/mean_terminated_length": 60.895833333333336,
1194
+ "completions/min_length": 60.75,
1195
+ "completions/min_terminated_length": 60.75,
1196
+ "epoch": 0.0019537841246159036,
1197
+ "frac_reward_zero_std": 0.75,
1198
+ "grad_norm": 0.12277305871248245,
1199
+ "kl": 0.1913923419266939,
1200
+ "learning_rate": 4.879998223840501e-08,
1201
+ "loss": 0.0014767789840698243,
1202
+ "num_tokens": 1038338.0,
1203
+ "reward": 0.0,
1204
+ "reward_std": 0.26726123690605164,
1205
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1206
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.26726123690605164,
1207
+ "step": 1100
1208
+ },
1209
+ {
1210
+ "clip_ratio/high_max": 0.0,
1211
+ "clip_ratio/high_mean": 0.0,
1212
+ "clip_ratio/low_mean": 0.0,
1213
+ "clip_ratio/low_min": 0.0,
1214
+ "clip_ratio/region_mean": 0.0,
1215
+ "completion_length": 77.15,
1216
+ "completions/clipped_ratio": 0.0,
1217
+ "completions/max_length": 75.6923076923077,
1218
+ "completions/max_terminated_length": 75.6923076923077,
1219
+ "completions/mean_length": 75.49038461538461,
1220
+ "completions/mean_terminated_length": 75.49038461538461,
1221
+ "completions/min_length": 75.23076923076923,
1222
+ "completions/min_terminated_length": 75.23076923076923,
1223
+ "epoch": 0.001998188309266265,
1224
+ "frac_reward_zero_std": 0.46153846153846156,
1225
+ "grad_norm": 5.799914106319193e-06,
1226
+ "kl": 0.1744626285880804,
1227
+ "learning_rate": 4.99100819253569e-08,
1228
+ "loss": 0.0021522629261016848,
1229
+ "num_tokens": 1065805.0,
1230
+ "reward": 4.584972675030048e-09,
1231
+ "reward_std": 0.575639587182265,
1232
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.584972675030048e-09,
1233
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.575639587182265,
1234
+ "step": 1125
1235
+ },
1236
+ {
1237
+ "clip_ratio/high_max": 0.0,
1238
+ "clip_ratio/high_mean": 0.0,
1239
+ "clip_ratio/low_mean": 0.0,
1240
+ "clip_ratio/low_min": 0.0,
1241
+ "clip_ratio/region_mean": 0.0,
1242
+ "completion_length": 58.0,
1243
+ "completions/clipped_ratio": 0.0,
1244
+ "completions/max_length": 59.083333333333336,
1245
+ "completions/max_terminated_length": 59.083333333333336,
1246
+ "completions/mean_length": 59.0,
1247
+ "completions/mean_terminated_length": 59.0,
1248
+ "completions/min_length": 58.833333333333336,
1249
+ "completions/min_terminated_length": 58.833333333333336,
1250
+ "epoch": 0.0020425924939166265,
1251
+ "frac_reward_zero_std": 0.6666666666666666,
1252
+ "grad_norm": 0.27839189767837524,
1253
+ "kl": 0.2037867596000433,
1254
+ "learning_rate": 5.102018161230878e-08,
1255
+ "loss": 0.0014832359552383423,
1256
+ "num_tokens": 1086317.0,
1257
+ "reward": 2.483526865641276e-09,
1258
+ "reward_std": 0.35634832580884296,
1259
+ "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
1260
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
1261
+ "step": 1150
1262
+ },
1263
+ {
1264
+ "clip_ratio/high_max": 0.0,
1265
+ "clip_ratio/high_mean": 0.0,
1266
+ "clip_ratio/low_mean": 0.0,
1267
+ "clip_ratio/low_min": 0.0,
1268
+ "clip_ratio/region_mean": 0.0,
1269
+ "completion_length": 72.6,
1270
+ "completions/clipped_ratio": 0.0,
1271
+ "completions/max_length": 73.3076923076923,
1272
+ "completions/max_terminated_length": 73.3076923076923,
1273
+ "completions/mean_length": 72.6923076923077,
1274
+ "completions/mean_terminated_length": 72.6923076923077,
1275
+ "completions/min_length": 72.23076923076923,
1276
+ "completions/min_terminated_length": 72.23076923076923,
1277
+ "epoch": 0.002086996678566988,
1278
+ "frac_reward_zero_std": 0.6153846153846154,
1279
+ "grad_norm": 0.00011730552068911493,
1280
+ "kl": 0.16116364367306232,
1281
+ "learning_rate": 5.213028129926067e-08,
1282
+ "loss": 0.001632765084505081,
1283
+ "num_tokens": 1113093.0,
1284
+ "reward": 3.4387295062725362e-09,
1285
+ "reward_std": 0.4111711337016179,
1286
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.4387295062725362e-09,
1287
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
1288
+ "step": 1175
1289
+ },
1290
+ {
1291
+ "clip_ratio/high_max": 0.0,
1292
+ "clip_ratio/high_mean": 0.0,
1293
+ "clip_ratio/low_mean": 0.0,
1294
+ "clip_ratio/low_min": 0.0,
1295
+ "clip_ratio/region_mean": 0.0,
1296
+ "completion_length": 62.91,
1297
+ "completions/clipped_ratio": 0.0,
1298
+ "completions/max_length": 62.5,
1299
+ "completions/max_terminated_length": 62.5,
1300
+ "completions/mean_length": 62.40625,
1301
+ "completions/mean_terminated_length": 62.40625,
1302
+ "completions/min_length": 62.333333333333336,
1303
+ "completions/min_terminated_length": 62.333333333333336,
1304
+ "epoch": 0.0021314008632173495,
1305
+ "frac_reward_zero_std": 0.75,
1306
+ "grad_norm": 0.0802081748843193,
1307
+ "kl": 0.19629520073533058,
1308
+ "learning_rate": 5.324038098621256e-08,
1309
+ "loss": 0.0019979089498519897,
1310
+ "num_tokens": 1134580.0,
1311
+ "reward": -2.483526865641276e-09,
1312
+ "reward_std": 0.2672612468401591,
1313
+ "rewards/TranscriptCorrectionGrpoReward/mean": -2.483526865641276e-09,
1314
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.26726123690605164,
1315
+ "step": 1200
1316
+ },
1317
+ {
1318
+ "clip_ratio/high_max": 0.0,
1319
+ "clip_ratio/high_mean": 0.0,
1320
+ "clip_ratio/low_mean": 0.0,
1321
+ "clip_ratio/low_min": 0.0,
1322
+ "clip_ratio/region_mean": 0.0,
1323
+ "completion_length": 68.7,
1324
+ "completions/clipped_ratio": 0.0,
1325
+ "completions/max_length": 68.46153846153847,
1326
+ "completions/max_terminated_length": 68.46153846153847,
1327
+ "completions/mean_length": 68.28846153846153,
1328
+ "completions/mean_terminated_length": 68.28846153846153,
1329
+ "completions/min_length": 68.07692307692308,
1330
+ "completions/min_terminated_length": 68.07692307692308,
1331
+ "epoch": 0.002175805047867711,
1332
+ "frac_reward_zero_std": 0.46153846153846156,
1333
+ "grad_norm": 0.0008079666295088828,
1334
+ "kl": 0.18986737683415414,
1335
+ "learning_rate": 5.435048067316445e-08,
1336
+ "loss": 0.0016339871287345887,
1337
+ "num_tokens": 1159506.0,
1338
+ "reward": 4.011851090651292e-09,
1339
+ "reward_std": 0.575639587182265,
1340
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.011851090651292e-09,
1341
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.575639587182265,
1342
+ "step": 1225
1343
+ },
1344
+ {
1345
+ "clip_ratio/high_max": 0.0,
1346
+ "clip_ratio/high_mean": 0.0,
1347
+ "clip_ratio/low_mean": 0.0,
1348
+ "clip_ratio/low_min": 0.0,
1349
+ "clip_ratio/region_mean": 0.0,
1350
+ "completion_length": 68.46,
1351
+ "completions/clipped_ratio": 0.0,
1352
+ "completions/max_length": 69.08333333333333,
1353
+ "completions/max_terminated_length": 69.08333333333333,
1354
+ "completions/mean_length": 68.89583333333333,
1355
+ "completions/mean_terminated_length": 68.89583333333333,
1356
+ "completions/min_length": 68.41666666666667,
1357
+ "completions/min_terminated_length": 68.41666666666667,
1358
+ "epoch": 0.0022202092325180724,
1359
+ "frac_reward_zero_std": 0.5833333333333334,
1360
+ "grad_norm": 0.1505577713251114,
1361
+ "kl": 0.17770596094429492,
1362
+ "learning_rate": 5.5460580360116337e-08,
1363
+ "loss": 0.0011485755443572999,
1364
+ "num_tokens": 1183152.0,
1365
+ "reward": 0.0,
1366
+ "reward_std": 0.44543540477752686,
1367
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1368
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
1369
+ "step": 1250
1370
+ },
1371
+ {
1372
+ "clip_ratio/high_max": 0.0,
1373
+ "clip_ratio/high_mean": 0.0,
1374
+ "clip_ratio/low_mean": 0.0,
1375
+ "clip_ratio/low_min": 0.0,
1376
+ "clip_ratio/region_mean": 0.0,
1377
+ "completion_length": 59.73,
1378
+ "completions/clipped_ratio": 0.0,
1379
+ "completions/max_length": 60.53846153846154,
1380
+ "completions/max_terminated_length": 60.53846153846154,
1381
+ "completions/mean_length": 60.40384615384615,
1382
+ "completions/mean_terminated_length": 60.40384615384615,
1383
+ "completions/min_length": 60.30769230769231,
1384
+ "completions/min_terminated_length": 60.30769230769231,
1385
+ "epoch": 0.002264613417168434,
1386
+ "frac_reward_zero_std": 0.6153846153846154,
1387
+ "grad_norm": 0.013753416948020458,
1388
+ "kl": 0.19532982856035233,
1389
+ "learning_rate": 5.6570680047068226e-08,
1390
+ "loss": -0.005677615404129028,
1391
+ "num_tokens": 1205850.0,
1392
+ "reward": 0.0,
1393
+ "reward_std": 0.4111711337016179,
1394
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1395
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4111711337016179,
1396
+ "step": 1275
1397
+ },
1398
+ {
1399
+ "clip_ratio/high_max": 0.0,
1400
+ "clip_ratio/high_mean": 0.0,
1401
+ "clip_ratio/low_mean": 0.0,
1402
+ "clip_ratio/low_min": 0.0,
1403
+ "clip_ratio/region_mean": 0.0,
1404
+ "completion_length": 69.81,
1405
+ "completions/clipped_ratio": 0.0,
1406
+ "completions/max_length": 69.5,
1407
+ "completions/max_terminated_length": 69.5,
1408
+ "completions/mean_length": 69.5,
1409
+ "completions/mean_terminated_length": 69.5,
1410
+ "completions/min_length": 69.5,
1411
+ "completions/min_terminated_length": 69.5,
1412
+ "epoch": 0.0023090176018187953,
1413
+ "frac_reward_zero_std": 1.0,
1414
+ "grad_norm": 4.925558823742904e-05,
1415
+ "kl": 0.1651756139844656,
1416
+ "learning_rate": 5.7680779734020115e-08,
1417
+ "loss": 0.009348901510238648,
1418
+ "num_tokens": 1230002.0,
1419
+ "reward": 0.0,
1420
+ "reward_std": 0.0,
1421
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1422
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.0,
1423
+ "step": 1300
1424
+ },
1425
+ {
1426
+ "clip_ratio/high_max": 0.0,
1427
+ "clip_ratio/high_mean": 0.0,
1428
+ "clip_ratio/low_mean": 0.0,
1429
+ "clip_ratio/low_min": 0.0,
1430
+ "clip_ratio/region_mean": 0.0,
1431
+ "completion_length": 57.67,
1432
+ "completions/clipped_ratio": 0.0,
1433
+ "completions/max_length": 57.15384615384615,
1434
+ "completions/max_terminated_length": 57.15384615384615,
1435
+ "completions/mean_length": 56.99038461538461,
1436
+ "completions/mean_terminated_length": 56.99038461538461,
1437
+ "completions/min_length": 56.76923076923077,
1438
+ "completions/min_terminated_length": 56.76923076923077,
1439
+ "epoch": 0.002353421786469157,
1440
+ "frac_reward_zero_std": 0.5384615384615384,
1441
+ "grad_norm": 7.053396257106215e-05,
1442
+ "kl": 0.21575049236416816,
1443
+ "learning_rate": 5.8790879420972004e-08,
1444
+ "loss": 0.0014468547701835632,
1445
+ "num_tokens": 1251097.0,
1446
+ "reward": -1.146243168757512e-09,
1447
+ "reward_std": 0.4934053604419415,
1448
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.146243168757512e-09,
1449
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.49340536961188686,
1450
+ "step": 1325
1451
+ },
1452
+ {
1453
+ "clip_ratio/high_max": 0.0,
1454
+ "clip_ratio/high_mean": 0.0,
1455
+ "clip_ratio/low_mean": 0.0,
1456
+ "clip_ratio/low_min": 0.0,
1457
+ "clip_ratio/region_mean": 0.0,
1458
+ "completion_length": 67.94,
1459
+ "completions/clipped_ratio": 0.0,
1460
+ "completions/max_length": 69.33333333333333,
1461
+ "completions/max_terminated_length": 69.33333333333333,
1462
+ "completions/mean_length": 69.10416666666667,
1463
+ "completions/mean_terminated_length": 69.10416666666667,
1464
+ "completions/min_length": 68.91666666666667,
1465
+ "completions/min_terminated_length": 68.91666666666667,
1466
+ "epoch": 0.0023978259711195183,
1467
+ "frac_reward_zero_std": 0.5833333333333334,
1468
+ "grad_norm": 0.18296878039836884,
1469
+ "kl": 0.17479414448142053,
1470
+ "learning_rate": 5.990097910792389e-08,
1471
+ "loss": 0.0015808692574501037,
1472
+ "num_tokens": 1274979.0,
1473
+ "reward": 6.20881716410319e-09,
1474
+ "reward_std": 0.4454353948434194,
1475
+ "rewards/TranscriptCorrectionGrpoReward/mean": 6.20881716410319e-09,
1476
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.44543540477752686,
1477
+ "step": 1350
1478
+ },
1479
+ {
1480
+ "clip_ratio/high_max": 0.0,
1481
+ "clip_ratio/high_mean": 0.0,
1482
+ "clip_ratio/low_mean": 0.0,
1483
+ "clip_ratio/low_min": 0.0,
1484
+ "clip_ratio/region_mean": 0.0,
1485
+ "completion_length": 68.77,
1486
+ "completions/clipped_ratio": 0.0,
1487
+ "completions/max_length": 70.3076923076923,
1488
+ "completions/max_terminated_length": 70.3076923076923,
1489
+ "completions/mean_length": 70.125,
1490
+ "completions/mean_terminated_length": 70.125,
1491
+ "completions/min_length": 70.0,
1492
+ "completions/min_terminated_length": 70.0,
1493
+ "epoch": 0.0024422301557698797,
1494
+ "frac_reward_zero_std": 0.5384615384615384,
1495
+ "grad_norm": 0.09749742597341537,
1496
+ "kl": 0.1687933798879385,
1497
+ "learning_rate": 6.101107879487578e-08,
1498
+ "loss": 0.0013929146528244018,
1499
+ "num_tokens": 1300904.0,
1500
+ "reward": -1.146243168757512e-09,
1501
+ "reward_std": 0.4934053604419415,
1502
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.146243168757512e-09,
1503
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
1504
+ "step": 1375
1505
+ },
1506
+ {
1507
+ "clip_ratio/high_max": 0.0,
1508
+ "clip_ratio/high_mean": 0.0,
1509
+ "clip_ratio/low_mean": 0.0,
1510
+ "clip_ratio/low_min": 0.0,
1511
+ "clip_ratio/region_mean": 0.0,
1512
+ "completion_length": 56.27,
1513
+ "completions/clipped_ratio": 0.0,
1514
+ "completions/max_length": 54.416666666666664,
1515
+ "completions/max_terminated_length": 54.416666666666664,
1516
+ "completions/mean_length": 54.28125,
1517
+ "completions/mean_terminated_length": 54.28125,
1518
+ "completions/min_length": 54.083333333333336,
1519
+ "completions/min_terminated_length": 54.083333333333336,
1520
+ "epoch": 0.002486634340420241,
1521
+ "frac_reward_zero_std": 0.4166666666666667,
1522
+ "grad_norm": 3.840520730591379e-05,
1523
+ "kl": 0.3089423456788063,
1524
+ "learning_rate": 6.212117848182767e-08,
1525
+ "loss": 0.0018638545274734498,
1526
+ "num_tokens": 1320219.0,
1527
+ "reward": 2.483526865641276e-09,
1528
+ "reward_std": 0.6236095527807871,
1529
+ "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
1530
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.6236095527807871,
1531
+ "step": 1400
1532
+ },
1533
+ {
1534
+ "clip_ratio/high_max": 0.0,
1535
+ "clip_ratio/high_mean": 0.0,
1536
+ "clip_ratio/low_mean": 0.0,
1537
+ "clip_ratio/low_min": 0.0,
1538
+ "clip_ratio/region_mean": 0.0,
1539
+ "completion_length": 58.14,
1540
+ "completions/clipped_ratio": 0.0,
1541
+ "completions/max_length": 57.76923076923077,
1542
+ "completions/max_terminated_length": 57.76923076923077,
1543
+ "completions/mean_length": 57.63461538461539,
1544
+ "completions/mean_terminated_length": 57.63461538461539,
1545
+ "completions/min_length": 57.53846153846154,
1546
+ "completions/min_terminated_length": 57.53846153846154,
1547
+ "epoch": 0.0025310385250706027,
1548
+ "frac_reward_zero_std": 0.7692307692307693,
1549
+ "grad_norm": 2.807289320116979e-06,
1550
+ "kl": 0.18068893946707248,
1551
+ "learning_rate": 6.323127816877955e-08,
1552
+ "loss": 0.001833959072828293,
1553
+ "num_tokens": 1341893.0,
1554
+ "reward": -1.146243168757512e-09,
1555
+ "reward_std": 0.24670268939091608,
1556
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.146243168757512e-09,
1557
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.24670268022097075,
1558
+ "step": 1425
1559
+ },
1560
+ {
1561
+ "clip_ratio/high_max": 0.0,
1562
+ "clip_ratio/high_mean": 0.0,
1563
+ "clip_ratio/low_mean": 0.0,
1564
+ "clip_ratio/low_min": 0.0,
1565
+ "clip_ratio/region_mean": 0.0,
1566
+ "completion_length": 69.21,
1567
+ "completions/clipped_ratio": 0.0,
1568
+ "completions/max_length": 70.66666666666667,
1569
+ "completions/max_terminated_length": 70.66666666666667,
1570
+ "completions/mean_length": 70.21875,
1571
+ "completions/mean_terminated_length": 70.21875,
1572
+ "completions/min_length": 69.91666666666667,
1573
+ "completions/min_terminated_length": 69.91666666666667,
1574
+ "epoch": 0.002575442709720964,
1575
+ "frac_reward_zero_std": 0.4166666666666667,
1576
+ "grad_norm": 3.650064002158615e-07,
1577
+ "kl": 0.17660438932478428,
1578
+ "learning_rate": 6.434137785573144e-08,
1579
+ "loss": 0.002189251184463501,
1580
+ "num_tokens": 1366362.0,
1581
+ "reward": -1.862645149230957e-09,
1582
+ "reward_std": 0.6236095428466797,
1583
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.862645149230957e-09,
1584
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.6236095527807871,
1585
+ "step": 1450
1586
+ },
1587
+ {
1588
+ "clip_ratio/high_max": 0.0,
1589
+ "clip_ratio/high_mean": 0.0,
1590
+ "clip_ratio/low_mean": 0.0,
1591
+ "clip_ratio/low_min": 0.0,
1592
+ "clip_ratio/region_mean": 0.0,
1593
+ "completion_length": 72.56,
1594
+ "completions/clipped_ratio": 0.0,
1595
+ "completions/max_length": 72.15384615384616,
1596
+ "completions/max_terminated_length": 72.15384615384616,
1597
+ "completions/mean_length": 72.0,
1598
+ "completions/mean_terminated_length": 72.0,
1599
+ "completions/min_length": 71.92307692307692,
1600
+ "completions/min_terminated_length": 71.92307692307692,
1601
+ "epoch": 0.0026198468943713256,
1602
+ "frac_reward_zero_std": 0.6923076923076923,
1603
+ "grad_norm": 3.0107104976195842e-05,
1604
+ "kl": 0.15023484855890273,
1605
+ "learning_rate": 6.545147754268333e-08,
1606
+ "loss": 0.002286182641983032,
1607
+ "num_tokens": 1393282.0,
1608
+ "reward": 1.146243168757512e-09,
1609
+ "reward_std": 0.3289369069612943,
1610
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.146243168757512e-09,
1611
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.3289369069612943,
1612
+ "step": 1475
1613
+ },
1614
+ {
1615
+ "clip_ratio/high_max": 0.0,
1616
+ "clip_ratio/high_mean": 0.0,
1617
+ "clip_ratio/low_mean": 0.0,
1618
+ "clip_ratio/low_min": 0.0,
1619
+ "clip_ratio/region_mean": 0.0,
1620
+ "completion_length": 64.04,
1621
+ "completions/clipped_ratio": 0.0,
1622
+ "completions/max_length": 65.0,
1623
+ "completions/max_terminated_length": 65.0,
1624
+ "completions/mean_length": 64.29166666666667,
1625
+ "completions/mean_terminated_length": 64.29166666666667,
1626
+ "completions/min_length": 63.75,
1627
+ "completions/min_terminated_length": 63.75,
1628
+ "epoch": 0.002664251079021687,
1629
+ "frac_reward_zero_std": 0.4166666666666667,
1630
+ "grad_norm": 0.19509506225585938,
1631
+ "kl": 0.21482937432825566,
1632
+ "learning_rate": 6.656157722963522e-08,
1633
+ "loss": 0.0038438504934310913,
1634
+ "num_tokens": 1415430.0,
1635
+ "reward": 4.967053731282552e-09,
1636
+ "reward_std": 0.6236095527807871,
1637
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
1638
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.6236095627148946,
1639
+ "step": 1500
1640
+ },
1641
+ {
1642
+ "clip_ratio/high_max": 0.0,
1643
+ "clip_ratio/high_mean": 0.0,
1644
+ "clip_ratio/low_mean": 0.0,
1645
+ "clip_ratio/low_min": 0.0,
1646
+ "clip_ratio/region_mean": 0.0,
1647
+ "completion_length": 72.58,
1648
+ "completions/clipped_ratio": 0.0,
1649
+ "completions/max_length": 73.76923076923077,
1650
+ "completions/max_terminated_length": 73.76923076923077,
1651
+ "completions/mean_length": 73.75,
1652
+ "completions/mean_terminated_length": 73.75,
1653
+ "completions/min_length": 73.6923076923077,
1654
+ "completions/min_terminated_length": 73.6923076923077,
1655
+ "epoch": 0.0027086552636720485,
1656
+ "frac_reward_zero_std": 0.9230769230769231,
1657
+ "grad_norm": 1.5856154277571477e-05,
1658
+ "kl": 0.16710449412465095,
1659
+ "learning_rate": 6.767167691658711e-08,
1660
+ "loss": 0.0014291378855705262,
1661
+ "num_tokens": 1442644.0,
1662
+ "reward": 0.0,
1663
+ "reward_std": 0.08223422674032357,
1664
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1665
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.08223422674032357,
1666
+ "step": 1525
1667
+ },
1668
+ {
1669
+ "clip_ratio/high_max": 0.0,
1670
+ "clip_ratio/high_mean": 0.0,
1671
+ "clip_ratio/low_mean": 0.0,
1672
+ "clip_ratio/low_min": 0.0,
1673
+ "clip_ratio/region_mean": 0.0,
1674
+ "completion_length": 68.38,
1675
+ "completions/clipped_ratio": 0.0,
1676
+ "completions/max_length": 67.08333333333333,
1677
+ "completions/max_terminated_length": 67.08333333333333,
1678
+ "completions/mean_length": 66.9375,
1679
+ "completions/mean_terminated_length": 66.9375,
1680
+ "completions/min_length": 66.83333333333333,
1681
+ "completions/min_terminated_length": 66.83333333333333,
1682
+ "epoch": 0.00275305944832241,
1683
+ "frac_reward_zero_std": 0.5,
1684
+ "grad_norm": 0.12691470980644226,
1685
+ "kl": 0.1746705377101898,
1686
+ "learning_rate": 6.8781776603539e-08,
1687
+ "loss": 0.0015592724084854126,
1688
+ "num_tokens": 1465798.0,
1689
+ "reward": 1.241763432820638e-09,
1690
+ "reward_std": 0.5345224738121033,
1691
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-09,
1692
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.5345224738121033,
1693
+ "step": 1550
1694
+ },
1695
+ {
1696
+ "clip_ratio/high_max": 0.0,
1697
+ "clip_ratio/high_mean": 0.0,
1698
+ "clip_ratio/low_mean": 0.0,
1699
+ "clip_ratio/low_min": 0.0,
1700
+ "clip_ratio/region_mean": 0.0,
1701
+ "completion_length": 70.47,
1702
+ "completions/clipped_ratio": 0.0,
1703
+ "completions/max_length": 71.38461538461539,
1704
+ "completions/max_terminated_length": 71.38461538461539,
1705
+ "completions/mean_length": 70.99038461538461,
1706
+ "completions/mean_terminated_length": 70.99038461538461,
1707
+ "completions/min_length": 70.53846153846153,
1708
+ "completions/min_terminated_length": 70.53846153846153,
1709
+ "epoch": 0.0027974636329727715,
1710
+ "frac_reward_zero_std": 0.46153846153846156,
1711
+ "grad_norm": 0.6687065958976746,
1712
+ "kl": 0.17096829459071158,
1713
+ "learning_rate": 6.989187629049088e-08,
1714
+ "loss": 0.01692964315414429,
1715
+ "num_tokens": 1491837.0,
1716
+ "reward": 3.4387295062725362e-09,
1717
+ "reward_std": 0.575639587182265,
1718
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.4387295062725362e-09,
1719
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.575639587182265,
1720
+ "step": 1575
1721
+ },
1722
+ {
1723
+ "clip_ratio/high_max": 0.0,
1724
+ "clip_ratio/high_mean": 0.0,
1725
+ "clip_ratio/low_mean": 0.0,
1726
+ "clip_ratio/low_min": 0.0,
1727
+ "clip_ratio/region_mean": 0.0,
1728
+ "completion_length": 61.91,
1729
+ "completions/clipped_ratio": 0.0,
1730
+ "completions/max_length": 61.416666666666664,
1731
+ "completions/max_terminated_length": 61.416666666666664,
1732
+ "completions/mean_length": 60.989583333333336,
1733
+ "completions/mean_terminated_length": 60.989583333333336,
1734
+ "completions/min_length": 60.666666666666664,
1735
+ "completions/min_terminated_length": 60.666666666666664,
1736
+ "epoch": 0.002841867817623133,
1737
+ "frac_reward_zero_std": 0.6666666666666666,
1738
+ "grad_norm": 2.47890693572117e-06,
1739
+ "kl": 0.189963039457798,
1740
+ "learning_rate": 7.100197597744277e-08,
1741
+ "loss": -0.01327080488204956,
1742
+ "num_tokens": 1513364.0,
1743
+ "reward": -4.346172014872233e-09,
1744
+ "reward_std": 0.35634830594062805,
1745
+ "rewards/TranscriptCorrectionGrpoReward/mean": -4.346172014872233e-09,
1746
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
1747
+ "step": 1600
1748
+ },
1749
+ {
1750
+ "clip_ratio/high_max": 0.0,
1751
+ "clip_ratio/high_mean": 0.0,
1752
+ "clip_ratio/low_mean": 0.0,
1753
+ "clip_ratio/low_min": 0.0,
1754
+ "clip_ratio/region_mean": 0.0,
1755
+ "completion_length": 71.91,
1756
+ "completions/clipped_ratio": 0.0,
1757
+ "completions/max_length": 72.0,
1758
+ "completions/max_terminated_length": 72.0,
1759
+ "completions/mean_length": 71.52884615384616,
1760
+ "completions/mean_terminated_length": 71.52884615384616,
1761
+ "completions/min_length": 70.84615384615384,
1762
+ "completions/min_terminated_length": 70.84615384615384,
1763
+ "epoch": 0.0028862720022734944,
1764
+ "frac_reward_zero_std": 0.5384615384615384,
1765
+ "grad_norm": 0.00014156986435409635,
1766
+ "kl": 0.16550406880676746,
1767
+ "learning_rate": 7.211207566439466e-08,
1768
+ "loss": 0.0033655133843421936,
1769
+ "num_tokens": 1539331.0,
1770
+ "reward": 3.4387295062725362e-09,
1771
+ "reward_std": 0.4934053604419415,
1772
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.4387295062725362e-09,
1773
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.49340536961188686,
1774
+ "step": 1625
1775
+ },
1776
+ {
1777
+ "clip_ratio/high_max": 0.0,
1778
+ "clip_ratio/high_mean": 0.0,
1779
+ "clip_ratio/low_mean": 0.0,
1780
+ "clip_ratio/low_min": 0.0,
1781
+ "clip_ratio/region_mean": 0.0,
1782
+ "completion_length": 60.05,
1783
+ "completions/clipped_ratio": 0.0,
1784
+ "completions/max_length": 60.333333333333336,
1785
+ "completions/max_terminated_length": 60.333333333333336,
1786
+ "completions/mean_length": 59.96875,
1787
+ "completions/mean_terminated_length": 59.96875,
1788
+ "completions/min_length": 59.5,
1789
+ "completions/min_terminated_length": 59.5,
1790
+ "epoch": 0.002930676186923856,
1791
+ "frac_reward_zero_std": 0.5,
1792
+ "grad_norm": 0.38339468836784363,
1793
+ "kl": 0.19749208971858023,
1794
+ "learning_rate": 7.322217535134655e-08,
1795
+ "loss": 0.004807085394859314,
1796
+ "num_tokens": 1559944.0,
1797
+ "reward": 6.20881716410319e-10,
1798
+ "reward_std": 0.5345224738121033,
1799
+ "rewards/TranscriptCorrectionGrpoReward/mean": 6.20881716410319e-10,
1800
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.5345224738121033,
1801
+ "step": 1650
1802
+ },
1803
+ {
1804
+ "clip_ratio/high_max": 0.0,
1805
+ "clip_ratio/high_mean": 0.0,
1806
+ "clip_ratio/low_mean": 0.0,
1807
+ "clip_ratio/low_min": 0.0,
1808
+ "clip_ratio/region_mean": 0.0,
1809
+ "completion_length": 66.99,
1810
+ "completions/clipped_ratio": 0.0,
1811
+ "completions/max_length": 69.0,
1812
+ "completions/max_terminated_length": 69.0,
1813
+ "completions/mean_length": 68.60576923076923,
1814
+ "completions/mean_terminated_length": 68.60576923076923,
1815
+ "completions/min_length": 68.0,
1816
+ "completions/min_terminated_length": 68.0,
1817
+ "epoch": 0.0029750803715742173,
1818
+ "frac_reward_zero_std": 0.5384615384615384,
1819
+ "grad_norm": 0.1574590653181076,
1820
+ "kl": 0.19060430452227592,
1821
+ "learning_rate": 7.433227503829843e-08,
1822
+ "loss": -0.003221469521522522,
1823
+ "num_tokens": 1585159.0,
1824
+ "reward": -2.292486337515024e-09,
1825
+ "reward_std": 0.4934053604419415,
1826
+ "rewards/TranscriptCorrectionGrpoReward/mean": -2.292486337515024e-09,
1827
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
1828
+ "step": 1675
1829
+ },
1830
+ {
1831
+ "clip_ratio/high_max": 0.0,
1832
+ "clip_ratio/high_mean": 0.0,
1833
+ "clip_ratio/low_mean": 0.0,
1834
+ "clip_ratio/low_min": 0.0,
1835
+ "clip_ratio/region_mean": 0.0,
1836
+ "completion_length": 67.44,
1837
+ "completions/clipped_ratio": 0.0,
1838
+ "completions/max_length": 65.91666666666667,
1839
+ "completions/max_terminated_length": 65.91666666666667,
1840
+ "completions/mean_length": 65.70833333333333,
1841
+ "completions/mean_terminated_length": 65.70833333333333,
1842
+ "completions/min_length": 65.5,
1843
+ "completions/min_terminated_length": 65.5,
1844
+ "epoch": 0.0030194845562245788,
1845
+ "frac_reward_zero_std": 0.5833333333333334,
1846
+ "grad_norm": 7.122421266103629e-06,
1847
+ "kl": 0.17307962857186796,
1848
+ "learning_rate": 7.544237472525032e-08,
1849
+ "loss": 0.008914402723312377,
1850
+ "num_tokens": 1607587.0,
1851
+ "reward": 0.0,
1852
+ "reward_std": 0.4454353948434194,
1853
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1854
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
1855
+ "step": 1700
1856
+ },
1857
+ {
1858
+ "clip_ratio/high_max": 0.0,
1859
+ "clip_ratio/high_mean": 0.0,
1860
+ "clip_ratio/low_mean": 0.0,
1861
+ "clip_ratio/low_min": 0.0,
1862
+ "clip_ratio/region_mean": 0.0,
1863
+ "completion_length": 63.88,
1864
+ "completions/clipped_ratio": 0.0,
1865
+ "completions/max_length": 64.15384615384616,
1866
+ "completions/max_terminated_length": 64.15384615384616,
1867
+ "completions/mean_length": 64.15384615384616,
1868
+ "completions/mean_terminated_length": 64.15384615384616,
1869
+ "completions/min_length": 64.15384615384616,
1870
+ "completions/min_terminated_length": 64.15384615384616,
1871
+ "epoch": 0.0030638887408749402,
1872
+ "frac_reward_zero_std": 0.8461538461538461,
1873
+ "grad_norm": 6.491336534963921e-05,
1874
+ "kl": 0.1862632592022419,
1875
+ "learning_rate": 7.655247441220221e-08,
1876
+ "loss": 0.0018509525060653688,
1877
+ "num_tokens": 1631387.0,
1878
+ "reward": -4.584972675030048e-09,
1879
+ "reward_std": 0.16446845348064715,
1880
+ "rewards/TranscriptCorrectionGrpoReward/mean": -4.584972675030048e-09,
1881
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.16446845348064715,
1882
+ "step": 1725
1883
+ },
1884
+ {
1885
+ "clip_ratio/high_max": 0.0,
1886
+ "clip_ratio/high_mean": 0.0,
1887
+ "clip_ratio/low_mean": 0.0,
1888
+ "clip_ratio/low_min": 0.0,
1889
+ "clip_ratio/region_mean": 0.0,
1890
+ "completion_length": 54.91,
1891
+ "completions/clipped_ratio": 0.0,
1892
+ "completions/max_length": 54.416666666666664,
1893
+ "completions/max_terminated_length": 54.416666666666664,
1894
+ "completions/mean_length": 54.239583333333336,
1895
+ "completions/mean_terminated_length": 54.239583333333336,
1896
+ "completions/min_length": 54.083333333333336,
1897
+ "completions/min_terminated_length": 54.083333333333336,
1898
+ "epoch": 0.0031082929255253017,
1899
+ "frac_reward_zero_std": 0.5833333333333334,
1900
+ "grad_norm": 0.0003962465561926365,
1901
+ "kl": 0.2070286213606596,
1902
+ "learning_rate": 7.76625740991541e-08,
1903
+ "loss": 0.0013501006364822388,
1904
+ "num_tokens": 1650058.0,
1905
+ "reward": 1.241763432820638e-09,
1906
+ "reward_std": 0.4454353948434194,
1907
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-09,
1908
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
1909
+ "step": 1750
1910
+ },
1911
+ {
1912
+ "clip_ratio/high_max": 0.0,
1913
+ "clip_ratio/high_mean": 0.0,
1914
+ "clip_ratio/low_mean": 0.0,
1915
+ "clip_ratio/low_min": 0.0,
1916
+ "clip_ratio/region_mean": 0.0,
1917
+ "completion_length": 63.11,
1918
+ "completions/clipped_ratio": 0.0,
1919
+ "completions/max_length": 62.53846153846154,
1920
+ "completions/max_terminated_length": 62.53846153846154,
1921
+ "completions/mean_length": 62.375,
1922
+ "completions/mean_terminated_length": 62.375,
1923
+ "completions/min_length": 62.23076923076923,
1924
+ "completions/min_terminated_length": 62.23076923076923,
1925
+ "epoch": 0.0031526971101756627,
1926
+ "frac_reward_zero_std": 0.5384615384615384,
1927
+ "grad_norm": 0.14643262326717377,
1928
+ "kl": 0.21996574357151985,
1929
+ "learning_rate": 7.877267378610598e-08,
1930
+ "loss": -0.010204806327819824,
1931
+ "num_tokens": 1673193.0,
1932
+ "reward": 5.73121584378756e-10,
1933
+ "reward_std": 0.49340536961188686,
1934
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.73121584378756e-10,
1935
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
1936
+ "step": 1775
1937
+ },
1938
+ {
1939
+ "clip_ratio/high_max": 0.0,
1940
+ "clip_ratio/high_mean": 0.0,
1941
+ "clip_ratio/low_mean": 0.0,
1942
+ "clip_ratio/low_min": 0.0,
1943
+ "clip_ratio/region_mean": 0.0,
1944
+ "completion_length": 74.83,
1945
+ "completions/clipped_ratio": 0.0,
1946
+ "completions/max_length": 76.58333333333333,
1947
+ "completions/max_terminated_length": 76.58333333333333,
1948
+ "completions/mean_length": 76.11458333333333,
1949
+ "completions/mean_terminated_length": 76.11458333333333,
1950
+ "completions/min_length": 75.83333333333333,
1951
+ "completions/min_terminated_length": 75.83333333333333,
1952
+ "epoch": 0.003197101294826024,
1953
+ "frac_reward_zero_std": 0.3333333333333333,
1954
+ "grad_norm": 2.9733921110164374e-05,
1955
+ "kl": 0.15571292936801912,
1956
+ "learning_rate": 7.988277347305787e-08,
1957
+ "loss": 0.01144843339920044,
1958
+ "num_tokens": 1699020.0,
1959
+ "reward": 0.0,
1960
+ "reward_std": 0.7126966218153635,
1961
+ "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
1962
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.7126966317494711,
1963
+ "step": 1800
1964
+ },
1965
+ {
1966
+ "clip_ratio/high_max": 0.0,
1967
+ "clip_ratio/high_mean": 0.0,
1968
+ "clip_ratio/low_mean": 0.00019230769947171212,
1969
+ "clip_ratio/low_min": 0.00019230769947171212,
1970
+ "clip_ratio/region_mean": 0.00019230769947171212,
1971
+ "completion_length": 71.58,
1972
+ "completions/clipped_ratio": 0.0,
1973
+ "completions/max_length": 71.15384615384616,
1974
+ "completions/max_terminated_length": 71.15384615384616,
1975
+ "completions/mean_length": 70.60576923076923,
1976
+ "completions/mean_terminated_length": 70.60576923076923,
1977
+ "completions/min_length": 70.15384615384616,
1978
+ "completions/min_terminated_length": 70.15384615384616,
1979
+ "epoch": 0.0032415054794763857,
1980
+ "frac_reward_zero_std": 0.5384615384615384,
1981
+ "grad_norm": 0.5476241707801819,
1982
+ "kl": 0.18457129560410976,
1983
+ "learning_rate": 8.099287316000976e-08,
1984
+ "loss": 0.0033506143093109133,
1985
+ "num_tokens": 1725115.0,
1986
+ "reward": -1.146243168757512e-09,
1987
+ "reward_std": 0.49340537878183216,
1988
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.146243168757512e-09,
1989
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
1990
+ "step": 1825
1991
+ },
1992
+ {
1993
+ "clip_ratio/high_max": 0.0,
1994
+ "clip_ratio/high_mean": 0.0,
1995
+ "clip_ratio/low_mean": 0.00010989011265337467,
1996
+ "clip_ratio/low_min": 0.00010989011265337467,
1997
+ "clip_ratio/region_mean": 0.00010989011265337467,
1998
+ "completion_length": 72.87,
1999
+ "completions/clipped_ratio": 0.0,
2000
+ "completions/max_length": 74.83333333333333,
2001
+ "completions/max_terminated_length": 74.83333333333333,
2002
+ "completions/mean_length": 73.97916666666667,
2003
+ "completions/mean_terminated_length": 73.97916666666667,
2004
+ "completions/min_length": 72.91666666666667,
2005
+ "completions/min_terminated_length": 72.91666666666667,
2006
+ "epoch": 0.003285909664126747,
2007
+ "frac_reward_zero_std": 0.3333333333333333,
2008
+ "grad_norm": 1.1398409469620674e-06,
2009
+ "kl": 0.21169950984418393,
2010
+ "learning_rate": 8.210297284696166e-08,
2011
+ "loss": -0.0035066825151443483,
2012
+ "num_tokens": 1750601.0,
2013
+ "reward": 3.725290298461914e-09,
2014
+ "reward_std": 0.7126966317494711,
2015
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.725290298461914e-09,
2016
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.7126966317494711,
2017
+ "step": 1850
2018
+ },
2019
+ {
2020
+ "clip_ratio/high_max": 0.0,
2021
+ "clip_ratio/high_mean": 0.0,
2022
+ "clip_ratio/low_mean": 0.0,
2023
+ "clip_ratio/low_min": 0.0,
2024
+ "clip_ratio/region_mean": 0.0,
2025
+ "completion_length": 62.34,
2026
+ "completions/clipped_ratio": 0.0,
2027
+ "completions/max_length": 61.15384615384615,
2028
+ "completions/max_terminated_length": 61.15384615384615,
2029
+ "completions/mean_length": 60.98076923076923,
2030
+ "completions/mean_terminated_length": 60.98076923076923,
2031
+ "completions/min_length": 60.76923076923077,
2032
+ "completions/min_terminated_length": 60.76923076923077,
2033
+ "epoch": 0.0033303138487771086,
2034
+ "frac_reward_zero_std": 0.5384615384615384,
2035
+ "grad_norm": 2.990056418639142e-05,
2036
+ "kl": 0.18480682276189328,
2037
+ "learning_rate": 8.321307253391355e-08,
2038
+ "loss": 0.0017520076036453248,
2039
+ "num_tokens": 1773559.0,
2040
+ "reward": 3.4387295062725362e-09,
2041
+ "reward_std": 0.49340536961188686,
2042
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.4387295062725362e-09,
2043
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
2044
+ "step": 1875
2045
+ },
2046
+ {
2047
+ "clip_ratio/high_max": 0.0,
2048
+ "clip_ratio/high_mean": 0.0,
2049
+ "clip_ratio/low_mean": 0.0,
2050
+ "clip_ratio/low_min": 0.0,
2051
+ "clip_ratio/region_mean": 0.0,
2052
+ "completion_length": 61.7,
2053
+ "completions/clipped_ratio": 0.0,
2054
+ "completions/max_length": 63.25,
2055
+ "completions/max_terminated_length": 63.25,
2056
+ "completions/mean_length": 63.145833333333336,
2057
+ "completions/mean_terminated_length": 63.145833333333336,
2058
+ "completions/min_length": 63.0,
2059
+ "completions/min_terminated_length": 63.0,
2060
+ "epoch": 0.00337471803342747,
2061
+ "frac_reward_zero_std": 0.5833333333333334,
2062
+ "grad_norm": 3.1972882425179705e-05,
2063
+ "kl": 0.17734202355146408,
2064
+ "learning_rate": 8.432317222086544e-08,
2065
+ "loss": 0.001646375060081482,
2066
+ "num_tokens": 1794909.0,
2067
+ "reward": 3.725290298461914e-09,
2068
+ "reward_std": 0.4454353948434194,
2069
+ "rewards/TranscriptCorrectionGrpoReward/mean": 3.725290298461914e-09,
2070
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4454353948434194,
2071
+ "step": 1900
2072
+ },
2073
+ {
2074
+ "clip_ratio/high_max": 0.0,
2075
+ "clip_ratio/high_mean": 0.0,
2076
+ "clip_ratio/low_mean": 0.0,
2077
+ "clip_ratio/low_min": 0.0,
2078
+ "clip_ratio/region_mean": 0.0,
2079
+ "completion_length": 66.37,
2080
+ "completions/clipped_ratio": 0.0,
2081
+ "completions/max_length": 66.84615384615384,
2082
+ "completions/max_terminated_length": 66.84615384615384,
2083
+ "completions/mean_length": 66.8173076923077,
2084
+ "completions/mean_terminated_length": 66.8173076923077,
2085
+ "completions/min_length": 66.6923076923077,
2086
+ "completions/min_terminated_length": 66.6923076923077,
2087
+ "epoch": 0.0034191222180778315,
2088
+ "frac_reward_zero_std": 0.5384615384615384,
2089
+ "grad_norm": 0.18009275197982788,
2090
+ "kl": 0.20137830704450607,
2091
+ "learning_rate": 8.543327190781733e-08,
2092
+ "loss": -0.009820441007614136,
2093
+ "num_tokens": 1819690.0,
2094
+ "reward": 4.584972675030048e-09,
2095
+ "reward_std": 0.4934053604419415,
2096
+ "rewards/TranscriptCorrectionGrpoReward/mean": 4.584972675030048e-09,
2097
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
2098
+ "step": 1925
2099
+ },
2100
+ {
2101
+ "clip_ratio/high_max": 0.0,
2102
+ "clip_ratio/high_mean": 0.0,
2103
+ "clip_ratio/low_mean": 0.0,
2104
+ "clip_ratio/low_min": 0.0,
2105
+ "clip_ratio/region_mean": 0.0,
2106
+ "completion_length": 68.86,
2107
+ "completions/clipped_ratio": 0.0,
2108
+ "completions/max_length": 68.83333333333333,
2109
+ "completions/max_terminated_length": 68.83333333333333,
2110
+ "completions/mean_length": 68.47916666666667,
2111
+ "completions/mean_terminated_length": 68.47916666666667,
2112
+ "completions/min_length": 68.25,
2113
+ "completions/min_terminated_length": 68.25,
2114
+ "epoch": 0.003463526402728193,
2115
+ "frac_reward_zero_std": 0.3333333333333333,
2116
+ "grad_norm": 0.3680269718170166,
2117
+ "kl": 0.16890673726797104,
2118
+ "learning_rate": 8.654337159476922e-08,
2119
+ "loss": 0.012883994579315185,
2120
+ "num_tokens": 1842984.0,
2121
+ "reward": 5.587935447692871e-09,
2122
+ "reward_std": 0.7126966118812561,
2123
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.587935447692871e-09,
2124
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.7126966317494711,
2125
+ "step": 1950
2126
+ },
2127
+ {
2128
+ "clip_ratio/high_max": 0.0,
2129
+ "clip_ratio/high_mean": 0.0,
2130
+ "clip_ratio/low_mean": 0.0,
2131
+ "clip_ratio/low_min": 0.0,
2132
+ "clip_ratio/region_mean": 0.0,
2133
+ "completion_length": 60.56,
2134
+ "completions/clipped_ratio": 0.0,
2135
+ "completions/max_length": 60.92307692307692,
2136
+ "completions/max_terminated_length": 60.92307692307692,
2137
+ "completions/mean_length": 60.53846153846154,
2138
+ "completions/mean_terminated_length": 60.53846153846154,
2139
+ "completions/min_length": 60.46153846153846,
2140
+ "completions/min_terminated_length": 60.46153846153846,
2141
+ "epoch": 0.0035079305873785545,
2142
+ "frac_reward_zero_std": 0.7692307692307693,
2143
+ "grad_norm": 0.0003735637292265892,
2144
+ "kl": 0.19156488351523876,
2145
+ "learning_rate": 8.765347128172109e-08,
2146
+ "loss": 0.0019974344968795776,
2147
+ "num_tokens": 1865592.0,
2148
+ "reward": 5.73121584378756e-10,
2149
+ "reward_std": 0.24670268939091608,
2150
+ "rewards/TranscriptCorrectionGrpoReward/mean": 5.73121584378756e-10,
2151
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.24670268939091608,
2152
+ "step": 1975
2153
+ },
2154
+ {
2155
+ "clip_ratio/high_max": 0.0,
2156
+ "clip_ratio/high_mean": 0.0,
2157
+ "clip_ratio/low_mean": 0.0,
2158
+ "clip_ratio/low_min": 0.0,
2159
+ "clip_ratio/region_mean": 0.0,
2160
+ "completion_length": 62.22,
2161
+ "completions/clipped_ratio": 0.0,
2162
+ "completions/max_length": 62.416666666666664,
2163
+ "completions/max_terminated_length": 62.416666666666664,
2164
+ "completions/mean_length": 62.3125,
2165
+ "completions/mean_terminated_length": 62.3125,
2166
+ "completions/min_length": 62.166666666666664,
2167
+ "completions/min_terminated_length": 62.166666666666664,
2168
+ "epoch": 0.003552334772028916,
2169
+ "frac_reward_zero_std": 0.6666666666666666,
2170
+ "grad_norm": 0.3915543258190155,
2171
+ "kl": 0.1927006433904171,
2172
+ "learning_rate": 8.876357096867298e-08,
2173
+ "loss": 0.001761952042579651,
2174
+ "num_tokens": 1887118.0,
2175
+ "reward": 1.241763432820638e-09,
2176
+ "reward_std": 0.35634830594062805,
2177
+ "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-09,
2178
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.35634831587473553,
2179
+ "step": 2000
2180
+ },
2181
+ {
2182
+ "clip_ratio/high_max": 0.0,
2183
+ "clip_ratio/high_mean": 0.0,
2184
+ "clip_ratio/low_mean": 0.0,
2185
+ "clip_ratio/low_min": 0.0,
2186
+ "clip_ratio/region_mean": 0.0,
2187
+ "completion_length": 73.29,
2188
+ "completions/clipped_ratio": 0.0,
2189
+ "completions/max_length": 74.3076923076923,
2190
+ "completions/max_terminated_length": 74.3076923076923,
2191
+ "completions/mean_length": 74.0,
2192
+ "completions/mean_terminated_length": 74.0,
2193
+ "completions/min_length": 73.6923076923077,
2194
+ "completions/min_terminated_length": 73.6923076923077,
2195
+ "epoch": 0.0035967389566792774,
2196
+ "frac_reward_zero_std": 0.5384615384615384,
2197
+ "grad_norm": 0.11132384091615677,
2198
+ "kl": 0.17609826825559138,
2199
+ "learning_rate": 8.987367065562487e-08,
2200
+ "loss": 0.0013741311430931092,
2201
+ "num_tokens": 1914230.0,
2202
+ "reward": -1.146243168757512e-09,
2203
+ "reward_std": 0.49340535127199614,
2204
+ "rewards/TranscriptCorrectionGrpoReward/mean": -1.146243168757512e-09,
2205
+ "rewards/TranscriptCorrectionGrpoReward/std": 0.4934053604419415,
2206
+ "step": 2025
2207
+ },
2208
+ {
2209
+ "eval_cer_subset": 0.014265474620157311,
2210
+ "eval_cer_subset_edit_distance": 876,
2211
+ "eval_cer_subset_groups": 250,
2212
+ "eval_cer_subset_items": 250,
2213
+ "eval_cer_subset_ref_chars": 61407,
2214
+ "eval_loss": 0.033980044345898006,
2215
+ "step": 2026
2216
+ }
2217
+ ],
2218
+ "logging_steps": 25,
2219
+ "max_steps": 2252040,
2220
+ "num_input_tokens_seen": 1914230,
2221
+ "num_train_epochs": 4,
2222
+ "save_steps": 1013,
2223
+ "stateful_callbacks": {
2224
+ "TrainerControl": {
2225
+ "args": {
2226
+ "should_epoch_stop": false,
2227
+ "should_evaluate": false,
2228
+ "should_log": false,
2229
+ "should_save": true,
2230
+ "should_training_stop": false
2231
+ },
2232
+ "attributes": {}
2233
+ }
2234
+ },
2235
+ "total_flos": 0.0,
2236
+ "train_batch_size": 1,
2237
+ "trial_name": null,
2238
+ "trial_params": null
2239
+ }
meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:28fd8dbecc91ea4d1302120052163b26bd82d9f2b04a5a72e1957eb427813a61
3
+ size 6737
meta-llama__meta-llama-3.1-8b/grpo/run_manifest.json CHANGED
@@ -9,17 +9,23 @@
9
  "best_metric_value": 0.01423290504339896,
10
  "best_step": 1013,
11
  "best_checkpoint_step": 1013,
12
- "latest_checkpoint_step": 1013,
13
  "checkpoints": [
14
  {
15
  "step": 1013,
16
  "epoch": 0.001799257562032646,
17
  "repo_path": "meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-1013",
18
  "timestamp": "2026-04-09T17:58:40.645060+00:00"
 
 
 
 
 
 
19
  }
20
  ],
21
  "epoch_artifacts": [],
22
  "resume_history": [],
23
  "created_at": "2026-04-09T17:24:46.169440+00:00",
24
- "updated_at": "2026-04-09T17:58:40.645077+00:00"
25
  }
 
9
  "best_metric_value": 0.01423290504339896,
10
  "best_step": 1013,
11
  "best_checkpoint_step": 1013,
12
+ "latest_checkpoint_step": 2026,
13
  "checkpoints": [
14
  {
15
  "step": 1013,
16
  "epoch": 0.001799257562032646,
17
  "repo_path": "meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-1013",
18
  "timestamp": "2026-04-09T17:58:40.645060+00:00"
19
+ },
20
+ {
21
+ "step": 2026,
22
+ "epoch": 0.003598515124065292,
23
+ "repo_path": "meta-llama__meta-llama-3.1-8b/grpo/checkpoints/checkpoint-2026",
24
+ "timestamp": "2026-04-09T18:30:35.889946+00:00"
25
  }
26
  ],
27
  "epoch_artifacts": [],
28
  "resume_history": [],
29
  "created_at": "2026-04-09T17:24:46.169440+00:00",
30
+ "updated_at": "2026-04-09T18:30:35.889960+00:00"
31
  }