Hugging Face's logo

Guen
/
tempgraphrag-grpo

Safetensors
Model card Files
xet
 Community
Guen commited on
Commit
3172870
·
verified ·
1 Parent(s): b37657c

sync 2026-04-23T23:10:06+00:00

Browse files
Files changed (24) hide show
  1. .gitattributes +2 -0
  2. grpo/README.md +67 -0
  3. grpo/checkpoint-1000/adapter_config.json +45 -0
  4. grpo/checkpoint-1000/adapter_model.safetensors +3 -0
  5. grpo/checkpoint-1000/chat_template.jinja +89 -0
  6. grpo/checkpoint-1000/generation_config.json +12 -0
  7. grpo/checkpoint-1000/optimizer.pt +3 -0
  8. grpo/checkpoint-1000/rng_state.pth +3 -0
  9. grpo/checkpoint-1000/scheduler.pt +3 -0
  10. grpo/checkpoint-1000/tokenizer.json +3 -0
  11. grpo/checkpoint-1000/tokenizer_config.json +34 -0
  12. grpo/checkpoint-1000/trainer_state.json +1584 -0
  13. grpo/checkpoint-1000/training_args.bin +3 -0
  14. grpo/checkpoint-500/adapter_config.json +45 -0
  15. grpo/checkpoint-500/adapter_model.safetensors +3 -0
  16. grpo/checkpoint-500/chat_template.jinja +89 -0
  17. grpo/checkpoint-500/generation_config.json +12 -0
  18. grpo/checkpoint-500/optimizer.pt +3 -0
  19. grpo/checkpoint-500/rng_state.pth +3 -0
  20. grpo/checkpoint-500/scheduler.pt +3 -0
  21. grpo/checkpoint-500/tokenizer.json +3 -0
  22. grpo/checkpoint-500/tokenizer_config.json +34 -0
  23. grpo/checkpoint-500/trainer_state.json +809 -0
  24. grpo/checkpoint-500/training_args.bin +3 -0
.gitattributes CHANGED
@@ -39,3 +39,5 @@ sft/checkpoint-490/tokenizer.json filter=lfs diff=lfs merge=lfs -text
39
  sft/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
40
  grpo-smoke/checkpoint-50/tokenizer.json filter=lfs diff=lfs merge=lfs -text
41
  grpo-smoke/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
 
39
  sft/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
40
  grpo-smoke/checkpoint-50/tokenizer.json filter=lfs diff=lfs merge=lfs -text
41
  grpo-smoke/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
42
+ grpo/checkpoint-1000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
43
+ grpo/checkpoint-500/tokenizer.json filter=lfs diff=lfs merge=lfs -text
grpo/README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3-8B
3
+ library_name: transformers
4
+ model_name: grpo
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - grpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for grpo
13
+
14
+ This model is a fine-tuned version of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).
15
+ It has been trained using [TRL](https://github.com/huggingface/trl).
16
+
17
+ ## Quick start
18
+
19
+ ```python
20
+ from transformers import pipeline
21
+
22
+ question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
23
+ generator = pipeline("text-generation", model="None", device="cuda")
24
+ output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
25
+ print(output["generated_text"])
26
+ ```
27
+
28
+ ## Training procedure
29
+
30
+
31
+
32
+
33
+
34
+ This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
35
+
36
+ ### Framework versions
37
+
38
+ - TRL: 1.2.0
39
+ - Transformers: 5.6.2
40
+ - Pytorch: 2.11.0
41
+ - Datasets: 4.8.4
42
+ - Tokenizers: 0.22.2
43
+
44
+ ## Citations
45
+
46
+ Cite GRPO as:
47
+
48
+ ```bibtex
49
+ @article{shao2024deepseekmath,
50
+ title = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
51
+ author = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
52
+ year = 2024,
53
+ eprint = {arXiv:2402.03300},
54
+ }
55
+ ```
56
+
57
+ Cite TRL as:
58
+
59
+ ```bibtex
60
+ @software{vonwerra2020trl,
61
+ title = {{TRL: Transformers Reinforcement Learning}},
62
+ author = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
63
+ license = {Apache-2.0},
64
+ url = {https://github.com/huggingface/trl},
65
+ year = {2020}
66
+ }
67
+ ```
grpo/checkpoint-1000/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-8B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "o_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
grpo/checkpoint-1000/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39bd69d1e3bcd055a95670f34d387d8dee0b706e4decb1754fdc0edb9dd95411
3
+ size 30709768
grpo/checkpoint-1000/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if message.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
grpo/checkpoint-1000/generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "temperature": 0.6,
9
+ "top_k": 20,
10
+ "top_p": 0.95,
11
+ "transformers_version": "5.6.2"
12
+ }
grpo/checkpoint-1000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d8e256c29d4e5f57372f980f6eb0d8c6877923876ab9d1c7d339e4bc76f7845
3
+ size 61588683
grpo/checkpoint-1000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:855955a10b80ad60f7262ddc90448e6842ed9e741c027a08c242c8feb27c816d
3
+ size 14645
grpo/checkpoint-1000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb69edc30acd035a479a8196251698f3cdea75f5be9dc4dad020892566d2a1c3
3
+ size 1465
grpo/checkpoint-1000/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
3
+ size 11422650
grpo/checkpoint-1000/tokenizer_config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "max_length": 512,
26
+ "model_max_length": 131072,
27
+ "pad_token": "<|endoftext|>",
28
+ "split_special_tokens": false,
29
+ "stride": 0,
30
+ "tokenizer_class": "Qwen2Tokenizer",
31
+ "truncation_side": "right",
32
+ "truncation_strategy": "longest_first",
33
+ "unk_token": null
34
+ }
grpo/checkpoint-1000/trainer_state.json ADDED
@@ -0,0 +1,1584 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.16404199475065617,
6
+ "eval_steps": 500,
7
+ "global_step": 1000,
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
+ "completions/clipped_ratio": 0.0,
19
+ "completions/max_length": 6.05,
20
+ "completions/max_terminated_length": 6.05,
21
+ "completions/mean_length": 4.61875,
22
+ "completions/mean_terminated_length": 4.61875,
23
+ "completions/min_length": 3.5,
24
+ "completions/min_terminated_length": 3.5,
25
+ "entropy": 0.8203600860819279,
26
+ "epoch": 0.0032808398950131233,
27
+ "frac_reward_zero_std": 0.3,
28
+ "grad_norm": 5.8125,
29
+ "learning_rate": 4.990500000000001e-06,
30
+ "loss": 0.06519373655319213,
31
+ "num_tokens": 5635.0,
32
+ "reward": 1.1270193845033645,
33
+ "reward_std": 0.04764379555272171,
34
+ "rewards/reward_grounded/mean": 1.0,
35
+ "rewards/reward_grounded/std": 0.0,
36
+ "rewards/reward_length/mean": -0.050634765625,
37
+ "rewards/reward_length/std": 0.015013676788657903,
38
+ "rewards/reward_semantic/mean": 0.43183712139725683,
39
+ "rewards/reward_semantic/std": 0.15772892236709596,
40
+ "step": 20,
41
+ "step_time": 1.6850911805999202
42
+ },
43
+ {
44
+ "clip_ratio/high_max": 0.0,
45
+ "clip_ratio/high_mean": 0.0,
46
+ "clip_ratio/low_mean": 0.0,
47
+ "clip_ratio/low_min": 0.0,
48
+ "clip_ratio/region_mean": 0.0,
49
+ "completions/clipped_ratio": 0.0,
50
+ "completions/max_length": 7.15,
51
+ "completions/max_terminated_length": 7.15,
52
+ "completions/mean_length": 5.00625,
53
+ "completions/mean_terminated_length": 5.00625,
54
+ "completions/min_length": 3.65,
55
+ "completions/min_terminated_length": 3.65,
56
+ "entropy": 0.929626553109847,
57
+ "epoch": 0.006561679790026247,
58
+ "frac_reward_zero_std": 0.2,
59
+ "grad_norm": 0.0,
60
+ "learning_rate": 4.9805e-06,
61
+ "loss": 0.07899624109268188,
62
+ "num_tokens": 11660.0,
63
+ "reward": 1.0556338340044022,
64
+ "reward_std": 0.017533331926097163,
65
+ "rewards/reward_grounded/mean": 1.0,
66
+ "rewards/reward_grounded/std": 0.0,
67
+ "rewards/reward_length/mean": -0.0642578125,
68
+ "rewards/reward_length/std": 0.013992739858804271,
69
+ "rewards/reward_semantic/mean": 0.19615575280040504,
70
+ "rewards/reward_semantic/std": 0.05699479430913925,
71
+ "step": 40,
72
+ "step_time": 1.79874156090018
73
+ },
74
+ {
75
+ "clip_ratio/high_max": 0.0,
76
+ "clip_ratio/high_mean": 0.0,
77
+ "clip_ratio/low_mean": 0.0,
78
+ "clip_ratio/low_min": 0.0,
79
+ "clip_ratio/region_mean": 0.0,
80
+ "completions/clipped_ratio": 0.0,
81
+ "completions/max_length": 5.5,
82
+ "completions/max_terminated_length": 5.5,
83
+ "completions/mean_length": 4.325,
84
+ "completions/mean_terminated_length": 4.325,
85
+ "completions/min_length": 3.5,
86
+ "completions/min_terminated_length": 3.5,
87
+ "entropy": 0.7027173407775991,
88
+ "epoch": 0.00984251968503937,
89
+ "frac_reward_zero_std": 0.3,
90
+ "grad_norm": 0.0,
91
+ "learning_rate": 4.9705e-06,
92
+ "loss": 0.023067036271095277,
93
+ "num_tokens": 17440.0,
94
+ "reward": 1.1290264517068862,
95
+ "reward_std": 0.03574381650541909,
96
+ "rewards/reward_grounded/mean": 1.0,
97
+ "rewards/reward_grounded/std": 0.0,
98
+ "rewards/reward_length/mean": -0.052685546875,
99
+ "rewards/reward_length/std": 0.012657346948981286,
100
+ "rewards/reward_semantic/mean": 0.4388690486550331,
101
+ "rewards/reward_semantic/std": 0.11865786015987397,
102
+ "step": 60,
103
+ "step_time": 1.6641022313501708
104
+ },
105
+ {
106
+ "clip_ratio/high_max": 0.0,
107
+ "clip_ratio/high_mean": 0.0,
108
+ "clip_ratio/low_mean": 0.0,
109
+ "clip_ratio/low_min": 0.0,
110
+ "clip_ratio/region_mean": 0.0,
111
+ "completions/clipped_ratio": 0.0,
112
+ "completions/max_length": 7.05,
113
+ "completions/max_terminated_length": 7.05,
114
+ "completions/mean_length": 4.89375,
115
+ "completions/mean_terminated_length": 4.89375,
116
+ "completions/min_length": 3.45,
117
+ "completions/min_terminated_length": 3.45,
118
+ "entropy": 0.8099277728608285,
119
+ "epoch": 0.013123359580052493,
120
+ "frac_reward_zero_std": 0.2,
121
+ "grad_norm": 2.578125,
122
+ "learning_rate": 4.9605e-06,
123
+ "loss": 0.12280056476593018,
124
+ "num_tokens": 23631.0,
125
+ "reward": 1.08654942214489,
126
+ "reward_std": 0.040316034654097166,
127
+ "rewards/reward_grounded/mean": 1.0,
128
+ "rewards/reward_grounded/std": 0.0,
129
+ "rewards/reward_length/mean": -0.06103515625,
130
+ "rewards/reward_length/std": 0.022364351549185812,
131
+ "rewards/reward_semantic/mean": 0.2986706353724003,
132
+ "rewards/reward_semantic/std": 0.1318330906331539,
133
+ "step": 80,
134
+ "step_time": 1.7635547111498453
135
+ },
136
+ {
137
+ "clip_ratio/high_max": 0.0,
138
+ "clip_ratio/high_mean": 0.0,
139
+ "clip_ratio/low_mean": 0.0,
140
+ "clip_ratio/low_min": 0.0,
141
+ "clip_ratio/region_mean": 0.0,
142
+ "completions/clipped_ratio": 0.0,
143
+ "completions/max_length": 6.75,
144
+ "completions/max_terminated_length": 6.75,
145
+ "completions/mean_length": 4.75,
146
+ "completions/mean_terminated_length": 4.75,
147
+ "completions/min_length": 3.4,
148
+ "completions/min_terminated_length": 3.4,
149
+ "entropy": 0.8052310052706162,
150
+ "epoch": 0.016404199475065617,
151
+ "frac_reward_zero_std": 0.2,
152
+ "grad_norm": 3.171875,
153
+ "learning_rate": 4.9505e-06,
154
+ "loss": 0.007156229019165039,
155
+ "num_tokens": 29599.0,
156
+ "reward": 1.0940562933683395,
157
+ "reward_std": 0.030761140065442304,
158
+ "rewards/reward_grounded/mean": 1.0,
159
+ "rewards/reward_grounded/std": 0.0,
160
+ "rewards/reward_length/mean": -0.06103515625,
161
+ "rewards/reward_length/std": 0.017728388286195697,
162
+ "rewards/reward_semantic/mean": 0.3236935347318649,
163
+ "rewards/reward_semantic/std": 0.1025692880153656,
164
+ "step": 100,
165
+ "step_time": 1.6173622274000992
166
+ },
167
+ {
168
+ "clip_ratio/high_max": 0.0,
169
+ "clip_ratio/high_mean": 0.0,
170
+ "clip_ratio/low_mean": 0.0,
171
+ "clip_ratio/low_min": 0.0,
172
+ "clip_ratio/region_mean": 0.0,
173
+ "completions/clipped_ratio": 0.0,
174
+ "completions/max_length": 5.6,
175
+ "completions/max_terminated_length": 5.6,
176
+ "completions/mean_length": 4.125,
177
+ "completions/mean_terminated_length": 4.125,
178
+ "completions/min_length": 3.15,
179
+ "completions/min_terminated_length": 3.15,
180
+ "entropy": 0.7014230243818019,
181
+ "epoch": 0.01968503937007874,
182
+ "frac_reward_zero_std": 0.3,
183
+ "grad_norm": 0.0,
184
+ "learning_rate": 4.9405e-06,
185
+ "loss": 0.07218956351280212,
186
+ "num_tokens": 35107.0,
187
+ "reward": 1.1007445603609085,
188
+ "reward_std": 0.028740251205454113,
189
+ "rewards/reward_grounded/mean": 1.0,
190
+ "rewards/reward_grounded/std": 0.0,
191
+ "rewards/reward_length/mean": -0.04814453125,
192
+ "rewards/reward_length/std": 0.010129839461296797,
193
+ "rewards/reward_semantic/mean": 0.34383928664028646,
194
+ "rewards/reward_semantic/std": 0.09484268203377724,
195
+ "step": 120,
196
+ "step_time": 1.5441937706006683
197
+ },
198
+ {
199
+ "clip_ratio/high_max": 0.0,
200
+ "clip_ratio/high_mean": 0.0,
201
+ "clip_ratio/low_mean": 0.0,
202
+ "clip_ratio/low_min": 0.0,
203
+ "clip_ratio/region_mean": 0.0,
204
+ "completions/clipped_ratio": 0.0,
205
+ "completions/max_length": 7.25,
206
+ "completions/max_terminated_length": 7.25,
207
+ "completions/mean_length": 4.9625,
208
+ "completions/mean_terminated_length": 4.9625,
209
+ "completions/min_length": 3.5,
210
+ "completions/min_terminated_length": 3.5,
211
+ "entropy": 0.7183661407629188,
212
+ "epoch": 0.022965879265091863,
213
+ "frac_reward_zero_std": 0.15,
214
+ "grad_norm": 2.609375,
215
+ "learning_rate": 4.9305000000000005e-06,
216
+ "loss": 0.05585476756095886,
217
+ "num_tokens": 40989.0,
218
+ "reward": 1.1004581451416016,
219
+ "reward_std": 0.03431669266719837,
220
+ "rewards/reward_grounded/mean": 1.0,
221
+ "rewards/reward_grounded/std": 0.0,
222
+ "rewards/reward_length/mean": -0.059765625,
223
+ "rewards/reward_length/std": 0.017409664555452765,
224
+ "rewards/reward_semantic/mean": 0.34482143111526964,
225
+ "rewards/reward_semantic/std": 0.11257617846131325,
226
+ "step": 140,
227
+ "step_time": 1.7615427238992196
228
+ },
229
+ {
230
+ "clip_ratio/high_max": 0.0,
231
+ "clip_ratio/high_mean": 0.0,
232
+ "clip_ratio/low_mean": 0.0,
233
+ "clip_ratio/low_min": 0.0,
234
+ "clip_ratio/region_mean": 0.0,
235
+ "completions/clipped_ratio": 0.0,
236
+ "completions/max_length": 7.05,
237
+ "completions/max_terminated_length": 7.05,
238
+ "completions/mean_length": 4.89375,
239
+ "completions/mean_terminated_length": 4.89375,
240
+ "completions/min_length": 3.5,
241
+ "completions/min_terminated_length": 3.5,
242
+ "entropy": 0.6915914016732131,
243
+ "epoch": 0.026246719160104987,
244
+ "frac_reward_zero_std": 0.1,
245
+ "grad_norm": 2.859375,
246
+ "learning_rate": 4.920500000000001e-06,
247
+ "loss": 0.0472392201423645,
248
+ "num_tokens": 47220.0,
249
+ "reward": 1.0850708901882171,
250
+ "reward_std": 0.06178898175858194,
251
+ "rewards/reward_grounded/mean": 1.0,
252
+ "rewards/reward_grounded/std": 0.0,
253
+ "rewards/reward_length/mean": -0.065576171875,
254
+ "rewards/reward_length/std": 0.01732458183541894,
255
+ "rewards/reward_semantic/mean": 0.2944990089163184,
256
+ "rewards/reward_semantic/std": 0.2051414344459772,
257
+ "step": 160,
258
+ "step_time": 1.7252537605003453
259
+ },
260
+ {
261
+ "clip_ratio/high_max": 0.0,
262
+ "clip_ratio/high_mean": 0.0,
263
+ "clip_ratio/low_mean": 0.0,
264
+ "clip_ratio/low_min": 0.0,
265
+ "clip_ratio/region_mean": 0.0,
266
+ "completions/clipped_ratio": 0.0,
267
+ "completions/max_length": 6.25,
268
+ "completions/max_terminated_length": 6.25,
269
+ "completions/mean_length": 4.55625,
270
+ "completions/mean_terminated_length": 4.55625,
271
+ "completions/min_length": 3.35,
272
+ "completions/min_terminated_length": 3.35,
273
+ "entropy": 0.8429647823562846,
274
+ "epoch": 0.02952755905511811,
275
+ "frac_reward_zero_std": 0.05,
276
+ "grad_norm": 3.078125,
277
+ "learning_rate": 4.910500000000001e-06,
278
+ "loss": 0.08106310367584228,
279
+ "num_tokens": 53053.0,
280
+ "reward": 1.0679614007472993,
281
+ "reward_std": 0.03796071938340902,
282
+ "rewards/reward_grounded/mean": 1.0,
283
+ "rewards/reward_grounded/std": 0.0,
284
+ "rewards/reward_length/mean": -0.0566650390625,
285
+ "rewards/reward_length/std": 0.014682336029363797,
286
+ "rewards/reward_semantic/mean": 0.23598214369267226,
287
+ "rewards/reward_semantic/std": 0.12518299631774427,
288
+ "step": 180,
289
+ "step_time": 1.7508945095991293
290
+ },
291
+ {
292
+ "clip_ratio/high_max": 0.0,
293
+ "clip_ratio/high_mean": 0.0,
294
+ "clip_ratio/low_mean": 0.0,
295
+ "clip_ratio/low_min": 0.0,
296
+ "clip_ratio/region_mean": 0.0,
297
+ "completions/clipped_ratio": 0.0,
298
+ "completions/max_length": 6.2,
299
+ "completions/max_terminated_length": 6.2,
300
+ "completions/mean_length": 4.5375,
301
+ "completions/mean_terminated_length": 4.5375,
302
+ "completions/min_length": 3.75,
303
+ "completions/min_terminated_length": 3.75,
304
+ "entropy": 0.5915892604841246,
305
+ "epoch": 0.03280839895013123,
306
+ "frac_reward_zero_std": 0.35,
307
+ "grad_norm": 3.71875,
308
+ "learning_rate": 4.9005e-06,
309
+ "loss": 0.0888818621635437,
310
+ "num_tokens": 58931.0,
311
+ "reward": 1.120530515909195,
312
+ "reward_std": 0.0405463597693597,
313
+ "rewards/reward_grounded/mean": 1.0,
314
+ "rewards/reward_grounded/std": 0.0,
315
+ "rewards/reward_length/mean": -0.0493896484375,
316
+ "rewards/reward_length/std": 0.008753356069792061,
317
+ "rewards/reward_semantic/mean": 0.4100000007078052,
318
+ "rewards/reward_semantic/std": 0.13480054624378682,
319
+ "step": 200,
320
+ "step_time": 1.713693901949955
321
+ },
322
+ {
323
+ "clip_ratio/high_max": 0.0,
324
+ "clip_ratio/high_mean": 0.0,
325
+ "clip_ratio/low_mean": 0.0,
326
+ "clip_ratio/low_min": 0.0,
327
+ "clip_ratio/region_mean": 0.0,
328
+ "completions/clipped_ratio": 0.0,
329
+ "completions/max_length": 6.45,
330
+ "completions/max_terminated_length": 6.45,
331
+ "completions/mean_length": 5.15,
332
+ "completions/mean_terminated_length": 5.15,
333
+ "completions/min_length": 4.2,
334
+ "completions/min_terminated_length": 4.2,
335
+ "entropy": 0.6395838386631112,
336
+ "epoch": 0.03608923884514436,
337
+ "frac_reward_zero_std": 0.35,
338
+ "grad_norm": 2.625,
339
+ "learning_rate": 4.8905e-06,
340
+ "loss": 0.06417204141616821,
341
+ "num_tokens": 65139.0,
342
+ "reward": 1.1249804645776749,
343
+ "reward_std": 0.027039075889479135,
344
+ "rewards/reward_grounded/mean": 1.0,
345
+ "rewards/reward_grounded/std": 0.0,
346
+ "rewards/reward_length/mean": -0.056640625,
347
+ "rewards/reward_length/std": 0.011077208817005158,
348
+ "rewards/reward_semantic/mean": 0.4260416656732559,
349
+ "rewards/reward_semantic/std": 0.08868714347481728,
350
+ "step": 220,
351
+ "step_time": 1.7278446534002796
352
+ },
353
+ {
354
+ "clip_ratio/high_max": 0.0,
355
+ "clip_ratio/high_mean": 0.0,
356
+ "clip_ratio/low_mean": 0.0,
357
+ "clip_ratio/low_min": 0.0,
358
+ "clip_ratio/region_mean": 0.0,
359
+ "completions/clipped_ratio": 0.0,
360
+ "completions/max_length": 5.4,
361
+ "completions/max_terminated_length": 5.4,
362
+ "completions/mean_length": 4.2625,
363
+ "completions/mean_terminated_length": 4.2625,
364
+ "completions/min_length": 3.0,
365
+ "completions/min_terminated_length": 3.0,
366
+ "entropy": 0.5652526923571713,
367
+ "epoch": 0.03937007874015748,
368
+ "frac_reward_zero_std": 0.35,
369
+ "grad_norm": 8.125,
370
+ "learning_rate": 4.8805e-06,
371
+ "loss": 0.07150443196296692,
372
+ "num_tokens": 71221.0,
373
+ "reward": 1.1466667413711549,
374
+ "reward_std": 0.03061878081643954,
375
+ "rewards/reward_grounded/mean": 1.0,
376
+ "rewards/reward_grounded/std": 0.0,
377
+ "rewards/reward_length/mean": -0.063330078125,
378
+ "rewards/reward_length/std": 0.015704452409408987,
379
+ "rewards/reward_semantic/mean": 0.4994440995156765,
380
+ "rewards/reward_semantic/std": 0.10041465926915408,
381
+ "step": 240,
382
+ "step_time": 1.6952891380005894
383
+ },
384
+ {
385
+ "clip_ratio/high_max": 0.0,
386
+ "clip_ratio/high_mean": 0.0,
387
+ "clip_ratio/low_mean": 0.0,
388
+ "clip_ratio/low_min": 0.0,
389
+ "clip_ratio/region_mean": 0.0,
390
+ "completions/clipped_ratio": 0.0,
391
+ "completions/max_length": 6.3,
392
+ "completions/max_terminated_length": 6.3,
393
+ "completions/mean_length": 4.46875,
394
+ "completions/mean_terminated_length": 4.46875,
395
+ "completions/min_length": 3.35,
396
+ "completions/min_terminated_length": 3.35,
397
+ "entropy": 0.672205779588694,
398
+ "epoch": 0.04265091863517061,
399
+ "frac_reward_zero_std": 0.2,
400
+ "grad_norm": 6.4375,
401
+ "learning_rate": 4.8705e-06,
402
+ "loss": 0.09235309958457946,
403
+ "num_tokens": 77416.0,
404
+ "reward": 1.0690006405115127,
405
+ "reward_std": 0.019543477255501783,
406
+ "rewards/reward_grounded/mean": 1.0,
407
+ "rewards/reward_grounded/std": 0.0,
408
+ "rewards/reward_length/mean": -0.0533203125,
409
+ "rewards/reward_length/std": 0.017377534485422074,
410
+ "rewards/reward_semantic/mean": 0.23888888889923693,
411
+ "rewards/reward_semantic/std": 0.0630335971713066,
412
+ "step": 260,
413
+ "step_time": 1.7545264535998286
414
+ },
415
+ {
416
+ "clip_ratio/high_max": 0.0,
417
+ "clip_ratio/high_mean": 0.0,
418
+ "clip_ratio/low_mean": 0.0,
419
+ "clip_ratio/low_min": 0.0,
420
+ "clip_ratio/region_mean": 0.0,
421
+ "completions/clipped_ratio": 0.0,
422
+ "completions/max_length": 5.3,
423
+ "completions/max_terminated_length": 5.3,
424
+ "completions/mean_length": 4.525,
425
+ "completions/mean_terminated_length": 4.525,
426
+ "completions/min_length": 3.65,
427
+ "completions/min_terminated_length": 3.65,
428
+ "entropy": 0.5580058551760885,
429
+ "epoch": 0.045931758530183726,
430
+ "frac_reward_zero_std": 0.45,
431
+ "grad_norm": 4.0625,
432
+ "learning_rate": 4.8605e-06,
433
+ "loss": 0.02879692018032074,
434
+ "num_tokens": 83332.0,
435
+ "reward": 1.1027960389852525,
436
+ "reward_std": 0.011529068512027152,
437
+ "rewards/reward_grounded/mean": 1.0,
438
+ "rewards/reward_grounded/std": 0.0,
439
+ "rewards/reward_length/mean": -0.0558349609375,
440
+ "rewards/reward_length/std": 0.008870826044585556,
441
+ "rewards/reward_semantic/mean": 0.3519593259319663,
442
+ "rewards/reward_semantic/std": 0.03782847560942173,
443
+ "step": 280,
444
+ "step_time": 1.5900321766996968
445
+ },
446
+ {
447
+ "clip_ratio/high_max": 0.0,
448
+ "clip_ratio/high_mean": 0.0,
449
+ "clip_ratio/low_mean": 0.0,
450
+ "clip_ratio/low_min": 0.0,
451
+ "clip_ratio/region_mean": 0.0,
452
+ "completions/clipped_ratio": 0.0,
453
+ "completions/max_length": 6.35,
454
+ "completions/max_terminated_length": 6.35,
455
+ "completions/mean_length": 4.56875,
456
+ "completions/mean_terminated_length": 4.56875,
457
+ "completions/min_length": 3.2,
458
+ "completions/min_terminated_length": 3.2,
459
+ "entropy": 0.8288409212742408,
460
+ "epoch": 0.04921259842519685,
461
+ "frac_reward_zero_std": 0.2,
462
+ "grad_norm": 0.0,
463
+ "learning_rate": 4.8505e-06,
464
+ "loss": 0.044028478860855105,
465
+ "num_tokens": 89567.0,
466
+ "reward": 1.1026858359575271,
467
+ "reward_std": 0.03983475278655533,
468
+ "rewards/reward_grounded/mean": 1.0,
469
+ "rewards/reward_grounded/std": 0.0,
470
+ "rewards/reward_length/mean": -0.062744140625,
471
+ "rewards/reward_length/std": 0.018717235408257694,
472
+ "rewards/reward_semantic/mean": 0.35274351127445697,
473
+ "rewards/reward_semantic/std": 0.13158041909337043,
474
+ "step": 300,
475
+ "step_time": 1.7004065297001034
476
+ },
477
+ {
478
+ "clip_ratio/high_max": 0.0,
479
+ "clip_ratio/high_mean": 0.0,
480
+ "clip_ratio/low_mean": 0.0,
481
+ "clip_ratio/low_min": 0.0,
482
+ "clip_ratio/region_mean": 0.0,
483
+ "completions/clipped_ratio": 0.0,
484
+ "completions/max_length": 5.65,
485
+ "completions/max_terminated_length": 5.65,
486
+ "completions/mean_length": 4.5875,
487
+ "completions/mean_terminated_length": 4.5875,
488
+ "completions/min_length": 3.5,
489
+ "completions/min_terminated_length": 3.5,
490
+ "entropy": 0.7240128130373705,
491
+ "epoch": 0.05249343832020997,
492
+ "frac_reward_zero_std": 0.2,
493
+ "grad_norm": 5.0625,
494
+ "learning_rate": 4.8405000000000005e-06,
495
+ "loss": 0.03284294605255127,
496
+ "num_tokens": 95157.0,
497
+ "reward": 1.0648824661970138,
498
+ "reward_std": 0.014824012622557347,
499
+ "rewards/reward_grounded/mean": 1.0,
500
+ "rewards/reward_grounded/std": 0.0,
501
+ "rewards/reward_length/mean": -0.055029296875,
502
+ "rewards/reward_length/std": 0.010819258564151824,
503
+ "rewards/reward_semantic/mean": 0.2254464277997613,
504
+ "rewards/reward_semantic/std": 0.04843873754143715,
505
+ "step": 320,
506
+ "step_time": 1.6396856073492927
507
+ },
508
+ {
509
+ "clip_ratio/high_max": 0.0,
510
+ "clip_ratio/high_mean": 0.0,
511
+ "clip_ratio/low_mean": 0.0,
512
+ "clip_ratio/low_min": 0.0,
513
+ "clip_ratio/region_mean": 0.0,
514
+ "completions/clipped_ratio": 0.0,
515
+ "completions/max_length": 6.25,
516
+ "completions/max_terminated_length": 6.25,
517
+ "completions/mean_length": 4.625,
518
+ "completions/mean_terminated_length": 4.625,
519
+ "completions/min_length": 3.35,
520
+ "completions/min_terminated_length": 3.35,
521
+ "entropy": 0.8246265778172528,
522
+ "epoch": 0.0557742782152231,
523
+ "frac_reward_zero_std": 0.35,
524
+ "grad_norm": 0.0,
525
+ "learning_rate": 4.830500000000001e-06,
526
+ "loss": 0.05039528012275696,
527
+ "num_tokens": 100857.0,
528
+ "reward": 1.0999272018671036,
529
+ "reward_std": 0.012020479088823777,
530
+ "rewards/reward_grounded/mean": 1.0,
531
+ "rewards/reward_grounded/std": 0.0,
532
+ "rewards/reward_length/mean": -0.0510986328125,
533
+ "rewards/reward_length/std": 0.013131194305606187,
534
+ "rewards/reward_semantic/mean": 0.3416071431711316,
535
+ "rewards/reward_semantic/std": 0.03859963417053223,
536
+ "step": 340,
537
+ "step_time": 1.6233859046002181
538
+ },
539
+ {
540
+ "clip_ratio/high_max": 0.0,
541
+ "clip_ratio/high_mean": 0.0,
542
+ "clip_ratio/low_mean": 0.0,
543
+ "clip_ratio/low_min": 0.0,
544
+ "clip_ratio/region_mean": 0.0,
545
+ "completions/clipped_ratio": 0.0,
546
+ "completions/max_length": 5.75,
547
+ "completions/max_terminated_length": 5.75,
548
+ "completions/mean_length": 4.49375,
549
+ "completions/mean_terminated_length": 4.49375,
550
+ "completions/min_length": 3.7,
551
+ "completions/min_terminated_length": 3.7,
552
+ "entropy": 0.49938802123069764,
553
+ "epoch": 0.05905511811023622,
554
+ "frac_reward_zero_std": 0.3,
555
+ "grad_norm": 0.0,
556
+ "learning_rate": 4.8205e-06,
557
+ "loss": 0.01368042677640915,
558
+ "num_tokens": 106664.0,
559
+ "reward": 1.1010554283857346,
560
+ "reward_std": 0.045361624303768625,
561
+ "rewards/reward_grounded/mean": 1.0,
562
+ "rewards/reward_grounded/std": 0.0,
563
+ "rewards/reward_length/mean": -0.0576416015625,
564
+ "rewards/reward_length/std": 0.010658931022044271,
565
+ "rewards/reward_semantic/mean": 0.3464583337306976,
566
+ "rewards/reward_semantic/std": 0.15053959861397742,
567
+ "step": 360,
568
+ "step_time": 1.6211198363504082
569
+ },
570
+ {
571
+ "clip_ratio/high_max": 0.0,
572
+ "clip_ratio/high_mean": 0.0,
573
+ "clip_ratio/low_mean": 0.0,
574
+ "clip_ratio/low_min": 0.0,
575
+ "clip_ratio/region_mean": 0.0,
576
+ "completions/clipped_ratio": 0.0,
577
+ "completions/max_length": 5.8,
578
+ "completions/max_terminated_length": 5.8,
579
+ "completions/mean_length": 4.74375,
580
+ "completions/mean_terminated_length": 4.74375,
581
+ "completions/min_length": 3.6,
582
+ "completions/min_terminated_length": 3.6,
583
+ "entropy": 0.6371999262647023,
584
+ "epoch": 0.062335958005249346,
585
+ "frac_reward_zero_std": 0.3,
586
+ "grad_norm": 9.1875,
587
+ "learning_rate": 4.8105e-06,
588
+ "loss": 0.03233610987663269,
589
+ "num_tokens": 112511.0,
590
+ "reward": 1.1021533191204071,
591
+ "reward_std": 0.02930783515548683,
592
+ "rewards/reward_grounded/mean": 1.0,
593
+ "rewards/reward_grounded/std": 0.0,
594
+ "rewards/reward_length/mean": -0.05693359375,
595
+ "rewards/reward_length/std": 0.012558965157950297,
596
+ "rewards/reward_semantic/mean": 0.3500000026077032,
597
+ "rewards/reward_semantic/std": 0.09658735543489456,
598
+ "step": 380,
599
+ "step_time": 1.6564469950499188
600
+ },
601
+ {
602
+ "clip_ratio/high_max": 0.0,
603
+ "clip_ratio/high_mean": 0.0,
604
+ "clip_ratio/low_mean": 0.0,
605
+ "clip_ratio/low_min": 0.0,
606
+ "clip_ratio/region_mean": 0.0,
607
+ "completions/clipped_ratio": 0.0,
608
+ "completions/max_length": 6.05,
609
+ "completions/max_terminated_length": 6.05,
610
+ "completions/mean_length": 4.33125,
611
+ "completions/mean_terminated_length": 4.33125,
612
+ "completions/min_length": 3.35,
613
+ "completions/min_terminated_length": 3.35,
614
+ "entropy": 0.7747498400121913,
615
+ "epoch": 0.06561679790026247,
616
+ "frac_reward_zero_std": 0.15,
617
+ "grad_norm": 16.0,
618
+ "learning_rate": 4.8005e-06,
619
+ "loss": 0.02885372340679169,
620
+ "num_tokens": 118188.0,
621
+ "reward": 1.1014230579137803,
622
+ "reward_std": 0.054096709901932626,
623
+ "rewards/reward_grounded/mean": 1.0,
624
+ "rewards/reward_grounded/std": 0.0,
625
+ "rewards/reward_length/mean": -0.054931640625,
626
+ "rewards/reward_length/std": 0.01498162968782708,
627
+ "rewards/reward_semantic/mean": 0.3472321443259716,
628
+ "rewards/reward_semantic/std": 0.17982315719127656,
629
+ "step": 400,
630
+ "step_time": 1.6521750739995695
631
+ },
632
+ {
633
+ "clip_ratio/high_max": 0.0,
634
+ "clip_ratio/high_mean": 0.0,
635
+ "clip_ratio/low_mean": 0.0,
636
+ "clip_ratio/low_min": 0.0,
637
+ "clip_ratio/region_mean": 0.0,
638
+ "completions/clipped_ratio": 0.0,
639
+ "completions/max_length": 5.55,
640
+ "completions/max_terminated_length": 5.55,
641
+ "completions/mean_length": 4.25,
642
+ "completions/mean_terminated_length": 4.25,
643
+ "completions/min_length": 3.25,
644
+ "completions/min_terminated_length": 3.25,
645
+ "entropy": 0.6093234239519916,
646
+ "epoch": 0.0688976377952756,
647
+ "frac_reward_zero_std": 0.2,
648
+ "grad_norm": 8.1875,
649
+ "learning_rate": 4.7905e-06,
650
+ "loss": 0.07366046905517579,
651
+ "num_tokens": 123980.0,
652
+ "reward": 1.088520696759224,
653
+ "reward_std": 0.02127447940329148,
654
+ "rewards/reward_grounded/mean": 1.0,
655
+ "rewards/reward_grounded/std": 0.0,
656
+ "rewards/reward_length/mean": -0.0520263671875,
657
+ "rewards/reward_length/std": 0.010646102635655552,
658
+ "rewards/reward_semantic/mean": 0.30374008174985645,
659
+ "rewards/reward_semantic/std": 0.06995791383087635,
660
+ "step": 420,
661
+ "step_time": 1.598766735350273
662
+ },
663
+ {
664
+ "clip_ratio/high_max": 0.0,
665
+ "clip_ratio/high_mean": 0.0,
666
+ "clip_ratio/low_mean": 0.0,
667
+ "clip_ratio/low_min": 0.0,
668
+ "clip_ratio/region_mean": 0.0,
669
+ "completions/clipped_ratio": 0.0,
670
+ "completions/max_length": 5.6,
671
+ "completions/max_terminated_length": 5.6,
672
+ "completions/mean_length": 4.4625,
673
+ "completions/mean_terminated_length": 4.4625,
674
+ "completions/min_length": 3.4,
675
+ "completions/min_terminated_length": 3.4,
676
+ "entropy": 0.6000309210250634,
677
+ "epoch": 0.07217847769028872,
678
+ "frac_reward_zero_std": 0.25,
679
+ "grad_norm": 6.25,
680
+ "learning_rate": 4.7805e-06,
681
+ "loss": 0.02538236975669861,
682
+ "num_tokens": 130310.0,
683
+ "reward": 1.1284031838178634,
684
+ "reward_std": 0.030537298787385224,
685
+ "rewards/reward_grounded/mean": 1.0,
686
+ "rewards/reward_grounded/std": 0.0,
687
+ "rewards/reward_length/mean": -0.0548828125,
688
+ "rewards/reward_length/std": 0.012618192343506963,
689
+ "rewards/reward_semantic/mean": 0.43715773820877074,
690
+ "rewards/reward_semantic/std": 0.10091984458267689,
691
+ "step": 440,
692
+ "step_time": 1.6581264014497719
693
+ },
694
+ {
695
+ "clip_ratio/high_max": 0.0,
696
+ "clip_ratio/high_mean": 0.0,
697
+ "clip_ratio/low_mean": 0.0,
698
+ "clip_ratio/low_min": 0.0,
699
+ "clip_ratio/region_mean": 0.0,
700
+ "completions/clipped_ratio": 0.0,
701
+ "completions/max_length": 6.75,
702
+ "completions/max_terminated_length": 6.75,
703
+ "completions/mean_length": 4.9,
704
+ "completions/mean_terminated_length": 4.9,
705
+ "completions/min_length": 3.45,
706
+ "completions/min_terminated_length": 3.45,
707
+ "entropy": 0.7827204793691636,
708
+ "epoch": 0.07545931758530183,
709
+ "frac_reward_zero_std": 0.15,
710
+ "grad_norm": 2.140625,
711
+ "learning_rate": 4.7705e-06,
712
+ "loss": 0.09432601332664489,
713
+ "num_tokens": 136206.0,
714
+ "reward": 1.061219373345375,
715
+ "reward_std": 0.04166477765902528,
716
+ "rewards/reward_grounded/mean": 1.0,
717
+ "rewards/reward_grounded/std": 0.0,
718
+ "rewards/reward_length/mean": -0.067041015625,
719
+ "rewards/reward_length/std": 0.016452802077401428,
720
+ "rewards/reward_semantic/mean": 0.21523809637874364,
721
+ "rewards/reward_semantic/std": 0.13769038021564484,
722
+ "step": 460,
723
+ "step_time": 1.763348830850282
724
+ },
725
+ {
726
+ "clip_ratio/high_max": 0.0,
727
+ "clip_ratio/high_mean": 0.0,
728
+ "clip_ratio/low_mean": 0.0,
729
+ "clip_ratio/low_min": 0.0,
730
+ "clip_ratio/region_mean": 0.0,
731
+ "completions/clipped_ratio": 0.0,
732
+ "completions/max_length": 5.95,
733
+ "completions/max_terminated_length": 5.95,
734
+ "completions/mean_length": 4.5125,
735
+ "completions/mean_terminated_length": 4.5125,
736
+ "completions/min_length": 3.3,
737
+ "completions/min_terminated_length": 3.3,
738
+ "entropy": 0.6544731822835275,
739
+ "epoch": 0.07874015748031496,
740
+ "frac_reward_zero_std": 0.25,
741
+ "grad_norm": 0.0,
742
+ "learning_rate": 4.7605e-06,
743
+ "loss": 0.095479816198349,
744
+ "num_tokens": 142168.0,
745
+ "reward": 1.110293486714363,
746
+ "reward_std": 0.03730325770447962,
747
+ "rewards/reward_grounded/mean": 1.0,
748
+ "rewards/reward_grounded/std": 0.0,
749
+ "rewards/reward_length/mean": -0.0526123046875,
750
+ "rewards/reward_length/std": 0.013677800679579378,
751
+ "rewards/reward_semantic/mean": 0.37641369104385375,
752
+ "rewards/reward_semantic/std": 0.12245006412267685,
753
+ "step": 480,
754
+ "step_time": 1.693691986000158
755
+ },
756
+ {
757
+ "clip_ratio/high_max": 0.0,
758
+ "clip_ratio/high_mean": 0.0,
759
+ "clip_ratio/low_mean": 0.0,
760
+ "clip_ratio/low_min": 0.0,
761
+ "clip_ratio/region_mean": 0.0,
762
+ "completions/clipped_ratio": 0.0,
763
+ "completions/max_length": 6.2,
764
+ "completions/max_terminated_length": 6.2,
765
+ "completions/mean_length": 4.5375,
766
+ "completions/mean_terminated_length": 4.5375,
767
+ "completions/min_length": 3.3,
768
+ "completions/min_terminated_length": 3.3,
769
+ "entropy": 0.7134705329648569,
770
+ "epoch": 0.08202099737532809,
771
+ "frac_reward_zero_std": 0.2,
772
+ "grad_norm": 0.0,
773
+ "learning_rate": 4.7505000000000005e-06,
774
+ "loss": 0.04971726834774017,
775
+ "num_tokens": 148190.0,
776
+ "reward": 1.059982505440712,
777
+ "reward_std": 0.029997232874302428,
778
+ "rewards/reward_grounded/mean": 1.0,
779
+ "rewards/reward_grounded/std": 0.0,
780
+ "rewards/reward_length/mean": -0.053564453125,
781
+ "rewards/reward_length/std": 0.012155310017988086,
782
+ "rewards/reward_semantic/mean": 0.20886905267834663,
783
+ "rewards/reward_semantic/std": 0.0987570121884346,
784
+ "step": 500,
785
+ "step_time": 1.7915035136496953
786
+ },
787
+ {
788
+ "clip_ratio/high_max": 0.0,
789
+ "clip_ratio/high_mean": 0.0,
790
+ "clip_ratio/low_mean": 0.0,
791
+ "clip_ratio/low_min": 0.0,
792
+ "clip_ratio/region_mean": 0.0,
793
+ "completions/clipped_ratio": 0.0,
794
+ "completions/max_length": 4.75,
795
+ "completions/max_terminated_length": 4.75,
796
+ "completions/mean_length": 3.85625,
797
+ "completions/mean_terminated_length": 3.85625,
798
+ "completions/min_length": 3.15,
799
+ "completions/min_terminated_length": 3.15,
800
+ "entropy": 0.489788422039328,
801
+ "epoch": 0.08530183727034121,
802
+ "frac_reward_zero_std": 0.35,
803
+ "grad_norm": 0.0,
804
+ "learning_rate": 4.740500000000001e-06,
805
+ "loss": 0.015317876636981965,
806
+ "num_tokens": 153895.0,
807
+ "reward": 1.14158952832222,
808
+ "reward_std": 0.03168491964024724,
809
+ "rewards/reward_grounded/mean": 1.0,
810
+ "rewards/reward_grounded/std": 0.0,
811
+ "rewards/reward_length/mean": -0.0512451171875,
812
+ "rewards/reward_length/std": 0.007844625366851687,
813
+ "rewards/reward_semantic/mean": 0.48050595819950104,
814
+ "rewards/reward_semantic/std": 0.10510801821947098,
815
+ "step": 520,
816
+ "step_time": 1.6148291991500627
817
+ },
818
+ {
819
+ "clip_ratio/high_max": 0.0,
820
+ "clip_ratio/high_mean": 0.0,
821
+ "clip_ratio/low_mean": 0.0,
822
+ "clip_ratio/low_min": 0.0,
823
+ "clip_ratio/region_mean": 0.0,
824
+ "completions/clipped_ratio": 0.0,
825
+ "completions/max_length": 6.15,
826
+ "completions/max_terminated_length": 6.15,
827
+ "completions/mean_length": 4.54375,
828
+ "completions/mean_terminated_length": 4.54375,
829
+ "completions/min_length": 3.6,
830
+ "completions/min_terminated_length": 3.6,
831
+ "entropy": 0.7991890870037424,
832
+ "epoch": 0.08858267716535433,
833
+ "frac_reward_zero_std": 0.3,
834
+ "grad_norm": 2.546875,
835
+ "learning_rate": 4.730500000000001e-06,
836
+ "loss": 0.05347582697868347,
837
+ "num_tokens": 159790.0,
838
+ "reward": 1.0938216686248778,
839
+ "reward_std": 0.01884553566815157,
840
+ "rewards/reward_grounded/mean": 1.0,
841
+ "rewards/reward_grounded/std": 0.0,
842
+ "rewards/reward_length/mean": -0.0449951171875,
843
+ "rewards/reward_length/std": 0.007818706298712641,
844
+ "rewards/reward_semantic/mean": 0.3202380955219269,
845
+ "rewards/reward_semantic/std": 0.061923586577177045,
846
+ "step": 540,
847
+ "step_time": 1.8538054024499615
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
+ "completions/clipped_ratio": 0.0,
856
+ "completions/max_length": 5.2,
857
+ "completions/max_terminated_length": 5.2,
858
+ "completions/mean_length": 4.2375,
859
+ "completions/mean_terminated_length": 4.2375,
860
+ "completions/min_length": 3.45,
861
+ "completions/min_terminated_length": 3.45,
862
+ "entropy": 0.5264639136769802,
863
+ "epoch": 0.09186351706036745,
864
+ "frac_reward_zero_std": 0.3,
865
+ "grad_norm": 4.40625,
866
+ "learning_rate": 4.720500000000001e-06,
867
+ "loss": 0.022994591295719145,
868
+ "num_tokens": 165812.0,
869
+ "reward": 1.130643653869629,
870
+ "reward_std": 0.035184593574376775,
871
+ "rewards/reward_grounded/mean": 1.0,
872
+ "rewards/reward_grounded/std": 0.0,
873
+ "rewards/reward_length/mean": -0.0594482421875,
874
+ "rewards/reward_length/std": 0.012201186269521713,
875
+ "rewards/reward_semantic/mean": 0.44538690596818925,
876
+ "rewards/reward_semantic/std": 0.11669721417129039,
877
+ "step": 560,
878
+ "step_time": 1.5765380310496766
879
+ },
880
+ {
881
+ "clip_ratio/high_max": 0.0,
882
+ "clip_ratio/high_mean": 0.0,
883
+ "clip_ratio/low_mean": 0.0,
884
+ "clip_ratio/low_min": 0.0,
885
+ "clip_ratio/region_mean": 0.0,
886
+ "completions/clipped_ratio": 0.0,
887
+ "completions/max_length": 5.9,
888
+ "completions/max_terminated_length": 5.9,
889
+ "completions/mean_length": 4.53125,
890
+ "completions/mean_terminated_length": 4.53125,
891
+ "completions/min_length": 3.6,
892
+ "completions/min_terminated_length": 3.6,
893
+ "entropy": 0.6579255037364419,
894
+ "epoch": 0.09514435695538058,
895
+ "frac_reward_zero_std": 0.35,
896
+ "grad_norm": 4.625,
897
+ "learning_rate": 4.710500000000001e-06,
898
+ "loss": 0.03634168207645416,
899
+ "num_tokens": 171769.0,
900
+ "reward": 1.1139779269695282,
901
+ "reward_std": 0.028173982925363817,
902
+ "rewards/reward_grounded/mean": 1.0,
903
+ "rewards/reward_grounded/std": 0.0,
904
+ "rewards/reward_length/mean": -0.0481201171875,
905
+ "rewards/reward_length/std": 0.011657711572479457,
906
+ "rewards/reward_semantic/mean": 0.38794642984867095,
907
+ "rewards/reward_semantic/std": 0.09298380687832833,
908
+ "step": 580,
909
+ "step_time": 1.6179808518501886
910
+ },
911
+ {
912
+ "clip_ratio/high_max": 0.0,
913
+ "clip_ratio/high_mean": 0.0,
914
+ "clip_ratio/low_mean": 0.0,
915
+ "clip_ratio/low_min": 0.0,
916
+ "clip_ratio/region_mean": 0.0,
917
+ "completions/clipped_ratio": 0.0,
918
+ "completions/max_length": 5.9,
919
+ "completions/max_terminated_length": 5.9,
920
+ "completions/mean_length": 4.25,
921
+ "completions/mean_terminated_length": 4.25,
922
+ "completions/min_length": 3.2,
923
+ "completions/min_terminated_length": 3.2,
924
+ "entropy": 0.7248269693967814,
925
+ "epoch": 0.0984251968503937,
926
+ "frac_reward_zero_std": 0.2,
927
+ "grad_norm": 10.3125,
928
+ "learning_rate": 4.700500000000001e-06,
929
+ "loss": 0.09872611761093139,
930
+ "num_tokens": 177649.0,
931
+ "reward": 1.0644860982894897,
932
+ "reward_std": 0.026886727681267077,
933
+ "rewards/reward_grounded/mean": 1.0,
934
+ "rewards/reward_grounded/std": 0.0,
935
+ "rewards/reward_length/mean": -0.0540283203125,
936
+ "rewards/reward_length/std": 0.014043591287918388,
937
+ "rewards/reward_semantic/mean": 0.22395833432674409,
938
+ "rewards/reward_semantic/std": 0.0882976733148098,
939
+ "step": 600,
940
+ "step_time": 1.6426984294495923
941
+ },
942
+ {
943
+ "clip_ratio/high_max": 0.0,
944
+ "clip_ratio/high_mean": 0.0,
945
+ "clip_ratio/low_mean": 0.0,
946
+ "clip_ratio/low_min": 0.0,
947
+ "clip_ratio/region_mean": 0.0,
948
+ "completions/clipped_ratio": 0.0,
949
+ "completions/max_length": 5.4,
950
+ "completions/max_terminated_length": 5.4,
951
+ "completions/mean_length": 4.5625,
952
+ "completions/mean_terminated_length": 4.5625,
953
+ "completions/min_length": 3.55,
954
+ "completions/min_terminated_length": 3.55,
955
+ "entropy": 0.4465312559736049,
956
+ "epoch": 0.10170603674540682,
957
+ "frac_reward_zero_std": 0.4,
958
+ "grad_norm": 0.0,
959
+ "learning_rate": 4.6905e-06,
960
+ "loss": -0.010341504216194152,
961
+ "num_tokens": 183835.0,
962
+ "reward": 1.1265694797039032,
963
+ "reward_std": 0.028417593393533024,
964
+ "rewards/reward_grounded/mean": 1.0,
965
+ "rewards/reward_grounded/std": 0.0,
966
+ "rewards/reward_length/mean": -0.0556640625,
967
+ "rewards/reward_length/std": 0.011586534441448748,
968
+ "rewards/reward_semantic/mean": 0.43117559552192686,
969
+ "rewards/reward_semantic/std": 0.0946267582476139,
970
+ "step": 620,
971
+ "step_time": 1.6642883875004373
972
+ },
973
+ {
974
+ "clip_ratio/high_max": 0.0,
975
+ "clip_ratio/high_mean": 0.0,
976
+ "clip_ratio/low_mean": 0.0,
977
+ "clip_ratio/low_min": 0.0,
978
+ "clip_ratio/region_mean": 0.0,
979
+ "completions/clipped_ratio": 0.0,
980
+ "completions/max_length": 6.8,
981
+ "completions/max_terminated_length": 6.8,
982
+ "completions/mean_length": 5.25,
983
+ "completions/mean_terminated_length": 5.25,
984
+ "completions/min_length": 3.8,
985
+ "completions/min_terminated_length": 3.8,
986
+ "entropy": 0.4096738053149238,
987
+ "epoch": 0.10498687664041995,
988
+ "frac_reward_zero_std": 0.4,
989
+ "grad_norm": 0.0,
990
+ "learning_rate": 4.6805e-06,
991
+ "loss": 0.05256671905517578,
992
+ "num_tokens": 190283.0,
993
+ "reward": 1.1175170987844467,
994
+ "reward_std": 0.015989218310278375,
995
+ "rewards/reward_grounded/mean": 1.0,
996
+ "rewards/reward_grounded/std": 0.0,
997
+ "rewards/reward_length/mean": -0.049658203125,
998
+ "rewards/reward_length/std": 0.008098985790275038,
999
+ "rewards/reward_semantic/mean": 0.4000000020489097,
1000
+ "rewards/reward_semantic/std": 0.05260697938501835,
1001
+ "step": 640,
1002
+ "step_time": 1.7202262992497708
1003
+ },
1004
+ {
1005
+ "clip_ratio/high_max": 0.0,
1006
+ "clip_ratio/high_mean": 0.0,
1007
+ "clip_ratio/low_mean": 0.0,
1008
+ "clip_ratio/low_min": 0.0,
1009
+ "clip_ratio/region_mean": 0.0,
1010
+ "completions/clipped_ratio": 0.0,
1011
+ "completions/max_length": 5.0,
1012
+ "completions/max_terminated_length": 5.0,
1013
+ "completions/mean_length": 4.40625,
1014
+ "completions/mean_terminated_length": 4.40625,
1015
+ "completions/min_length": 3.75,
1016
+ "completions/min_terminated_length": 3.75,
1017
+ "entropy": 0.44533549200182276,
1018
+ "epoch": 0.10826771653543307,
1019
+ "frac_reward_zero_std": 0.45,
1020
+ "grad_norm": 28.0,
1021
+ "learning_rate": 4.6705000000000004e-06,
1022
+ "loss": 0.000417201966047287,
1023
+ "num_tokens": 196284.0,
1024
+ "reward": 1.1231877475976944,
1025
+ "reward_std": 0.02121658924515941,
1026
+ "rewards/reward_grounded/mean": 1.0,
1027
+ "rewards/reward_grounded/std": 0.0,
1028
+ "rewards/reward_length/mean": -0.0481201171875,
1029
+ "rewards/reward_length/std": 0.007935813744552433,
1030
+ "rewards/reward_semantic/mean": 0.4186458334326744,
1031
+ "rewards/reward_semantic/std": 0.06994377076625824,
1032
+ "step": 660,
1033
+ "step_time": 1.5930430018996957
1034
+ },
1035
+ {
1036
+ "clip_ratio/high_max": 0.0,
1037
+ "clip_ratio/high_mean": 0.0,
1038
+ "clip_ratio/low_mean": 0.0,
1039
+ "clip_ratio/low_min": 0.0,
1040
+ "clip_ratio/region_mean": 0.0,
1041
+ "completions/clipped_ratio": 0.0,
1042
+ "completions/max_length": 5.95,
1043
+ "completions/max_terminated_length": 5.95,
1044
+ "completions/mean_length": 5.04375,
1045
+ "completions/mean_terminated_length": 5.04375,
1046
+ "completions/min_length": 3.95,
1047
+ "completions/min_terminated_length": 3.95,
1048
+ "entropy": 0.4285797670326247,
1049
+ "epoch": 0.1115485564304462,
1050
+ "frac_reward_zero_std": 0.25,
1051
+ "grad_norm": 8.1875,
1052
+ "learning_rate": 4.6605000000000005e-06,
1053
+ "loss": -0.04365907907485962,
1054
+ "num_tokens": 202619.0,
1055
+ "reward": 1.0846716284751892,
1056
+ "reward_std": 0.020432764168072028,
1057
+ "rewards/reward_grounded/mean": 1.0,
1058
+ "rewards/reward_grounded/std": 0.0,
1059
+ "rewards/reward_length/mean": -0.0378173828125,
1060
+ "rewards/reward_length/std": 0.01130076235276647,
1061
+ "rewards/reward_semantic/mean": 0.28854166716337204,
1062
+ "rewards/reward_semantic/std": 0.06747942492365837,
1063
+ "step": 680,
1064
+ "step_time": 1.6218179324499942
1065
+ },
1066
+ {
1067
+ "clip_ratio/high_max": 0.0,
1068
+ "clip_ratio/high_mean": 0.0,
1069
+ "clip_ratio/low_mean": 0.0,
1070
+ "clip_ratio/low_min": 0.0,
1071
+ "clip_ratio/region_mean": 0.0,
1072
+ "completions/clipped_ratio": 0.0,
1073
+ "completions/max_length": 5.15,
1074
+ "completions/max_terminated_length": 5.15,
1075
+ "completions/mean_length": 4.6375,
1076
+ "completions/mean_terminated_length": 4.6375,
1077
+ "completions/min_length": 4.3,
1078
+ "completions/min_terminated_length": 4.3,
1079
+ "entropy": 0.15305614546541618,
1080
+ "epoch": 0.11482939632545931,
1081
+ "frac_reward_zero_std": 0.7,
1082
+ "grad_norm": 6.15625,
1083
+ "learning_rate": 4.650500000000001e-06,
1084
+ "loss": -0.022796785831451415,
1085
+ "num_tokens": 208433.0,
1086
+ "reward": 1.115874022245407,
1087
+ "reward_std": 0.013541911324136891,
1088
+ "rewards/reward_grounded/mean": 1.0,
1089
+ "rewards/reward_grounded/std": 0.0,
1090
+ "rewards/reward_length/mean": -0.03251953125,
1091
+ "rewards/reward_length/std": 0.0023872115067206322,
1092
+ "rewards/reward_semantic/mean": 0.391666666790843,
1093
+ "rewards/reward_semantic/std": 0.045165380090475084,
1094
+ "step": 700,
1095
+ "step_time": 1.5669768184995518
1096
+ },
1097
+ {
1098
+ "clip_ratio/high_max": 0.0,
1099
+ "clip_ratio/high_mean": 0.0,
1100
+ "clip_ratio/low_mean": 0.0,
1101
+ "clip_ratio/low_min": 0.0,
1102
+ "clip_ratio/region_mean": 0.0,
1103
+ "completions/clipped_ratio": 0.0,
1104
+ "completions/max_length": 5.6,
1105
+ "completions/max_terminated_length": 5.6,
1106
+ "completions/mean_length": 4.7125,
1107
+ "completions/mean_terminated_length": 4.7125,
1108
+ "completions/min_length": 3.95,
1109
+ "completions/min_terminated_length": 3.95,
1110
+ "entropy": 0.3805655787051364,
1111
+ "epoch": 0.11811023622047244,
1112
+ "frac_reward_zero_std": 0.4,
1113
+ "grad_norm": 0.0,
1114
+ "learning_rate": 4.640500000000001e-06,
1115
+ "loss": 0.004567959904670715,
1116
+ "num_tokens": 214115.0,
1117
+ "reward": 1.0923460215330123,
1118
+ "reward_std": 0.030167541204718873,
1119
+ "rewards/reward_grounded/mean": 1.0,
1120
+ "rewards/reward_grounded/std": 0.0,
1121
+ "rewards/reward_length/mean": -0.0387939453125,
1122
+ "rewards/reward_length/std": 0.00844412127044052,
1123
+ "rewards/reward_semantic/mean": 0.3142857149243355,
1124
+ "rewards/reward_semantic/std": 0.10010053664445877,
1125
+ "step": 720,
1126
+ "step_time": 1.6804628341502394
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
+ "completions/clipped_ratio": 0.0,
1135
+ "completions/max_length": 5.4,
1136
+ "completions/max_terminated_length": 5.4,
1137
+ "completions/mean_length": 4.93125,
1138
+ "completions/mean_terminated_length": 4.93125,
1139
+ "completions/min_length": 4.5,
1140
+ "completions/min_terminated_length": 4.5,
1141
+ "entropy": 0.25949509698621115,
1142
+ "epoch": 0.12139107611548557,
1143
+ "frac_reward_zero_std": 0.55,
1144
+ "grad_norm": 21.125,
1145
+ "learning_rate": 4.630500000000001e-06,
1146
+ "loss": 0.022067296504974365,
1147
+ "num_tokens": 220184.0,
1148
+ "reward": 1.1069529592990874,
1149
+ "reward_std": 0.022339693034882657,
1150
+ "rewards/reward_grounded/mean": 1.0,
1151
+ "rewards/reward_grounded/std": 0.0,
1152
+ "rewards/reward_length/mean": -0.0404052734375,
1153
+ "rewards/reward_length/std": 0.006306704942835495,
1154
+ "rewards/reward_semantic/mean": 0.3632440477609634,
1155
+ "rewards/reward_semantic/std": 0.07467387840151787,
1156
+ "step": 740,
1157
+ "step_time": 1.6020566637997036
1158
+ },
1159
+ {
1160
+ "clip_ratio/high_max": 0.0,
1161
+ "clip_ratio/high_mean": 0.0,
1162
+ "clip_ratio/low_mean": 0.0,
1163
+ "clip_ratio/low_min": 0.0,
1164
+ "clip_ratio/region_mean": 0.0,
1165
+ "completions/clipped_ratio": 0.0,
1166
+ "completions/max_length": 5.75,
1167
+ "completions/max_terminated_length": 5.75,
1168
+ "completions/mean_length": 5.2375,
1169
+ "completions/mean_terminated_length": 5.2375,
1170
+ "completions/min_length": 4.85,
1171
+ "completions/min_terminated_length": 4.85,
1172
+ "entropy": 0.13271254739520372,
1173
+ "epoch": 0.12467191601049869,
1174
+ "frac_reward_zero_std": 0.65,
1175
+ "grad_norm": 0.0,
1176
+ "learning_rate": 4.620500000000001e-06,
1177
+ "loss": -0.011591804772615432,
1178
+ "num_tokens": 225926.0,
1179
+ "reward": 1.1052490204572678,
1180
+ "reward_std": 0.012424617203214439,
1181
+ "rewards/reward_grounded/mean": 1.0,
1182
+ "rewards/reward_grounded/std": 0.0,
1183
+ "rewards/reward_length/mean": -0.03251953125,
1184
+ "rewards/reward_length/std": 0.0069050868391059336,
1185
+ "rewards/reward_semantic/mean": 0.35625,
1186
+ "rewards/reward_semantic/std": 0.04110970795154571,
1187
+ "step": 760,
1188
+ "step_time": 1.5845169639495906
1189
+ },
1190
+ {
1191
+ "clip_ratio/high_max": 0.0,
1192
+ "clip_ratio/high_mean": 0.0,
1193
+ "clip_ratio/low_mean": 0.0,
1194
+ "clip_ratio/low_min": 0.0,
1195
+ "clip_ratio/region_mean": 0.0,
1196
+ "completions/clipped_ratio": 0.0,
1197
+ "completions/max_length": 5.85,
1198
+ "completions/max_terminated_length": 5.85,
1199
+ "completions/mean_length": 5.29375,
1200
+ "completions/mean_terminated_length": 5.29375,
1201
+ "completions/min_length": 4.7,
1202
+ "completions/min_terminated_length": 4.7,
1203
+ "entropy": 0.2635622876246998,
1204
+ "epoch": 0.1279527559055118,
1205
+ "frac_reward_zero_std": 0.6,
1206
+ "grad_norm": 0.0,
1207
+ "learning_rate": 4.6105e-06,
1208
+ "loss": -0.012393039464950562,
1209
+ "num_tokens": 231541.0,
1210
+ "reward": 1.0538557916879654,
1211
+ "reward_std": 0.024872867923113516,
1212
+ "rewards/reward_grounded/mean": 1.0,
1213
+ "rewards/reward_grounded/std": 0.0,
1214
+ "rewards/reward_length/mean": -0.03642578125,
1215
+ "rewards/reward_length/std": 0.006841688393615186,
1216
+ "rewards/reward_semantic/mean": 0.1855902776122093,
1217
+ "rewards/reward_semantic/std": 0.08292527124285698,
1218
+ "step": 780,
1219
+ "step_time": 1.5691503484998066
1220
+ },
1221
+ {
1222
+ "clip_ratio/high_max": 0.0,
1223
+ "clip_ratio/high_mean": 0.0,
1224
+ "clip_ratio/low_mean": 0.0,
1225
+ "clip_ratio/low_min": 0.0,
1226
+ "clip_ratio/region_mean": 0.0,
1227
+ "completions/clipped_ratio": 0.0,
1228
+ "completions/max_length": 5.65,
1229
+ "completions/max_terminated_length": 5.65,
1230
+ "completions/mean_length": 5.26875,
1231
+ "completions/mean_terminated_length": 5.26875,
1232
+ "completions/min_length": 4.95,
1233
+ "completions/min_terminated_length": 4.95,
1234
+ "entropy": 0.1539801352791983,
1235
+ "epoch": 0.13123359580052493,
1236
+ "frac_reward_zero_std": 0.8,
1237
+ "grad_norm": 0.0,
1238
+ "learning_rate": 4.6005e-06,
1239
+ "loss": 0.02297772765159607,
1240
+ "num_tokens": 237712.0,
1241
+ "reward": 1.0723930418491363,
1242
+ "reward_std": 0.006485555377730634,
1243
+ "rewards/reward_grounded/mean": 1.0,
1244
+ "rewards/reward_grounded/std": 0.0,
1245
+ "rewards/reward_length/mean": -0.033984375,
1246
+ "rewards/reward_length/std": 0.004696746869012713,
1247
+ "rewards/reward_semantic/mean": 0.2469742067158222,
1248
+ "rewards/reward_semantic/std": 0.021238823607563972,
1249
+ "step": 800,
1250
+ "step_time": 1.5607441842005756
1251
+ },
1252
+ {
1253
+ "clip_ratio/high_max": 0.0,
1254
+ "clip_ratio/high_mean": 0.0,
1255
+ "clip_ratio/low_mean": 0.0,
1256
+ "clip_ratio/low_min": 0.0,
1257
+ "clip_ratio/region_mean": 0.0,
1258
+ "completions/clipped_ratio": 0.0,
1259
+ "completions/max_length": 5.85,
1260
+ "completions/max_terminated_length": 5.85,
1261
+ "completions/mean_length": 5.025,
1262
+ "completions/mean_terminated_length": 5.025,
1263
+ "completions/min_length": 4.5,
1264
+ "completions/min_terminated_length": 4.5,
1265
+ "entropy": 0.22530270127426774,
1266
+ "epoch": 0.13451443569553806,
1267
+ "frac_reward_zero_std": 0.5,
1268
+ "grad_norm": 0.0,
1269
+ "learning_rate": 4.5905e-06,
1270
+ "loss": -0.00012428686022758484,
1271
+ "num_tokens": 243900.0,
1272
+ "reward": 1.0480462223291398,
1273
+ "reward_std": 0.014764534048299537,
1274
+ "rewards/reward_grounded/mean": 1.0,
1275
+ "rewards/reward_grounded/std": 0.0,
1276
+ "rewards/reward_length/mean": -0.0357421875,
1277
+ "rewards/reward_length/std": 0.010764729115180671,
1278
+ "rewards/reward_semantic/mean": 0.16611111108213664,
1279
+ "rewards/reward_semantic/std": 0.049135804176330566,
1280
+ "step": 820,
1281
+ "step_time": 1.6110364599507192
1282
+ },
1283
+ {
1284
+ "clip_ratio/high_max": 0.0,
1285
+ "clip_ratio/high_mean": 0.0,
1286
+ "clip_ratio/low_mean": 0.0,
1287
+ "clip_ratio/low_min": 0.0,
1288
+ "clip_ratio/region_mean": 0.0,
1289
+ "completions/clipped_ratio": 0.0,
1290
+ "completions/max_length": 6.2,
1291
+ "completions/max_terminated_length": 6.2,
1292
+ "completions/mean_length": 5.25625,
1293
+ "completions/mean_terminated_length": 5.25625,
1294
+ "completions/min_length": 4.8,
1295
+ "completions/min_terminated_length": 4.8,
1296
+ "entropy": 0.31118827113505176,
1297
+ "epoch": 0.1377952755905512,
1298
+ "frac_reward_zero_std": 0.5,
1299
+ "grad_norm": 0.0,
1300
+ "learning_rate": 4.5805000000000004e-06,
1301
+ "loss": 0.01610923409461975,
1302
+ "num_tokens": 250069.0,
1303
+ "reward": 1.0784188359975815,
1304
+ "reward_std": 0.02915856411345885,
1305
+ "rewards/reward_grounded/mean": 1.0,
1306
+ "rewards/reward_grounded/std": 0.0,
1307
+ "rewards/reward_length/mean": -0.0343017578125,
1308
+ "rewards/reward_length/std": 0.008795349719002843,
1309
+ "rewards/reward_semantic/mean": 0.26711309850215914,
1310
+ "rewards/reward_semantic/std": 0.09740542992949486,
1311
+ "step": 840,
1312
+ "step_time": 1.6490229548499884
1313
+ },
1314
+ {
1315
+ "clip_ratio/high_max": 0.0,
1316
+ "clip_ratio/high_mean": 0.0,
1317
+ "clip_ratio/low_mean": 0.0,
1318
+ "clip_ratio/low_min": 0.0,
1319
+ "clip_ratio/region_mean": 0.0,
1320
+ "completions/clipped_ratio": 0.0,
1321
+ "completions/max_length": 6.25,
1322
+ "completions/max_terminated_length": 6.25,
1323
+ "completions/mean_length": 5.48125,
1324
+ "completions/mean_terminated_length": 5.48125,
1325
+ "completions/min_length": 4.8,
1326
+ "completions/min_terminated_length": 4.8,
1327
+ "entropy": 0.19009512941393042,
1328
+ "epoch": 0.1410761154855643,
1329
+ "frac_reward_zero_std": 0.6,
1330
+ "grad_norm": 0.0,
1331
+ "learning_rate": 4.5705000000000005e-06,
1332
+ "loss": -0.048175308108329776,
1333
+ "num_tokens": 256186.0,
1334
+ "reward": 1.1095454066991806,
1335
+ "reward_std": 0.03285904193180613,
1336
+ "rewards/reward_grounded/mean": 1.0,
1337
+ "rewards/reward_grounded/std": 0.0,
1338
+ "rewards/reward_length/mean": -0.0268798828125,
1339
+ "rewards/reward_length/std": 0.009483330580405892,
1340
+ "rewards/reward_semantic/mean": 0.36963141039013864,
1341
+ "rewards/reward_semantic/std": 0.11041291058063507,
1342
+ "step": 860,
1343
+ "step_time": 1.84969353199931
1344
+ },
1345
+ {
1346
+ "clip_ratio/high_max": 0.0,
1347
+ "clip_ratio/high_mean": 0.0,
1348
+ "clip_ratio/low_mean": 0.0,
1349
+ "clip_ratio/low_min": 0.0,
1350
+ "clip_ratio/region_mean": 0.0,
1351
+ "completions/clipped_ratio": 0.0,
1352
+ "completions/max_length": 5.7,
1353
+ "completions/max_terminated_length": 5.7,
1354
+ "completions/mean_length": 4.79375,
1355
+ "completions/mean_terminated_length": 4.79375,
1356
+ "completions/min_length": 4.4,
1357
+ "completions/min_terminated_length": 4.4,
1358
+ "entropy": 0.21118192300355076,
1359
+ "epoch": 0.14435695538057744,
1360
+ "frac_reward_zero_std": 0.6,
1361
+ "grad_norm": 0.0,
1362
+ "learning_rate": 4.560500000000001e-06,
1363
+ "loss": 0.050315862894058226,
1364
+ "num_tokens": 262425.0,
1365
+ "reward": 1.0997518181800843,
1366
+ "reward_std": 0.022761242519482038,
1367
+ "rewards/reward_grounded/mean": 1.0,
1368
+ "rewards/reward_grounded/std": 0.0,
1369
+ "rewards/reward_length/mean": -0.031689453125,
1370
+ "rewards/reward_length/std": 0.00647321484866552,
1371
+ "rewards/reward_semantic/mean": 0.3377877000719309,
1372
+ "rewards/reward_semantic/std": 0.07582005113363266,
1373
+ "step": 880,
1374
+ "step_time": 1.6589296972999363
1375
+ },
1376
+ {
1377
+ "clip_ratio/high_max": 0.0,
1378
+ "clip_ratio/high_mean": 0.0,
1379
+ "clip_ratio/low_mean": 0.0,
1380
+ "clip_ratio/low_min": 0.0,
1381
+ "clip_ratio/region_mean": 0.0,
1382
+ "completions/clipped_ratio": 0.0,
1383
+ "completions/max_length": 5.8,
1384
+ "completions/max_terminated_length": 5.8,
1385
+ "completions/mean_length": 4.91875,
1386
+ "completions/mean_terminated_length": 4.91875,
1387
+ "completions/min_length": 4.15,
1388
+ "completions/min_terminated_length": 4.15,
1389
+ "entropy": 0.3470641552201414,
1390
+ "epoch": 0.14763779527559054,
1391
+ "frac_reward_zero_std": 0.5,
1392
+ "grad_norm": 0.0,
1393
+ "learning_rate": 4.550500000000001e-06,
1394
+ "loss": 0.012325967103242874,
1395
+ "num_tokens": 268468.0,
1396
+ "reward": 1.0776826739311218,
1397
+ "reward_std": 0.018389373533136678,
1398
+ "rewards/reward_grounded/mean": 1.0,
1399
+ "rewards/reward_grounded/std": 0.0,
1400
+ "rewards/reward_length/mean": -0.0361083984375,
1401
+ "rewards/reward_length/std": 0.010244967485778033,
1402
+ "rewards/reward_semantic/mean": 0.26496031805872916,
1403
+ "rewards/reward_semantic/std": 0.060299596190452574,
1404
+ "step": 900,
1405
+ "step_time": 1.6155126945997835
1406
+ },
1407
+ {
1408
+ "clip_ratio/high_max": 0.0,
1409
+ "clip_ratio/high_mean": 0.0,
1410
+ "clip_ratio/low_mean": 0.0,
1411
+ "clip_ratio/low_min": 0.0,
1412
+ "clip_ratio/region_mean": 0.0,
1413
+ "completions/clipped_ratio": 0.0,
1414
+ "completions/max_length": 6.0,
1415
+ "completions/max_terminated_length": 6.0,
1416
+ "completions/mean_length": 5.2,
1417
+ "completions/mean_terminated_length": 5.2,
1418
+ "completions/min_length": 4.45,
1419
+ "completions/min_terminated_length": 4.45,
1420
+ "entropy": 0.2223609389172907,
1421
+ "epoch": 0.15091863517060367,
1422
+ "frac_reward_zero_std": 0.5,
1423
+ "grad_norm": 29.5,
1424
+ "learning_rate": 4.540500000000001e-06,
1425
+ "loss": 0.008905045688152313,
1426
+ "num_tokens": 274548.0,
1427
+ "reward": 1.0319956064224243,
1428
+ "reward_std": 0.01447186817167676,
1429
+ "rewards/reward_grounded/mean": 1.0,
1430
+ "rewards/reward_grounded/std": 0.0,
1431
+ "rewards/reward_length/mean": -0.027587890625,
1432
+ "rewards/reward_length/std": 0.004147578129777685,
1433
+ "rewards/reward_semantic/mean": 0.1112500011920929,
1434
+ "rewards/reward_semantic/std": 0.04789424017071724,
1435
+ "step": 920,
1436
+ "step_time": 1.6041553608993127
1437
+ },
1438
+ {
1439
+ "clip_ratio/high_max": 0.0,
1440
+ "clip_ratio/high_mean": 0.0,
1441
+ "clip_ratio/low_mean": 0.0,
1442
+ "clip_ratio/low_min": 0.0,
1443
+ "clip_ratio/region_mean": 0.0,
1444
+ "completions/clipped_ratio": 0.0,
1445
+ "completions/max_length": 6.25,
1446
+ "completions/max_terminated_length": 6.25,
1447
+ "completions/mean_length": 5.5,
1448
+ "completions/mean_terminated_length": 5.5,
1449
+ "completions/min_length": 4.8,
1450
+ "completions/min_terminated_length": 4.8,
1451
+ "entropy": 0.15990385009681632,
1452
+ "epoch": 0.1541994750656168,
1453
+ "frac_reward_zero_std": 0.65,
1454
+ "grad_norm": 25.75,
1455
+ "learning_rate": 4.5305e-06,
1456
+ "loss": 0.0012000683695077897,
1457
+ "num_tokens": 280548.0,
1458
+ "reward": 1.0634800434112548,
1459
+ "reward_std": 0.01882710763893556,
1460
+ "rewards/reward_grounded/mean": 1.0,
1461
+ "rewards/reward_grounded/std": 0.0,
1462
+ "rewards/reward_length/mean": -0.02861328125,
1463
+ "rewards/reward_length/std": 0.0077044784789904956,
1464
+ "rewards/reward_semantic/mean": 0.21636904776096344,
1465
+ "rewards/reward_semantic/std": 0.06269824169576169,
1466
+ "step": 940,
1467
+ "step_time": 1.591113177501029
1468
+ },
1469
+ {
1470
+ "clip_ratio/high_max": 0.0,
1471
+ "clip_ratio/high_mean": 0.0,
1472
+ "clip_ratio/low_mean": 0.0,
1473
+ "clip_ratio/low_min": 0.0,
1474
+ "clip_ratio/region_mean": 0.0,
1475
+ "completions/clipped_ratio": 0.0,
1476
+ "completions/max_length": 6.0,
1477
+ "completions/max_terminated_length": 6.0,
1478
+ "completions/mean_length": 5.48125,
1479
+ "completions/mean_terminated_length": 5.48125,
1480
+ "completions/min_length": 5.0,
1481
+ "completions/min_terminated_length": 5.0,
1482
+ "entropy": 0.1008574430443332,
1483
+ "epoch": 0.15748031496062992,
1484
+ "frac_reward_zero_std": 0.7,
1485
+ "grad_norm": 0.0,
1486
+ "learning_rate": 4.5205e-06,
1487
+ "loss": 0.0621263861656189,
1488
+ "num_tokens": 286377.0,
1489
+ "reward": 1.1116095036268234,
1490
+ "reward_std": 0.022756787104299292,
1491
+ "rewards/reward_grounded/mean": 1.0,
1492
+ "rewards/reward_grounded/std": 0.0,
1493
+ "rewards/reward_length/mean": -0.02861328125,
1494
+ "rewards/reward_length/std": 0.004306197509868071,
1495
+ "rewards/reward_semantic/mean": 0.3768005946651101,
1496
+ "rewards/reward_semantic/std": 0.07597103714942932,
1497
+ "step": 960,
1498
+ "step_time": 1.6255918225498136
1499
+ },
1500
+ {
1501
+ "clip_ratio/high_max": 0.0,
1502
+ "clip_ratio/high_mean": 0.0,
1503
+ "clip_ratio/low_mean": 0.0,
1504
+ "clip_ratio/low_min": 0.0,
1505
+ "clip_ratio/region_mean": 0.0,
1506
+ "completions/clipped_ratio": 0.0,
1507
+ "completions/max_length": 5.7,
1508
+ "completions/max_terminated_length": 5.7,
1509
+ "completions/mean_length": 5.4625,
1510
+ "completions/mean_terminated_length": 5.4625,
1511
+ "completions/min_length": 4.9,
1512
+ "completions/min_terminated_length": 4.9,
1513
+ "entropy": 0.12628535750282027,
1514
+ "epoch": 0.16076115485564305,
1515
+ "frac_reward_zero_std": 0.65,
1516
+ "grad_norm": 0.0,
1517
+ "learning_rate": 4.5105e-06,
1518
+ "loss": 0.0023237094283103943,
1519
+ "num_tokens": 292251.0,
1520
+ "reward": 1.0658237963914872,
1521
+ "reward_std": 0.012343478866387158,
1522
+ "rewards/reward_grounded/mean": 1.0,
1523
+ "rewards/reward_grounded/std": 0.0,
1524
+ "rewards/reward_length/mean": -0.03173828125,
1525
+ "rewards/reward_length/std": 0.007525579165667295,
1526
+ "rewards/reward_semantic/mean": 0.2247023830190301,
1527
+ "rewards/reward_semantic/std": 0.040503402799367906,
1528
+ "step": 980,
1529
+ "step_time": 1.583995696999318
1530
+ },
1531
+ {
1532
+ "clip_ratio/high_max": 0.0,
1533
+ "clip_ratio/high_mean": 0.0,
1534
+ "clip_ratio/low_mean": 0.0,
1535
+ "clip_ratio/low_min": 0.0,
1536
+ "clip_ratio/region_mean": 0.0,
1537
+ "completions/clipped_ratio": 0.0,
1538
+ "completions/max_length": 5.55,
1539
+ "completions/max_terminated_length": 5.55,
1540
+ "completions/mean_length": 5.24375,
1541
+ "completions/mean_terminated_length": 5.24375,
1542
+ "completions/min_length": 4.85,
1543
+ "completions/min_terminated_length": 4.85,
1544
+ "entropy": 0.09762425903263647,
1545
+ "epoch": 0.16404199475065617,
1546
+ "frac_reward_zero_std": 0.7,
1547
+ "grad_norm": 17.375,
1548
+ "learning_rate": 4.5005e-06,
1549
+ "loss": -0.0012902652844786644,
1550
+ "num_tokens": 297890.0,
1551
+ "reward": 1.0763329923152924,
1552
+ "reward_std": 0.013358409999636934,
1553
+ "rewards/reward_grounded/mean": 1.0,
1554
+ "rewards/reward_grounded/std": 0.0,
1555
+ "rewards/reward_length/mean": -0.02958984375,
1556
+ "rewards/reward_length/std": 0.006560613529291004,
1557
+ "rewards/reward_semantic/mean": 0.259375,
1558
+ "rewards/reward_semantic/std": 0.04419417306780815,
1559
+ "step": 1000,
1560
+ "step_time": 1.56278663364983
1561
+ }
1562
+ ],
1563
+ "logging_steps": 20,
1564
+ "max_steps": 10000,
1565
+ "num_input_tokens_seen": 297890,
1566
+ "num_train_epochs": 2,
1567
+ "save_steps": 500,
1568
+ "stateful_callbacks": {
1569
+ "TrainerControl": {
1570
+ "args": {
1571
+ "should_epoch_stop": false,
1572
+ "should_evaluate": false,
1573
+ "should_log": false,
1574
+ "should_save": true,
1575
+ "should_training_stop": false
1576
+ },
1577
+ "attributes": {}
1578
+ }
1579
+ },
1580
+ "total_flos": 0.0,
1581
+ "train_batch_size": 1,
1582
+ "trial_name": null,
1583
+ "trial_params": null
1584
+ }
grpo/checkpoint-1000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:200955a61f7b161c5c840b30b5e94f7cdbc4ef6fd71e4ae199aac51700a74d8c
3
+ size 7249
grpo/checkpoint-500/adapter_config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "Qwen/Qwen3-8B",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 32,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 16,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "o_proj"
37
+ ],
38
+ "target_parameters": null,
39
+ "task_type": "CAUSAL_LM",
40
+ "trainable_token_indices": null,
41
+ "use_bdlora": null,
42
+ "use_dora": false,
43
+ "use_qalora": false,
44
+ "use_rslora": false
45
+ }
grpo/checkpoint-500/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c495b6f773c28ea7b3db53201bf8e483e052f1d92170c2f8d33f1c5c00402099
3
+ size 30709768
grpo/checkpoint-500/chat_template.jinja ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if message.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant\n' }}
86
+ {%- if enable_thinking is defined and enable_thinking is false %}
87
+ {{- '<think>\n\n</think>\n\n' }}
88
+ {%- endif %}
89
+ {%- endif %}
grpo/checkpoint-500/generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "temperature": 0.6,
9
+ "top_k": 20,
10
+ "top_p": 0.95,
11
+ "transformers_version": "5.6.2"
12
+ }
grpo/checkpoint-500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b1ceb19343f5364b22a706ecea8a1dc051df0b10328b411583336a69f4b90460
3
+ size 61588683
grpo/checkpoint-500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1473cfce6c23f70272186c04464bb84276cd9e0e268fdcc23fdcee01911ba6dd
3
+ size 14645
grpo/checkpoint-500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c0d03881872f5a83f8f0fb959642a078dd9a1c90856f4def3d6354832b319893
3
+ size 1465
grpo/checkpoint-500/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
3
+ size 11422650
grpo/checkpoint-500/tokenizer_config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "local_files_only": false,
25
+ "max_length": 512,
26
+ "model_max_length": 131072,
27
+ "pad_token": "<|endoftext|>",
28
+ "split_special_tokens": false,
29
+ "stride": 0,
30
+ "tokenizer_class": "Qwen2Tokenizer",
31
+ "truncation_side": "right",
32
+ "truncation_strategy": "longest_first",
33
+ "unk_token": null
34
+ }
grpo/checkpoint-500/trainer_state.json ADDED
@@ -0,0 +1,809 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.08202099737532809,
6
+ "eval_steps": 500,
7
+ "global_step": 500,
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
+ "completions/clipped_ratio": 0.0,
19
+ "completions/max_length": 6.05,
20
+ "completions/max_terminated_length": 6.05,
21
+ "completions/mean_length": 4.61875,
22
+ "completions/mean_terminated_length": 4.61875,
23
+ "completions/min_length": 3.5,
24
+ "completions/min_terminated_length": 3.5,
25
+ "entropy": 0.8203600860819279,
26
+ "epoch": 0.0032808398950131233,
27
+ "frac_reward_zero_std": 0.3,
28
+ "grad_norm": 5.8125,
29
+ "learning_rate": 4.990500000000001e-06,
30
+ "loss": 0.06519373655319213,
31
+ "num_tokens": 5635.0,
32
+ "reward": 1.1270193845033645,
33
+ "reward_std": 0.04764379555272171,
34
+ "rewards/reward_grounded/mean": 1.0,
35
+ "rewards/reward_grounded/std": 0.0,
36
+ "rewards/reward_length/mean": -0.050634765625,
37
+ "rewards/reward_length/std": 0.015013676788657903,
38
+ "rewards/reward_semantic/mean": 0.43183712139725683,
39
+ "rewards/reward_semantic/std": 0.15772892236709596,
40
+ "step": 20,
41
+ "step_time": 1.6850911805999202
42
+ },
43
+ {
44
+ "clip_ratio/high_max": 0.0,
45
+ "clip_ratio/high_mean": 0.0,
46
+ "clip_ratio/low_mean": 0.0,
47
+ "clip_ratio/low_min": 0.0,
48
+ "clip_ratio/region_mean": 0.0,
49
+ "completions/clipped_ratio": 0.0,
50
+ "completions/max_length": 7.15,
51
+ "completions/max_terminated_length": 7.15,
52
+ "completions/mean_length": 5.00625,
53
+ "completions/mean_terminated_length": 5.00625,
54
+ "completions/min_length": 3.65,
55
+ "completions/min_terminated_length": 3.65,
56
+ "entropy": 0.929626553109847,
57
+ "epoch": 0.006561679790026247,
58
+ "frac_reward_zero_std": 0.2,
59
+ "grad_norm": 0.0,
60
+ "learning_rate": 4.9805e-06,
61
+ "loss": 0.07899624109268188,
62
+ "num_tokens": 11660.0,
63
+ "reward": 1.0556338340044022,
64
+ "reward_std": 0.017533331926097163,
65
+ "rewards/reward_grounded/mean": 1.0,
66
+ "rewards/reward_grounded/std": 0.0,
67
+ "rewards/reward_length/mean": -0.0642578125,
68
+ "rewards/reward_length/std": 0.013992739858804271,
69
+ "rewards/reward_semantic/mean": 0.19615575280040504,
70
+ "rewards/reward_semantic/std": 0.05699479430913925,
71
+ "step": 40,
72
+ "step_time": 1.79874156090018
73
+ },
74
+ {
75
+ "clip_ratio/high_max": 0.0,
76
+ "clip_ratio/high_mean": 0.0,
77
+ "clip_ratio/low_mean": 0.0,
78
+ "clip_ratio/low_min": 0.0,
79
+ "clip_ratio/region_mean": 0.0,
80
+ "completions/clipped_ratio": 0.0,
81
+ "completions/max_length": 5.5,
82
+ "completions/max_terminated_length": 5.5,
83
+ "completions/mean_length": 4.325,
84
+ "completions/mean_terminated_length": 4.325,
85
+ "completions/min_length": 3.5,
86
+ "completions/min_terminated_length": 3.5,
87
+ "entropy": 0.7027173407775991,
88
+ "epoch": 0.00984251968503937,
89
+ "frac_reward_zero_std": 0.3,
90
+ "grad_norm": 0.0,
91
+ "learning_rate": 4.9705e-06,
92
+ "loss": 0.023067036271095277,
93
+ "num_tokens": 17440.0,
94
+ "reward": 1.1290264517068862,
95
+ "reward_std": 0.03574381650541909,
96
+ "rewards/reward_grounded/mean": 1.0,
97
+ "rewards/reward_grounded/std": 0.0,
98
+ "rewards/reward_length/mean": -0.052685546875,
99
+ "rewards/reward_length/std": 0.012657346948981286,
100
+ "rewards/reward_semantic/mean": 0.4388690486550331,
101
+ "rewards/reward_semantic/std": 0.11865786015987397,
102
+ "step": 60,
103
+ "step_time": 1.6641022313501708
104
+ },
105
+ {
106
+ "clip_ratio/high_max": 0.0,
107
+ "clip_ratio/high_mean": 0.0,
108
+ "clip_ratio/low_mean": 0.0,
109
+ "clip_ratio/low_min": 0.0,
110
+ "clip_ratio/region_mean": 0.0,
111
+ "completions/clipped_ratio": 0.0,
112
+ "completions/max_length": 7.05,
113
+ "completions/max_terminated_length": 7.05,
114
+ "completions/mean_length": 4.89375,
115
+ "completions/mean_terminated_length": 4.89375,
116
+ "completions/min_length": 3.45,
117
+ "completions/min_terminated_length": 3.45,
118
+ "entropy": 0.8099277728608285,
119
+ "epoch": 0.013123359580052493,
120
+ "frac_reward_zero_std": 0.2,
121
+ "grad_norm": 2.578125,
122
+ "learning_rate": 4.9605e-06,
123
+ "loss": 0.12280056476593018,
124
+ "num_tokens": 23631.0,
125
+ "reward": 1.08654942214489,
126
+ "reward_std": 0.040316034654097166,
127
+ "rewards/reward_grounded/mean": 1.0,
128
+ "rewards/reward_grounded/std": 0.0,
129
+ "rewards/reward_length/mean": -0.06103515625,
130
+ "rewards/reward_length/std": 0.022364351549185812,
131
+ "rewards/reward_semantic/mean": 0.2986706353724003,
132
+ "rewards/reward_semantic/std": 0.1318330906331539,
133
+ "step": 80,
134
+ "step_time": 1.7635547111498453
135
+ },
136
+ {
137
+ "clip_ratio/high_max": 0.0,
138
+ "clip_ratio/high_mean": 0.0,
139
+ "clip_ratio/low_mean": 0.0,
140
+ "clip_ratio/low_min": 0.0,
141
+ "clip_ratio/region_mean": 0.0,
142
+ "completions/clipped_ratio": 0.0,
143
+ "completions/max_length": 6.75,
144
+ "completions/max_terminated_length": 6.75,
145
+ "completions/mean_length": 4.75,
146
+ "completions/mean_terminated_length": 4.75,
147
+ "completions/min_length": 3.4,
148
+ "completions/min_terminated_length": 3.4,
149
+ "entropy": 0.8052310052706162,
150
+ "epoch": 0.016404199475065617,
151
+ "frac_reward_zero_std": 0.2,
152
+ "grad_norm": 3.171875,
153
+ "learning_rate": 4.9505e-06,
154
+ "loss": 0.007156229019165039,
155
+ "num_tokens": 29599.0,
156
+ "reward": 1.0940562933683395,
157
+ "reward_std": 0.030761140065442304,
158
+ "rewards/reward_grounded/mean": 1.0,
159
+ "rewards/reward_grounded/std": 0.0,
160
+ "rewards/reward_length/mean": -0.06103515625,
161
+ "rewards/reward_length/std": 0.017728388286195697,
162
+ "rewards/reward_semantic/mean": 0.3236935347318649,
163
+ "rewards/reward_semantic/std": 0.1025692880153656,
164
+ "step": 100,
165
+ "step_time": 1.6173622274000992
166
+ },
167
+ {
168
+ "clip_ratio/high_max": 0.0,
169
+ "clip_ratio/high_mean": 0.0,
170
+ "clip_ratio/low_mean": 0.0,
171
+ "clip_ratio/low_min": 0.0,
172
+ "clip_ratio/region_mean": 0.0,
173
+ "completions/clipped_ratio": 0.0,
174
+ "completions/max_length": 5.6,
175
+ "completions/max_terminated_length": 5.6,
176
+ "completions/mean_length": 4.125,
177
+ "completions/mean_terminated_length": 4.125,
178
+ "completions/min_length": 3.15,
179
+ "completions/min_terminated_length": 3.15,
180
+ "entropy": 0.7014230243818019,
181
+ "epoch": 0.01968503937007874,
182
+ "frac_reward_zero_std": 0.3,
183
+ "grad_norm": 0.0,
184
+ "learning_rate": 4.9405e-06,
185
+ "loss": 0.07218956351280212,
186
+ "num_tokens": 35107.0,
187
+ "reward": 1.1007445603609085,
188
+ "reward_std": 0.028740251205454113,
189
+ "rewards/reward_grounded/mean": 1.0,
190
+ "rewards/reward_grounded/std": 0.0,
191
+ "rewards/reward_length/mean": -0.04814453125,
192
+ "rewards/reward_length/std": 0.010129839461296797,
193
+ "rewards/reward_semantic/mean": 0.34383928664028646,
194
+ "rewards/reward_semantic/std": 0.09484268203377724,
195
+ "step": 120,
196
+ "step_time": 1.5441937706006683
197
+ },
198
+ {
199
+ "clip_ratio/high_max": 0.0,
200
+ "clip_ratio/high_mean": 0.0,
201
+ "clip_ratio/low_mean": 0.0,
202
+ "clip_ratio/low_min": 0.0,
203
+ "clip_ratio/region_mean": 0.0,
204
+ "completions/clipped_ratio": 0.0,
205
+ "completions/max_length": 7.25,
206
+ "completions/max_terminated_length": 7.25,
207
+ "completions/mean_length": 4.9625,
208
+ "completions/mean_terminated_length": 4.9625,
209
+ "completions/min_length": 3.5,
210
+ "completions/min_terminated_length": 3.5,
211
+ "entropy": 0.7183661407629188,
212
+ "epoch": 0.022965879265091863,
213
+ "frac_reward_zero_std": 0.15,
214
+ "grad_norm": 2.609375,
215
+ "learning_rate": 4.9305000000000005e-06,
216
+ "loss": 0.05585476756095886,
217
+ "num_tokens": 40989.0,
218
+ "reward": 1.1004581451416016,
219
+ "reward_std": 0.03431669266719837,
220
+ "rewards/reward_grounded/mean": 1.0,
221
+ "rewards/reward_grounded/std": 0.0,
222
+ "rewards/reward_length/mean": -0.059765625,
223
+ "rewards/reward_length/std": 0.017409664555452765,
224
+ "rewards/reward_semantic/mean": 0.34482143111526964,
225
+ "rewards/reward_semantic/std": 0.11257617846131325,
226
+ "step": 140,
227
+ "step_time": 1.7615427238992196
228
+ },
229
+ {
230
+ "clip_ratio/high_max": 0.0,
231
+ "clip_ratio/high_mean": 0.0,
232
+ "clip_ratio/low_mean": 0.0,
233
+ "clip_ratio/low_min": 0.0,
234
+ "clip_ratio/region_mean": 0.0,
235
+ "completions/clipped_ratio": 0.0,
236
+ "completions/max_length": 7.05,
237
+ "completions/max_terminated_length": 7.05,
238
+ "completions/mean_length": 4.89375,
239
+ "completions/mean_terminated_length": 4.89375,
240
+ "completions/min_length": 3.5,
241
+ "completions/min_terminated_length": 3.5,
242
+ "entropy": 0.6915914016732131,
243
+ "epoch": 0.026246719160104987,
244
+ "frac_reward_zero_std": 0.1,
245
+ "grad_norm": 2.859375,
246
+ "learning_rate": 4.920500000000001e-06,
247
+ "loss": 0.0472392201423645,
248
+ "num_tokens": 47220.0,
249
+ "reward": 1.0850708901882171,
250
+ "reward_std": 0.06178898175858194,
251
+ "rewards/reward_grounded/mean": 1.0,
252
+ "rewards/reward_grounded/std": 0.0,
253
+ "rewards/reward_length/mean": -0.065576171875,
254
+ "rewards/reward_length/std": 0.01732458183541894,
255
+ "rewards/reward_semantic/mean": 0.2944990089163184,
256
+ "rewards/reward_semantic/std": 0.2051414344459772,
257
+ "step": 160,
258
+ "step_time": 1.7252537605003453
259
+ },
260
+ {
261
+ "clip_ratio/high_max": 0.0,
262
+ "clip_ratio/high_mean": 0.0,
263
+ "clip_ratio/low_mean": 0.0,
264
+ "clip_ratio/low_min": 0.0,
265
+ "clip_ratio/region_mean": 0.0,
266
+ "completions/clipped_ratio": 0.0,
267
+ "completions/max_length": 6.25,
268
+ "completions/max_terminated_length": 6.25,
269
+ "completions/mean_length": 4.55625,
270
+ "completions/mean_terminated_length": 4.55625,
271
+ "completions/min_length": 3.35,
272
+ "completions/min_terminated_length": 3.35,
273
+ "entropy": 0.8429647823562846,
274
+ "epoch": 0.02952755905511811,
275
+ "frac_reward_zero_std": 0.05,
276
+ "grad_norm": 3.078125,
277
+ "learning_rate": 4.910500000000001e-06,
278
+ "loss": 0.08106310367584228,
279
+ "num_tokens": 53053.0,
280
+ "reward": 1.0679614007472993,
281
+ "reward_std": 0.03796071938340902,
282
+ "rewards/reward_grounded/mean": 1.0,
283
+ "rewards/reward_grounded/std": 0.0,
284
+ "rewards/reward_length/mean": -0.0566650390625,
285
+ "rewards/reward_length/std": 0.014682336029363797,
286
+ "rewards/reward_semantic/mean": 0.23598214369267226,
287
+ "rewards/reward_semantic/std": 0.12518299631774427,
288
+ "step": 180,
289
+ "step_time": 1.7508945095991293
290
+ },
291
+ {
292
+ "clip_ratio/high_max": 0.0,
293
+ "clip_ratio/high_mean": 0.0,
294
+ "clip_ratio/low_mean": 0.0,
295
+ "clip_ratio/low_min": 0.0,
296
+ "clip_ratio/region_mean": 0.0,
297
+ "completions/clipped_ratio": 0.0,
298
+ "completions/max_length": 6.2,
299
+ "completions/max_terminated_length": 6.2,
300
+ "completions/mean_length": 4.5375,
301
+ "completions/mean_terminated_length": 4.5375,
302
+ "completions/min_length": 3.75,
303
+ "completions/min_terminated_length": 3.75,
304
+ "entropy": 0.5915892604841246,
305
+ "epoch": 0.03280839895013123,
306
+ "frac_reward_zero_std": 0.35,
307
+ "grad_norm": 3.71875,
308
+ "learning_rate": 4.9005e-06,
309
+ "loss": 0.0888818621635437,
310
+ "num_tokens": 58931.0,
311
+ "reward": 1.120530515909195,
312
+ "reward_std": 0.0405463597693597,
313
+ "rewards/reward_grounded/mean": 1.0,
314
+ "rewards/reward_grounded/std": 0.0,
315
+ "rewards/reward_length/mean": -0.0493896484375,
316
+ "rewards/reward_length/std": 0.008753356069792061,
317
+ "rewards/reward_semantic/mean": 0.4100000007078052,
318
+ "rewards/reward_semantic/std": 0.13480054624378682,
319
+ "step": 200,
320
+ "step_time": 1.713693901949955
321
+ },
322
+ {
323
+ "clip_ratio/high_max": 0.0,
324
+ "clip_ratio/high_mean": 0.0,
325
+ "clip_ratio/low_mean": 0.0,
326
+ "clip_ratio/low_min": 0.0,
327
+ "clip_ratio/region_mean": 0.0,
328
+ "completions/clipped_ratio": 0.0,
329
+ "completions/max_length": 6.45,
330
+ "completions/max_terminated_length": 6.45,
331
+ "completions/mean_length": 5.15,
332
+ "completions/mean_terminated_length": 5.15,
333
+ "completions/min_length": 4.2,
334
+ "completions/min_terminated_length": 4.2,
335
+ "entropy": 0.6395838386631112,
336
+ "epoch": 0.03608923884514436,
337
+ "frac_reward_zero_std": 0.35,
338
+ "grad_norm": 2.625,
339
+ "learning_rate": 4.8905e-06,
340
+ "loss": 0.06417204141616821,
341
+ "num_tokens": 65139.0,
342
+ "reward": 1.1249804645776749,
343
+ "reward_std": 0.027039075889479135,
344
+ "rewards/reward_grounded/mean": 1.0,
345
+ "rewards/reward_grounded/std": 0.0,
346
+ "rewards/reward_length/mean": -0.056640625,
347
+ "rewards/reward_length/std": 0.011077208817005158,
348
+ "rewards/reward_semantic/mean": 0.4260416656732559,
349
+ "rewards/reward_semantic/std": 0.08868714347481728,
350
+ "step": 220,
351
+ "step_time": 1.7278446534002796
352
+ },
353
+ {
354
+ "clip_ratio/high_max": 0.0,
355
+ "clip_ratio/high_mean": 0.0,
356
+ "clip_ratio/low_mean": 0.0,
357
+ "clip_ratio/low_min": 0.0,
358
+ "clip_ratio/region_mean": 0.0,
359
+ "completions/clipped_ratio": 0.0,
360
+ "completions/max_length": 5.4,
361
+ "completions/max_terminated_length": 5.4,
362
+ "completions/mean_length": 4.2625,
363
+ "completions/mean_terminated_length": 4.2625,
364
+ "completions/min_length": 3.0,
365
+ "completions/min_terminated_length": 3.0,
366
+ "entropy": 0.5652526923571713,
367
+ "epoch": 0.03937007874015748,
368
+ "frac_reward_zero_std": 0.35,
369
+ "grad_norm": 8.125,
370
+ "learning_rate": 4.8805e-06,
371
+ "loss": 0.07150443196296692,
372
+ "num_tokens": 71221.0,
373
+ "reward": 1.1466667413711549,
374
+ "reward_std": 0.03061878081643954,
375
+ "rewards/reward_grounded/mean": 1.0,
376
+ "rewards/reward_grounded/std": 0.0,
377
+ "rewards/reward_length/mean": -0.063330078125,
378
+ "rewards/reward_length/std": 0.015704452409408987,
379
+ "rewards/reward_semantic/mean": 0.4994440995156765,
380
+ "rewards/reward_semantic/std": 0.10041465926915408,
381
+ "step": 240,
382
+ "step_time": 1.6952891380005894
383
+ },
384
+ {
385
+ "clip_ratio/high_max": 0.0,
386
+ "clip_ratio/high_mean": 0.0,
387
+ "clip_ratio/low_mean": 0.0,
388
+ "clip_ratio/low_min": 0.0,
389
+ "clip_ratio/region_mean": 0.0,
390
+ "completions/clipped_ratio": 0.0,
391
+ "completions/max_length": 6.3,
392
+ "completions/max_terminated_length": 6.3,
393
+ "completions/mean_length": 4.46875,
394
+ "completions/mean_terminated_length": 4.46875,
395
+ "completions/min_length": 3.35,
396
+ "completions/min_terminated_length": 3.35,
397
+ "entropy": 0.672205779588694,
398
+ "epoch": 0.04265091863517061,
399
+ "frac_reward_zero_std": 0.2,
400
+ "grad_norm": 6.4375,
401
+ "learning_rate": 4.8705e-06,
402
+ "loss": 0.09235309958457946,
403
+ "num_tokens": 77416.0,
404
+ "reward": 1.0690006405115127,
405
+ "reward_std": 0.019543477255501783,
406
+ "rewards/reward_grounded/mean": 1.0,
407
+ "rewards/reward_grounded/std": 0.0,
408
+ "rewards/reward_length/mean": -0.0533203125,
409
+ "rewards/reward_length/std": 0.017377534485422074,
410
+ "rewards/reward_semantic/mean": 0.23888888889923693,
411
+ "rewards/reward_semantic/std": 0.0630335971713066,
412
+ "step": 260,
413
+ "step_time": 1.7545264535998286
414
+ },
415
+ {
416
+ "clip_ratio/high_max": 0.0,
417
+ "clip_ratio/high_mean": 0.0,
418
+ "clip_ratio/low_mean": 0.0,
419
+ "clip_ratio/low_min": 0.0,
420
+ "clip_ratio/region_mean": 0.0,
421
+ "completions/clipped_ratio": 0.0,
422
+ "completions/max_length": 5.3,
423
+ "completions/max_terminated_length": 5.3,
424
+ "completions/mean_length": 4.525,
425
+ "completions/mean_terminated_length": 4.525,
426
+ "completions/min_length": 3.65,
427
+ "completions/min_terminated_length": 3.65,
428
+ "entropy": 0.5580058551760885,
429
+ "epoch": 0.045931758530183726,
430
+ "frac_reward_zero_std": 0.45,
431
+ "grad_norm": 4.0625,
432
+ "learning_rate": 4.8605e-06,
433
+ "loss": 0.02879692018032074,
434
+ "num_tokens": 83332.0,
435
+ "reward": 1.1027960389852525,
436
+ "reward_std": 0.011529068512027152,
437
+ "rewards/reward_grounded/mean": 1.0,
438
+ "rewards/reward_grounded/std": 0.0,
439
+ "rewards/reward_length/mean": -0.0558349609375,
440
+ "rewards/reward_length/std": 0.008870826044585556,
441
+ "rewards/reward_semantic/mean": 0.3519593259319663,
442
+ "rewards/reward_semantic/std": 0.03782847560942173,
443
+ "step": 280,
444
+ "step_time": 1.5900321766996968
445
+ },
446
+ {
447
+ "clip_ratio/high_max": 0.0,
448
+ "clip_ratio/high_mean": 0.0,
449
+ "clip_ratio/low_mean": 0.0,
450
+ "clip_ratio/low_min": 0.0,
451
+ "clip_ratio/region_mean": 0.0,
452
+ "completions/clipped_ratio": 0.0,
453
+ "completions/max_length": 6.35,
454
+ "completions/max_terminated_length": 6.35,
455
+ "completions/mean_length": 4.56875,
456
+ "completions/mean_terminated_length": 4.56875,
457
+ "completions/min_length": 3.2,
458
+ "completions/min_terminated_length": 3.2,
459
+ "entropy": 0.8288409212742408,
460
+ "epoch": 0.04921259842519685,
461
+ "frac_reward_zero_std": 0.2,
462
+ "grad_norm": 0.0,
463
+ "learning_rate": 4.8505e-06,
464
+ "loss": 0.044028478860855105,
465
+ "num_tokens": 89567.0,
466
+ "reward": 1.1026858359575271,
467
+ "reward_std": 0.03983475278655533,
468
+ "rewards/reward_grounded/mean": 1.0,
469
+ "rewards/reward_grounded/std": 0.0,
470
+ "rewards/reward_length/mean": -0.062744140625,
471
+ "rewards/reward_length/std": 0.018717235408257694,
472
+ "rewards/reward_semantic/mean": 0.35274351127445697,
473
+ "rewards/reward_semantic/std": 0.13158041909337043,
474
+ "step": 300,
475
+ "step_time": 1.7004065297001034
476
+ },
477
+ {
478
+ "clip_ratio/high_max": 0.0,
479
+ "clip_ratio/high_mean": 0.0,
480
+ "clip_ratio/low_mean": 0.0,
481
+ "clip_ratio/low_min": 0.0,
482
+ "clip_ratio/region_mean": 0.0,
483
+ "completions/clipped_ratio": 0.0,
484
+ "completions/max_length": 5.65,
485
+ "completions/max_terminated_length": 5.65,
486
+ "completions/mean_length": 4.5875,
487
+ "completions/mean_terminated_length": 4.5875,
488
+ "completions/min_length": 3.5,
489
+ "completions/min_terminated_length": 3.5,
490
+ "entropy": 0.7240128130373705,
491
+ "epoch": 0.05249343832020997,
492
+ "frac_reward_zero_std": 0.2,
493
+ "grad_norm": 5.0625,
494
+ "learning_rate": 4.8405000000000005e-06,
495
+ "loss": 0.03284294605255127,
496
+ "num_tokens": 95157.0,
497
+ "reward": 1.0648824661970138,
498
+ "reward_std": 0.014824012622557347,
499
+ "rewards/reward_grounded/mean": 1.0,
500
+ "rewards/reward_grounded/std": 0.0,
501
+ "rewards/reward_length/mean": -0.055029296875,
502
+ "rewards/reward_length/std": 0.010819258564151824,
503
+ "rewards/reward_semantic/mean": 0.2254464277997613,
504
+ "rewards/reward_semantic/std": 0.04843873754143715,
505
+ "step": 320,
506
+ "step_time": 1.6396856073492927
507
+ },
508
+ {
509
+ "clip_ratio/high_max": 0.0,
510
+ "clip_ratio/high_mean": 0.0,
511
+ "clip_ratio/low_mean": 0.0,
512
+ "clip_ratio/low_min": 0.0,
513
+ "clip_ratio/region_mean": 0.0,
514
+ "completions/clipped_ratio": 0.0,
515
+ "completions/max_length": 6.25,
516
+ "completions/max_terminated_length": 6.25,
517
+ "completions/mean_length": 4.625,
518
+ "completions/mean_terminated_length": 4.625,
519
+ "completions/min_length": 3.35,
520
+ "completions/min_terminated_length": 3.35,
521
+ "entropy": 0.8246265778172528,
522
+ "epoch": 0.0557742782152231,
523
+ "frac_reward_zero_std": 0.35,
524
+ "grad_norm": 0.0,
525
+ "learning_rate": 4.830500000000001e-06,
526
+ "loss": 0.05039528012275696,
527
+ "num_tokens": 100857.0,
528
+ "reward": 1.0999272018671036,
529
+ "reward_std": 0.012020479088823777,
530
+ "rewards/reward_grounded/mean": 1.0,
531
+ "rewards/reward_grounded/std": 0.0,
532
+ "rewards/reward_length/mean": -0.0510986328125,
533
+ "rewards/reward_length/std": 0.013131194305606187,
534
+ "rewards/reward_semantic/mean": 0.3416071431711316,
535
+ "rewards/reward_semantic/std": 0.03859963417053223,
536
+ "step": 340,
537
+ "step_time": 1.6233859046002181
538
+ },
539
+ {
540
+ "clip_ratio/high_max": 0.0,
541
+ "clip_ratio/high_mean": 0.0,
542
+ "clip_ratio/low_mean": 0.0,
543
+ "clip_ratio/low_min": 0.0,
544
+ "clip_ratio/region_mean": 0.0,
545
+ "completions/clipped_ratio": 0.0,
546
+ "completions/max_length": 5.75,
547
+ "completions/max_terminated_length": 5.75,
548
+ "completions/mean_length": 4.49375,
549
+ "completions/mean_terminated_length": 4.49375,
550
+ "completions/min_length": 3.7,
551
+ "completions/min_terminated_length": 3.7,
552
+ "entropy": 0.49938802123069764,
553
+ "epoch": 0.05905511811023622,
554
+ "frac_reward_zero_std": 0.3,
555
+ "grad_norm": 0.0,
556
+ "learning_rate": 4.8205e-06,
557
+ "loss": 0.01368042677640915,
558
+ "num_tokens": 106664.0,
559
+ "reward": 1.1010554283857346,
560
+ "reward_std": 0.045361624303768625,
561
+ "rewards/reward_grounded/mean": 1.0,
562
+ "rewards/reward_grounded/std": 0.0,
563
+ "rewards/reward_length/mean": -0.0576416015625,
564
+ "rewards/reward_length/std": 0.010658931022044271,
565
+ "rewards/reward_semantic/mean": 0.3464583337306976,
566
+ "rewards/reward_semantic/std": 0.15053959861397742,
567
+ "step": 360,
568
+ "step_time": 1.6211198363504082
569
+ },
570
+ {
571
+ "clip_ratio/high_max": 0.0,
572
+ "clip_ratio/high_mean": 0.0,
573
+ "clip_ratio/low_mean": 0.0,
574
+ "clip_ratio/low_min": 0.0,
575
+ "clip_ratio/region_mean": 0.0,
576
+ "completions/clipped_ratio": 0.0,
577
+ "completions/max_length": 5.8,
578
+ "completions/max_terminated_length": 5.8,
579
+ "completions/mean_length": 4.74375,
580
+ "completions/mean_terminated_length": 4.74375,
581
+ "completions/min_length": 3.6,
582
+ "completions/min_terminated_length": 3.6,
583
+ "entropy": 0.6371999262647023,
584
+ "epoch": 0.062335958005249346,
585
+ "frac_reward_zero_std": 0.3,
586
+ "grad_norm": 9.1875,
587
+ "learning_rate": 4.8105e-06,
588
+ "loss": 0.03233610987663269,
589
+ "num_tokens": 112511.0,
590
+ "reward": 1.1021533191204071,
591
+ "reward_std": 0.02930783515548683,
592
+ "rewards/reward_grounded/mean": 1.0,
593
+ "rewards/reward_grounded/std": 0.0,
594
+ "rewards/reward_length/mean": -0.05693359375,
595
+ "rewards/reward_length/std": 0.012558965157950297,
596
+ "rewards/reward_semantic/mean": 0.3500000026077032,
597
+ "rewards/reward_semantic/std": 0.09658735543489456,
598
+ "step": 380,
599
+ "step_time": 1.6564469950499188
600
+ },
601
+ {
602
+ "clip_ratio/high_max": 0.0,
603
+ "clip_ratio/high_mean": 0.0,
604
+ "clip_ratio/low_mean": 0.0,
605
+ "clip_ratio/low_min": 0.0,
606
+ "clip_ratio/region_mean": 0.0,
607
+ "completions/clipped_ratio": 0.0,
608
+ "completions/max_length": 6.05,
609
+ "completions/max_terminated_length": 6.05,
610
+ "completions/mean_length": 4.33125,
611
+ "completions/mean_terminated_length": 4.33125,
612
+ "completions/min_length": 3.35,
613
+ "completions/min_terminated_length": 3.35,
614
+ "entropy": 0.7747498400121913,
615
+ "epoch": 0.06561679790026247,
616
+ "frac_reward_zero_std": 0.15,
617
+ "grad_norm": 16.0,
618
+ "learning_rate": 4.8005e-06,
619
+ "loss": 0.02885372340679169,
620
+ "num_tokens": 118188.0,
621
+ "reward": 1.1014230579137803,
622
+ "reward_std": 0.054096709901932626,
623
+ "rewards/reward_grounded/mean": 1.0,
624
+ "rewards/reward_grounded/std": 0.0,
625
+ "rewards/reward_length/mean": -0.054931640625,
626
+ "rewards/reward_length/std": 0.01498162968782708,
627
+ "rewards/reward_semantic/mean": 0.3472321443259716,
628
+ "rewards/reward_semantic/std": 0.17982315719127656,
629
+ "step": 400,
630
+ "step_time": 1.6521750739995695
631
+ },
632
+ {
633
+ "clip_ratio/high_max": 0.0,
634
+ "clip_ratio/high_mean": 0.0,
635
+ "clip_ratio/low_mean": 0.0,
636
+ "clip_ratio/low_min": 0.0,
637
+ "clip_ratio/region_mean": 0.0,
638
+ "completions/clipped_ratio": 0.0,
639
+ "completions/max_length": 5.55,
640
+ "completions/max_terminated_length": 5.55,
641
+ "completions/mean_length": 4.25,
642
+ "completions/mean_terminated_length": 4.25,
643
+ "completions/min_length": 3.25,
644
+ "completions/min_terminated_length": 3.25,
645
+ "entropy": 0.6093234239519916,
646
+ "epoch": 0.0688976377952756,
647
+ "frac_reward_zero_std": 0.2,
648
+ "grad_norm": 8.1875,
649
+ "learning_rate": 4.7905e-06,
650
+ "loss": 0.07366046905517579,
651
+ "num_tokens": 123980.0,
652
+ "reward": 1.088520696759224,
653
+ "reward_std": 0.02127447940329148,
654
+ "rewards/reward_grounded/mean": 1.0,
655
+ "rewards/reward_grounded/std": 0.0,
656
+ "rewards/reward_length/mean": -0.0520263671875,
657
+ "rewards/reward_length/std": 0.010646102635655552,
658
+ "rewards/reward_semantic/mean": 0.30374008174985645,
659
+ "rewards/reward_semantic/std": 0.06995791383087635,
660
+ "step": 420,
661
+ "step_time": 1.598766735350273
662
+ },
663
+ {
664
+ "clip_ratio/high_max": 0.0,
665
+ "clip_ratio/high_mean": 0.0,
666
+ "clip_ratio/low_mean": 0.0,
667
+ "clip_ratio/low_min": 0.0,
668
+ "clip_ratio/region_mean": 0.0,
669
+ "completions/clipped_ratio": 0.0,
670
+ "completions/max_length": 5.6,
671
+ "completions/max_terminated_length": 5.6,
672
+ "completions/mean_length": 4.4625,
673
+ "completions/mean_terminated_length": 4.4625,
674
+ "completions/min_length": 3.4,
675
+ "completions/min_terminated_length": 3.4,
676
+ "entropy": 0.6000309210250634,
677
+ "epoch": 0.07217847769028872,
678
+ "frac_reward_zero_std": 0.25,
679
+ "grad_norm": 6.25,
680
+ "learning_rate": 4.7805e-06,
681
+ "loss": 0.02538236975669861,
682
+ "num_tokens": 130310.0,
683
+ "reward": 1.1284031838178634,
684
+ "reward_std": 0.030537298787385224,
685
+ "rewards/reward_grounded/mean": 1.0,
686
+ "rewards/reward_grounded/std": 0.0,
687
+ "rewards/reward_length/mean": -0.0548828125,
688
+ "rewards/reward_length/std": 0.012618192343506963,
689
+ "rewards/reward_semantic/mean": 0.43715773820877074,
690
+ "rewards/reward_semantic/std": 0.10091984458267689,
691
+ "step": 440,
692
+ "step_time": 1.6581264014497719
693
+ },
694
+ {
695
+ "clip_ratio/high_max": 0.0,
696
+ "clip_ratio/high_mean": 0.0,
697
+ "clip_ratio/low_mean": 0.0,
698
+ "clip_ratio/low_min": 0.0,
699
+ "clip_ratio/region_mean": 0.0,
700
+ "completions/clipped_ratio": 0.0,
701
+ "completions/max_length": 6.75,
702
+ "completions/max_terminated_length": 6.75,
703
+ "completions/mean_length": 4.9,
704
+ "completions/mean_terminated_length": 4.9,
705
+ "completions/min_length": 3.45,
706
+ "completions/min_terminated_length": 3.45,
707
+ "entropy": 0.7827204793691636,
708
+ "epoch": 0.07545931758530183,
709
+ "frac_reward_zero_std": 0.15,
710
+ "grad_norm": 2.140625,
711
+ "learning_rate": 4.7705e-06,
712
+ "loss": 0.09432601332664489,
713
+ "num_tokens": 136206.0,
714
+ "reward": 1.061219373345375,
715
+ "reward_std": 0.04166477765902528,
716
+ "rewards/reward_grounded/mean": 1.0,
717
+ "rewards/reward_grounded/std": 0.0,
718
+ "rewards/reward_length/mean": -0.067041015625,
719
+ "rewards/reward_length/std": 0.016452802077401428,
720
+ "rewards/reward_semantic/mean": 0.21523809637874364,
721
+ "rewards/reward_semantic/std": 0.13769038021564484,
722
+ "step": 460,
723
+ "step_time": 1.763348830850282
724
+ },
725
+ {
726
+ "clip_ratio/high_max": 0.0,
727
+ "clip_ratio/high_mean": 0.0,
728
+ "clip_ratio/low_mean": 0.0,
729
+ "clip_ratio/low_min": 0.0,
730
+ "clip_ratio/region_mean": 0.0,
731
+ "completions/clipped_ratio": 0.0,
732
+ "completions/max_length": 5.95,
733
+ "completions/max_terminated_length": 5.95,
734
+ "completions/mean_length": 4.5125,
735
+ "completions/mean_terminated_length": 4.5125,
736
+ "completions/min_length": 3.3,
737
+ "completions/min_terminated_length": 3.3,
738
+ "entropy": 0.6544731822835275,
739
+ "epoch": 0.07874015748031496,
740
+ "frac_reward_zero_std": 0.25,
741
+ "grad_norm": 0.0,
742
+ "learning_rate": 4.7605e-06,
743
+ "loss": 0.095479816198349,
744
+ "num_tokens": 142168.0,
745
+ "reward": 1.110293486714363,
746
+ "reward_std": 0.03730325770447962,
747
+ "rewards/reward_grounded/mean": 1.0,
748
+ "rewards/reward_grounded/std": 0.0,
749
+ "rewards/reward_length/mean": -0.0526123046875,
750
+ "rewards/reward_length/std": 0.013677800679579378,
751
+ "rewards/reward_semantic/mean": 0.37641369104385375,
752
+ "rewards/reward_semantic/std": 0.12245006412267685,
753
+ "step": 480,
754
+ "step_time": 1.693691986000158
755
+ },
756
+ {
757
+ "clip_ratio/high_max": 0.0,
758
+ "clip_ratio/high_mean": 0.0,
759
+ "clip_ratio/low_mean": 0.0,
760
+ "clip_ratio/low_min": 0.0,
761
+ "clip_ratio/region_mean": 0.0,
762
+ "completions/clipped_ratio": 0.0,
763
+ "completions/max_length": 6.2,
764
+ "completions/max_terminated_length": 6.2,
765
+ "completions/mean_length": 4.5375,
766
+ "completions/mean_terminated_length": 4.5375,
767
+ "completions/min_length": 3.3,
768
+ "completions/min_terminated_length": 3.3,
769
+ "entropy": 0.7134705329648569,
770
+ "epoch": 0.08202099737532809,
771
+ "frac_reward_zero_std": 0.2,
772
+ "grad_norm": 0.0,
773
+ "learning_rate": 4.7505000000000005e-06,
774
+ "loss": 0.04971726834774017,
775
+ "num_tokens": 148190.0,
776
+ "reward": 1.059982505440712,
777
+ "reward_std": 0.029997232874302428,
778
+ "rewards/reward_grounded/mean": 1.0,
779
+ "rewards/reward_grounded/std": 0.0,
780
+ "rewards/reward_length/mean": -0.053564453125,
781
+ "rewards/reward_length/std": 0.012155310017988086,
782
+ "rewards/reward_semantic/mean": 0.20886905267834663,
783
+ "rewards/reward_semantic/std": 0.0987570121884346,
784
+ "step": 500,
785
+ "step_time": 1.7915035136496953
786
+ }
787
+ ],
788
+ "logging_steps": 20,
789
+ "max_steps": 10000,
790
+ "num_input_tokens_seen": 148190,
791
+ "num_train_epochs": 2,
792
+ "save_steps": 500,
793
+ "stateful_callbacks": {
794
+ "TrainerControl": {
795
+ "args": {
796
+ "should_epoch_stop": false,
797
+ "should_evaluate": false,
798
+ "should_log": false,
799
+ "should_save": true,
800
+ "should_training_stop": false
801
+ },
802
+ "attributes": {}
803
+ }
804
+ },
805
+ "total_flos": 0.0,
806
+ "train_batch_size": 1,
807
+ "trial_name": null,
808
+ "trial_params": null
809
+ }
grpo/checkpoint-500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:200955a61f7b161c5c840b30b5e94f7cdbc4ef6fd71e4ae199aac51700a74d8c
3
+ size 7249