Hugging Face's logo

KRAFTON
/
OpenThinker3-Offline-GRPO-7B

Safetensors
qwen2
Model card Files
xet
Community
1
gmkim commited on
Commit
c01c7d1
·
verified ·
1 Parent(s): 85a5c97

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -6
README.md CHANGED
@@ -15,14 +15,16 @@ For more details, please refer to our blog
15
  - [English Version](https://krafton-ai.github.io/blog/llm_post_training_en/)
16
  - [Korean Version](https://krafton-ai.github.io/blog/llm_post_training_kr/)
17
 
 
 
18
 
19
  ## Results
20
 
21
  | Model | Method | AIME25 | AMC23 | LiveCodeBench | GPQA-Diamond | IFEval |
22
  |--------------------------------|--------------------------------|--------|-------|---------------|--------------|--------|
23
- | Openthinker3-7B | Base | 57.2915 | 92.617 | 63.968 | 50.947 | 50.09 |
24
- | | Offline GRPO (+bias) | 59.5315 | 93.516 | 64.995 | 49.684 | 51.66 |
25
- | Openthinker2-7B | Base | 39.792 | 88.633 | 56.115 | 45.833 | 53.3 |
26
- | | Offline GRPO (+bias) | 40.3645 | 87.656 | 55.944 | 46.843 | 52.20 |
27
- | AceReason-Nemetron-1.1-7B | Base | 64.635 | 92.93 | 72.383 | 52.462 | 36.02 |
28
- | | Offline GRPO (+bias) | 65.521 | 93.164 | 72.603 | 54.356 | 38.23 |
 
15
  - [English Version](https://krafton-ai.github.io/blog/llm_post_training_en/)
16
  - [Korean Version](https://krafton-ai.github.io/blog/llm_post_training_kr/)
17
 
18
+ Additionally, we have released the code used for training
19
+ - [Github](https://github.com/krafton-ai/Offline-GRPO)
20
 
21
  ## Results
22
 
23
  | Model | Method | AIME25 | AMC23 | LiveCodeBench | GPQA-Diamond | IFEval |
24
  |--------------------------------|--------------------------------|--------|-------|---------------|--------------|--------|
25
+ | Openthinker3-7B | Base | 57.292 | 92.617 | 63.968 | 50.947 | 50.09 |
26
+ | | Offline GRPO (+bias) | 59.532 | 93.516 | 65.435 | 50.947 | 51.14 |
27
+ | Openthinker2-7B | Base | 39.792 | 88.633 | 56.115 | 45.833 | 53.30 |
28
+ | | Offline GRPO (+bias) | 40.573 | 88.359 | 56.115 | 46.717 | 52.62 |
29
+ | AceReason-Nemetron-1.1-7B | Base | 64.635 | 92.930 | 72.383 | 52.462 | 36.02 |
30
+ | | Offline GRPO (+bias) | 65.573 | 93.203 | 71.673 | 52.146 | 37.38 |