初始化项目,由ModelHub XC社区提供模型

Model: divelab/DAPO_E2H-math-cosine
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-28 06:48:18 +08:00
commit 8164dc222c
15 changed files with 152134 additions and 0 deletions

7
.hydra/overrides.yaml Normal file
View File

@@ -0,0 +1,7 @@
- mode=train
- task=math
- algorithm=grpo
- algorithm.training.curriculum_schedule=cosine
- model=qwen15
- algorithm.training.max_steps=1600
- algorithm.training.vllm_mode=colocate