jvonrad's picture
Add README.md
2a636b6 verified
|
Raw
History Blame
590 Bytes
metadata
base_model: Qwen/Qwen2.5-7B
tags:
  - grpo
  - cross-lingual
  - ablation

Qwen-2.5-7B-grpo-bonus5-ablation

GRPO all_correct_bonus ablation on Qwen2.5-7B, trained on the WIKI-FACT / PolyFact cross-lingual factual-recall task (see jvonrad/Lost-in-Mistranslation). --all_correct_bonus 5.0 (all other hyperparameters match the paper's main GRPO run). Evaluated in results/qwen-2.5-7b-grpo-bonus5-ablation_polyfact_consistency.json and results/qwen-2.5-7b-grpo-bonus5-ablation_gmmlu_lite_consistency.json in the repo above.