melodyhorse/gpt-oss-20b-triviaqa-rl

GPT-OSS-20B TriviaQA RL

LoRA adapter for GPT-OSS-20B trained with reinforcement learning on TriviaQA.

Training Details

• Base model: openai/gpt-oss-20b
• Method: GRPO reinforcement learning
• Dataset: TriviaQA (train split)
• LoRA rank: 32
• Target modules: all-linear
• Learning rate: 2e-5
• Group size: 8
• Groups per batch: 32
• Checkpoint: step 1240