Pinned Loading

1. nanbeige4-3b-cold-start-sft nanbeige4-3b-cold-start-sft Public

   LoRA cold-start thinking SFT experiments on Nanbeige4-3B-Base using reasoning traces distilled from frontier models

   Python

2. qwen2.5-1.5b-gsm8k-grpo-rlvr qwen2.5-1.5b-gsm8k-grpo-rlvr Public

   RLVR experiments on GSM8K using Hugging Face TRL GRPO, Qwen2.5-1.5B-Instruct, and Modal to study reward design in practice.

   Python

3. gsm8k-grpo-prime-lab gsm8k-grpo-prime-lab Public

   Prime Intellect Lab scaffold repo for my second RL experiment, mainly centered on the GSM8K GRPO config used in the final run.