R2EGym-7B-Agent-Coder-Instruct (checkpoint-800)

• Base: Qwen/Qwen2.5-Coder-7B-Instruct
• Training: SFT with DeepSpeed ZeRO-3
• Checkpoint: checkpoint-800

Notes

• This repo includes ZeRO optimizer states in global_step800/ for resuming training.
• For inference, use the model-0000*-of-00004.safetensors shards and tokenizer files.