Qwen2.5-1.5B-Instruct_BF16_open-r1-DAPO-Math-17k-Processed_588_FlashRL_G4-L2048_new

This repository contains a checkpoint trained with GRPO on open-r1/DAPO-Math-17k-Processed starting from Qwen/Qwen2.5-1.5B-Instruct.
This snapshot corresponds to training step 588.

Contents include:

• Model weights (.safetensors)
• Config files (config.json, generation_config.json)
• Tokenizer files (tokenizer.json, tokenizer_config.json, vocab.json, merges.txt, special_tokens_map.json, added_tokens.json)
• Optional chat template (chat_template.jinja)

Training artifacts (optimizer/scheduler states and RNG) have been intentionally excluded.