| role : SFT round-2 (the RL actor checkpoint) | |
| format : HF (merged) + tokenizer/processor | |
| source : /mnt/faster_nas/benyou.mys/medixrl/checkpoints/sft_qwen3vl_mix3_p40to60_0826 | |
| (root files; raw training checkpoints stay at source, e.g. checkpoint-824) | |
| role : SFT round-2 (the RL actor checkpoint) | |
| format : HF (merged) + tokenizer/processor | |
| source : /mnt/faster_nas/benyou.mys/medixrl/checkpoints/sft_qwen3vl_mix3_p40to60_0826 | |
| (root files; raw training checkpoints stay at source, e.g. checkpoint-824) | |