GitHub

Qwen2.5-VL cast to Megatron format ERROR

CUDA_VISIBLE_DEVICES=0 \ swift export \ --model Qwen/Qwen2.5-VL-7B-Instruct \ --to_mcore true \ --torch_dtype bfloat16 \ --output_dir Qwen2.5-VL-7B-Instruct-mcore ...
GitHub

How to convert megatron Qwen2.5-VL checkpoints(*.distcp) to huggingface format?

I trained Qwen2.5-VL-7B with the megatron backend using examples/grpo_trainer/run_qwen2_5_vl-7b-megatron.sh. Checkpoints were saved as *.distcp. I can’t find vlm ...