Saraswathy/vlm-mix-resume-stem60-social40-step100

stem60-social40: complete EasyR1 resume checkpoint

This public repository contains the complete EasyR1 training-resume state at step 100, including FSDP model and optimizer shards, extra state, dataloader state, and the LoRA adapter. Verify every file against SHA256SUMS.json before resuming. It is not a merged standalone model; the base is Qwen/Qwen3-VL-4B-Instruct.