Modelpublic

zhiyuanhucs/genshin-llava-qwen3.5-9b-128k

sourceHugging Faceupdated 2mo agoView on Hugging Face

0likes1downloads

Model Card

Genshin + LLaVA Qwen3.5-9B 128K

Full-parameter Megatron-SWIFT SFT with TP=4, MTP=1 and 128K packing. Training mixture: Genshin 3 passes + LLaVA-OneVision 0.162 passes.

Each step-N branch contains the merged Hugging Face safetensors export for checkpoint N.