Wan 2.2 nvfp4 (uncalibrated)
These are quick nvfp4 quants of a couple wan 2.2 models, nvfp4 mixed uses standard torch fp8_e4m3 for some layers (attn.v and ffn.2) to avoid ghosting and other quality drops.
These are quick nvfp4 quants of a couple wan 2.2 models, nvfp4 mixed uses standard torch fp8_e4m3 for some layers (attn.v and ffn.2) to avoid ghosting and other quality drops.
wan2.2_i2v_high_noise_14B_nvfp4_mixed.safetensors 11 GBwan2.2_i2v_low_noise_14B_nvfp4_mixed.safetensors 11 GB有文件?提交种子。
提交种子Wan-AI/Wan2.2-I2V-A14B