SearchingMan's picture
add pruned-24 NVFP4 AWQ encoder
310d8db verified
Raw
History Blame Contribute Delete
619 Bytes
This repository contains modified derivatives of Qwen3-VL by the Qwen team.
Upstream sources:
- Qwen/Qwen3-VL-8B-Instruct, revision 0c351dd01ed87e9c1b53cbc748cba10e6187ff3b
- Qwen/Qwen3-VL-32B-Instruct, revision 0cfaf48183f594c314753d30a4c4974bc75f3ccb
Modifications include language-model extraction, physical layer pruning, NVFP4/AWQ and INT8 ConvRot packaging, trained ARA residual weights, and a trained 4096-to-5120 conditioning adapter. These modifications are not produced or endorsed by the Qwen team.
MiniMax-H3 is the compatible downstream model. No MiniMax-H3 diffusion-model or VAE weights are included.