[diffusion] feat: support loading mixed w4a8 text encoders (#36037)

This commit is contained in:
Mick
2026-08-24 13:31:27 +08:00
committed by GitHub
parent 344613c159
commit 8df3b9eff9
9 changed files with 260 additions and 34 deletions
@@ -958,6 +958,17 @@ linear layers use FP8 while embeddings, normalization, and the vision tower
remain BF16. This is an approximate serve-time choice and is incompatible with
the strict `quality="high"` deployment contract.
The mixed W4A8 community encoder follows the same component-path contract:
```bash Overlay
--component-paths.text_encoder \
Winnougan/MiniMax-H3-INT4_Convrot_ComfyUI/qwen3vl_32b_minimax_h3-w4a8_convrot.safetensors
```
Install `comfy-kitchen>=0.2.27` and omit `--quantization`. SGLang automatically
loads its W4A8 language linears and tensorwise INT8 embedding; the unmarked
vision tower remains BF16.
</Tab>
<Tab title="Encoder scheduling">