[diffusion] refactor: reuse srt quantization contracts and mxfp8 kernels (#36063)

2026-08-24 09:26:54 +08:00
parent 167c339c8e
commit 1c1c9d9b4e
16 changed files with 277 additions and 511 deletions
@@ -291,6 +291,13 @@ the fused ConvRot kernel without requantizing them. TP1/2/4 and sequence or
 layerwise offload are supported; TP8 violates the checkpoint's 256-element
 ConvRot group boundary, and FSDP is rejected.

+Self-describing community MXFP8 files use the same
+`--transformer-weights-path <repo/file.safetensors>` form and need no
+quantization flag. SGLang reads their global format metadata, keeps unmarked H3
+layers in their original dtype, and reuses SRT's MXFP8 dense kernels. The
+selected SRT backend must support MXFP8 on the target GPU; FSDP is rejected for
+this mixed per-layer layout.
+
 ### Advanced: precomputed AdaLN cache

 The [model card](https://huggingface.co/MiniMaxAI/MiniMax-H3) notes that about