[Quantization] [Refactor] Create special "GptOssMxfp4MoeMethod"#39604

Merged

mgoin merged 10 commits intovllm-project:mainfrom

zyongye:rename_gpt_oss_mxfp4

Apr 13, 2026

Commits on Apr 11, 2026

[Quantization] Rename mxfp4 quant layer and oracle to gpt_oss_mxfp4
zyongye
committed

Commits on Apr 12, 2026

[Quantization] Add GptOssMxfp4Config subclass and gpt_oss_mxfp4 quant method
zyongye
committed
[Quantization] Rename make_mxfp4_moe_* to make_gpt_oss_mxfp4_moe_* in oracle
zyongye
committed
[Quantization] Rename mxfp4 oracle functions to gpt_oss_mxfp4 scope
zyongye
committed
[Quantization] Rename GptOssMxfp4MoeBackend back to Mxfp4MoeBackend
zyongye
committed
[Quantization] Rename oracle/gpt_oss_mxfp4.py back to oracle/mxfp4.py
zyongye
committed
[Quantization] Rename gpt_oss-specific oracle functions
zyongye
committed
[Quantization] Improve error messages and document normalization paths
zyongye
committed
[Quantization] Fix Mxfp4Config base class and MXFP4 variant detection
zyongye
committed
[Quantization] Fix type annotations and doc cross-reference
zyongye
committed