vllm.model_executor.layers.quantization.online
¶
Modules:
-
base– -
fp8– -
int4–Online int4 weight-only quantization.
-
int8– -
legacy_env–Deprecated environment variables redirected to
--quantization-config. -
lm_head–Online lm_head quantization: weight-only methods for the model's own
-
loading– -
lut_b–Online LUT-B quantization of routed MoE experts (MXFP8 activations).
-
moe_base– -
mxfp4–Online MXFP4 (microscaling FP4, block-32) quantization methods.
-
mxfp8–Online MXFP8 (microscaling FP8, block-32) quantization methods.
-
nvfp4– -
requantization– -
shared_expert–