[Qwen3.5][MTP] Preserve online NVFP4 draft quantization for mixed checkpoints (#35545)

This commit is contained in:
YAMY
2026-08-19 11:12:32 -07:00
committed by GitHub
parent 157d8ad27a
commit 03cf2de2e3
+3 -3
View File
@@ -1516,14 +1516,14 @@ class ModelConfig:
"quant_method", "" if not self.quantization else self.quantization
).lower()
# ModelOpt FP4 checkpoints quantize only the target model; an
# embedded MTP draft may stay unquantized, so an explicit
# ModelOpt FP4 and mixed checkpoints can quantize only the target
# model; an embedded MTP draft may stay unquantized, so an explicit
# nvfp4_online opt-in for the draft wins over checkpoint detection.
# The online loader rejects already-packed weights at load time.
preserve_online_draft_quantization = (
self.is_draft_model
and self.quantization == "nvfp4_online"
and quant_method == "modelopt_fp4"
and quant_method in ("modelopt_fp4", "modelopt_mixed")
)
# An explicit online-requantization request (e.g. quark_mxfp4 on top
# of an NVFP4/mixed checkpoint) must not be overridden back to the