[Bugfix][Qwen4Exp] Load PLE tables unquantized under Quark checkpoints (#59443)

Signed-off-by: Mikko Tukiainen <Mikko.Tukiainen@amd.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
This commit is contained in:
Mikko Tukiainen
2026-10-05 10:40:02 +00:00
committed by GitHub
co-authored by Cursor mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> TJian
parent c04c79e50d
commit edca360f13
2 changed files with 15 additions and 0 deletions
@@ -32,6 +32,7 @@ from vllm.model_executor.layers.quantization.modelopt import (
ModelOptMixedPrecisionConfig,
ModelOptQuantConfigBase,
)
from vllm.model_executor.layers.quantization.quark.quark import QuarkConfig
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
create_fp8_scale_parameter,
)
@@ -196,6 +197,9 @@ class Qwen4ExpPLEEmbeddingMethod(QuantizeMethodBase):
and not quant_config.config_parser.resolve(None, prefix).quantized
):
return Qwen4ExpPLEUnquantizedEmbeddingMethod()
# Quark quantizes only Linear and MoE layers; PLE tables stay BF16.
if isinstance(quant_config, QuarkConfig):
return Qwen4ExpPLEUnquantizedEmbeddingMethod()
if not isinstance(quant_config, Fp8Config):
raise NotImplementedError(
"Qwen4Exp PLE embedding does not support quantization config "