A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[UX] Remove "quantization is not fully optimized yet" log (#25012)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Sep 16, 2025 at 23:57 UTC
67532a1a6855e8262b3e1c9512c85e2fc934b3c0
1 file changed
-21
vllm/config/__init__.py
-21
@@ -1086,22 +1086,6 @@ class ModelConfig:
1086
1087
def _verify_quantization(self) -> None:
1088
supported_quantization = me_quant.QUANTIZATION_METHODS
1089
- optimized_quantization_methods = [
1090
- "fp8",
1091
- "modelopt",
1092
- "gptq_marlin_24",
1093
- "gptq_marlin",
1094
- "awq_marlin",
1095
- "fbgemm_fp8",
1096
- "compressed-tensors",
1097
- "experts_int8",
1098
- "quark",
1099
- "modelopt_fp4",
1100
- "bitblas",
1101
- "gptq_bitblas",
1102
- "inc",
1103
- "petit_nvfp4",
1104
- ]
1089
if self.quantization is not None:
1090
self.quantization = cast(me_quant.QuantizationMethods,
1091
self.quantization)
@@ -1183,11 +1167,6 @@ class ModelConfig:
1167
f"be one of {supported_quantization}.")
1168
from vllm.platforms import current_platform
1169
current_platform.verify_quantization(self.quantization)
1186
- if self.quantization not in optimized_quantization_methods:
1187
- logger.warning(
1188
- "%s quantization is not fully "
1189
- "optimized yet. The speed can be slower than "
1190
- "non-quantized models.", self.quantization)
1170
1171
def _verify_cuda_graph(self) -> None:
1172
# The `max_seq_len_to_capture` was incorrectly