[Bugfix] Fix six quantization exception messages split across positional args (#50479)
Signed-off-by: ErenAta16 <erena6466@gmail.com>
ErenAta16 committed
Aug 21, 2026 at 19:06 UTC
fbb17e780b1b6e91b3eb9777eb44d3288c2c30b6
3 files changed
+17
-17
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py
+8
-8
@@ -384,17 +384,17 @@ class CompressedTensorsConfig(QuantizationConfig):
384
supported = capability == min_capability
385
if error and not supported:
386
raise RuntimeError(
387
- "Quantization scheme is not supported for ",
388
- "the current GPU. Required capability: ",
389
- f"{min_capability}. Current capability: {capability}.",
387
+ "Quantization scheme is not supported for the current GPU. "
388
+ f"Required capability: {min_capability}. "
389
+ f"Current capability: {capability}."
390
)
391
else:
392
supported = capability >= min_capability
393
if error and not supported:
394
raise RuntimeError(
395
- "Quantization scheme is not supported for ",
396
- f"the current GPU. Min capability: {min_capability}. ",
397
- f"Current capability: {capability}.",
395
+ "Quantization scheme is not supported for the current GPU. "
396
+ f"Min capability: {min_capability}. "
397
+ f"Current capability: {capability}."
398
)
399
return supported
400
else:
@@ -737,8 +737,8 @@ class CompressedTensorsConfig(QuantizationConfig):
737
738
if not self._is_nvfp4_format(input_quant):
739
raise ValueError(
740
- "For NVFP4 weights, input quantization must also be NVFP4 format, ",
741
- "None for NVFP4A16",
740
+ "For NVFP4 weights, input quantization must also be NVFP4 "
741
+ "format, None for NVFP4A16"
742
)
743
return CompressedTensorsW4A4Fp4()
744
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe/compressed_tensors_moe.py
+6
-6
@@ -81,11 +81,11 @@ class CompressedTensorsMoEMethod(FusedMoEMethodBase):
81
82
if not valid_format_and_bits:
83
raise ValueError(
84
- "For Fused MoE layers, only format: ",
85
- f"{CompressionFormat.pack_quantized.value} ",
86
- f" and bits: {WNA16_SUPPORTED_BITS} is supported ",
84
+ "For Fused MoE layers, only format: "
85
+ f"{CompressionFormat.pack_quantized.value} "
86
+ f"and bits: {WNA16_SUPPORTED_BITS} is supported "
87
f"but got format: {CompressionFormat.pack_quantized.value} "
88
- f" and bits: {weight_quant.num_bits}",
88
+ f"and bits: {weight_quant.num_bits}"
89
)
90
91
# Native ROCm HIP kernels (RDNA3, etc.)
@@ -145,8 +145,8 @@ class CompressedTensorsMoEMethod(FusedMoEMethodBase):
145
)
146
if not _is_valid_nvfp4_activations:
147
raise ValueError(
148
- "For NVFP4 weights, input quantization must also be NVFP4 format ",
149
- f"or None for NVFP4A16, found {input_quant}",
148
+ "For NVFP4 weights, input quantization must also be NVFP4 "
149
+ f"format or None for NVFP4A16, found {input_quant}"
150
)
151
return CompressedTensorsW4A4Nvfp4MoEMethod(
152
layer.moe_config, layer_name, use_a16=(input_quant is None)
vllm/model_executor/layers/quantization/quark/quark.py
+3
-3
@@ -280,9 +280,9 @@ class QuarkConfig(QuantizationConfig):
280
supported = capability >= min_capability
281
if error and not supported:
282
raise RuntimeError(
283
- "Quantization scheme is not supported for ",
284
- f"the current GPU. Min capability: {min_capability}. ",
285
- f"Current capability: {capability}.",
283
+ "Quantization scheme is not supported for the current GPU. "
284
+ f"Min capability: {min_capability}. "
285
+ f"Current capability: {capability}."
286
)
287
return supported
288
else: