diff --git a/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py b/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py index 9be801225a..86a01434e5 100644 --- a/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py +++ b/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py @@ -384,17 +384,17 @@ class CompressedTensorsConfig(QuantizationConfig): supported = capability == min_capability if error and not supported: raise RuntimeError( - "Quantization scheme is not supported for ", - "the current GPU. Required capability: ", - f"{min_capability}. Current capability: {capability}.", + "Quantization scheme is not supported for the current GPU. " + f"Required capability: {min_capability}. " + f"Current capability: {capability}." ) else: supported = capability >= min_capability if error and not supported: raise RuntimeError( - "Quantization scheme is not supported for ", - f"the current GPU. Min capability: {min_capability}. ", - f"Current capability: {capability}.", + "Quantization scheme is not supported for the current GPU. " + f"Min capability: {min_capability}. " + f"Current capability: {capability}." ) return supported else: @@ -737,8 +737,8 @@ class CompressedTensorsConfig(QuantizationConfig): if not self._is_nvfp4_format(input_quant): raise ValueError( - "For NVFP4 weights, input quantization must also be NVFP4 format, ", - "None for NVFP4A16", + "For NVFP4 weights, input quantization must also be NVFP4 " + "format, None for NVFP4A16" ) return CompressedTensorsW4A4Fp4() diff --git a/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe/compressed_tensors_moe.py b/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe/compressed_tensors_moe.py index 78d70d492d..bcc930471d 100644 --- a/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe/compressed_tensors_moe.py +++ b/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe/compressed_tensors_moe.py @@ -81,11 +81,11 @@ class CompressedTensorsMoEMethod(FusedMoEMethodBase): if not valid_format_and_bits: raise ValueError( - "For Fused MoE layers, only format: ", - f"{CompressionFormat.pack_quantized.value} ", - f" and bits: {WNA16_SUPPORTED_BITS} is supported ", + "For Fused MoE layers, only format: " + f"{CompressionFormat.pack_quantized.value} " + f"and bits: {WNA16_SUPPORTED_BITS} is supported " f"but got format: {CompressionFormat.pack_quantized.value} " - f" and bits: {weight_quant.num_bits}", + f"and bits: {weight_quant.num_bits}" ) # Native ROCm HIP kernels (RDNA3, etc.) @@ -145,8 +145,8 @@ class CompressedTensorsMoEMethod(FusedMoEMethodBase): ) if not _is_valid_nvfp4_activations: raise ValueError( - "For NVFP4 weights, input quantization must also be NVFP4 format ", - f"or None for NVFP4A16, found {input_quant}", + "For NVFP4 weights, input quantization must also be NVFP4 " + f"format or None for NVFP4A16, found {input_quant}" ) return CompressedTensorsW4A4Nvfp4MoEMethod( layer.moe_config, layer_name, use_a16=(input_quant is None) diff --git a/vllm/model_executor/layers/quantization/quark/quark.py b/vllm/model_executor/layers/quantization/quark/quark.py index 912f61ac38..6c877b1a63 100644 --- a/vllm/model_executor/layers/quantization/quark/quark.py +++ b/vllm/model_executor/layers/quantization/quark/quark.py @@ -280,9 +280,9 @@ class QuarkConfig(QuantizationConfig): supported = capability >= min_capability if error and not supported: raise RuntimeError( - "Quantization scheme is not supported for ", - f"the current GPU. Min capability: {min_capability}. ", - f"Current capability: {capability}.", + "Quantization scheme is not supported for the current GPU. " + f"Min capability: {min_capability}. " + f"Current capability: {capability}." ) return supported else: