Various cosmetic/comment fixes (#12089)

Signed-off-by: mgoin <michael@neuralmagic.com>

vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_24.py

@@ -42,7 +42,7 @@ class CompressedTensors24(CompressedTensorsScheme):
 
         if not sparse_cutlass_supported():
             raise ValueError(
-                "Sparse CUTLASS not supported. vLLM must be built with"
+                "Sparse CUTLASS not supported. vLLM must be built with "
                 "CUDA 12.2 or later to use this feature")
 
         self.output_dtype = params_dtype

vllm/model_executor/models/aria.py

@@ -390,8 +390,7 @@ class AriaMoELMModel(LlamaModel):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/commandr.py

@@ -440,8 +440,7 @@ class CohereForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/dbrx.py

@@ -452,8 +452,7 @@ class DbrxForCausalLM(nn.Module, SupportsPP):
         for name, loaded_weight in weights:
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/exaone.py

@@ -533,8 +533,7 @@ class ExaoneForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/gpt_j.py

@@ -316,8 +316,7 @@ class GPTJForCausalLM(nn.Module, SupportsPP):
 
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/granite.py

@@ -475,8 +475,7 @@ class GraniteForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/llama.py

@@ -105,9 +105,9 @@ class LlamaAttention(nn.Module):
                  max_position_embeddings: int = 8192,
                  quant_config: Optional[QuantizationConfig] = None,
                  bias: bool = False,
+                 bias_o_proj: bool = False,
                  cache_config: Optional[CacheConfig] = None,
-                 prefix: str = "",
-                 bias_o_proj: bool = False) -> None:
+                 prefix: str = "") -> None:
         super().__init__()
         layer_idx = extract_layer_index(prefix)
         self.hidden_size = hidden_size
@@ -397,8 +397,7 @@ class LlamaModel(nn.Module):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/mixtral.py

@@ -431,8 +431,7 @@ class MixtralForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/mllama.py

@@ -1432,8 +1432,7 @@ class MllamaForConditionalGeneration(nn.Module, SupportsMultiModal):
                 loaded_weight = loaded_weight.view(loaded_weight.shape[0], -1)
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/nemotron.py

@@ -492,8 +492,7 @@ class NemotronForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/phimoe.py

@@ -626,8 +626,7 @@ class PhiMoEForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/qwen2.py

@@ -367,8 +367,7 @@ class Qwen2Model(nn.Module):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)

vllm/model_executor/models/solar.py

@@ -492,8 +492,7 @@ class SolarForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                 continue
             if (self.quant_config is not None and
                 (scale_name := self.quant_config.get_cache_scale(name))):
-                # Loading kv cache scales for quark and
-                # compressed-tensors quantization
+                # Loading kv cache quantization scales
                 param = params_dict[scale_name]
                 weight_loader = getattr(param, "weight_loader",
                                         default_weight_loader)