[Bugfix] Fix extra whitespace in strings caused by newline (#23272)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

benchmarks/benchmark_dataset.py

@@ -958,8 +958,10 @@ class InstructCoderDataset(HuggingFaceDataset):
         for i, item in enumerate(self.data):
             if len(sampled_requests) >= num_requests:
                 break
-            prompt = f"{item['input']}\n\n{item['instruction']} Just output \
+            prompt = (
-            the code, do not include any explanation."
+                f"{item['input']}\n\n{item['instruction']} Just output "
+                "the code, do not include any explanation."
+            )
 
             # apply template
             prompt = tokenizer.apply_chat_template(

examples/offline_inference/vision_language.py

@@ -283,8 +283,10 @@ def run_glm4v(questions: list[str], modality: str) -> ModelRequestData:
     )
 
     prompts = [
-        f"<|user|>\n<|begin_of_image|><|endoftext|><|end_of_image|>\
+        (
-        {question}<|assistant|>"
+            "<|user|>\n<|begin_of_image|><|endoftext|><|end_of_image|>"
+            f"{question}<|assistant|>"
+        )
         for question in questions
     ]
 
@@ -767,15 +769,13 @@ def run_llava_next_video(questions: list[str], modality: str) -> ModelRequestDat
 def run_llava_onevision(questions: list[str], modality: str) -> ModelRequestData:
     if modality == "video":
         prompts = [
-            f"<|im_start|>user <video>\n{question}<|im_end|> \
+            f"<|im_start|>user <video>\n{question}<|im_end|><|im_start|>assistant\n"
-        <|im_start|>assistant\n"
             for question in questions
         ]
 
     elif modality == "image":
         prompts = [
-            f"<|im_start|>user <image>\n{question}<|im_end|> \
+            f"<|im_start|>user <image>\n{question}<|im_end|><|im_start|>assistant\n"
-        <|im_start|>assistant\n"
             for question in questions
         ]
 
@@ -998,8 +998,7 @@ def run_molmo(questions: list[str], modality: str) -> ModelRequestData:
     )
 
     prompts = [
-        f"<|im_start|>user <image>\n{question}<|im_end|> \
+        f"<|im_start|>user <image>\n{question}<|im_end|><|im_start|>assistant\n"
-        <|im_start|>assistant\n"
         for question in questions
     ]
 

vllm/benchmarks/datasets.py

@@ -1289,8 +1289,10 @@ class InstructCoderDataset(HuggingFaceDataset):
         for i, item in enumerate(self.data):
             if len(sampled_requests) >= num_requests:
                 break
-            prompt = f"{item['input']}\n\n{item['instruction']} Just output \
+            prompt = (
-            the code, do not include any explanation."
+                f"{item['input']}\n\n{item['instruction']} Just output "
+                "the code, do not include any explanation."
+            )
 
             # apply template
             prompt = tokenizer.apply_chat_template(

vllm/model_executor/model_loader/tpu.py

@@ -98,14 +98,15 @@ class TPUModelLoader(DefaultModelLoader):
 
         # Check parameters
         for name, param in model.named_parameters():
-            assert param.device.type == device_type, f"Parameter {name} is on \
+            assert param.device.type == device_type, (
-                {param.device.type} instead of {device_type}"
+                f"Parameter {name} is on {param.device.type} "
+                f"instead of {device_type}")
 
         # Check buffers
         for name, buffer in model.named_buffers():
-            assert buffer.device.type == device_type, \
+            assert buffer.device.type == device_type, (
-                f"Buffer {name} is on {buffer.device.type} instead of \
+                f"Buffer {name} is on {buffer.device.type} "
-                    {device_type}"
+                f"instead of {device_type}")
 
         for module in model.modules():
             if (mesh is not None) and (get_fqn(module) == 'QKVParallelLinear'):

vllm/model_executor/models/hyperclovax_vision.py

@@ -929,8 +929,8 @@ class HCXVisionForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
                 target_group_size = 0
 
             elif video_group_size < target_group_size:
-                raise RuntimeError(f"video_group_size < target_group_size!! \
+                raise RuntimeError(
-                        [{video_group_size} < {target_group_size}]")
+                    f"{video_group_size=} < {target_group_size=}")
 
         assert len(target_features
                    ) == 0, f"target_features is not empty!! {target_features}"
@@ -1114,9 +1114,8 @@ def reshape_and_unpad_image_features(
     base_image_feature = image_feature[0]
     image_feature = image_feature[1:]
 
-    assert (height * width == base_image_feature.shape[0]
+    assert height * width == base_image_feature.shape[0], (
-            ), f"height: {height}, width: {width}, \
+        f"{height=} * {width=} != {base_image_feature.shape[0]=}")
-        base_image_feature.shape[0]: {base_image_feature.shape[0]}"
 
     num_patch_width, num_patch_height = get_anyres_image_grid_shape(
         image_size, possible_resolutions, grid_size)

vllm/model_executor/models/phi4mm.py

@@ -262,9 +262,9 @@ class Phi4MMImageEncoder(nn.Module):
             img_features.shape[1]))
         assert base_feat_height == base_feat_height_target \
             and base_feat_width == base_feat_height_target, \
-                f'base_feat_height: {base_feat_height},"\
+                (f"base_feat_height: {base_feat_height}, "
-                f" base_feat_width: {base_feat_width}, "\
+                 f"base_feat_width: {base_feat_width}, "
-                f"expect {base_feat_height_target} features for hd transform'
+                 f"expect {base_feat_height_target} features for hd transform")
 
         # bs x max_num_crops x (24x24) x C
         img_features = img_features.view(bs, -1,

vllm/transformers_utils/configs/eagle.py

@@ -61,8 +61,8 @@ class EAGLEConfig(PretrainedConfig):
                 else f"Eagle3{arch}" for arch in self.model.architectures
             ]
         else:
-            raise ValueError(f"Invalid method {method}. \
+            raise ValueError(f"Invalid method {method}. "
-                Supported methods are eagle and eagle3.")
+                             "Supported methods are eagle and eagle3.")
 
         super().__init__(**kwargs)