add .isort.cfg (#378)

python/sglang/srt/models/commandr.py

@@ -20,13 +20,10 @@
 
 # This file is based on the LLama model definition file in transformers
 """PyTorch Cohere model."""
-from typing import List, Optional, Tuple
+from typing import Optional, Tuple
 
 import torch
 import torch.utils.checkpoint
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from torch.nn.parameter import Parameter
 from transformers import PretrainedConfig
@@ -49,6 +46,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 
 @torch.compile
 def layer_norm_func(hidden_states, weight, variance_epsilon):

python/sglang/srt/models/dbrx.py

@@ -5,10 +5,6 @@ from typing import Optional
 
 import torch
 import torch.nn as nn
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
-from sglang.srt.models.dbrx_config import DbrxConfig
 from vllm.model_executor.layers.fused_moe import fused_moe
 from vllm.model_executor.layers.linear import (
     LinearMethodBase,
@@ -35,6 +31,11 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+from sglang.srt.models.dbrx_config import DbrxConfig
+
 
 class DbrxRouter(nn.Module):
     """A Router implementation for DBRX that returns logits for each expert

python/sglang/srt/models/gemma.py

@@ -4,9 +4,6 @@
 from typing import Optional, Tuple
 
 import torch
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from transformers import PretrainedConfig
 from vllm.config import LoRAConfig
@@ -28,6 +25,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 
 class GemmaMLP(nn.Module):
     def __init__(

python/sglang/srt/models/llama2.py

@@ -1,12 +1,9 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/671af2b1c0b3ed6d856d37c21a561cc429a10701/vllm/model_executor/models/llama.py#L1
 """Inference-only LLaMA model compatible with HuggingFace weights."""
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, Optional, Tuple
 
 import torch
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from transformers import LlamaConfig
 from vllm.model_executor.layers.activation import SiluAndMul
@@ -30,6 +27,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 
 class LlamaMLP(nn.Module):
     def __init__(

python/sglang/srt/models/llava.py

@@ -4,6 +4,15 @@ from typing import List, Optional
 
 import numpy as np
 import torch
+from torch import nn
+from transformers import CLIPVisionModel, LlavaConfig
+from transformers.models.llava.modeling_llava import LlavaMultiModalProjector
+from vllm.model_executor.layers.linear import LinearMethodBase
+from vllm.model_executor.weight_utils import (
+    default_weight_loader,
+    hf_model_weights_iterator,
+)
+
 from sglang.srt.managers.router.infer_batch import ForwardMode
 from sglang.srt.managers.router.model_runner import InputMetadata
 from sglang.srt.mm_utils import (
@@ -12,14 +21,6 @@ from sglang.srt.mm_utils import (
     unpad_image_shape,
 )
 from sglang.srt.models.llama2 import LlamaForCausalLM
-from torch import nn
-from transformers import CLIPVisionModel, LlamaConfig, LlavaConfig
-from transformers.models.llava.modeling_llava import LlavaMultiModalProjector
-from vllm.model_executor.layers.linear import LinearMethodBase
-from vllm.model_executor.weight_utils import (
-    default_weight_loader,
-    hf_model_weights_iterator,
-)
 
 
 class LlavaLlamaForCausalLM(nn.Module):

python/sglang/srt/models/mixtral.py

@@ -1,14 +1,11 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/d0215a58e78572d91dadafe9d832a2db89b09a13/vllm/model_executor/models/mixtral.py#L1
 """Inference-only Mixtral model."""
-from typing import List, Optional, Tuple
+from typing import Optional
 
 import numpy as np
 import torch
 import torch.nn.functional as F
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from transformers import MixtralConfig
 from vllm.model_executor.layers.layernorm import RMSNorm
@@ -35,6 +32,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 
 class MixtralMLP(nn.Module):
     def __init__(

python/sglang/srt/models/qwen.py

@@ -1,9 +1,6 @@
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, Optional
 
 import torch
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from transformers import PretrainedConfig
 from vllm.model_executor.layers.activation import SiluAndMul
@@ -27,6 +24,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 
 class QWenMLP(nn.Module):
     def __init__(

python/sglang/srt/models/qwen2.py

@@ -1,12 +1,9 @@
 # Adapted from llama2.py
 # Modify details for the adaptation of Qwen2 model.
 """Inference-only Qwen2 model compatible with HuggingFace weights."""
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, Optional, Tuple
 
 import torch
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.layernorm import RMSNorm
@@ -29,6 +26,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 Qwen2Config = None
 
 

python/sglang/srt/models/stablelm.py

@@ -5,9 +5,6 @@ model compatible with HuggingFace weights."""
 from typing import Optional, Tuple
 
 import torch
-from sglang.srt.layers.logits_processor import LogitsProcessor
-from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.router.model_runner import InputMetadata
 from torch import nn
 from transformers import PretrainedConfig
 from vllm.model_executor.layers.activation import SiluAndMul
@@ -30,6 +27,10 @@ from vllm.model_executor.weight_utils import (
     hf_model_weights_iterator,
 )
 
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+
 
 class StablelmMLP(nn.Module):
     def __init__(

python/sglang/srt/models/yivl.py

@@ -1,21 +1,20 @@
 """Inference-only Yi-VL model."""
 
-import os
-from typing import List, Optional
+from typing import Optional
 
 import torch
 import torch.nn as nn
-from sglang.srt.models.llava import (
-    LlavaLlamaForCausalLM,
-    clip_vision_embed_forward,
-    monkey_path_clip_vision_embed_forward,
-)
 from transformers import CLIPVisionModel, LlavaConfig
 from vllm.model_executor.weight_utils import (
     default_weight_loader,
     hf_model_weights_iterator,
 )
 
+from sglang.srt.models.llava import (
+    LlavaLlamaForCausalLM,
+    monkey_path_clip_vision_embed_forward,
+)
+
 
 class YiVLForCausalLM(LlavaLlamaForCausalLM):
     def __init__(self, *args, **kwargs):