Clean up imports (#5467)

python/sglang/srt/lora/backend/__init__.py

@@ -1,25 +0,0 @@
-from sglang.srt.lora.backend.base_backend import BaseLoRABackend
-
-
-def get_backend_from_name(name: str) -> BaseLoRABackend:
-    """
-    Get corresponding backend class from backend's name
-    """
-    if name == "triton":
-        from sglang.srt.lora.backend.triton_backend import TritonLoRABackend
-
-        return TritonLoRABackend
-    elif name == "flashinfer":
-        from sglang.srt.lora.backend.flashinfer_backend import FlashInferLoRABackend
-
-        return FlashInferLoRABackend
-    else:
-        raise ValueError(f"Invalid backend: {name}")
-
-
-__all__ = [
-    "BaseLoRABackend",
-    "FlashInferLoRABackend",
-    "TritonLoRABackend",
-    "get_backend_from_name",
-]

python/sglang/srt/lora/backend/base_backend.py

@@ -75,7 +75,7 @@ class BaseLoRABackend:
         qkv_lora_a: torch.Tensor,
         qkv_lora_b: Union[torch.Tensor, Tuple[torch.Tensor]],
         *args,
-        **kwargs
+        **kwargs,
     ) -> torch.Tensor:
         """Run the lora pass for QKV Layer.
 
@@ -98,7 +98,7 @@ class BaseLoRABackend:
         gate_up_lora_a: torch.Tensor,
         gate_up_lora_b: Union[torch.Tensor, Tuple[torch.Tensor]],
         *args,
-        **kwargs
+        **kwargs,
     ) -> torch.Tensor:
         """Run the lora pass for gate_up_proj, usually attached to MergedColumnParallelLayer.
 
@@ -115,3 +115,19 @@ class BaseLoRABackend:
 
     def set_batch_info(self, batch_info: LoRABatchInfo):
         self.batch_info = batch_info
+
+
+def get_backend_from_name(name: str) -> BaseLoRABackend:
+    """
+    Get corresponding backend class from backend's name
+    """
+    if name == "triton":
+        from sglang.srt.lora.backend.triton_backend import TritonLoRABackend
+
+        return TritonLoRABackend
+    elif name == "flashinfer":
+        from sglang.srt.lora.backend.flashinfer_backend import FlashInferLoRABackend
+
+        return FlashInferLoRABackend
+    else:
+        raise ValueError(f"Invalid backend: {name}")

python/sglang/srt/lora/backend/flashinfer_backend.py

@@ -2,7 +2,7 @@ from typing import Tuple
 
 import torch
 
-from sglang.srt.lora.backend import BaseLoRABackend
+from sglang.srt.lora.backend.base_backend import BaseLoRABackend
 from sglang.srt.lora.utils import LoRABatchInfo
 from sglang.srt.utils import is_flashinfer_available
 

python/sglang/srt/lora/backend/triton_backend.py

@@ -1,6 +1,6 @@
 import torch
 
-from sglang.srt.lora.backend import BaseLoRABackend
+from sglang.srt.lora.backend.base_backend import BaseLoRABackend
 from sglang.srt.lora.triton_ops import (
     gate_up_lora_b_fwd,
     qkv_lora_b_fwd,

python/sglang/srt/lora/layers.py

@@ -16,7 +16,7 @@ from sglang.srt.layers.linear import (
     RowParallelLinear,
 )
 from sglang.srt.layers.vocab_parallel_embedding import VocabParallelEmbedding
-from sglang.srt.lora.backend import BaseLoRABackend
+from sglang.srt.lora.backend.base_backend import BaseLoRABackend
 
 
 class BaseLayerWithLoRA(nn.Module):

python/sglang/srt/lora/lora.py

@@ -27,7 +27,7 @@ from torch import nn
 
 from sglang.srt.configs.load_config import LoadConfig
 from sglang.srt.hf_transformers_utils import AutoConfig
-from sglang.srt.lora.backend import BaseLoRABackend
+from sglang.srt.lora.backend.base_backend import BaseLoRABackend
 from sglang.srt.lora.lora_config import LoRAConfig
 from sglang.srt.model_loader.loader import DefaultModelLoader
 

python/sglang/srt/lora/lora_manager.py

@@ -22,7 +22,7 @@ import torch
 
 from sglang.srt.configs.load_config import LoadConfig
 from sglang.srt.hf_transformers_utils import AutoConfig
-from sglang.srt.lora.backend import BaseLoRABackend, get_backend_from_name
+from sglang.srt.lora.backend.base_backend import BaseLoRABackend, get_backend_from_name
 from sglang.srt.lora.layers import BaseLayerWithLoRA, get_lora_layer
 from sglang.srt.lora.lora import LoRAAdapter
 from sglang.srt.lora.lora_config import LoRAConfig