[Platform] Move get_punica_wrapper() function to Platform (#11516)

Signed-off-by: Shanshan Shen <467638484@qq.com>

vllm/lora/punica_wrapper/punica_selector.py

@@ -1,5 +1,6 @@
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
+from vllm.utils import resolve_obj_by_qualname
 
 from .punica_base import PunicaWrapperBase
 
@@ -7,20 +8,11 @@ logger = init_logger(__name__)
 
 
 def get_punica_wrapper(*args, **kwargs) -> PunicaWrapperBase:
-    if current_platform.is_cuda_alike():
+    punica_wrapper_qualname = current_platform.get_punica_wrapper()
-        # Lazy import to avoid ImportError
+    punica_wrapper_cls = resolve_obj_by_qualname(punica_wrapper_qualname)
-        from vllm.lora.punica_wrapper.punica_gpu import PunicaWrapperGPU
+    punica_wrapper = punica_wrapper_cls(*args, **kwargs)
-        logger.info_once("Using PunicaWrapperGPU.")
+    assert punica_wrapper is not None, \
-        return PunicaWrapperGPU(*args, **kwargs)
+        "the punica_wrapper_qualname(" + punica_wrapper_qualname + ") is wrong."
-    elif current_platform.is_cpu():
+    logger.info_once("Using " + punica_wrapper_qualname.rsplit(".", 1)[1] +
-        # Lazy import to avoid ImportError
+                     ".")
-        from vllm.lora.punica_wrapper.punica_cpu import PunicaWrapperCPU
+    return punica_wrapper
-        logger.info_once("Using PunicaWrapperCPU.")
-        return PunicaWrapperCPU(*args, **kwargs)
-    elif current_platform.is_hpu():
-        # Lazy import to avoid ImportError
-        from vllm.lora.punica_wrapper.punica_hpu import PunicaWrapperHPU
-        logger.info_once("Using PunicaWrapperHPU.")
-        return PunicaWrapperHPU(*args, **kwargs)
-    else:
-        raise NotImplementedError

vllm/platforms/cpu.py

@@ -109,3 +109,7 @@ class CpuPlatform(Platform):
     def is_pin_memory_available(cls) -> bool:
         logger.warning("Pin memory is not supported on CPU.")
         return False
+
+    @classmethod
+    def get_punica_wrapper(cls) -> str:
+        return "vllm.lora.punica_wrapper.punica_cpu.PunicaWrapperCPU"

vllm/platforms/cuda.py

@@ -218,6 +218,10 @@ class CudaPlatformBase(Platform):
         logger.info("Using Flash Attention backend.")
         return "vllm.attention.backends.flash_attn.FlashAttentionBackend"
 
+    @classmethod
+    def get_punica_wrapper(cls) -> str:
+        return "vllm.lora.punica_wrapper.punica_gpu.PunicaWrapperGPU"
+
 
 # NVML utils
 # Note that NVML is not affected by `CUDA_VISIBLE_DEVICES`,

vllm/platforms/hpu.py

@@ -63,3 +63,7 @@ class HpuPlatform(Platform):
     def is_pin_memory_available(cls):
         logger.warning("Pin memory is not supported on HPU.")
         return False
+
+    @classmethod
+    def get_punica_wrapper(cls) -> str:
+        return "vllm.lora.punica_wrapper.punica_hpu.PunicaWrapperHPU"

vllm/platforms/interface.py

@@ -276,6 +276,13 @@ class Platform:
             return False
         return True
 
+    @classmethod
+    def get_punica_wrapper(cls) -> str:
+        """
+        Return the punica wrapper for current platform.
+        """
+        raise NotImplementedError
+
 
 class UnspecifiedPlatform(Platform):
     _enum = PlatformEnum.UNSPECIFIED

vllm/platforms/rocm.py

@@ -153,3 +153,7 @@ class RocmPlatform(Platform):
                 "Using AWQ quantization with ROCm, but VLLM_USE_TRITON_AWQ"
                 " is not set, enabling VLLM_USE_TRITON_AWQ.")
         envs.VLLM_USE_TRITON_AWQ = True
+
+    @classmethod
+    def get_punica_wrapper(cls) -> str:
+        return "vllm.lora.punica_wrapper.punica_gpu.PunicaWrapperGPU"