[fix]修复模型注册失败及其他报错

vllm/model_executor/layers/fused_moe/fused_moe.py

@@ -1174,7 +1174,7 @@ def invoke_fused_moe_kernel(A: torch.Tensor,
             expert_ids,
             num_tokens_post_padded,
             B.shape[1] if not use_nn_moe else B.shape[2],
-            B.shape[2],
+            A.shape[1],
             EM,
             topk_ids.numel(),
             A.stride(0),
@@ -1551,7 +1551,7 @@ def inplace_fused_experts(hidden_states: torch.Tensor,
                           w2: torch.Tensor,
                           topk_weights: torch.Tensor,
                           topk_ids: torch.Tensor,
-                          activation: str = "silu",
+                          activation: Optional[str] = None,
                           use_fp8_w8a8: bool = False,
                           use_int8_w8a8: bool = False,
                           use_int8_w8a16: bool = False,
@@ -1583,7 +1583,7 @@ def inplace_fused_experts_fake(
         w2: torch.Tensor,
         topk_weights: torch.Tensor,
         topk_ids: torch.Tensor,
-        activation: str = "silu",
+        activation: Optional[str] = None,
         use_fp8_w8a8: bool = False,
         use_int8_w8a8: bool = False,
         use_int8_w8a16: bool = False,
@@ -1618,7 +1618,7 @@ def outplace_fused_experts(
         w2: torch.Tensor,
         topk_weights: torch.Tensor,
         topk_ids: torch.Tensor,
-        activation: str = "silu",
+        activation: Optional[str] = None,
         use_fp8_w8a8: bool = False,
         use_int8_w8a8: bool = False,
         use_int8_w8a16: bool = False,
@@ -1651,7 +1651,7 @@ def outplace_fused_experts_fake(
         w2: torch.Tensor,
         topk_weights: torch.Tensor,
         topk_ids: torch.Tensor,
-        activation: str = "silu",
+        activation: Optional[str] = None,
         use_fp8_w8a8: bool = False,
         use_int8_w8a8: bool = False,
         use_int8_w8a16: bool = False,

vllm/platforms/rocm.py

@@ -62,12 +62,12 @@ _ROCM_PARTIALLY_SUPPORTED_MODELS: Dict[str, str] = {
 }
 
 # Prevent use of clashing `{CUDA/HIP}_VISIBLE_DEVICES``
-if "HIP_VISIBLE_DEVICES" in os.environ:
-    val = os.environ["HIP_VISIBLE_DEVICES"]
-    if cuda_val := os.environ.get("CUDA_VISIBLE_DEVICES", None):
-        assert val == cuda_val
-    else:
-        os.environ["CUDA_VISIBLE_DEVICES"] = val
+# if "HIP_VISIBLE_DEVICES" in os.environ:
+#     val = os.environ["HIP_VISIBLE_DEVICES"]
+#     if cuda_val := os.environ.get("CUDA_VISIBLE_DEVICES", None):
+#         assert val == cuda_val
+#     else:
+#         os.environ["CUDA_VISIBLE_DEVICES"] = val
 
 # AMDSMI utils
 # Note that NVML is not affected by `{CUDA/HIP}_VISIBLE_DEVICES`,