Merge branch 'v0.11.0-dev-tmp' into 'v0.11.0-dev'

The gfx928 architecture forces the use of the Triton gemm.

See merge request dcutoolkit/deeplearing/vllm!428

vllm/envs.py

@@ -6,6 +6,7 @@ import json
 import os
 import sys
 import tempfile
+import torch
 from typing import TYPE_CHECKING, Any, Callable, Literal, Optional, Union
 
 if TYPE_CHECKING:
@@ -1704,7 +1705,9 @@ environment_variables: dict[str, Callable[[], Any]] = {
     # cutlass: 2 (will remove in the future)
     # blaslt: 3 (default)
     # rocblas: others
-    "VLLM_W8A8_BACKEND": lambda: int(os.getenv("VLLM_W8A8_BACKEND", "3")),
+    "VLLM_W8A8_BACKEND": lambda: int(
+        1 if "gfx928" in torch.cuda.get_device_properties("cuda").gcnArchName.split(':')[0] else os.getenv("VLLM_W8A8_BACKEND", "3")),
+
     # Force using Triton MoE path (disable Marlin W16A16 MoE).
     "VLLM_USE_MOE_W16A16_TRITON":
         lambda: (os.environ.get("VLLM_USE_MOE_W16A16_TRITON", "0").lower() in

vllm/model_executor/layers/quantization/slimquant_w4a8.py

@@ -92,8 +92,8 @@ class SlimQuantW4A8Int8LinearMethod(LinearMethodBase):
 
     def __init__(self, quantization_config: SlimQuantW4A8Int8Config):
         self.quantization_config = quantization_config
-        self.tritonsingleton= W8a8GetCacheJSON()
-        self.w8a8_strategy=int(os.getenv('W8A8_SUPPORT_METHODS', '1'))
+        self.tritonsingleton = W8a8GetCacheJSON()
+        self.w8a8_strategy = envs.VLLM_W8A8_BACKEND
 
     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
         n=layer.weight.shape[0]