增加线性int8 gemm配置

vllm/model_executor/layers/quantization/utils/int8_utils.py

 # SPDX-License-Identifier: Apache-2.0
 
-# Adapted from https://github.com/sgl-project/sglang/pull/3730
 import functools
 import json
 import logging
@@ -336,23 +335,60 @@ def w8a8_block_int8_matmul(
     C_shape = A.shape[:-1] + (N,)
     C = A.new_empty(C_shape, dtype=output_dtype)
 
-    configs = get_w8a8_block_int8_configs(N, K, block_size[0], block_size[1])
-    if configs:
-        # If an optimal configuration map has been found, look up the
-        # optimal config
-        config = configs[min(configs.keys(), key=lambda x: abs(x - M))]
-    else:
+    #configs = get_w8a8_block_int8_configs(N, K, block_size[0], block_size[1])
+    #if configs:
+    #    # If an optimal configuration map has been found, look up the
+    #    # optimal config
+    #    config = configs[min(configs.keys(), key=lambda x: abs(x - M))]
+    #else:
         # Default config
         # Block-wise quant: BLOCK_SIZE_K must be divisable by block_size[1]
         #print("block_size[0]:{},block_size[1]:{}".format(block_size[0],block_size[1]))
+    #    config = {
+    #        "BLOCK_SIZE_M": 32, #64
+    #        "BLOCK_SIZE_N": block_size[0],
+    #        "BLOCK_SIZE_K": block_size[1],
+    #        "GROUP_SIZE_M": 32,
+    #        "num_warps": 4,
+    #        "num_stages": 3,
+    #    }
+
+    if M<=64:
         config = {
-            "BLOCK_SIZE_M": 32, #64
-            "BLOCK_SIZE_N": block_size[0],
-            "BLOCK_SIZE_K": block_size[1],
-            "GROUP_SIZE_M": 32,
+            "BLOCK_SIZE_M": 16, #64
+            "BLOCK_SIZE_N": 64,
+            "BLOCK_SIZE_K": 128,
+            "GROUP_SIZE_M": 2,
             "num_warps": 4,
-            "num_stages": 3,
+            "num_stages": 0,
         }
+    elif M<128:
+        config = {
+            "BLOCK_SIZE_M": 32, #64
+            "BLOCK_SIZE_N": 64,
+            "BLOCK_SIZE_K": 128,
+            "GROUP_SIZE_M": 2,
+            "num_warps": 4,
+            "num_stages": 0,
+        }   
+    elif M<=256:
+        config = {
+            "BLOCK_SIZE_M": 64, #64
+            "BLOCK_SIZE_N": 64,
+            "BLOCK_SIZE_K": 128,
+            "GROUP_SIZE_M": 2,
+            "num_warps": 4,
+            "num_stages": 0,
+        }                     
+    else :
+        config = {
+            "BLOCK_SIZE_M": 64, #64
+            "BLOCK_SIZE_N": 128,
+            "BLOCK_SIZE_K": 128,
+            "GROUP_SIZE_M": 8,
+            "num_warps": 8,
+            "num_stages": 0,
+        }     
 
     def grid(META):
         return (
@@ -514,4 +550,4 @@ def block_dequant(
                 i * block_k : min((i + 1) * block_k, k),
             ] *= x_s[j][i]
 
-    return x_dq_block
\ No newline at end of file
+    return x_dq_block