resolve diff for mixtral8x7B configs (ROCm#437)

Signed-off-by: Divakar Verma <[email protected]>

Author: divakar-amd

vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json

@@ -96,7 +96,7 @@
         "num_stages": 2,
         "waves_per_eu": 0,
         "matrix_instr_nonkdim": 16,
-        "kpack": 1
+        "kpack": 2
     },
     "96": {
         "BLOCK_SIZE_M": 32,
@@ -123,7 +123,7 @@
     "256": {
         "BLOCK_SIZE_M": 128,
         "BLOCK_SIZE_N": 128,
-        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_K": 64,
         "GROUP_SIZE_M": 4,
         "num_warps": 8,
         "num_stages": 2,

vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json

@@ -8,7 +8,7 @@
         "num_stages": 2,
         "waves_per_eu": 0,
         "matrix_instr_nonkdim": 16,
-        "kpack": 1
+        "kpack": 2
     },
     "2": {
         "BLOCK_SIZE_M": 16,

vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json

@@ -8,7 +8,7 @@
         "num_stages": 2,
         "waves_per_eu": 0,
         "matrix_instr_nonkdim": 16,
-        "kpack": 1
+        "kpack": 2
     },
     "2": {
         "BLOCK_SIZE_M": 16,
@@ -63,7 +63,7 @@
         "num_stages": 2,
         "waves_per_eu": 0,
         "matrix_instr_nonkdim": 16,
-        "kpack": 2
+        "kpack": 1
     },
     "32": {
         "BLOCK_SIZE_M": 16,