From a8820c1a234e44ad45a3e51f4b737147bbeb81bd Mon Sep 17 00:00:00 2001
From: Divakar Verma <divakar.verma@amd.com>
Date: Tue, 18 Feb 2025 18:00:52 -0600
Subject: [PATCH] resolve diff for mixtral8x7B

Signed-off-by: Divakar Verma <divakar.verma@amd.com>
---
 .../E=8,N=14336,device_name=AMD_Instinct_MI300X.json      | 8 ++++----
 .../E=8,N=1792,device_name=AMD_Instinct_MI300X.json       | 2 +-
 .../E=8,N=3584,device_name=AMD_Instinct_MI300X.json       | 4 ++--
 .../E=8,N=7168,device_name=AMD_Instinct_MI300X.json       | 6 +++---
 4 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
index 022d5ece7f87..66f9106bd1be 100644
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
@@ -45,8 +45,8 @@
     },
     "16": {
         "BLOCK_SIZE_M": 16,
-        "BLOCK_SIZE_N": 64,
-        "BLOCK_SIZE_K": 64,
+        "BLOCK_SIZE_N": 16,
+        "BLOCK_SIZE_K": 256,
         "GROUP_SIZE_M": 1,
         "num_warps": 2,
         "num_stages": 2,
@@ -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,
diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
index de2320e4b28c..1b46cb571651 100644
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
+++ b/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,
diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
index 5a3f415d5414..ed5b655d8993 100644
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
@@ -45,8 +45,8 @@
     },
     "16": {
         "BLOCK_SIZE_M": 16,
-        "BLOCK_SIZE_N": 64,
-        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 256,
         "GROUP_SIZE_M": 1,
         "num_warps": 2,
         "num_stages": 2,
diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
index 8dec5e3afaba..822f04e33e87 100644
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
+++ b/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,
@@ -128,7 +128,7 @@
         "num_warps": 8,
         "num_stages": 2,
         "waves_per_eu": 0,
-        "matrix_instr_nonkdim": 16,
+        "matrix_instr_nonkdim": 32,
         "kpack": 2
     },
     "512": {