From d87daadc2f6b68d64b42db78a12c1f254921b410 Mon Sep 17 00:00:00 2001
From: SZCJW <792430652@qq.com>
Date: Fri, 17 Apr 2026 00:09:11 +0800
Subject: [PATCH] x

---
 config.json        | 4 ++--
 docker-compose.yml | 4 +++-
 2 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/config.json b/config.json
index 5a583a6..ebd41ce 100644
--- a/config.json
+++ b/config.json
@@ -42,10 +42,10 @@
         "ctx": "131072",
         "trust_remote": true,
         "valid_tp": [1, 2],
-        "max_num_seqs": 12,
+        "max_num_seqs": 32,
         "max_tokens": "16384",
         "enforce_eager": false,
-        "gpu_util": "0.94",
+        "gpu_util": "0.90",
         "kv_cache_dtype": "fp8",
         "enable_prefix_caching": true,
         "max_num_batched_tokens": 16384,
diff --git a/docker-compose.yml b/docker-compose.yml
index d742e23..1f374be 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -24,7 +24,9 @@ services:
       # PYTHONUNBUFFERED: "1"
       # # 启用 AITER 加速（R9700 属于 gfx1201 架构）
       VLLM_ROCM_USE_AITER: "1"
-      VLLM_ROCM_USE_AITER_MHA: "0"
+      VLLM_ROCM_USE_AITER_MHA: "1"
+      VLLM_V1_USE_PREFILL_DECODE_ATTENTION: "0"
+      NCCL_MIN_NCHANNELS: "112"
       # # 增强 All-Reduce 性能（双卡通信优化）
       VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT4"
     devices: