From d84cdc00e532043edb173542d7af121ecc88653d Mon Sep 17 00:00:00 2001
From: SZCJW <792430652@qq.com>
Date: Sun, 29 Mar 2026 17:41:28 +0800
Subject: [PATCH] x

---
 config.json | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/config.json b/config.json
index c2c1c53..6f82e80 100644
--- a/config.json
+++ b/config.json
@@ -24,7 +24,7 @@
         "ctx": "24576",
         "trust_remote": true,
         "valid_tp": [2],
-        "max_num_seqs": "32",
+        "max_num_seqs": "8",
         "max_tokens": "16384",
         "gpu_util": "0.98",
         "enforce_eager": false,
@@ -41,7 +41,7 @@
         "ctx": "32768",
         "trust_remote": true,
         "valid_tp": [1, 2],
-        "max_num_seqs": "64",
+        "max_num_seqs": "8",
         "max_tokens": "32768",
         "gpu_util": "0.98",
         "tool_call_parser": "qwen3_xml",
@@ -54,7 +54,7 @@
         "ctx": "65536",
         "trust_remote": true,
         "valid_tp": [1, 2],
-        "max_num_seqs": "64",
+        "max_num_seqs": "8",
         "max_tokens": "32768",
         "gpu_util": "0.98",
         "tool_call_parser": "qwen3_xml",
@@ -69,9 +69,9 @@
         "ctx": "65536",
         "trust_remote": true,
         "valid_tp": [1, 2],
-        "max_num_seqs": "64",
-        "max_tokens": "32768",
-        "gpu_util": "0.98",
+        "max_num_seqs": "8",
+        "max_tokens": "4096",
+        "gpu_util": "0.92",
         "tool_call_parser": "qwen3_xml",
         "enable_auto_tool_choice": true,
         "served_model_name": "Qwen3.5-35B-A3B-GPTQ-Int4",
@@ -83,7 +83,7 @@
         "ctx": "65536",
         "trust_remote": true,
         "valid_tp": [1, 2],
-        "max_num_seqs": "64",
+        "max_num_seqs": "8",
         "max_tokens": "32768",
         "gpu_util": "0.98",
         "tool_call_parser": "qwen3_xml",