From 2d6c3dcb825b7f2aa5d5c994c48a6a66e178cda5 Mon Sep 17 00:00:00 2001
From: chenjw28 <792430652@qq.com>
Date: Mon, 18 May 2026 15:50:14 +0800
Subject: [PATCH] x

---
 config.json | 125 +---------------------------------------------------
 1 file changed, 1 insertion(+), 124 deletions(-)

diff --git a/config.json b/config.json
index b48d7f6..d818156 100644
--- a/config.json
+++ b/config.json
@@ -93,131 +93,8 @@
           "method": "dflash",
           "model": "Qwen3.6-27B-DFlash",
           "num_speculative_tokens": 8,
-          "draft_tensor_parallel_size": 1
+          "draft_tensor_parallel_size": 2
         }
-      },
-      "Qwen3.5-35B-A3B-FP8": {
-        "local_path": "Qwen3.5-35B-A3B-FP8",
-        "dtype": "float16",
-        "ctx": "131072",
-        "enforce_eager": false,
-        "trust_remote": true,
-        "valid_tp": [
-          1,
-          2
-        ],
-        "max_num_seqs": "8",
-        "max_tokens": "16384",
-        "gpu_util": "0.94",
-        "kv_cache_dtype": "fp8",
-        "enable_prefix_caching": true,
-        "max_num_batched_tokens": 16384,
-        "tool_call_parser": "qwen3_coder",
-        "reasoning_parser": "qwen3",
-        "enable_auto_tool_choice": true,
-        "served_model_name": "Qwen3.5-35B-A3B-FP8",
-        "hf_model_id": "Qwen/Qwen3.5-35B-A3B-FP8"
-      },
-      "Qwen3.5-35B-A3B-GPTQ-Int4": {
-        "local_path": "Qwen3.5-35B-A3B-GPTQ-Int4",
-        "dtype": "float16",
-        "quantization": "gptq",
-        "ctx": "131072",
-        "trust_remote": true,
-        "valid_tp": [
-          2
-        ],
-        "max_num_seqs": "6",
-        "max_tokens": "4096",
-        "gpu_util": "0.90",
-        "enforce_eager": false,
-        "tool_call_parser": "qwen3_coder",
-        "reasoning_parser": "qwen3",
-        "enable_auto_tool_choice": true,
-        "served_model_name": "Qwen3.5-35B-A3B-GPTQ-Int4",
-        "hf_model_id": "Qwen/Qwen3.5-35B-A3B-GPTQ-Int4",
-        "env": {
-          "VLLM_RPC_TIMEOUT": "300"
-        }
-      },
-      "Qwen3.5-27B-FP8": {
-        "local_path": "Qwen3.5-27B-FP8",
-        "dtype": "auto",
-        "ctx": "131072",
-        "trust_remote": true,
-        "valid_tp": [
-          1,
-          2
-        ],
-        "max_num_seqs": 32,
-        "max_tokens": "16384",
-        "enforce_eager": false,
-        "gpu_util": "0.90",
-        "kv_cache_dtype": "fp8",
-        "enable_prefix_caching": true,
-        "max_num_batched_tokens": 16384,
-        "tool_call_parser": "qwen3_coder",
-        "reasoning_parser": "qwen3",
-        "enable_auto_tool_choice": true,
-        "served_model_name": "Qwen3.5-27B-FP8",
-        "hf_model_id": "RedHatAI/Qwen3.5-27B-FP8-dynamic"
-      },
-      "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": {
-        "local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
-        "dtype": "auto",
-        "ctx": "131072",
-        "trust_remote": true,
-        "valid_tp": [
-          1,
-          2
-        ],
-        "max_num_seqs": "8",
-        "max_tokens": "65536",
-        "max_num_batched_tokens": 16384,
-        "gpu_util": "0.94",
-        "enforce_eager": false,
-        "enable_prefix_caching": true,
-        "tool_call_parser": "qwen3_coder",
-        "reasoning_parser": "qwen3",
-        "enable_auto_tool_choice": true,
-        "served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
-        "hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit"
-      },
-      "GLM-4.7-Flash-AWQ": {
-        "local_path": "GLM-4.7-Flash-AWQ",
-        "ctx": "32768",
-        "trust_remote": true,
-        "valid_tp": [
-          1,
-          2
-        ],
-        "max_num_seqs": "16",
-        "max_tokens": "32768",
-        "gpu_util": "0.98",
-        "tool_call_parser": "qwen3_coder",
-        "reasoning_parser": "qwen3",
-        "enable_auto_tool_choice": true,
-        "served_model_name": "GLM-4.7-Flash-AWQ",
-        "hf_model_id": "THUDM/GLM-4.7-Flash-AWQ"
-      },
-      "Gemma-4-26B-A4B": {
-        "local_path": "gemma-4-26B-A4B",
-        "dtype": "auto",
-        "ctx": "65536",
-        "trust_remote": true,
-        "valid_tp": [
-          1,
-          2
-        ],
-        "max_num_seqs": 16,
-        "max_tokens": "8192",
-        "enforce_eager": false,
-        "gpu_util": "0.85",
-        "kv_cache_dtype": "fp8",
-        "enable_prefix_caching": true,
-        "max_num_batched_tokens": 16384,
-        "served_model_name": "Gemma-4-26B-A4B",
-        "hf_model_id": "google/gemma-4-26B-A4B"
       }
     }
   }