From 6574035969eed805cb1c3dec1bb69f4ec8755562 Mon Sep 17 00:00:00 2001 From: SZCJW <792430652@qq.com> Date: Wed, 1 Apr 2026 01:19:14 +0800 Subject: [PATCH] x --- config.json | 36 ++++++++++++++++++------------------ 1 file changed, 18 insertions(+), 18 deletions(-) diff --git a/config.json b/config.json index 315509e..cbd5c9a 100644 --- a/config.json +++ b/config.json @@ -22,24 +22,6 @@ "default": "Qwen3.5-35B-A3B-GPTQ-Int4", "selected": "Qwen3.5-35B-A3B-GPTQ-Int4", "profiles": { - "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": { - "local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", - "ctx": "24576", - "trust_remote": true, - "valid_tp": [2], - "max_num_seqs": "8", - "max_tokens": "16384", - "gpu_util": "0.98", - "enforce_eager": false, - "env": { - "VLLM_USE_TRITON_AWQ": "1" - }, - "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", @@ -102,6 +84,24 @@ "enable_auto_tool_choice": true, "served_model_name": "Qwen3.5-35B-A3B-FP8", "hf_model_id": "Qwen/Qwen3.5-35B-A3B-FP8" + }, + "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": { + "local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", + "ctx": "65536", + "trust_remote": true, + "valid_tp": [2], + "max_num_seqs": "4", + "max_tokens": "8192", + "gpu_util": "0.90", + "enforce_eager": false, + "env": { + "VLLM_USE_TRITON_AWQ": "1" + }, + "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" } } }