From fc86b4426b18c68c79531708e1ab28e184ca8e46 Mon Sep 17 00:00:00 2001 From: chenjw28 <792430652@qq.com> Date: Thu, 2 Apr 2026 14:57:07 +0800 Subject: [PATCH] x --- .gitignore | 2 ++ config.json | 11 ++++++++--- 2 files changed, 10 insertions(+), 3 deletions(-) create mode 100644 .gitignore diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..ae945fa --- /dev/null +++ b/.gitignore @@ -0,0 +1,2 @@ +.idea/ +venv/ diff --git a/config.json b/config.json index 7283fa4..61a0456 100644 --- a/config.json +++ b/config.json @@ -87,10 +87,11 @@ }, "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": { "local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", - "ctx": "262144", + "dtype": "auto", + "ctx": "131072", "trust_remote": true, "valid_tp": [2], - "max_num_seqs": "4", + "max_num_seqs": "6", "max_tokens": "4096", "gpu_util": "0.90", "enforce_eager": false, @@ -101,7 +102,11 @@ "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" + "hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", + "temperature": 0.7, + "top_p": 0.8, + "top_k": 20, + "min_p": 0 } } }