x

2026-03-29 02:57:21 +08:00
parent 3792606cf9
commit 8802e9b226
4 changed files with 19 additions and 6 deletions
@@ -35,6 +35,9 @@
      "max_num_seqs": "32",
      "max_tokens": "65536",
      "gpu_util": "0.95",
+      "tool_call_parser": "qwen3_xml",
+      "enable_auto_tool_choice": false,
+      "served_model_name": "Qwen3-Coder-30B-A3B-Instruct-GPTQ-4bit",
      "hf_model_id": "cpatonn/Qwen3-Coder-30B-A3B-Instruct-GPTQ-4bit"
    },
    "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": {
@@ -47,6 +50,8 @@
      "enforce_eager": false,
      "env": {"VLLM_USE_TRITON_AWQ": "1"},
      "tool_call_parser": "qwen3_xml",
+      "enable_auto_tool_choice": false,
+      "served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
      "hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit"
    },
    "gemma-3-27b-it-FP8-dynamic": {