diff --git a/config.json b/config.json index 4ca2f46..abbb6bc 100644 --- a/config.json +++ b/config.json @@ -20,7 +20,7 @@ "revision": "", "models": { "default": "Qwen3.5-35B-A3B-GPTQ-Int4", - "selected": "Qwen3.5-35B-A3B-FP8", + "selected": "Qwen3.5-27B-FP8", "profiles": { "GLM-4.7-Flash-AWQ": { "local_path": "GLM-4.7-Flash-AWQ", @@ -44,6 +44,7 @@ "valid_tp": [1, 2], "max_num_seqs": "4", "max_tokens": "8192", + "enforce_eager": false, "gpu_util": "0.94", "tool_call_parser": "qwen3_coder", "reasoning_parser": "qwen3",