diff --git a/config.json b/config.json index c88c4fc..ed06f1f 100644 --- a/config.json +++ b/config.json @@ -29,12 +29,12 @@ "hf_model_id": "RedHatAI/Qwen3-14B-FP8-dynamic" }, "Qwen3-Coder-30B-A3B-Instruct-GPTQ-4bit": { - "ctx": "131072", // 128K + "ctx": "131072", "trust_remote": true, "valid_tp": [1, 2], - "max_num_seqs": "32", // 适当减少以节省显存 + "max_num_seqs": "32", "max_tokens": "65536", - "gpu_util": "0.95", // 适当降低以预留更多显存 + "gpu_util": "0.95", "hf_model_id": "cpatonn/Qwen3-Coder-30B-A3B-Instruct-GPTQ-4bit" }, "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": {