glm-0414原始模型融合出现问题

[yawzhe]

1.使用huggface上下载的模型和训练好的模型进行融合UDA_VISIBLE_DEVICES=0 mergekit-yaml /workspace/yaml/linear.yml /merged_model_dir/glm4-0414_32_516/chekpoint-1482-55-merged --trust-remote-code --copy-tokenizer 命令如上 是需要修改东西吗 使用上面的进行融合后 发现vllm推理为空， 测试命令 curl -s http://localhost:8411/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer your_api_key_here" - d "{"model": "", "messages": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Your prompt here"}], "temperature
": 0.7, "top_p": 0.8, "repetition_penalty": 1.05, "max_tokens": 512}" 返回结果 {"id":"chatcmpl-3639f9f3ab9c485d8afa4f0a105e5052","object":"chat.completion","created":1747711875,"model":"/model_path","choices":[{"index":0,"message":{"role":"assistant","reasoning_content":null,"content":"","tool_calls":[]},"logprobs":null,"finish_reason":"length","stop_reason":null}],"usage":{"prompt_tokens":16,"total_tokens":528,"completion_tokens":512,"prompt_tokens_details":null},"prompt_logprobs":null} ，之前使用原版或者训练好的模型进行vllm推理都没有问题. 模型参数融合models:

• model: /model_merged/glm4-0414 parameters: weight: 0.5
• model: /model_merged/glm4-0414_sft/checkpoint-1482 parameters: weight: 0.5 merge_method: linear dtype: float16