使用huggingface.js调用textGeneration时出现Error fetching blob错误求助

问题背景 #

之前基于OpenAI GPT API搭建的项目，转用huggingface.js API时，将原postChatGPTMessage替换为postHFMessage函数后，调用出现Error: An error occurred while fetching the blob错误。

原代码 #

const postHFMessage = async (contextMessage, conversation) => {
    // build the conversation for context
    const { textGeneration } = await import('@huggingface/inference');
    
    // Prepare the payload for the text generation request
    const payload = {
        inputs: contextMessage,
        parameters: {
            max_length: 100,
            temperature: 0.1,
            // Add any other parameters as needed
            },
        };

    // Call the textGeneration function with your payload
    try {
        const response = await textGeneration({
            accessToken: process.env.HUGGINGFACE_API_KEY,
            model: USED_MODEL,
            inputs: payload.inputs,
            parameters: payload.parameters,
        });
        if (response) {
            const responseData = await response.json();
            console.log('Received response:', responseData);
            return response;
        }
        else {
            console.log("no response from textGeneration");
        }
    }
    catch(error) {
        console.log("Could not get response from textGeneration: ", error);
    }
};

报错信息 #

Could not get response from textGeneration:  Error: An error occurred while fetching the blob
    at request (file:///Users/o*******/node_modules/@huggingface/inference/dist/index.js:207:11)

解决方案 #

这个错误既不需要把参数转为Blob，也不是权限问题，核心是错误处理了textGeneration的返回值：huggingface.js的textGeneration方法内部已经完成了响应解析，直接返回JSON格式的结果，不需要再调用.json()方法。

修改后的代码 #

const postHFMessage = async (contextMessage, conversation) => {
    const { textGeneration } = await import('@huggingface/inference');
    
    const payload = {
        inputs: contextMessage,
        parameters: {
            max_length: 100,
            temperature: 0.1,
        },
    };

    try {
        // textGeneration直接返回解析后的响应数据，无需再调用.json()
        const responseData = await textGeneration({
            accessToken: process.env.HUGGINGFACE_API_KEY,
            model: USED_MODEL,
            inputs: payload.inputs,
            parameters: payload.parameters,
        });
        console.log('Received response:', responseData);
        return responseData;
    } catch(error) {
        console.log("Could not get response from textGeneration: ", error);
    }
};

额外验证点 #

• 确认USED_MODEL是支持文本生成任务的模型（如gpt2、distilgpt2、meta-llama/Llama-2-7b-chat-hf等），且你的API密钥拥有该模型的访问权限
• 免费API密钥有调用次数与速率限制，若超出额度也会引发异常，可登录Hugging Face账户查看使用情况

内容的提问来源于stack exchange，提问作者Omer Sela Rothenberg