多模态图片请求报错 - 最大 20M 限制

[leing2021]

问题描述

使用 CLIProxyAPI 代理服务时，提交带图片的多模态请求会报错。

复现步骤

1. 使用 CC Switch 切换到 CLI Proxy API
2. 发送包含图片的问题（图片大小约 1M）
3. 返回错误：请求大小超过 20M 限制

测试结果

• 官方 API + glm-5 + 1M 图片 → ✅ 正常
• CLI Proxy API + glm-5 + 1M 图片 → ❌ 报 20M 限制错误

请求示例

{
  "model": "glm-5",
  "messages": [{
    "role": "user",
    "content": [
      {"type": "text", "text": "描述这张图片"},
      {"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,..."}}
    ]
  }]
}

请检查请求大小计算的逻辑，可能是代理层在计算请求体大小时存在 bug。

[luispater]

Thanks for the report. I reviewed the current repository and could not find a matching code path that enforces a 20 MB request limit for OpenAI-style multimodal requests. In the OpenAI HTTP path, the handler reads the raw request body directly, the OpenAI-compatible executor forwards the payload to the upstream provider, and the OpenAI-to-OpenAI translator only rewrites the model field. For a roughly 1 MiB image, the base64-encoded JSON payload is still around 1.4 MiB, so this report does not currently map to a confirmed code bug in this repository. If you believe this is still caused by CLIProxyAPI, please reopen with the exact upstream error body, the relevant openai-compat/provider configuration, whether custom payload rules are enabled, and a request log excerpt showing the outbound upstream request and response.