Repository navigation
Bug 反馈报告:Aicode 调用英伟达(NVIDIA)模型流式请求超时 #58
RyanZGY307
started this conversation in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
1. 基本信息与环境

涉及模型:Kimi K3、DeepSeek V4.1 flash、GLM 5.3
供应商/接入点:NVIDIA API(https://integrate.api.nvidia.com/v1)
网络环境:直连(未使用代理或 VPN)
复现场景:既出现在带有上下文的旧对话中,也出现在全新的空白对话中。
2. 错误现象与日志核心信息
界面报错:LLM 调用失败: First content timeout after 120000ms,伴随重试机制触发(如 正在重试 2/5 ...)。
底层日志:
SocketTimeoutException: First content timeout after 120000ms(首包响应超时,等待时间达到 120 秒上限)。
IOException: unexpected end of stream(流式传输过程中连接意外中断)。
All reactions