Repository navigation
Degraded response generation and poor compatibility with Gemini models #6783
Unanswered
yakushevhk
asked this question in
Q&A
Replies: 1 comment
先说清 DSH 对模型端点的要求,再说"Gemini 这一路"最可能卡在哪1. 本仓只适配两类协议,而且没有 Gemini 专用适配器
没有 Gemini 原生协议。 所以你必然是"Gemini 直连"或"某个 OpenAI 兼容中转",而这两条路都会在下面几处与预期不同。 2. 具体会在哪里退化(你说的三个症状都能对上)
3. "空响应"这条也值得知道有一个已知的漏网路径:DSH 会把"正常结束但一个内容块都没有"的响应当成可重试的 4. 请补这几样,才能定位到具体一层
有了 1 和 2,基本就能判断是"协议翻译错了"还是"模型行为如此"。我不建议在没看到这两个之前就把它当成 DSH 的 bug——因为这条路本身不在本仓适配范围内,而症状又高度符合"中转层的协议翻译不完整"。 |
0 replies
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.
Description
Running deepseek-harness with Gemini models (or proxy endpoints forwarding to Gemini) exhibits poor response quality and instability.
Symptoms
All reactions