Open jobobby04 opened 10 months ago
I've observed the same behavior with Miku in latest source when using Ollama, Oobabooga, and TabbyAPI.
After configuring Miku to work with each of these engines, I found that sending a single chat message (such as "Hello") to a Chatbot caused Miku to continuously send repeated requests to the /v1/completions endpoint.
These POST requests persisted indefinitely, as shown in the Ollama server logs below, yet the chatbot failed to return/display a proper response:
[GIN] 2024/10/08 - 19:11:17 | 200 | 40.1179ms | 192.168.10.159 | POST "/v1/completions" [GIN] 2024/10/08 - 19:11:18 | 200 | 41.3535ms | 192.168.10.159 | POST "/v1/completions" [GIN] 2024/10/08 - 19:11:18 | 200 | 43.2242ms | 192.168.10.159 | POST "/v1/completions" ... Etc.
Aphrodite was the only engine that worked successfully, suggesting that this issue may lie in Miku's handling of API requests across other OpenAI-compatible engines.
It just keeps looping requests, not letting my server get any time to generate text. I have to kill miku to stop it.