Broken pipe error

[Asherathe]

When using RisuAI 136.0.1 as the frontend and Kobold.cpp 1.76 as the backend with a Yi 34B based model, text generation will cut off after around 400-500 tokens. Terminal shows the error message: "Token streaming was interrupted or aborted! [Errno 32] Broken pipe". Streaming is enabled in Risu, max response tokens is set to 4096, and API selected is OpenAI-compatible. Kobold.cpp is started with the command line "./koboldcpp-linux-x64-nocuda --usevulkan --gpulayers 24 --threads 7 --contextsize 8192".

I made a bug report on Kobold's github, but the dev said he thinks Risu is closing the connection.

Running up-to-date openSuse Tumbleweed, i7-12700KF cpu with 32GB DDR4 RAM and AMD RX 6800 card.

[Asherathe]

I'm still experiencing this error in RisuAI 138.1.3 + Kobold.cpp 1.77 with a Gemma 2 27B-based model. Text generation cuts off after around 400 tokens with: "Token streaming was interrupted or aborted! [Errno 32] Broken pipe".