[BUG]: Live Transcription fails on GPU
How are you running AnythingLLM?
AnythingLLM desktop app
What happened?
When using Meeting Assistant (v1.15.0-r2, but also existed in v1.14) with live transcription, the transcription is showing during the meeting, and continues after the meeting (it is slower than real time, despite a GPU, as it keeps loading the model in and out of the GPU -> this is a separate issue). Unfortunately, after a few minutes and before transcription succeeded, the transcription state switches to
" Meeting is processing...
You can leave this page. We will send a notification when the transcription and summary are complete. "
but it never finishes.
A "Retranscribe" after closing and reopening AnythingLLM works and produces a full transcription without failing.
Are there known steps to reproduce?
Record a long meeting in AnythingLLM with "live transcription" enabled (maybe also GPU usage must be activated, not sure). Watch the transcription fail. I use "Full Diarization", but it also fails with "Basic Diarization".
LLM Provider & Model (if applicable)
Embedder Provider & Model (if applicable)
Source: Mintplex-Labs/anything-llm