Loading from the unpinned main branch let an upstream repo restructure
(dropped vocoder_streaming.safetensors) crash-loop the tts container on
every boot for days; pin to the last snapshot with the full file set.
entrypoint.sh now honors MTP_N_PARALLEL (default 1, unchanged) for
-np instead of hardcoding 1, so open-webui's docker-compose.mtp.yaml
can request multiple llama-server slots.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
mtp-relaxed-decoding/ holds the patch, build, and guide for a custom
llama-server with relaxed-acceptance MTP speculative decoding
(see its README for the full writeup). Bumps the open-webui submodule
to the commit that adds it as a new service, connection, and the
ollama-auth proxy swap.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
spotify-voice-assistant was still using the git.hetherman.cloud https
remote in .gitmodules instead of the SSH fork URL used by open-webui.
Also explicitly ignore .venv/ and freeze its packages for recreation.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- open-webui: submodule, LAN OIDC/Authentik-fronted, GPU-passthrough ollama,
ollama-auth proxy, asr/tts wired via docker-compose.audio.yaml
- spotify-voice-assistant: submodule, Home Assistant custom integration
- tts-server / asr-server: plain directories, built as sidecars by
open-webui/docker-stack.sh (no separate git history)
bhetherman
