Skip to content

v0.10.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 19 Oct 03:15
· 279 commits to main since this release
v0.10.0
bd3492b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Minor Changes

  • #41 c48aeb5 Use the GPU where it pays. docker-compose.gpu.yml hands an NVIDIA card to Speaches (SPEACHES_IMAGE=ghcr.io/speaches-ai/speaches:latest-cuda, so faster-whisper-large-v3 transcribes faster than small did on the CPU), to Ollama in a container, and to the app, where ffmpeg now encodes clips, intros and outros with h264_nvenc when FFMPEG_ENCODER=auto finds it at boot (VideoToolbox on a Mac outside Docker); a hardware encoder that fails mid-render falls back to libx264 for that clip. ONNX_EXECUTION_PROVIDER moves the face, text and alignment models to CUDA, CoreML or DirectML with a CPU fallback (CPU stays the default: measured faster on Apple silicon). Settings → System → Overview and the new reelsaw machine command show the GPU the process sees, the encoder in use and why, and where each on-device model runs.

  • #41 9324f80 The one-command installers (install.sh, install.ps1) now look at the machine before doing anything: RAM, an NVIDIA GPU and its VRAM, Docker's memory, free disk, an Ollama already running and the models it has, and the port. They print a plan and ask three questions (local Ollama or a hosted provider, which model, go ahead), install Ollama when it is missing and you chose local, reuse a good model Ollama already has, pick the model by VRAM when a GPU serves it, move a busy port, and put Ollama, Speaches and the app on the GPU when the NVIDIA Container Toolkit is present. Hosted providers (OpenAI, Anthropic, Google, Groq, Vercel AI Gateway) are a choice in the same prompt. REELSAW_YES=1 takes the proposed answers, REELSAW_DRY_RUN=1 prints the plan and stops; re-running upgrades an existing install.

Patch Changes

  • #44 9ba8669 Detect the GPU on Docker Desktop (WSL2) via /dev/dxg when nvidia-smi and /dev/nvidia0 are missing, document that Windows .env must use ; in COMPOSE_FILE (a colon is the drive-letter separator), and keep pnpm docs:gen defaults as POSIX paths on Windows.

Install or upgrade: reelsaw-0.10.0.zip below holds the compose file and installer pinned to this release; an existing install runs docker compose pull && docker compose up -d. Details in deploy/README.md. Full history: CHANGELOG.md.

Assets 3
Loading