zeekayandHanzo Dev
ecd717c6fb
A fork of meetily (MIT, Zackriya Solutions) rebranded to Hanzo. The Python/FastAPI tier, its Docker files, its whisper-server scripts and the whisper.cpp submodule are deleted rather than rebranded. Upstream's own CLAUDE.md says that tier is archived and unsupported and tells you not to add endpoints to it; the shipping product is one Tauri binary that transcribes in-process with whisper-rs and Parakeet. Rebranding a corpse would have doubled the surface and taught the next reader that a dead service is a live one. Going with it: the vestigial localhost:5167 gate the sidebar checked before reading its own database. .github/workflows goes too. This repo lives on git.hanzo.ai, which reads .hanzo/workflows, and we do not use GitHub-hosted runners — six workflow files that cannot run read like CI that passes. Nothing replaces them yet, because a pipeline that has never run is worse than an honest absence. Summaries are written by api.hanzo.ai, our own stack, and it is the default. The other providers still compile and should go in a follow-up. The upstream MIT notice stays and the Hanzo copyright joins it. About and the README say plainly what this is a fork of. Left alone deliberately, and each is a decision someone must still make: the minisign updater key is upstream's and we cannot re-sign, so the updater points at a release path that does not exist yet; the ffmpeg binaries still come from upstream's GitHub; the Parakeet models still come from upstream's host. Co-authored-by: Hanzo Dev <[email protected]>
2.1 KiB
GPU Acceleration Guide
Minutes supports GPU acceleration for transcription, which can significantly improve performance. This guide provides detailed information on how to set up and configure GPU acceleration for your system.
Supported Backends
Minutes uses the whisper-rs library, which supports several GPU acceleration backends:
- CUDA: For NVIDIA GPUs.
- Metal: For Apple Silicon and modern Intel-based Macs.
- Core ML: An additional acceleration layer for Apple Silicon.
- Vulkan: A cross-platform solution for modern AMD and Intel GPUs.
- OpenBLAS: A CPU-based optimization that can provide a significant speed-up over standard CPU processing.
Automatic GPU Detection
The build scripts (dev-gpu.sh, build-gpu.sh) are designed to automatically detect your GPU and enable the appropriate feature flag during the build process. The detection is handled by the scripts/auto-detect-gpu.js script.
Here's the detection priority:
- CUDA (NVIDIA)
- Metal (Apple)
- Vulkan (AMD/Intel)
- OpenBLAS (CPU)
If no GPU is detected, the application will fall back to CPU-only processing.
Manual Configuration
If you want to manually configure the GPU acceleration backend, you can do so by enabling the corresponding feature flag in the frontend/src-tauri/Cargo.toml file.
For example, to enable CUDA, you would modify the [features] section as follows:
[features]
default = ["cuda"]
# ... other features
cuda = ["whisper-rs/cuda"]
Then, you would build the application using the standard pnpm tauri:build command.
Platform-Specific Instructions
Linux
For detailed instructions on setting up GPU acceleration on Linux, please refer to the Linux build instructions.
macOS
On macOS, Metal GPU acceleration is enabled by default. No additional configuration is required.
Windows
To enable GPU acceleration on Windows, you will need to install the appropriate toolkit for your GPU (e.g., the CUDA Toolkit for NVIDIA GPUs) and then build the application with the corresponding feature flag enabled.