Releases: yanun0323/deepseek_ssd
Releases · yanun0323/deepseek_ssd
Release list
DeepSeekV4SSD 1.0.5
What's new in 1.0.5
Reliability
- The app now uses Metal's recommended working-set limit by default. This avoids startup failures when a fixed 48 GiB limit exceeds the value that Metal allows. You can still set a lower memory limit.
- Fixed generation failures when consecutive HTTP requests run on different worker threads. MLX generation now keeps the correct stream across requests.
Packaging and updates
- Added a repeatable workflow for Developer ID signing, Apple notarization, Sparkle appcasts, and GitHub Releases.
- Added checks for the App and ZIP signatures, notarization tickets, English, Simplified Chinese, and Traditional Chinese localization files, and isolated App startup.
- Fixed localization loading in the packaged App. The App now loads files from
Contents/Resourcesbefore it uses the Swift package resource bundle.
Documentation
- Clarified request size limits and setup instructions in the English, Simplified Chinese, Traditional Chinese, Japanese, and Korean README files.
Pull requests
- #1 Make MLX memory limit adaptive by @hugoncosta
- #2 Fix MLX generation across HTTP worker threads by @yanun0323
- #4 Improve the README by @webbrain-one
Validation
- Swift: 25 passed, 3 skipped, and 0 failed.
- Python: 70 passed and 0 failed.
- The release workflow verifies signed and notarized artifacts before publication. It downloads the published artifacts and verifies them again.
DeepSeekV4SSD 1.0.4
Important
Version 1.0.3 cannot install this update automatically because the previous
Sparkle signing key is no longer available. Quit the app, download
DeepSeekV4SSD-macOS-arm64.zip from this release, and replace the existing
app manually. Automatic updates work again after you install version 1.0.4.
What's New in 1.0.4
New Features
- Added a power-saving mode that limits the SSD read rate.
- Added persistent chat history, installed model selection, UI state, and server settings.
- Added secure API key storage in Keychain.
- Added the accumulated output token count to the performance view.
Improvements
- Improved runtime tracing and CLI diagnostics.
- Improved reproducibility for profiling prompts and token hashes.
- Improved app packaging for the bundled Python framework and libraries.
- Improved code signing for nested app components.
- Improved the layout and alignment of the power-saving controls.
- Reduced the minimum app window width to 960 points.
- Standardized the main content width for better readability.
Documentation
- Reorganized the current project documentation.
- Added architecture, performance, research, and validation documents.
- Added reproducible benchmark artifacts.
- Moved superseded research documents to the archive.
DeepSeekV4SSD 1.0.3
DeepSeekV4SSD 1.0.2
DeepSeekV4SSD 1.0.1
Changes
- Improved runtime performance for long prompts and token generation.
- Added OpenAI Responses API and tool-call compatibility.
- Added English, Simplified Chinese, and Traditional Chinese UI languages.
- Redesigned the app layout and model management flow.
- Added live and historical per-second performance metrics.
- Changed the default API port to 11434.
- Fixed streaming, localization, packaging, and startup issues.