Dette er en ældre version. Den nyeste version er 1.2.0+2. Bionic 1.2.0+2 · Udgivet:
Installationsfiler 1.1.10+8
| Arkitektur | Type | Omfang | Filstørrelse | Download | SHA256 |
|---|---|---|---|---|---|
| 64-bit | EXE (NSIS) | — | 782,9 MB | bionic-installers.lmstudio.ai | VirusTotal |
Link og SHA256 stammer fra fællesskabets manifest i Windows Package Manager. Sammenlign hashen for den downloadede fil, før du kører den.
Udgivelsesnoter
- In-app terminal now available in public Bionic builds.
- oMLX engine for Apple Silicon Macs.
- Excalidraw file viewing previews in chat.
- Faster project loading with side-chat drafts loaded on demand.
- llama.cpp 2.66.0 extension packs.
- SafeTensors model support for speculative decoding.
- Speculative decoding for MLX models.
- Fixed tab-strip scrolling changing direction during trackpad gestures.
- Better runtime compatibility and download selection for M5 Macs.
- Additional compatible LLM backend downloads in Runtime settings.
- Embedding and decision models in the model catalog.
- Experimental MoE expert GPU cache setting for llama.cpp MoE models.
- Restored terminal sessions start automatically when selected.
- Fixed ROCm runtime compatibility detection for AMD Radeon 8065S graphics.
- Code diff syntax highlighting improvements.
- Configurable shell profiles for the in-app terminal.
- More reliable context compaction when conversations exceed model limits.
- Fixed keyboard shortcuts, clipboard handling, saved shell selections, and IME input in the in-app terminal.
- Voice model auto-loading defaults off on Macs with less than 32 GB of memory.
- On-device Read Aloud support on Linux and Windows, in addition to Mac.
- Fixed misleading context limit guidance when multiple requests exhaust model context.
- Fixed multiline shell text being rendered as math.
- Read Aloud button styling now matches other message actions.
- Lower memory usage and detailed progress for large session exports and moves.
- Screen reader support for in-app terminal output.
- Cleaner transcript image previews and message styling.
- Fixed stale draft-model selections when switching to bundled MTP.
- Seven additional offline Read Aloud voices.
