dwarfstar4 (llm/dwarfstar4) Updated: 1 week, 2 days ago Add to my watchlist
DeepSeek V4 Flash-specific native inference engineDwarfStar 4 is a small native inference engine specific for DeepSeek V4 Flash, with a CLI, HTTP server, Metal backend, KV state handling, and model download helper.
Version: 20260514 License: MIT
GitHub
| Maintainers | No Maintainer |
| Categories | llm |
| Homepage | https://github.com/antirez/ds4 |
| Platforms | darwin |
| Variants |
|
"dwarfstar4" depends on
run (1)
build (1)
Ports that depend on "dwarfstar4"
No ports
Port notes
The upstream model download script has been installed as:
${prefix}/bin/ds4-download-model
Run it before using ds4 to run a supported GGUF model. By default, the
script stores models under ~/.local/share/ds4/gguf and updates the symlink:
~/.local/share/ds4/ds4flash.gguf
The ds4 tools still default to looking for ds4flash.gguf in the current working
directory. Use -m to point at the downloaded model, for example:
ds4-download-model q2-imatrix
ds4 -m ~/.local/share/ds4/ds4flash.gguf
Port Health:
Loading Port Health