This repo is less of a “raw model drop” and more of a practical, privacy-friendly tool: a Polish assistant that rewrites dense, bureaucratic documents into plainer Polish (“prosty język”). It’s built on speakleash/Bielik-4.5B-v3.0-Instruct and is published under Apache-2.0.
The interesting part is packaging. Instead of requiring Python/Transformers, it ships as a llamafile-based local app (Windows/Linux/macOS), with optional CUDA/ROCm acceleration when available. That makes it a nice reference if you’re trying to deliver LLM functionality to non-technical users (or regulated environments) where “upload the document to a SaaS” is a non-starter. If you work with government, legal, or health orgs in Poland, this is also a concrete example of a narrow, language-specific workflow where a mid-sized local model can be genuinely useful.
Quick stats from the listing feed: pipeline: text-generation · 8 likes · 26 downloads.
Source listing: https://huggingface.co/models?sort=modified