Qwen-4B Sagui (Sagui-4B)
Copyright 2026 sysrobson

Also published as Sagui-4B-Vision-Thinking-Code-GGUF. "Qwen-4B" refers to the Qwen
model family and the 4 billion parameter size, not to a Qwen 4 release.

This product is a modified version of Qwen3.5-4B
(https://huggingface.co/Qwen/Qwen3.5-4B), released by the Qwen team under the
Apache License 2.0. The GGUF conversion used as source was published by Unsloth
(https://huggingface.co/unsloth).

Modifications made to the original work:
- Vocabulary reduced from 248,320 to 140,761 tokens by removing tokens outside
  Latin scripts and technical symbols, along with their embedding rows.
- Quantized to IQ3_XXS with an importance matrix calibrated on TypeScript.
- GGUF identity metadata changed (name, repository, quantizer, tags).
- Default chat template and sampling parameters provided for Ollama.

"Qwen" is used only to describe the origin of this work. Sagui-4B is an
independent community project, not affiliated with or endorsed by the Qwen team
or Alibaba Cloud.