---
base_model: Qwen/Qwen3.5-2B
library_name: gguf
tags:
- gguf
- qwen
- bash
- shell
- linux
- llama.cpp
- text-generation
---

# Qwen3.5-2B ShellCommand-Linux GGUF

This repository contains merged GGUF exports of the current best `Qwen3.5-2B` ShellCommand-Linux LoRA.

## Source

- adapter source: `https://huggingface.co/louisguthmann/qwen3.5-2b-shellcommand-linux-lora`
- GitHub repo: `https://github.com/GuthL/bitnet-nl2sh`

## Files

- `Qwen3.5-2B-shellcommand-linux-F16.gguf`
- `Qwen3.5-2B-shellcommand-linux-Q4_K_M.gguf`
- `Qwen3.5-2B-shellcommand-linux-Q4_K_S.gguf`

## Inherited Eval Snapshot

These metrics come from the source LoRA adapter before GGUF quantization.

- score: `276.5033`
- verifier ok rate: `0.7750`
- verifier command rate: `0.7604`
- verifier ask rate: `0.7500`
- verifier cannot rate: `1.0000`
- exact any-exact rate: `0.2500`
- exact parse-ok rate: `0.9800`

## Recommended Deployment Variants

- `Q4_K_M`: safer default if you want more quality headroom
- `Q4_K_S`: leaner option if memory or latency is tighter

## CX23 Benchmarking

See the GitHub docs for the exact benchmark commands used for `llama.cpp` on Hetzner `CX23`.