dale-9b (GGUF)

dale-9b (GGUF)

Turns a Spanish or English request into one Linux bash command, returned as a small JSON object. For terminal users and developers who want a local command helper that runs on CPU or GPU.

Made by manuelgscom on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: dale-9b is a fine-tuned Qwen3.5-9B that reads a request written in Spanish or English and answers with one bash command for Linux. The answer comes back as a small JSON object. This copy is in GGUF format, so it can run on a CPU or GPU through llama.cpp tools.

Who it suits: It suits people who live in a Linux terminal and cannot always recall the exact command, and developers who want a local command generator inside their own tool.

Industry | Who uses it | How they use it | Example
Software | Developers | Describe a shell task in plain words and get a command | Asking for a way to find large files
IT operations | Sysadmins | Draft Linux commands offline before reviewing them | Turning a note in Spanish into one command
Education | Linux learners | See which command matches a plain request | Comparing the output with a manual page
Tooling | App builders | Serve the model with llama-server behind their own app | Adding a command helper to a notes tool

You can skip it if you do not use Linux, or if you want a model that writes anything other than a single command.

Where to use it: The model works in the dale command line tool or in any app that talks to llama-server.

• The dale command line tool, which runs it on a GPU through Vulkan
• Any other app that can call llama-server, running on CPU or GPU
• Linux, since the commands it writes are for Linux

Use case | Who | What they get back
Plain request to command | Terminal users | One bash command in a JSON answer
Spanish or English asks | Bilingual teams | The same kind of answer in either language
Local helper inside an app | App builders | A model served on your own machine

How to set it up: You need Python for the download tool and enough disk and memory for a model of about 5.6 GB. The dale tool uses this model on a dedicated GPU with at least 8 GiB of free video memory, and you choose it at the first-run question.

• Python, for the Hugging Face download tool
• About 5.6 GB of disk for the Q4_K_M file
• A GPU with 8 GiB of free video memory, if you use it through dale

1. Install the Hugging Face download tool.
2. Download the files of this repository.
3. Run the model with the dale tool or with llama-server.

pip install -U huggingface_hub
hf download manuelgscom/dale-9b-GGUF

How to use it: Write your request in Spanish or English and let the model answer with its JSON object. The model was trained with a fixed prompt and with reasoning switched off, so keep that format. Outside dale there is no risk check, so read each command before you run it.

Tool | What it does
dale | Command line tool that runs the model offline, explains the command, rates the risk and runs it only after you confirm
llama-server | Serves the GGUF file to any other application
hf download | Fetches the model files to your computer

Example: You ask in Spanish for a command that lists the files in a folder, sorted by size. The model returns a small JSON object holding a single bash command. Inside dale you also get an explanation and a risk rating, and nothing runs until you confirm.

For newcomers and experts: If you are new, treat it as a translator from plain words to a shell command. Try a harmless request first, and read the command before running it.

It scored 89.2 percent on the author's public 471-case test through llama-server, against 62.8 percent for the untuned base. Through dale on the GPU it scored 89.6 percent.

Plain request in Spanish or English
Fixed prompt, reasoning off
dale-9b writes a JSON answer
One bash command comes back
You review it, then run it

Good to know: The dale family has four sizes of the same model. Only the 4B and 9B ones use the GPU in the dale tool.

• License: Apache 2.0
• Cost to run: no API key, account or paid service is named, but the dale tool wants about 5.4 GiB of video memory for this model
• Data it reaches: runs offline on your machine, and the commands it suggests can touch whatever your shell can reach once you run them
• Not for: running commands unchecked outside dale, or tasks beyond writing one Linux command

Title: Test v2 accuracy by dale size
dale-0.8b: 79.4%
dale-2b: 81.7%
dale-4b: 87.0%
dale-9b: 89.2%

On Hugging Face, 13 November 2026

• Task: text generation
• Parameters: 9B
• Weights: about 4.5 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 292
• Likes: 3
• License: apache-2.0
• Languages: es, en
• Based on: Qwen/Qwen3.5-9B
• Tags: gguf, bash, shell, nl2bash, llama.cpp, vulkan, text-generation, model-index, imatrix, conversational

Links

• Model: https://huggingface.co/manuelgscom/dale-9b-GGUF
• Files: https://huggingface.co/manuelgscom/dale-9b-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download manuelgscom/dale-9b-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.