Instructions to use a11yequitas/a11yequitas-gr with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use a11yequitas/a11yequitas-gr with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="a11yequitas/a11yequitas-gr",
	filename="a11yequitas-gr-rhel10-3b-q4_k_m.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use a11yequitas/a11yequitas-gr with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf a11yequitas/a11yequitas-gr:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf a11yequitas/a11yequitas-gr:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf a11yequitas/a11yequitas-gr:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf a11yequitas/a11yequitas-gr:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf a11yequitas/a11yequitas-gr:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf a11yequitas/a11yequitas-gr:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf a11yequitas/a11yequitas-gr:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf a11yequitas/a11yequitas-gr:Q4_K_M

Use Docker

docker model run hf.co/a11yequitas/a11yequitas-gr:Q4_K_M

LM Studio
Jan
Ollama
How to use a11yequitas/a11yequitas-gr with Ollama:
```
ollama run hf.co/a11yequitas/a11yequitas-gr:Q4_K_M
```

Unsloth Studio

How to use a11yequitas/a11yequitas-gr with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for a11yequitas/a11yequitas-gr to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for a11yequitas/a11yequitas-gr to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for a11yequitas/a11yequitas-gr to start chatting

How to use a11yequitas/a11yequitas-gr with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf a11yequitas/a11yequitas-gr:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "a11yequitas/a11yequitas-gr:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use a11yequitas/a11yequitas-gr with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf a11yequitas/a11yequitas-gr:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default a11yequitas/a11yequitas-gr:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use a11yequitas/a11yequitas-gr with Docker Model Runner:
```
docker model run hf.co/a11yequitas/a11yequitas-gr:Q4_K_M
```

Lemonade

How to use a11yequitas/a11yequitas-gr with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull a11yequitas/a11yequitas-gr:Q4_K_M

Run and chat with the model

lemonade run user.a11yequitas-gr-Q4_K_M

List all available models

lemonade list

a11yequitas-gr (Granite line)

A11y-Equitas Granite line — fine-tuned technical assistants built on IBM Granite 4.1 3B.

Three specialized variants available as Ollama tags:

Tag	Expertise	Size
`3b`	WCAG 2.2 AA accessibility (axe-core violations)	~2.1 GB
`rhel10`	WCAG 2.2 AA + RHEL 10 / Podman 5 / Ansible	~2.0 GB
`rhel9`	WCAG 2.2 AA + RHEL 9 / Podman 4 / Ansible	~2.0 GB

a11yequitas-gr:3b

Accessibility assistant for content editors, developers, and QA teams working on WCAG 2.2 AA compliance.

Explains axe-core violations in plain language: what is broken, who is affected, whether the fix belongs in the CMS or needs a developer, what to change, what to leave alone, and how to verify.

a11yequitas-gr:rhel10

Hybrid assistant combining WCAG 2.2 AA accessibility knowledge with RHEL 10 / Podman 5 / ansible-core 2.17 system administration expertise.

For RHEL 10 questions: answers use DNF5 (not yum/dnf4), ansible.builtin.dnf5, and Podman 5 quadlet files (not podman generate systemd, which was removed in Podman 5). Structures answers as COMMAND / EXPLAIN / WATCH OUT / VERIFY.

a11yequitas-gr:rhel9

Hybrid assistant combining WCAG 2.2 AA accessibility knowledge with RHEL 9 / Podman 4 / ansible-core 2.16 system administration expertise.

For RHEL 9 questions: answers use DNF (not yum or dnf5), ansible.builtin.dnf, and podman generate systemd for container systemd integration. Structures answers as COMMAND / EXPLAIN / WATCH OUT / VERIFY.

Files

File	Size	Description
`a11yequitas-gr3b-v111-q4_k_m.gguf`	~2.1 GB	gr:3b v1.1.1 — C7 ARIA name patch, Q4_K_M
`a11yequitas-gr-rhel10-v111-q4_k_m.gguf`	~2.0 GB	gr:rhel10 v1.1.1 — C7 ARIA name patch, Q4_K_M
`a11yequitas-gr-rhel9-v111-q4_k_m.gguf`	~2.0 GB	gr:rhel9 v1.1.1 — C7 ARIA name patch, Q4_K_M
`a11yequitas-gr3b-v11-q4_k_m.gguf`	~2.1 GB	gr:3b v1.1.0 — 1835-row WCAG dataset, Q4_K_M
`a11yequitas-gr3b-v10-q4_k_m.gguf`	~2.1 GB	gr:3b v1.0.0 — 867-row dataset, Q4_K_M
`a11yequitas-gr-rhel10-3b-q4_k_m.gguf`	~2.0 GB	gr:rhel10 v1.0.0 — 2257-row hybrid, Q4_K_M
`a11yequitas-gr-rhel9-3b-q4_k_m.gguf`	~2.0 GB	gr:rhel9 v1.0.0 — 1975-row hybrid, Q4_K_M

Exam scores (axe Fix-It, 50 Q, 100 pts, LLM-judged)

Model	Version	Score	Auto-fails	Gate
gr:rhel9	v1.1.1	98/100	1	≥85% ✓
gr:3b	v1.1.1	96/100	1	≥85% ✓
gr:rhel10	v1.1.1	94/100	2	≥85% ✓
gr:rhel9	v1.0.0	98/100	1	≥85% ✓
gr:3b	v1.1.0	96/100	2	≥85% ✓
gr:rhel10	v1.0.0	94/100	2	≥85% ✓
granite4.1:3b base	—	~60/100	~10	—

v1.1.1 patch: fixes C7 ARIA accessible name precedence (aria-label overrides visible label).

Use with Ollama

ollama pull rockypod/a11yequitas-gr:3b
ollama pull rockypod/a11yequitas-gr:rhel10
ollama pull rockypod/a11yequitas-gr:rhel9

Use with llama.cpp

llama-cli -m a11yequitas-gr3b-v111-q4_k_m.gguf \
  --chat-template granite \
  -sys "You are a11yequitas-gr, an accessibility assistant trained on WCAG 2.2. ..." \
  -i

Technical

Base: IBM Granite 4.1 3B (ibm-granite/granite-4.1-3b-base)
Chat template: Granite native (<|start_of_role|> / <|end_of_role|> / <|end_of_text|>)
Fine-tuning: Unsloth LoRA, r=32, alpha=32, RTX 3090 Ti
Quantization: Q4_K_M

gr:3b v1.1.1 dataset

1835 WCAG 2.2 AA pairs + 20 C7 ARIA name patch rows = 1855 total

gr:rhel10 v1.1.1 dataset

2257-row hybrid (1835 WCAG + 422 RHEL10) + 20 C7 patch = 2277 total

gr:rhel9 v1.1.1 dataset

1975-row hybrid (1835 WCAG + 140 RHEL9) + 20 C7 patch = 1995 total

Scope

Accessibility (all variants): WCAG 2.2 Level AA. Full axe-core rule set: color contrast · image alt · form labels · link/button names · document structure · lists · tables · ARIA · keyboard · frames · media · in-CMS prioritization.

RHEL 10 (gr:rhel10 only): DNF5 · Podman 5 + quadlet · ansible-core 2.17 · SELinux · firewalld.

RHEL 9 (gr:rhel9 only): DNF · Podman 4 + podman generate systemd · ansible-core 2.16 · SELinux · firewalld.

Project

Built by A11y Equitas — an NPO working to make public-sector digital infrastructure accessible by default.

License: MIT

Downloads last month: 150

GGUF

Model size

3B params

Architecture

granite

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for a11yequitas/a11yequitas-gr

Base model

ibm-granite/granite-4.1-3b-base

Quantized

(4)

this model