Instructions to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4",
	filename="hexa-forge-code-3b-v0.2.0-r4.Q5_K_M.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M
# Run inference directly in the terminal:
llama-cli -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M
# Run inference directly in the terminal:
llama-cli -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_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 dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M
# Run inference directly in the terminal:
./llama-cli -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_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 dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M

Use Docker

docker model run hf.co/dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M

LM Studio
Jan
Ollama
How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with Ollama:
```
ollama run hf.co/dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M
```

Unsloth Studio new

How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 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 dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 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 dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 to start chatting

Pi new

How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_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": "dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_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 dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M

Run Hermes

hermes

Docker Model Runner
How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with Docker Model Runner:
```
docker model run hf.co/dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M
```

Lemonade

How to use dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4:Q5_K_M

Run and chat with the model

lemonade run user.hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4-Q5_K_M

List all available models

lemonade list

hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4

GGUF Q5_K_M quantization of the v0.2.0-r4 LoRA-merged Qwen2.5-Coder-3B model.

Size: 2.07 GB (5.75 BPW) — 64% smaller than F16, runs on 4 GB VRAM.

Eval baselines (STRICT, real hexa-cc compile + spec matchers)

Same r4 adapter as the F16 sibling — the quantization is applied AFTER LoRA merge so the eval numbers carry over (with sub-percent Q5_K_M drift).

bench	r4 (FP16 baseline)	gate target
hexa-eval Mk.0.1	60.71% (17/28)	≥80% (v1.0.0)
5-NL Mk.0.1	92% (23/25)	≥70% ✅

Quantization details

type:        Q5_K_M
size:        2,224,814,240 bytes (2.07 GB)
bpw:         5.75 bits per weight
src:         hexa-forge-code-3b-v0.2.0-r4.f16.gguf (6.17 GB)
tool:        llama-quantize from llama.cpp HEAD (built 2026-05-12)
quant time:  28.6 s

Inference

./llama-cli -m hexa-forge-code-3b-v0.2.0-r4.Q5_K_M.gguf \
    -p "### User:\nWrite a hexa function add(a: i32, b: i32) -> i32.\n### Assistant:\n"

Lineage

base: Qwen/Qwen2.5-Coder-3B
adapter: dancinlab/hexa-forge-code-3b-qwen2.5-lora-r16-v0.2.0-r4
f16 GGUF: dancinlab/hexa-forge-code-3b-GGUF-f16-v0.2.0-r4

Downloads last month: 133

GGUF

Model size

3B params

Architecture

qwen2

Hardware compatibility

5-bit

Inference Providers NEW

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

Model tree for dancinlab/hexa-forge-code-3b-Q5_K_M-GGUF-v0.2.0-r4

Base model

Qwen/Qwen2.5-3B

Finetuned

Qwen/Qwen2.5-Coder-3B

Adapter

(35)

this model