File size: 2,764 Bytes
8f18ff9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 | ---
title: Worth Doing AI
short_description: High-quality GGUF quantizations for local Mac inference
thumbnail: https://cdn-avatars.huggingface.co/v1/production/uploads/69dd63f1bb6154c42dfa1d49/4SVRV_yi-MhH6kz4yAY31.png
---
# Worth Doing AI
We provide **high-quality GGUF quantizations** of the best open-source language models, optimized for **local inference on Apple Silicon Macs**.
## What We Do
We select the best small general-purpose models and quantize them using `llama.cpp` with carefully chosen quantization levels. Every model is tested on Apple Silicon hardware before release.
**Our focus:**
- Best-in-class small models (1.7B to 7B parameters)
- GGUF format compatible with Ollama, LM Studio, and llama.cpp
- Optimized for Apple Silicon (Metal GPU acceleration)
- Multiple quantization levels to fit any hardware
## Available Models
| Model | Parameters | Q4_K_M | Q5_K_M | Q8_0 |
|-------|-----------|--------|--------|------|
| [Qwen2.5-7B-Instruct-GGUF](https://huggingface.co/worthdoing/Qwen2.5-7B-Instruct-GGUF) | 7B | 4.4 GB | 5.1 GB | 7.5 GB |
| [Mistral-7B-Instruct-v0.3-GGUF](https://huggingface.co/worthdoing/Mistral-7B-Instruct-v0.3-GGUF) | 7B | 4.1 GB | 4.8 GB | 7.2 GB |
| [Phi-4-mini-GGUF](https://huggingface.co/worthdoing/Phi-4-mini-GGUF) | 3.8B | 2.3 GB | 2.6 GB | 3.8 GB |
| [Qwen2.5-3B-Instruct-GGUF](https://huggingface.co/worthdoing/Qwen2.5-3B-Instruct-GGUF) | 3B | 1.8 GB | 2.1 GB | 3.1 GB |
| [SmolLM2-1.7B-Instruct-GGUF](https://huggingface.co/worthdoing/SmolLM2-1.7B-Instruct-GGUF) | 1.7B | 1.0 GB | 1.1 GB | 1.7 GB |
## Quantization Levels
| Type | Bits per Weight | Best For |
|------|----------------|----------|
| **Q4_K_M** | ~4.6 bpw | **Recommended** - Best quality/size ratio for everyday use |
| **Q5_K_M** | ~5.3 bpw | Higher quality with minimal size increase |
| **Q8_0** | ~8.0 bpw | Near-original quality for maximum accuracy |
## Quick Start
### Ollama
```bash
# Download a GGUF file, then:
cat > Modelfile <<'EOF'
FROM ./qwen2.5-7b-instruct-Q4_K_M-worthdoing.gguf
EOF
ollama create qwen2.5-7b -f Modelfile
ollama run qwen2.5-7b
```
### llama.cpp
```bash
llama-cli -m qwen2.5-7b-instruct-Q4_K_M-worthdoing.gguf -p "Your prompt" -ngl 99
```
### LM Studio
Download any GGUF file and import it directly into LM Studio.
## Hardware Recommendations
| RAM | Recommended Models |
|-----|-------------------|
| **8 GB** | SmolLM2-1.7B (any quant), Qwen2.5-3B Q4_K_M/Q5_K_M |
| **16 GB** | Any 3-4B model (any quant), 7B models Q4_K_M |
| **32 GB+** | Any model, any quantization |
## About
Worth Doing AI is focused on making high-quality AI accessible for local, private use. All quantizations are performed with `llama.cpp` and verified on Apple Silicon hardware.
**Contact:** admin@worthdoing.ai
|