Hi, I'm Saheed Olaide (@SirSahOl) 👋

I build and optimize AI models for local edge inference on Apple Silicon, specializing in Apple MLX quantization, benchmarking, and deployment tooling.

---

🌟 Featured Collection

Check out my curated collection of native MLX conversions: 👉 MLX Models — Optimized for Apple Silicon

---

🛠️ Open Source Tooling

• mlx-foundry: A professional CLI pipeline for converting, benchmarking, and publishing HuggingFace models to Apple MLX format with publication-grade model cards.

---

📊 Models by SirSahOl

Qwen3 Series (Native MLX Conversions)

• Qwen3-0.6B-chat-mlx-4bit — 109.6 tok/s on M1 Apple Silicon (~550 MB RAM)
• Qwen3-0.6B-chat-mlx-8bit — 68.3 tok/s on M1 Apple Silicon
• Qwen3-0.6B-chat-mlx-16bit — 39.9 tok/s on M1 Apple Silicon

GLM Edge Series

• GLM-Edge-1.5B-Chat-mlx-16bit
• GLM-Edge-1.5B-Chat-mlx-8bit
• GLM-Edge-1.5B-Chat-mlx-4bit
• GLM-Edge-4B-Chat-mlx-4bit
• GLM-Edge-4B-Chat-mlx-8bit

---

💻 Hardware & Benchmarking Methodology

All models are verified and benchmarked on Apple M1 (8GB unified memory) and profiled for:

• Token generation throughput (tok/s)
• Time to first token (TTFT)
• Peak unified memory usage (RSS MB)

---

📬 Connect

• GitHub: github.com/sirsahol
• HuggingFace: huggingface.co/SirSahOl