Model Hub & MLX Foundry Overview

Creative Tech Studios optimizes and distributes frontier open-weight models specifically tuned for Apple Silicon’s unified memory architecture.

Through our open-source research initiative on Hugging Face (SirSahOl & Creative-Tech-Studios), we publish dozens of optimized architectures—including DeepSeek-R1, Qwen 2.5, GLM-Edge, and SmolLM2.

Official Hugging Face Collection

MLX Models by SirSahOl — Optimized for Apple Silicon
Access the primary curated collection on Hugging Face containing our tested and verified AWQ model conversions for Apple Silicon unified memory.

---

Technical Disciplines

Apple Silicon Unified Memory

Quantized for zero-copy memory access between CPU and Metal GPU on M-series chips (M1–M4).

4-Bit & 8-Bit AWQ Precision

Activation-aware Weight Quantization (AWQ) preserves perplexity scores within 0.8% of 16-bit FP baselines.

Benchmarked Performance

Every published model includes empirical tokens/sec benchmarks on real M-series hardware in its model card.

Hugging Face Integration

Direct compatibility with mlx-lm, LM Studio, Ollama, and local Python inference pipelines.

---

Documentation Tracks

Featured Model CatalogBrowse the flagship quantized models available for immediate pull.

Quantization PipelineUnderstand our automated AWQ conversion and validation pipeline.

Hardware BenchmarksReview real-world tokens/second metrics across M-series processors.