Testing a 27B Model Locally on Mac Studio
MaziyarPanahi · x · 2026-07-15
- This post demonstrates running prism-ml/Ternary-Bonsai-27B-gguf on a Mac Studio. It utilizes Hugging Face Inference Providers, hosted by Together AI. The author emphasizes that "you don't need a Mac Studio to use the same model," pointing to inference-as-a-service rather than hardware dependency.
- The quoted content provides more context: the model runs locally using llama.cpp + Metal, is about 7.2GB in size under the Apache-2.0 license, and processes a patient medical record spanning roughly 3 years and comprising 292 visit notes.
- During this test, GLM-5.2 acted as the questioner while Bonsai provided the answers. The model answered 3 questions in about 2 seconds while maintaining a cache of 19,398 tokens. The author notes it ultimately uncovered an issue that had been buried for 17 months, highlighting the practical utility of long-context and local privacy scenarios.
Related event: Mac Studio Runs 27B Model Locally to Process 3-Year Medical Records(2 posts)→
More from Infra
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11