AI BriefingM2 MacBook Pro 32GB Qwen 27B
·2026.04.29 06:01
Key point
On an M2 MacBook Pro with 32GB RAM, Qwen 3.6 27B's processing speed plummeted to 4t/s at a 52,000-token context.
Details
Ran Qwen 3.6 27B on an M2 MacBook Pro with 32GB RAM using 4bit XS unsloth quant.
Used the model files Qwen3.6-27B-IQ4_XS.gguf and Qwen3.6-27B-mmproj-BF16.gguf.
Initially it was prompt processing 80 t/s and generation 7.9 t/s, but once the context reached 52,000 tokens, it dropped to 4 t/s pp and 3.1 t/s tg.
- No signs of swapping were observed, and memory pressure never exceeded the yellow level.
llama-serverwas run with the settings-c 131072,-batch-size 256,-ngl 99,-ctk q8_0,-ctv q8_0,--spec-type ngram-mod,--spec.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Apple Releases LensVLM-9B for Processing Compressed Text Images
Underdog Releases 'Husky', a Model-Specific Inference Engine Up to 4.5x Faster than MLX
Jared Palmer Releases 'Kev', a Lightweight Decision-Making Model Based on Qwen3.5