Running 32B-120B models locally on M1 Max 64GB: errors and model picks
Food4Lessy · reddit · 2026-10-21
After 9 months of relying on cloud models (Gemini, Claude, DeepSeek, Groq, Cerebras), the author wants local models for sensitive prototyping work.
Their 7B Llama via MLX hit 45 tokens/s but is too weak; Qwen 32B throws errors in LM Studio. They ask for the best 32B-120B options on a 64GB M1 Max for research, light coding, CAD/PCB and image generation workflows.
More from Infra
- DeepSeek reportedly teaching ultra-efficient inference on Huawei Ascend 950 — no first-party evidence yet — teortaxesTex · 2026-10-21
- Exclusive: ByteDance, Alibaba, Tencent place over $35 billion in Nvidia chip orders — Scobleizer · 2026-10-21
- DeepSeek's own numbers reveal ~82% gross margin: 155T tokens/day per cluster at $38/GPU-hour — ssh4net · 2026-10-21
- vLLM says it can reproduce DeepSeek's inference performance, yielding $39/GPU/hr revenue — vllm_project · 2026-10-21
- AMD brings EAGLE-4 speculative decoding to Instinct MI355X day 0, 1.6x faster than built-in MTP — hongyangzh · 2026-10-21
- TokenGate: learned token pruning speeds up LLM prefill up to 11.3x — Liana Patel · 2026-10-21