Yes — with 6.0 GB to spare
DeepSeek-R1-Distill-Qwen 32B at Q4_K_M fits your M3 Pro · 36 GB entirely in unified memory at 8K context, at an estimated 4.5 tokens per second. Past 31K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully in unified memory
8K context
Q4_K_M · 18.4 GB
MIT
Released Jan 2025
MIT-licensed and close to the 70B distill on maths. A 24 GB card handles it at Q4.
The VRAM budget
weights 18.4 GB
Weights 18.4 GB
KV cache @ 8K 2.00 GB
Runtime overhead 0.6 GB
Free 6.0 GB of 27.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 32.5 GB | 35.1 GB | — | ~2.6 | −0.1% ppl | 8.1 GB over |
| Q6_K | 25.0 GB | 27.6 GB | 5K | ~3.3 | −0.4% ppl | 0.6 GB over |
| Q5_K_M | 21.7 GB | 24.3 GB | 18K | 3.9 | −0.8% ppl | Long context |
| Q4_K_M | 18.4 GB | 21.0 GB | 31K | 4.5 | −1.9% ppl | Recommended |
| Q3_K_M | 14.9 GB | 17.5 GB | 45K | 5.6 | −5.4% ppl | Long context |
| Q2_K | 12.8 GB | 15.4 GB | 54K | 6.6 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/DeepSeek-R1-Distill-Qwen-32B-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.
01Download is 18.4 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02macOS caps what the GPU may wire down at about 27.0 GB of 36 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03There is room to go to 31K context on this card.