Yes — with 46.5 GB to spare

Llama 3.2 1B Instruct at Q4_K_M fits your M3 Max · 64 GB entirely in unified memory at 8K context, at an estimated 321 tokens per second. There is room for its full 128K window.

Fully in unified memory 8K context Q4_K_M · 0.7 GB Llama 3.2 Community Released Sep 2024

The smallest Llama worth running. Fits anywhere, including phones and 4 GB cards.

What hardware do I need for Llama 3.2 1B Instruct? →

The VRAM budget

weights 0.7 GB
Weights 0.7 GB KV cache @ 8K 0.25 GB Runtime overhead 0.6 GB Free 46.5 GB of 48.0 GB

Quantisation ladder

QuantWeightsTotal @ 8KMax contextTok/sQualityFit
F16 2.3 GB 3.2 GB 128K 97 Reference Long context
Q8_0 1.2 GB 2.1 GB 128K 182 −0.1% ppl Long context
Q6_K 0.9 GB 1.8 GB 128K 236 −0.4% ppl Long context
Q5_K_M 0.8 GB 1.7 GB 128K 273 −0.8% ppl Long context
Q4_K_M 0.7 GB 1.5 GB 128K 321 −1.9% ppl Recommended
Q3_K_M 0.6 GB 1.4 GB 128K 396 −5.4% ppl Long context
Q2_K 0.5 GB 1.3 GB 128K 462 −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

terminal
$ pip install mlx-lm
$ mlx_lm.generate --model mlx-community/Llama-3.2-1B-Instruct-4bit \
    --max-tokens 512 --prompt "Hello"

Apple's own array framework. The fastest path on Apple Silicon. More on MLX.

01Download is 0.7 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 48.0 GB of 64 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03There is room to go to the model's full 128K context on this card.
See all models for this rig Compare with another model