Can It Run?
HomeModels › Llama 3.2 3B

Llama 3.2 3B on a Mac

Meta · 3.21B parameters · Llama 3.2 Community · released 2024-09

Weights @ Q4_K_M
1.8 GiB
the usual download size
KV cache
112 KB
per token, FP16
Max context
128K
tokens
Smallest Mac
8 GB
M1
Short answerA M1 8GB is the smallest Apple Silicon machine that loads Llama 3.2 3B at Q4_K_M with an 8K context, at roughly 28 tokens/sec. MacBook Air (M1) is the cheapest way to get that configuration.

Every Mac, ranked

MacMemoryNeedsTok/sVerdict
M1 8GB8 GB3.6 GiB28Yes, comfortably
M1 16GB16 GB3.6 GiB28Yes, comfortably
M2 16GB16 GB3.6 GiB41Yes, comfortably
M4 16GB16 GB3.6 GiB49Yes, comfortably
M1 Pro 16GB16 GB3.6 GiB81Yes, comfortably
M3 Pro 18GB18 GB3.6 GiB61Yes, comfortably
M2 24GB24 GB3.6 GiB41Yes, comfortably
M4 24GB24 GB3.6 GiB49Yes, comfortably
M4 Pro 24GB24 GB3.6 GiB111Yes, comfortably
M4 32GB32 GB3.6 GiB49Yes, comfortably
M1 Max 32GB32 GB3.6 GiB163Yes, comfortably
M2 Max 32GB32 GB3.6 GiB163Yes, comfortably
M3 Max 36GB36 GB3.6 GiB122Yes, comfortably
M4 Max 36GB36 GB3.6 GiB167Yes, comfortably
M4 Pro 48GB48 GB3.6 GiB111Yes, comfortably
M4 Max 48GB48 GB3.6 GiB222Yes, comfortably
M4 Pro 64GB64 GB3.6 GiB111Yes, comfortably
M1 Max 64GB64 GB3.6 GiB163Yes, comfortably
M4 Max 64GB64 GB3.6 GiB222Yes, comfortably
M2 Ultra 64GB64 GB3.6 GiB326Yes, comfortably
M2 Max 96GB96 GB3.6 GiB163Yes, comfortably
M3 Ultra 96GB96 GB3.6 GiB333Yes, comfortably
M3 Max 128GB128 GB3.6 GiB163Yes, comfortably
M4 Max 128GB128 GB3.6 GiB222Yes, comfortably
M2 Ultra 192GB192 GB3.6 GiB326Yes, comfortably
M3 Ultra 256GB256 GB3.6 GiB333Yes, comfortably
M3 Ultra 512GB512 GB3.6 GiB333Yes, comfortably

Download size by quantisation

Weight memory scales linearly with bits per parameter. Everything below assumes an 8K context on top.

QuantisationWeightsTotal @ 8KNotes
FP166.0 GiB8.0 GiBFull precision. Reference quality, twice the memory of Q8.
Q8_03.2 GiB5.0 GiBIndistinguishable from FP16 in practice, at half the size.
Q6_K2.5 GiB4.3 GiBNear-lossless. A good stop when you have memory to spare.
Q5_K_M2.1 GiB3.9 GiBSlightly better than Q4_K_M, noticeably bigger.
Q4_K_M1.8 GiB3.6 GiBThe default. Best quality-per-gigabyte for most people.
Q3_K_M1.5 GiB3.2 GiBVisible quality loss. Use to squeeze one size class up.
Q2_K1.1 GiB2.9 GiBLast resort. Often worse than a smaller model at Q4.