Can I run Llama 3.2 1B on Apple M4 Pro (48GB)?
Yes: Llama 3.2 1B runs on Apple M4 Pro (48GB) with F16 at roughly 65 tokens/s. Estimated memory 3.2 GiB at 8,192 tokens of context.
With F16 you can push the context to about 131,072 tokens before memory runs out (F16 KV cache).
Memory by quantisation (8,192-token context)
| Quantisation | Memory | Verdict | Est. speed |
|---|---|---|---|
| F16Unquantized reference | 3.2 GiB | Fits | ~65 tok/s |
| Q8_0Near-lossless | 2.2 GiB | Fits | ~112 tok/s |
| Q6_KVery close to F16 | 1.9 GiB | Fits | ~138 tok/s |
| Q5_K_MGreat quality/size | 1.8 GiB | Fits | ~154 tok/s |
| Q4_K_MBest default for most | 1.6 GiB | Fits | ~174 tok/s |
| IQ4_XSSmaller 4-bit (imatrix) | 1.6 GiB | Fits | ~191 tok/s |
| Q3_K_MNoticeable loss | 1.5 GiB | Fits | ~203 tok/s |
| Q2_KLast resort; big loss | 1.4 GiB | Fits | ~225 tok/s |
Apple M4 Pro (48GB) offers about 48 GiB of memory (75% usable for a model) at 273 GB/s. Speed is a bandwidth-bound estimate of single-stream generation at 65% efficiency; real results vary by runtime, drivers and settings, so measure on your machine.
How this is calculated
Weights ≈ parameters × bits per weight ÷ 8. KV cache ≈ 2 × 16 layers × 8 KV heads × 64 head dim × 2 bytes per token. Decode speed ≈ efficiency × memory bandwidth ÷ bytes read per token. Learn the maths in VRAM math and memory bandwidth, or try the VRAM calculator.
Llama 3.2 1B on other hardware
Similar models on Apple M4 Pro (48GB)
Running slowly or out of memory? See CUDA out of memory and the quantisation lesson.