Mistral on Apple M4

Can I run Mistral Small 24B on an Mac Mini M4 Pro 48GB?

Yes. Mistral Small 24B at Q4_K_M uses 15.2 GB of the 36.0 GB available on a Mac Mini M4 Pro 48GB, and runs at about 12.5 tokens per second. You can push the context to 32K.

Runs comfortably

15.2 GB of 36.0 GB · 42%
036 GB
Weights 13.4 GB
KV cache 1.3 GB
Runtime overhead 0.5 GB

Mistral Small 24B at Q4_K_M leaves 20.8 GB spare on a Mac Mini M4 Pro 48GB. There is room to raise the context length or move up a quantisation level.

Generation12.5tok/s
Prompt processing151tok/s
Max context32Ktokens
KV per 1K tokens0GB

Every quantisation of Mistral Small 24B on a Mac Mini M4 Pro 48GB

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 43.9 GB 45.7 GB Won't fit 2.27
INT8 / W8A8 8.50 23.0 GB 24.8 GB Runs comfortably 32K 7.42
Q8_0 (GGUF) 8.50 23.0 GB 24.8 GB Runs comfortably 32K 7.42
FP8 (E4M3) 8.00 22.0 GB 23.7 GB Runs comfortably 32K 7.77
Q6_K 6.56 18.0 GB 19.8 GB Runs comfortably 32K 9.41
Q5_K_M 5.67 15.6 GB 17.3 GB Runs comfortably 32K 10.8
Q5_K_S 5.52 15.1 GB 16.9 GB Runs comfortably 32K 11.1
AWQ 4-bit 4.25 13.5 GB 15.3 GB Runs comfortably 32K 12.4
GPTQ 4-bit 4.25 13.5 GB 15.3 GB Runs comfortably 32K 12.4
MXFP4 4.25 13.5 GB 15.3 GB Runs comfortably 32K 12.4
Q4_K_M 4.85 13.4 GB 15.2 GB Runs comfortably 32K 12.5
Q4_K_S 4.58 12.7 GB 14.5 GB Runs comfortably 32K 13.1
Q4_0 4.55 12.6 GB 14.4 GB Runs comfortably 32K 13.2
IQ4_XS 4.25 11.9 GB 13.6 GB Runs comfortably 32K 14.0
Q3_K_M 3.91 11.0 GB 12.8 GB Runs comfortably 32K 15.1
IQ3_M 3.70 10.4 GB 12.2 GB Runs comfortably 32K 15.8
IQ3_XXS 3.06 8.8 GB 10.6 GB Runs comfortably 32K 18.6
Q2_K 2.63 7.7 GB 9.5 GB Runs comfortably 32K 21.1
IQ2_XXS 2.06 6.2 GB 8.0 GB Runs comfortably 32K 25.6
IQ1_M 1.75 5.4 GB 7.2 GB Runs comfortably 32K 29.0

Also worth checking