How much VRAM does functionary-medium-v3.2 need?

Its published quantizations range from 15.60 GiB to 69.83 GiB in weights alone, before KV cache and runtime overhead.

Can I run functionary-medium-v3.2 on a Apple M3 Pro?

Q: Can I run functionary-medium-v3.2 on a Apple M3 Pro?

No. No published quantization of functionary-medium-v3.2 fits a Apple M3 Pro at any context with q8_0 KV cache.

Not at these settings. No indexed quantization of functionary-medium-v3.2 fits Apple M3 Pro at any context we compute, with q8_0 KV. The smallest shipped quantization is 15.60 GiB in weights alone, against 12.56 GiB usable. CPU offload can still run it, slowly.

From the file· weights summed from filesFrom the file· KV computed per layerPredicted· speed and compute buffer

KV cache dtypef16 q8_0 q4_0— 18 GB card, 12.56 GiB usable after overhead

Every quantization at every context

total memory required; green fits, red does not

Quant	Weights●	4K◐	8K◐	16K◐	32K◐	64K◐	128K◐
Q8_0	69.83 GiB	71.2	71.8	73.2	75.8	81.1	91.8
Q6_K	53.91 GiB	55.3	55.9	57.2	59.9	65.2	75.8
Q5_K_M	46.52 GiB	47.9	48.5	49.9	52.5	57.8	68.4
Q4_K_L	40.33 GiB	41.7	42.3	43.7	46.3	51.6	62.3
Q4_K_M	39.60 GiB	40.9	41.6	42.9	45.6	50.9	61.5
Q4_K_S	37.58 GiB	38.9	39.6	40.9	43.6	48.9	59.5
IQ4_XS	35.30 GiB	36.6	37.3	38.6	41.3	46.6	57.2
Q3_K_L	34.59 GiB	35.9	36.6	37.9	40.6	45.9	56.5
Q3_K_M	31.91 GiB	33.3	33.9	35.2	37.9	43.2	53.8
IQ3_M	29.74 GiB	31.1	31.7	33.1	35.7	41.0	51.7
Q3_K_S	28.79 GiB	30.1	30.8	32.1	34.8	40.1	50.7
IQ3_XXS	25.58 GiB	26.9	27.6	28.9	31.6	36.9	47.5
Q2_K_L	25.52 GiB	26.9	27.5	28.9	31.5	36.8	47.4
Q2_K	24.56 GiB	25.9	26.6	27.9	30.6	35.9	46.5
IQ2_M	22.46 GiB	23.8	24.5	25.8	28.5	33.8	44.4
IQ2_XS	19.69 GiB	21.0	21.7	23.0	25.7	31.0	41.6
IQ2_XXS	17.79 GiB	19.1	19.8	21.1	23.8	29.1	39.7
IQ1_M	15.60 GiB	16.9	17.6	18.9	21.6	26.9	37.5

●From the file◐Predictedwhat these mean

Figures are GiB of total memory: weights plus KV cache plus compute buffer and backend overhead. Weights and KV are near-exact; the overhead term is modeled. Hover any cell for the breakdown.

Why other calculators disagree

A parameters × bits ÷ 8 estimate ignores two things that dominate at long context. First, the weights themselves are not the nominal rate — quantizations are mixtures, so the real file is consistently larger than the label implies. Second, most of this model's layers cache only a 8,192-token window rather than the full context.