How much VRAM does dolphin-2.9.2-Phi-3-Medium need?

IQ1_S is exactly 3,054,259,616 bytes (2.84 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

dphn · text

dolphin-2.9.2-Phi-3-Medium

dphn/dolphin-2.9.2-Phi-3-Medium

dolphin-2.9.2-Phi-3-Medium at IQ1_S is exactly 3,054,259,616 bytes (2.84 GiB / 3.05 GB) — an effective 1.750 bits per weight, not the nominal 1.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

14.0B

Architecture

llama

40 layers

Context

4,096

native (config.json)

License

mit

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	2.84 GiB	3,054,259,616	1.750	—	legraphista
IQ1_M	3.10 GiB	3,332,275,616	1.910	—	legraphista
IQ2_XXS	3.53 GiB	3,795,635,616	2.175	—	legraphista
IQ2_XS	3.91 GiB	4,199,091,616	2.406	—	legraphista
IQ2_S	4.11 GiB	4,409,812,896	2.527	—	legraphista
IQ2_M	4.45 GiB	4,780,500,896	2.740	—	legraphista
Q2_K_S	4.50 GiB	4,832,735,136	2.769	—	legraphista
Q2_K	4.85 GiB	5,204,447,136	2.982	—	legraphista
IQ3_XXS	5.05 GiB	5,418,452,896	3.105	—	legraphista
IQ3_XS	5.38 GiB	5,780,224,416	3.312	—	legraphista
Q3_K_S	5.65 GiB	6,064,896,416	3.475	—	legraphista
IQ3_S	5.68 GiB	6,099,712,416	3.495	—	legraphista
IQ3_M	5.87 GiB	6,299,904,416	3.610	—	legraphista
Q3_K	6.29 GiB	6,750,976,416	3.869	—	legraphista
Q3_K_L	6.84 GiB	7,348,992,416	4.211	—	legraphista
IQ4_XS	6.99 GiB	7,506,977,696	4.302	—	legraphista
IQ4_NL	7.39 GiB	7,929,899,936	4.544	—	legraphista
Q4_K_S	7.41 GiB	7,957,752,736	4.560	—	legraphista
Q4_K	7.83 GiB	8,404,011,936	4.816	—	legraphista
Q5_K_S	8.96 GiB	9,621,588,608	5.514	—	legraphista
Q5_K	9.20 GiB	9,882,708,608	5.663	—	legraphista
Q6_K	10.67 GiB	11,453,824,128	6.564	—	legraphista
Q8_0	13.82 GiB	14,834,718,848	8.501	—	legraphista
BF16	26.00 GiB	27,922,053,248	16.001	—	legraphista

KV cache by context

unresolved

This model declares a 2,048-token sliding window, but we could not establish which layers use it. Its architecture publishes the layout as a per-layer array inside the model file rather than as a period in config.json, and we have not yet ingested that array.

A flat context × layers × heads figure would be substantially too high, so we are not showing one. This is tracked as a known gap rather than filled with a guess.

Compare with

same modality, comparable size

vs gemma-4-12B-it-qat-q4_0-unquantized vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 vs Qwen3-4B vs Qwen3-8B vs Llama-3.2-1B-Instruct All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts IQ1_S at roughly 7.31 GiB. The real file is 2.84 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

128

Hidden size

5120

Vocab

32,064

Sliding window

2048

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does dolphin-2.9.2-Phi-3-Medium need?: IQ1_S is exactly 3,054,259,616 bytes (2.84 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of dolphin-2.9.2-Phi-3-Medium should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.