microsoft · text

Phi-3-medium-128k-instruct

microsoft/Phi-3-medium-128k-instruct

Phi-3-medium-128k-instruct at Q4_K_M is exactly 8,566,821,408 bytes (7.98 GiB / 8.57 GB) — an effective 4.909 bits per weight, not the nominal 4. Its KV cache at 32K is 6.25 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
14.0B
Architecture
phi3
40 layers
Context
131,072
native (config.json)
License
mit

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
IQ1_S2.75 GiB2,957,997,0561.695—legraphista
IQ1_S2.75 GiB2,957,997,0881.695—bartowski
IQ1_M3.02 GiB3,242,157,0561.858—legraphista
IQ1_M3.02 GiB3,242,157,0881.858—bartowski
IQ2_XXS3.46 GiB3,715,757,0562.129—legraphista
IQ2_XXS3.46 GiB3,715,757,6322.129—bartowski
IQ2_XS3.84 GiB4,127,405,0562.365—legraphista
IQ2_XS3.84 GiB4,127,405,0882.365—bartowski
IQ2_XS3.84 GiB4,127,405,0882.365—bartowski
IQ2_S4.04 GiB4,338,126,3362.486—legraphista
IQ2_S4.04 GiB4,338,126,3682.486—bartowski
IQ2_S4.04 GiB4,338,126,3682.486—bartowski
IQ2_M4.39 GiB4,717,006,3362.703—legraphista
IQ2_M4.39 GiB4,717,006,3682.703—bartowski
IQ2_M4.39 GiB4,717,006,9122.703—bartowski
Q2_K_S4.44 GiB4,771,288,5762.734—legraphista
Q2_K4.79 GiB5,143,000,5762.947—legraphista
Q2_K4.79 GiB5,143,000,6082.947—bartowski
Q2_K4.79 GiB5,143,001,1522.947—bartowski
Q2_K_L4.94 GiB5,303,321,1523.039—bartowski
IQ3_XXS5.08 GiB5,453,262,3363.125—legraphista
IQ3_XXS5.08 GiB5,453,262,3683.125—bartowski
IQ3_XXS5.08 GiB5,453,262,3683.125—bartowski
IQ3_XS5.41 GiB5,806,841,8563.328—legraphista
IQ3_XS5.41 GiB5,806,841,8883.328—bartowski
IQ3_XS5.41 GiB5,806,842,4323.328—bartowski
Q3_K_S5.65 GiB6,064,889,8563.475—legraphista
IQ3_S5.65 GiB6,064,889,8563.475—legraphista
Q3_K_S5.65 GiB6,064,889,8883.475—bartowski
IQ3_S5.65 GiB6,064,889,8883.475—bartowski
Q3_K_S5.65 GiB6,064,890,4323.475—bartowski
IQ3_M6.03 GiB6,473,977,8563.710—legraphista
IQ3_M6.03 GiB6,473,977,8883.710—bartowski
IQ3_M6.03 GiB6,473,978,4323.710—bartowski
Q3_K6.45 GiB6,923,411,4563.967—legraphista
Q3_K_M6.45 GiB6,923,411,4883.967—bartowski
Q3_K_M6.45 GiB6,923,412,0323.967—bartowski
IQ4_XS6.95 GiB7,466,011,1364.278—legraphista
IQ4_XS6.95 GiB7,466,011,1684.278—bartowski
IQ4_XS6.95 GiB7,466,011,7124.278—bartowski

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.78 GiB0.78 GiB—40 / 0 / 0
8,1921.56 GiB1.56 GiB—40 / 0 / 0
16,3843.13 GiB3.13 GiB—40 / 0 / 0
32,7686.25 GiB6.25 GiB—40 / 0 / 0
65,53612.50 GiB12.50 GiB—40 / 0 / 0
131,07225.00 GiB25.00 GiB—40 / 0 / 0

Compare with

same modality, comparable size
vs gemma-4-12B-it-qat-q4_0-unquantizedvs gemma-4-E4B-itvs Qwen3-4Bvs Qwen3-8Bvs Llama-3.2-1B-Instructvs llama-3-youko-8bAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 7.31 GiB. The real file is 7.98 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
40
Attention heads
40
KV heads
10
Head dim
128
Hidden size
5120
Vocab
32,064
Sliding window
131072
SWA period
1
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does Phi-3-medium-128k-instruct need?
Q4_K_M is exactly 8,566,821,408 bytes (7.98 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Phi-3-medium-128k-instruct's KV cache?
6.25 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of Phi-3-medium-128k-instruct 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.