How much VRAM does SmolLM2-1.7B-Instruct need?

Q4_K_M is exactly 1,055,609,536 bytes (0.98 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is SmolLM2-1.7B-Instruct's KV cache?

6.00 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

HuggingFaceTB · text

SmolLM2-1.7B-Instruct

HuggingFaceTB/SmolLM2-1.7B-Instruct

SmolLM2-1.7B-Instruct at Q4_K_M is exactly 1,055,609,536 bytes (0.98 GiB / 1.06 GB) — an effective 4.935 bits per weight, not the nominal 4. Its KV cache at 32K is 6.00 GiB.

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

Parameters

1.7B

Architecture

llama

24 layers

Context

8,192

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	0.63 GiB	674,583,520	3.153	—	bartowski
Q2_K_L	0.65 GiB	698,962,912	3.267	—	bartowski
IQ3_XS	0.69 GiB	739,070,944	3.455	—	bartowski
Q3_K_S	0.72 GiB	776,819,680	3.631	—	bartowski
IQ3_M	0.75 GiB	810,243,040	3.788	—	bartowski
Q3_K_M	0.80 GiB	860,181,472	4.021	218	bartowski
Q3_K_L	0.87 GiB	932,532,928	4.359	—	lmstudio-community
Q3_K_L	0.87 GiB	932,533,216	4.359	—	bartowski
IQ4_XS	0.88 GiB	940,397,536	4.396	218	bartowski
Q4_0	0.93 GiB	993,874,912	4.646	218	bartowski
Q4_K_S	0.93 GiB	999,117,792	4.670	—	bartowski
Q4_K_M	0.98 GiB	1,055,609,536	4.935	—	HuggingFaceTB
Q4_K_M	0.98 GiB	1,055,609,536	4.935	—	lmstudio-community
Q4_K_M	0.98 GiB	1,055,609,824	4.935	218	bartowski
Q4_K_L	1.01 GiB	1,079,989,216	5.048	—	bartowski
Q5_K_S	1.11 GiB	1,192,055,776	5.572	—	bartowski
Q5_K_M	1.14 GiB	1,225,479,136	5.729	—	bartowski
Q5_K_L	1.16 GiB	1,249,858,528	5.843	—	bartowski
Q6_K	1.31 GiB	1,405,964,992	6.572	218	lmstudio-community
Q6_K	1.31 GiB	1,405,965,280	6.572	—	bartowski
Q6_K_L	1.33 GiB	1,430,344,672	6.686	—	bartowski
Q8_0	1.70 GiB	1,820,414,656	8.510	218	lmstudio-community
Q8_0	1.70 GiB	1,820,414,944	8.510	—	bartowski
F16	3.19 GiB	3,424,735,936	16.009	218	bartowski

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.75 GiB	0.75 GiB	—	24 / 0 / 0
8,192	1.50 GiB	1.50 GiB	—	24 / 0 / 0
16,384	3.00 GiB	3.00 GiB	—	24 / 0 / 0
32,768	6.00 GiB	6.00 GiB	—	24 / 0 / 0
65,536	12.00 GiB	12.00 GiB	—	24 / 0 / 0
131,072	24.00 GiB	24.00 GiB	—	24 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-4B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 0.90 GiB. The real file is 0.98 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

Hidden size

2048

Vocab

49,152

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does SmolLM2-1.7B-Instruct need?: Q4_K_M is exactly 1,055,609,536 bytes (0.98 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is SmolLM2-1.7B-Instruct's KV cache?: 6.00 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 SmolLM2-1.7B-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.