How much VRAM does starcoder2-7b need?

Q4_K_M is exactly 4,402,887,488 bytes (4.10 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

bigcode · text

starcoder2-7b

bigcode/starcoder2-7b

starcoder2-7b at Q4_K_M is exactly 4,402,887,488 bytes (4.10 GiB / 4.40 GB) — an effective 4.910 bits per weight, not the nominal 4.

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

Parameters

7.2B

Architecture

starcoder2

32 layers

Context

16,384

native (config.json)

License

bigcode-openrail-m

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	2.54 GiB	2,724,543,296	3.038	—	second-state
Q3_K_S	2.88 GiB	3,090,234,176	3.446	—	second-state
Q3_K_M	3.35 GiB	3,594,533,696	4.008	—	second-state
Q3_K_L	3.71 GiB	3,985,587,008	4.444	—	second-state
Q4_0	3.77 GiB	4,042,799,936	4.508	—	second-state
Q4_K_S	3.85 GiB	4,128,914,240	4.604	—	second-state
Q4_K_M	4.10 GiB	4,402,887,488	4.910	—	second-state
Q5_0	4.60 GiB	4,939,332,416	5.508	—	second-state
Q5_K_S	4.60 GiB	4,939,332,416	5.508	—	second-state
Q5_K_M	4.77 GiB	5,124,832,064	5.715	—	second-state
Q6_K	5.49 GiB	5,891,898,176	6.570	—	second-state
Q8_0	7.10 GiB	7,628,929,856	8.507	—	second-state

KV cache by context

unresolved

This model declares a 4,096-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-E4B-it vs Qwen3-4B vs Qwen3-8B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs llama-3-youko-8b 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 3.76 GiB. The real file is 4.10 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

4608

Vocab

49,152

Sliding window

4096

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does starcoder2-7b need?: Q4_K_M is exactly 4,402,887,488 bytes (4.10 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of starcoder2-7b 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.