How much VRAM does deepseek-coder-1.3b-base need?

Q4_K_M is exactly 873,453,568 bytes (0.81 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is deepseek-coder-1.3b-base'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.

deepseek-ai · text

deepseek-coder-1.3b-base

deepseek-ai/deepseek-coder-1.3b-base

deepseek-coder-1.3b-base at Q4_K_M is exactly 873,453,568 bytes (0.81 GiB / 0.87 GB) — an effective 5.190 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

1.3B

Architecture

llama

24 layers

Context

16,384

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	0.40 GiB	424,818,688	2.524	—	bartowski
IQ1_M	0.41 GiB	442,218,496	2.627	—	bartowski
IQ2_XXS	0.44 GiB	471,218,176	2.800	—	bartowski
IQ2_XS	0.46 GiB	497,563,648	2.956	—	bartowski
IQ2_S	0.49 GiB	528,652,288	3.141	—	bartowski
IQ2_M	0.51 GiB	551,852,032	3.279	—	bartowski
Q2_K	0.52 GiB	559,716,352	3.325	—	bartowski
Q2_K	0.52 GiB	559,716,832	3.325	—	QuantFactory
IQ3_XXS	0.54 GiB	580,950,016	3.452	—	bartowski
IQ3_XS	0.57 GiB	613,365,760	3.644	—	bartowski
Q2_K	0.59 GiB	631,705,632	3.753	—	TheBloke
Q3_K_S	0.60 GiB	641,824,768	3.813	—	bartowski
IQ3_S	0.60 GiB	641,824,768	3.813	—	bartowski
Q3_K_S	0.60 GiB	641,825,248	3.813	—	QuantFactory
Q3_K_S	0.61 GiB	658,862,112	3.915	—	TheBloke
IQ3_M	0.63 GiB	672,790,528	3.997	—	bartowski
Q3_K_M	0.66 GiB	704,133,120	4.184	—	bartowski
Q3_K_M	0.66 GiB	704,133,600	4.184	—	QuantFactory
Q3_K_M	0.66 GiB	704,966,688	4.189	—	TheBloke
IQ4_XS	0.69 GiB	744,380,416	4.423	—	bartowski
Q3_K_L	0.69 GiB	744,454,144	4.423	—	bartowski
Q3_K_L	0.69 GiB	744,454,624	4.423	—	QuantFactory
Q3_K_L	0.69 GiB	744,583,200	4.424	—	TheBloke
IQ4_NL	0.72 GiB	775,936,000	4.610	—	bartowski
Q4_0	0.72 GiB	775,936,480	4.610	—	QuantFactory
Q4_0	0.72 GiB	776,065,056	4.611	—	TheBloke
Q4_K_S	0.76 GiB	813,963,264	4.836	—	bartowski
Q4_K_S	0.76 GiB	813,963,744	4.836	—	QuantFactory
Q4_K_S	0.76 GiB	814,796,832	4.841	—	TheBloke
Q4_1	0.80 GiB	855,955,936	5.086	—	QuantFactory
Q4_K_M	0.81 GiB	873,453,568	5.190	—	bartowski
Q4_K_M	0.81 GiB	873,454,048	5.190	—	QuantFactory
Q4_K_M	0.81 GiB	873,582,624	5.190	—	TheBloke
Q5_0	0.87 GiB	935,975,392	5.561	—	QuantFactory
Q5_0	0.87 GiB	936,103,968	5.562	—	TheBloke
Q5_K_S	0.89 GiB	952,883,200	5.662	—	bartowski
Q5_K_S	0.89 GiB	952,883,680	5.662	—	QuantFactory
Q5_K_S	0.89 GiB	953,012,256	5.662	—	TheBloke
Q5_K_M	0.93 GiB	1,001,838,592	5.952	—	bartowski
Q5_K_M	0.93 GiB	1,001,839,072	5.952	—	QuantFactory

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 Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m vs Llama-3.2-3B-Instruct 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 0.71 GiB. The real file is 0.81 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

2048

Vocab

32,256

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does deepseek-coder-1.3b-base need?: Q4_K_M is exactly 873,453,568 bytes (0.81 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is deepseek-coder-1.3b-base'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 deepseek-coder-1.3b-base 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.