What are the hardware requirements for DeepSeek V2 Lite?

DeepSeek V2 Lite requires approximately 31.4 GB of VRAM and supports a context window of 160K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed DeepSeek V2 Lite and how large is it?

DeepSeek V2 Lite is developed by deepseek-ai, a model with 15.7b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate DeepSeek V2 Lite?

DeepSeek V2 Lite is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

DeepSeek V2 Lite by deepseek-ai — VRAM 31.4GB, 160K context

Name: DeepSeek V2 Lite
Author: deepseek-ai

DeepSeek V2 Lite is an open-source language model by deepseek-ai. Features: 15.7b LLM, VRAM: 31.4GB, Context: 160K, License: other, LLM Explorer Score: 0.22.

Arxiv:2405.04434 Conversational Custom code Deepseek v2 Endpoints compatible Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/deepseek-ai/DeepSeek-V2-Lite

DeepSeek V2 Lite Parameters and Internals

Model Type

Mixture-of-Experts, Language Model

Additional Notes

MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector.

Supported Languages

English (High), Chinese (High)

Training Details

Data Volume:

5.7T tokens

Methodology:

Multi-head Latent Attention, DeepSeekMoE

Context Length:

128000

Model Architecture:

27 layers, hidden dimension of 2048, MLA, 16 attention heads, KV compression to latent vector.

Input Output

Accepted Modalities:

text

Release Notes

Version:

DeepSeek-V2

Date:

2024-05-06

Notes:

Released the main model with 236B total parameters.

Version:

DeepSeek-V2-Lite

Date:

2024-05-16

Notes:

Introduced a smaller version with 16B total params.

LLM Name	DeepSeek V2 Lite
Repository 🤗	https://huggingface.co/deepseek-ai/DeepSeek-V2-Lite
Model Size	15.7b
Required VRAM	31.4 GB
Updated	2026-08-01
Maintainer	deepseek-ai
Model Type	deepseek_v2
Model Files	8.6 GB: 1-of-4 8.6 GB: 2-of-4 8.6 GB: 3-of-4 5.6 GB: 4-of-4
Model Architecture	DeepseekV2ForCausalLM
License	other
Context Length	163840
Model Max Length	163840
Transformers Version	4.33.1
Tokenizer Class	LlamaTokenizerFast
Beginning of Sentence Token	<｜begin▁of▁sentence｜>
End of Sentence Token	<｜end▁of▁sentence｜>
Vocabulary Size	102400
Torch Data Type	bfloat16

Rank the DeepSeek V2 Lite capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to DeepSeek V2 Lite

Best Alternatives	Context / RAM	Downloads	Likes
...eepSeek Coder V2 Lite Instruct	160K / 31.4 GB	542062	628
DeepSeek V2 Lite Chat	160K / 31.4 GB	1354449	144
...e Sft Math7k 1epoch 1e 4 Gamma	160K / 31.4 GB	11	1
Dsv2 Lite Coder Codeio Pp	160K / 31.4 GB	2	2
DeepSeek Coder V2 Lite Base	160K / 31.4 GB	5350	114
KwaiCoder DS V2 Lite Base	160K / 31.4 GB	37	7
...2 Lite Instruct FlashAttnPatch	160K / 31.4 GB	24	0
ESFT Vanilla Lite	4K / 31.4 GB	148	20
...eek Coder V2 Lite Instruct AWQ	160K / 9.1 GB	7853	11
LunarAI	160K / 31.4 GB	24	0

Note: green Score (e.g. "73.2") means that the model is better than deepseek-ai/DeepSeek-V2-Lite.