DeepSeek-Lite

1. Introduction

Last week, the release and buzz around DeepSeek-V2 have ignited widespread interest in MLA (Multi-head Latent Attention)! Many in the community suggested open-sourcing a smaller MoE model for in-depth research. And now DeepSeek-V2-Lite comes out:

• 16B total params, 2.4B active params, scratch training with 5.7T tokens
• Outperforms 7B dense and 16B MoE on many English & Chinese benchmarks
• Deployable on single 40G GPU, fine-tunable on 8x80G GPUs

DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. DeepSeek-V2 adopts innovative architectures including Multi-head Latent Attention (MLA) and DeepSeekMoE. MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector, while DeepSeekMoE enables training strong models at an economical cost through sparse computation.

2. News

• 2024.05.16: We released the DeepSeek-V2-Lite.
• 2024.05.06: We released the DeepSeek-V2.

3. Model Downloads

With DeepSeek-V2, we are open-sourcing base and chat models across two sizes:

Due to the constraints of HuggingFace, the open-source code currently experiences slower performance than our internal codebase when running on GPUs with Huggingface. To facilitate the efficient execution of our model, we offer a dedicated vllm solution that optimizes performance for running our model effectively.

4. Evaluation Results

Base Model

Standard Benchmark

For more evaluation details, such as few-shot settings and prompts, please check our paper.

Chat Model

Standard Benchmark

5. Model Architecture

DeepSeek-V2 adopts innovative architectures to guarantee economical training and efficient inference：

• For attention, we design MLA (Multi-head Latent Attention), which utilizes low-rank key-value union compression to eliminate the bottleneck of inference-time key-value cache, thus supporting efficient inference.
• For Feed-Forward Networks (FFNs), we adopt DeepSeekMoE architecture, a high-performance MoE architecture that enables training stronger models at lower costs.

DeepSeek-V2-Lite has 27 layers and a hidden dimension of 2048. It also employs MLA and has 16 attention heads, where each head has a dimension of 128. Its KV compression dimension is 512, but slightly different from DeepSeek-V2, it does not compress the queries. For the decoupled queries and key, it has a per-head dimension of 64. DeepSeek-V2-Lite also employs DeepSeekMoE, and all FFNs except for the first layer are replaced with MoE layers. Each MoE layer consists of 2 shared experts and 64 routed experts, where the intermediate hidden dimension of each expert is 1408. Among the routed experts, 6 experts will be activated for each token. Under this configuration, DeepSeek-V2-Lite comprises 15.7B total parameters, of which 2.4B are activated for each token.

6. Training Details

DeepSeek-V2-Lite is also trained from scratch on the same pre-training corpus of DeepSeek-V2, which is not polluted by any SFT data. It uses the AdamW optimizer with hyper-parameters set to $\beta_1=0.9$, $\beta_2=0.95$, and $\mathrm{weight_decay}=0.1$. The learning rate is scheduled using a warmup-and-step-decay strategy. Initially, the learning rate l