c980904204
* add initial structure * doc fixes, add model base logic * update init files * some fixes to config and modular * some improvements for attention * format * remove unused attn * some fixes for moe layer and for decoder * adapt _compute_yarn_parameters for deepseek * format * small fix * fix for decoder forward * add tests, small refactoring * fix dummies * fix init * fix doc * fix config docs * add sequce doc, fix init for gate * fix issues in tests * fix config doc * remove unused args * some fixes and refactoring after review * fix doc for config * small fixes for config args * revert config refactoring * small refactoring * minor fixes after rebase * small fix after merge * fix modular * remove rotaryembd from public init * small test fix * some rotary pos calculation improvement * fix format * some improvements and fixes * fix config * some refactoring * adjust some unit tests * skip test * small fixes and tests adjustment * reapply modular * fix all tests except Integration * fix integration testzs * cleanup BC stuff * rope * fix integrations tests based on a10 * style --------- Co-authored-by: Cyril Vallez <cyril.vallez@huggingface.co> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com>
3.0 KiB
DeepSeek-V2
Overview
The DeepSeek-V2 model was proposed in DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model by DeepSeek-AI Team.
The abstract from the paper is the following: We present DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token, and supports a context length of 128K tokens. 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. Compared with DeepSeek 67B, DeepSeek-V2 achieves significantly stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock its potential. Evaluation results show that, even with only 21B activated parameters, DeepSeek-V2 and its chat versions still achieve top-tier performance among open-source models.
This model was contributed by VladOS95-cyber. The original code can be found here.
Usage tips
The model uses Multi-head Latent Attention (MLA) and DeepSeekMoE architectures for efficient inference and cost-effective training. It employs an auxiliary-loss-free strategy for load balancing and multi-token prediction training objective. The model can be used for various language tasks after being pre-trained on 14.8 trillion tokens and going through Supervised Fine-Tuning and Reinforcement Learning stages.
DeepseekV2Config
autodoc DeepseekV2Config
DeepseekV2Model
autodoc DeepseekV2Model - forward
DeepseekV2ForCausalLM
autodoc DeepseekV2ForCausalLM - forward
DeepseekV2ForSequenceClassification
autodoc DeepseekV2ForSequenceClassification - forward