DeepSeek-V2-Chat-0628
An advanced dialogue generation model
- Ranked 11th overall on the LMSYS Chatbot Arena Leaderboard, 3rd in programming tasks, and 3rd in challenging prompts.
- Excellent performance on multiple evaluation metrics, such as HumanEval, MATH, BBH, IFEval, and Arena Hard.
- Optimized the command compliance capability in the field of "system" and improved user experience.
- Supports local operation and requires an 80GB * 8 GPU.
- Model inference can be performed through Huggingface's Transformers.
- Recommend using vLLM for model inference to provide higher efficiency and flexibility.
- Supporting commercial use, suitable for enterprises and developers who require efficient dialogue generation.
Product Details
DeepSeek-V2-Chat-0628 is an improved version of the DeepSeek-V2 series, designed specifically for dialogue generation tasks. It performed well on the LMSYS Chatbot Arena Leaderboard, ranking 11th overall, particularly in programming tasks and challenging prompts. This model shows significant improvements in multiple evaluation metrics, such as HumanEval, MATH, BBH, IFEval, and Arena Hard. In addition, its ability to follow instructions in the "system" field has been optimized, significantly improving the user experience.