Technow: Secret-Llama, DeepSeek-V2, PuLID
Introducing Secret Llama, a fully private, in-browser chatbot that keeps your data local. Meanwhile, DeepSeek-V2 is making waves with its top-tier performance in reasoning tasks. And don’t miss PuLID, a tuning-free ID customization method for text-to-image generation. These innovations are pushing the boundaries of AI!
Secret-Llama
Secret Llama is an entirely-in-browser, fully private chatbot supporting various models including Llama 3, Mistral and other open-source models.
This means all conversation data is only stored locally, even if the chatbot runs in a browser like Google Chrome or Microsoft Edge. Secret Llama does not need a server or installation, and can also work offline if needed. It has an easy-to-use interface similar to ChatGPT’s.
DeepSeek-V2
DeepSeek-V2, a Mixture-of-Experts language model, offers 236B total parameters, activating 21B per token, with a 128K context window capability. It places in the top three on AlignBench, demonstrating competitive performance against models like GPT-4.
In MT-Bench, it outperforms models such as Mixtral 8x22B. This model specializes in math, code, and reasoning tasks, showing a marked performance advantage, particularly in environments that demand high reasoning capabilities.
PuLID
PuLID is a tuning-free ID customization method for text-to-image generation that is performant and highly editable.
Using one or multiple initial images and a prompt, the model is capable of generating between one and four new images that integrate both the face of in the image and the style of the prompt, without further training needed.
PuLID actually introduces both contrastive alignment loss and accurate ID loss, minimizing disruption to the original model and ensuring high ID fidelity, while integrating the prompt.
