News roundup 49
Created on January 30|Last edited on January 30
Comment
DeepSeek hit by cyberattack, restricts new user registrations
The Chinese AI startup DeepSeek limited sign-ups after a major cyberattack, raising security concerns.
Qwen2.5-Max sets new benchmark for AI performance
With 325 billion parameters, Qwen2.5-Max outperforms top models in reasoning, coding, and general tasks.
Unsloth shrinks DeepSeek R1 by 80% with advanced quantization
Unsloth’s 1.58-bit quantization reduces DeepSeek R1’s size while maintaining high performance for local deployment.
https://wandb.ai/byyoung3/ml-news/reports/Unsloth-unveils-1-58-bit-DeepSeek-R1--VmlldzoxMTEyMjU3Mg
Janus Pro pushes boundaries in text-to-image AI
DeepSeek’s Janus Pro enhances multimodal understanding and image generation with a powerful 7B-parameter model.
Mistral AI introduces Mistral Small 3, a high-performance 24B open-source model
The 24B-parameter model rivals larger AI models while delivering faster response times and efficiency.
Add a comment