- AI
- LLM
- MultimodalAI
- Innovation
- OpenSource
- MachineLearning
- Research
•
•
•
•
•
•
-
Janus-Pro ; DeepSeek’s Next-Gen Multimodal Model for Vision & Text-to-Image 🖼️🤖
Janus-Pro from DeepSeek AI sets new standards in multimodal understanding and text-to-image generation, with open access for the research community.
-
Memory Layers in Large Language Models ; Boosting LLM Performance 🧠
Meta’s research on memory layers in LLMs shows how trainable key-value lookups dramatically improve factual accuracy and efficiency.
-
Large Concept Models ; Advancing Abstract Reasoning in Language Modeling 🤖💡
Meta’s Large Concept Models (LCMs) introduce semantic-level language modeling and diffusion-based generation for improved reasoning and multilingual versatility.
-
CosyVoice 2 ; Streaming Speech Synthesis with Human-Like Naturalness 🎤
CosyVoice 2 from Alibaba Group advances streaming speech synthesis with LLM-powered, low-latency, multilingual, and near human-parity speech generation.
-
DeepSeek-V3 ; 671B-Parameter MoE LLM Setting New AI Benchmarks 🌟🤖
DeepSeek-V3 is a 671B Mixture-of-Experts LLM introducing new architectural and training strategies—excelling at code, math, and beyond.