Tsinghua's Latest Research! How to Theoretically Unify SFT and RL, and the Efficient Adaptive Algorithm Hybrid Post-Training

Post-training of large language models has long been clearly divided into two paradigms: supervised fine-tuning (SFT) centered on imitation and reinforcement learning (RL) driven by exploration.

Hongjing Technology Obtains New Patent: Innovative Technology for Enhancing Multi-Turn Dialogue Quality

What is Multi-Turn Dialogue Optimization Technology? Multi-turn dialogue optimization technology refers to how to improve the quality and coherence of machine dialogue during multiple exchanges with ...
Semiconductor Engineering

Analog Plus 3D Optics to Accelerate AI inference and Combinatorial Optimization (Microsoft, Cambridge)

A new technical paper titled “Analog optical computer for AI inference and combinatorial optimization” was published by ...
news.ucsb.edu

Innovative hardware for rapidly solving optimization problems

The rise of AI, graphic processing, combinatorial optimization and other data-intensive applications has resulted in data-processing bottlenecks, as ever greater amounts of data must be shuttled back ...
Semiconductor Engineering

Thermally Aware Chiplet Placement Algorithm Based on Automatic Differentiation (MIT, IBM)

A new technical paper titled “DiffChip: Thermally Aware Chip Placement with Automatic Differentiation” was published by researchers at MIT and IBM. “Chiplets are modular integrated circuits that can ...