Tags

• Topics: Artificial Intelligence Large language model Natural language processing Transformer Speed optimization Open source
• Additional: SotA

Significance

• Extremely reducing the size of LLMs and losing as little accuracy as possible

Intuitive summary

Technical summary

• a [[BERT]] variant using 0.3% of neurons needed for same performance using [[fast feedforward networks]]

Main resources

• [2311.10770] Exponentially Faster Language Modelling

Deep dive

Brain storming

Additional resources

Related

AI

Metadata

• paper #tool #processing #long #focus