Tag: 小模型 (2 articles)

LiquidAI's LFM2.5-2.6B uses innovative Agentic RL to outperform models 4x larger on tool use and instruction following, enabling capable, privacy-preserving agents to run locally on everyday devices.

Dharma AI's experiment shows a 3-billion-parameter specialized OCR model outperformed all commercial frontier models on a specific enterprise task at 50 times lower cost, revealing a new trend where 'specialization' matters more than 'scale' in AI procurement.