MiniCPM4 - Efficient Edge-Side Large Model
๐ MiniCPM4 is here! 5x faster on end devices ๐ฅ
โจ What's new:
๐๏ธ Efficient Model Architecture
- InfLLM v2 -- Trainable Sparse Attention Mechanism
๐ง Efficient Learning Algorithms
- Model Wind Tunnel 2.0 -- Efficient Predictable Scaling
- BitCPM -- Ultimate Ternary Quantization
๐ High-Quality Training Data
- UltraClean -- High-quality Pre-training Data Filtering and Generation
- UltraChat v2 -- High-quality Supervised Fine-tuning Data Generation
โก Efficient Inference System:
- CPM.cu -- Light Lightweight and Efficient CUDA Inference Framework
- ArkInfer -- Cross-Platform Deployment Frameworโฆ
Watch on YouTube โ
(saves to browser)
DeepCamp AI