Knowledge note
Training
Navigation hub for model training, parallelism, model families, efficient attention, adaptation, and architectural mechanisms.
Training
- Training Parallelism — Tensor, pipeline, sequence, and distributed training methods.
- Foundation Models — GPT and LLaMA foundation-model papers.
- Model Families — DeepSeek and Kimi model reports.
- Efficient Attention Training — Sparse, recurrent, sliding-window, and shared-cache attention.
- Model Adaptation — Fine-tuning, transfer, and self-evolving agent methods.
- Training Architectures — Architectural mechanisms motivated by training stability and scale.