accuracy degradation
Accuracy degradation describes the decline in a model's performance when it encounters new data that differs from the training distribution. Recognizing the causes of accuracy degradation is essential for improving model robustness.
- Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts
- LoTA-QAF: Lossless Ternary Adaptation for Quantization-Aware Fine-Tuning
- Quartet: Native FP4 Training Can Be Optimal for Large Language Models
- Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning
- SALS: Sparse Attention in Latent Space for KV Cache Compression
- Sim-LLM: Optimizing LLM Inference at the Edge through Inter-Task KV Reuse