- llm
- nlp
- reasoning
- rl
- rlhf
- ssm
- theory
- information-theory
- representation-learning
- optimization
- rag
- alignment
- paper-review
- survey
- notes
•
•
•
•
•
•
•
•
•
•
•
•
•
•
-
REFRAG: Rethinking RAG-Based Decoding
Compress-sense-expand over retrieved passages for 31× TTFT speedup in RAG systems.
-
[Survey] Recent LLM Technical Reports
A collection of recent technical reports from Google DeepMind, xAI, AI21Labs, Databricks, HyperCLOVA, and others.
-
[Survey] Recent approaches on Super-Alignment
A collection of recent papers on weak-to-strong generalization, RLHF, and alignment for large models.
-
[Survey] Recent approaches on Efficient ML
A collection of recent papers on PEFT, quantization, and pruning for large language models.
-
Sliced Mutual Information for Memorization and Generalization
Using sliced mutual information as a tractable analytical framework for studying neural network memorization vs. generalization.