Papers
Browse the full research library.
Ten papers per page, with summaries, diagrams, formulas, concepts and citations kept close to the source.
RAG Survey
A systematic survey organizing RAG architectures, training strategies, applications, evaluation, and open challenges.
Best used as a map of the RAG design space before diving into individual retrieval and generation papers.
Yunfan Gao, Yun Xiong, Xinyu Gao
Read paper
PaperQA
A retrieval-augmented research agent that searches scientific literature, evaluates source relevance, and synthesizes evidence-backed answers.
Strong foundation for building research RAG systems with retrieval, source assessment, and synthesis.
Jakub Lála, Odhran O'Donoghue, Aleksandar Shtedritski
Read paper
Generative Agents
An agent architecture combining memory, retrieval, reflection, and planning to simulate believable long-horizon behavior.
Useful for understanding why long-running agents need memory beyond the immediate context window.
Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai
Read paper
LFM2.5-24B-A2B
LFM2.5-24B-A2B is a 24B-parameter hybrid MoE reasoning model with low active compute, a 131K context window, and training optimized for local tool-using agents.
Useful for AI engineers building local agents: study its MoE architecture, tool-use benchmarks, deployment speed, licensing, and reward-hacking lessons.
Liquid AI
Read paper
WikiSkill
WikiSkill gives agent skill evolution a persistent memory of what worked, what failed, and why.
Read this if you are building self-improving agents, agent memory, reusable skills, or production feedback loops. The most important idea is the asymmetric rollback: failed skill edits are discarded, but the experience behind them remains available for future evolution.
Liyan Tang, Cyrus Rashtchian, Chun-Sung Ferng
Read paper
OmniQuant
OmniQuant makes PTQ parameters learnable through Learnable Weight Clipping and Learnable Equivalent Transformation, supporting both weight-only and weight-activation LLM quantization.
Essential for understanding learnable PTQ, LWC, LET, block-wise reconstruction, W2A16, W4A4, and modern low-bit calibration.
Wenqi Shao, Mengzhao Chen, Zhaoyang Zhang
Read paper
QuIP
QuIP makes two-bit post-training quantization viable for large language models by combining adaptive rounding with incoherence processing.
Essential for understanding why incoherence and second-order information enable accurate 2-bit LLM quantization.
Jerry Chee, Yaohui Cai, Volodymyr Kuleshov
Read paper
SpQR
SpQR combines sparse high-precision outlier storage with 3–4-bit group quantization to achieve near-lossless LLM weight compression.
Essential for understanding sparse outlier handling, hierarchical quantization, metadata overhead, and near-lossless 3–4 bit LLM deployment.
Tim Dettmers, Ruslan Svirschevski, Vage Egiazarian
Read paper
AWQ
AWQ uses activation statistics to identify salient weight channels and scale them before low-bit weight-only quantization, preserving accuracy while enabling hardware-efficient LLM inference.
A foundational low-bit quantization paper that connects activation statistics, quantization error, hardware efficiency, calibration robustness, and edge inference.
Ji Lin, Jiaming Tang, Haotian Tang
Read paper
QLoRA
QLoRA combines 4-bit NF4 quantization with LoRA adapters to fine-tune very large language models under a single-GPU memory budget.
Essential for understanding modern 4-bit fine-tuning: NF4, double quantization, paged optimizers, LoRA, memory accounting, and evaluation.
Tim Dettmers, Artidoro Pagnoni, Ari Holtzman
Read paper