Mellea is a library for writing generative programs.

🥯 [ACL 2026 Findings] Bayesian Active Learning with Gaussian Processes Guided by LLM Relevance Scoring for Dense Passage Retrieval

[ICLR 2026 Oral] Official Implementation of the paper "MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interactions"

[ACL'26 Findings] Official Repository of "Hybrid-Vector Retrieval for Visually Rich Documents: Combining Single-Vector Efficiency and Multi-Vector Accuracy"

[ICLR 2026] LookaheadKV: Fast and Accurate KV Cache Eviction by Glimpsing into the Future without Generation

LLM-as-a-Verifier is a general-purpose framework that provides fine-grained feedback for any agent without requiring additional training. It achieves SOTA performance across coding, robotics, and m…

AI agents running research on single-GPU nanochat training automatically

Paper list for Efficient Reasoning.

Official JAX implementation of End-to-End Test-Time Training for Long Context

[AAAI26]: DS SERVE: The Largest Open Vector Store over Pretain Data; A Framework for Efficient and Scalable Neural Retrieval

A collection of token reduction (token pruning, merging, clustering, etc.) techniques for ML/AI

[NeurIPS'25 Oral] Query-agnostic KV cache eviction: 3–4× reduction in memory and 2× decrease in latency (Qwen3/2.5, Gemma3, LLaMA3)

[SIGIR 2025] The official repo for "Scaling Sparse and Dense Retrieval in Decoder-Only LLMs"

Official Implementation of the paper "Jointly Reinforcing Diversity and Quality in Language Model Generations"

MICA (Multiple Intelligent Conversational Agents) is designed to simplify the development of customer service bots.

Search-R1: An Efficient, Scalable RL Training Framework for Reasoning & Search Engine Calling interleaved LLM based on veRL

On the Theoretical Limitations of Embedding-Based Retrieval

Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation (NeurIPS 2025)

[COLM 2025] DEL: Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding

[EACL 2026] Detecting Training Data of Large Language Models via Expectation Maximization

Modifying Large Language Models Post-training for Diverse Creative Writing

A curated list of personalized alignment resources (continually updated).

📰 Must-read papers and blogs on Speculative Decoding ⚡️

Fully open data curation for reasoning models