Projects

Our lab is actively engaged in projects focused on creating language model agents that translate language instructions into executable actions across real-world domains such as databases (data agent), web applications (plugins/web agent), and the physical world (robotic agent) etc. We are currently developing an open-source framework to facilitate the construction and assessment of these agents, starting with XLang Agent demos. In the coming months, we'll open-source essential projects like frameworks, models, methods, and benchmarks, aiming to establish a robust community dedicated to building capable multifunctional agents.

Selected Projects

OpenAgents: An Open Platform for Language Agents in the Wild

ChatGPT Plus replica for researcher, developers and general users, 2.5k github stars

Lemur: Open Foundation Models for Language Agents

Lemur-70B and Lemur-70B-chat, Open & SOTA Foundation Models for Language Agents

Text2Reward: Automated Dense Reward Function Generation for Reinforcement Learning

One of the earliest works on using LLMs for RL reward function generation

Instructor Embeddings: One Embedder, Any Task

Over 500k downloads, 1k github stars, and used by 500+ projects, including LangChain, Pytorch serve

DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation

1,000 natural, diverse, realistic data science questions over Python libraries

Binder: Binding Language Models in Symbolic Languages

One of the earliest works empowering LLMs with tools: integrating LLM calls in programming languages

UnifiedSKG: A Unified Framework for Structured Knowledge Grounding

An overall summary of structure knowledge grounding before LLM era

Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL Task

One of the most popular complex text-to-SQL benchmarks with >200 submissions from leading research labs, including Google, Facebook, UCB, CMU, UW