Foundational Research
Exploring AI's limits, solving tough problems, and generating new knowledge for our company, our customers, and the world. We publish in journals, release open-source projects, and integrate findings into our Salesforce products.
Our Research Areas
Time Series Intelligence
Analyzing data patterns over time for informed decision - making insights.
Data Synthesis and Generation
Generate high - quality synthetic data that closely matches the real data our customers use in their businesses.
RLHF / DPO
Aligning LLMs with human expectations using optimization, reinforcement learning, and feedback.
Theoretical Deep Learning
Neural networks processing data, emulating human learning, solving complex problems.
Code Generation & Understanding
Improving productivity and enhancing software quality.
Large Action Model
Understand human requests and execute complex tasks
Multi-Agent Framework
Improving LLM-based task agent systems for seamless coordination via a manager agent.
Retrieval-Augmented Generation
Framework integrating retrieval and generation for enhanced content creation
LLM & Multimodal LLM
Pushing the boundaries of language understanding and generation.
Video Generation & Editing
Revolutionizing visual storytelling and content creation
Time Series Intelligence
Analyzing data patterns over time for informed decision - making insights.
Data Synthesis and Generation
Generate high - quality synthetic data that closely matches the real data our customers use in their businesses.
RLHF / DPO
Aligning LLMs with human expectations using optimization, reinforcement learning, and feedback.
Theoretical Deep Learning
Neural networks processing data, emulating human learning, solving complex problems.
Code Generation & Understanding
Improving productivity and enhancing software quality.
Large Action Model
Understand human requests and execute complex tasks
Multi-Agent Framework
Improving LLM-based task agent systems for seamless coordination via a manager agent.
Retrieval-Augmented Generation
Framework integrating retrieval and generation for enhanced content creation
LLM & Multimodal LLM
Pushing the boundaries of language understanding and generation.
Video Generation & Editing
Revolutionizing visual storytelling and content creation
Time Series Intelligence
Analyzing data patterns over time for informed decision - making insights.
Data Synthesis and Generation
Generate high - quality synthetic data that closely matches the real data our customers use in their businesses.
RLHF / DPO
Aligning LLMs with human expectations using optimization, reinforcement learning, and feedback.
Theoretical Deep Learning
Neural networks processing data, emulating human learning, solving complex problems.
Discover the latest in AI Research
AgentLite: Multi-Agent Network
Learn MorexGen-MM: The Latest Large Multimodal Models
Learn MorexLAM: Salesforce’s Family of in-house Large Action Models
Learn MoreCheck out our Open Source Projects
LAVIS: A Library for Language - Vision Intelligence
LAVIS is a Python deep learning library for LAnguage - and - VISion intelligence research and applications.
AgentLite
AgentLite is a research - oriented library designed for building and advancing LLM - based task - oriented agent systems.
Diffusion Model Alignment Using Direct Preference Optimization
This directory provides LoRA implementations of Diffusion DPO, a method to align diffusion models to human preferences by directly optimizing on human comparison data.
Natural Language Decathlon(decaNLP)
The Natural Language Decathlon is a multitask challenge that spans ten tasks.The framework itself is designed in a way that should make single - task training, transfer learning, and zero - shot evaluation simple.
xGen-7B
We trained a series of 7B LLMs named XGen-7B with standard dense attention on up to 8K sequence length for up to 1.5T tokens.
Conversational AI Programming with CodeGen
Our large-scale language model, CodeGen, which turns simple English prompts into executable code.
Merlion: A Machine Learning Library for Time Series
A Python library for time series intelligence that provides an end-to-end machine learning framework.
Moirai: A Time Series Foundation Model for Universal Forecasting
Moirai is a time series foundation model, offering capabilities such as forecasting tasks across multiple domains, frequencies, and variables in a zero - shot manner.
LAVIS: A Library for Language - Vision Intelligence
LAVIS is a Python deep learning library for LAnguage - and - VISion intelligence research and applications.
AgentLite
AgentLite is a research - oriented library designed for building and advancing LLM - based task - oriented agent systems.
Diffusion Model Alignment Using Direct Preference Optimization
This directory provides LoRA implementations of Diffusion DPO, a method to align diffusion models to human preferences by directly optimizing on human comparison data.
Natural Language Decathlon(decaNLP)
The Natural Language Decathlon is a multitask challenge that spans ten tasks.The framework itself is designed in a way that should make single - task training, transfer learning, and zero - shot evaluation simple.
xGen-7B
We trained a series of 7B LLMs named XGen-7B with standard dense attention on up to 8K sequence length for up to 1.5T tokens.
Conversational AI Programming with CodeGen
Our large-scale language model, CodeGen, which turns simple English prompts into executable code.
Merlion: A Machine Learning Library for Time Series
A Python library for time series intelligence that provides an end-to-end machine learning framework.
Moirai: A Time Series Foundation Model for Universal Forecasting
Moirai is a time series foundation model, offering capabilities such as forecasting tasks across multiple domains, frequencies, and variables in a zero - shot manner.
LAVIS: A Library for Language - Vision Intelligence
LAVIS is a Python deep learning library for LAnguage - and - VISion intelligence research and applications.
AgentLite
AgentLite is a research - oriented library designed for building and advancing LLM - based task - oriented agent systems.
Diffusion Model Alignment Using Direct Preference Optimization
This directory provides LoRA implementations of Diffusion DPO, a method to align diffusion models to human preferences by directly optimizing on human comparison data.
Natural Language Decathlon(decaNLP)
The Natural Language Decathlon is a multitask challenge that spans ten tasks.The framework itself is designed in a way that should make single - task training, transfer learning, and zero - shot evaluation simple.