Snowflake AI Research

Think. Execute. Excel: Arctic Text2SQL with Execution-Guided CoT

Snowflake Engineering Blog Featured Image

Snowflake Arctic Embed Joins ArcticTraining: Simple And Scalable Embedding Model Training

Benchmarking LLM-as-a-Judge for the RAG Triad Metrics

view all blog posts

59 Results

ARC ‘Challenge’ Is Not That Challenging

Publication

STUN (Structured-Then-Unstructured) to deploy large models efficiently

Publication

Arctic-TILT. Business Document Understanding at Sub-Billion Scale

Publication

Arctic Inference with Shift Parallelism: Fast and Efficient Open Source Inference System for Enterprise AI

Publication

Universal Checkpointing: A Flexible and Efficient Distributed Checkpointing System for Large-Scale DNN Training with Reconfigurable Parallelis

Publication

Snowpark: Performant, Secure, User-Friendly Data Engineering and AI/ML Next To Your Data

Access paper

view all publications

Explore our projects and contributions

Github

Arctic Inference

Arctic Inference combines Shift Parallelism, SwiftKV, fast speculative decoding, and optimized embedding inference to deliver state-of-the-art open-source AI performance.

Github

Arctic Training

Snowflake’s framework for optimizing large language models using scalable, enterprise-grade techniques like execution-based feedback, retrieval-augmented learning, and efficient fine-tuning.

Hugging face

Arctic Models

Download Snowflake’s enterprise-ready OSS models and frameworks for state-of-the-art retrieval, inference, speculative decoding, and Text2SQL optimization.

Github

Trulens

Tracing, evals, experiment tracking for LLM applications. Created and maintained by researchers & developers now primarily with Snowflake.

Github

DeepSpeed

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

Github

vLLM

A high-throughput and memory-efficient inference and serving engine for LLMs. Snowflake researchers and engineers actively contribute to vLLM.