Seen on the The Wall Street Journal ! Join us at snowflake.com/summit/

The Snowflake AI Team is gradually releasing the Arctic Cookbooks.

Why is this important?

These articles have valuable insights and tutorials; they don't just discuss their models but provide practical guidance.

They just published data sourcing and composition article.

#SnowflakeArctic from @snowflakedb is a big deal not just because it's an amazing model but because it moves the ecosystem of AI developers forward with real insights from a crack team.

And of course, no day of mine is complete without a cartoon from chatGPT. We want to teach

❄️ We're excited to introduce Snowflake’s innovative large language model, Arctic, into our model catalog. Discover Arctic in Microsoft Azure AI: msft.it/6018YKJDn

Congrats Snowflake sridhar Vivek Raghunathan Arctic Instruct is almost tied with GPT4-Turbo at 2.6% Hallucination Rate while still achieving a 100% Answer Rate, which is pretty hard to do. Kudos to your team.

Snowflake launches Arctic, an open ‘mixture-of-experts’ LLM to take on DBRX, Llama 3 venturebeat.com/data-infrastru…

VentureBeat Love the speed of innovation in the open source LLM world. This new model just took the #2 spot on Vectara's HHEM with an incredibly low 2.6% hallucination rate: huggingface.co/spaces/vectara…

Wow, Llama3 has been surpassed by Snowflake Arctic! Three interesting takeaways:

1. It only uses 3T tokens with 3-stage training, compared to 15 tokens used in llama3. This at least indicates that dataset curriculum learning (3 stage with different types of model capability)

Exciting news from Snowflake! We're excited to be partners with them in transforming apps and workflows through AI.

+ stay tuned for more to come soon. 👀

Thrilled to share the release of #Snowflake Arctic , an enterprise focused foundation model from #Snowflake , that is efficiently intelligent and truly open.

I joined Snowflake in December and immediately hit the ground running on Arctic. After 5+ years at Microsoft building DeepSpeed I was excited for a new challenge. We started on day one completely from scratch, which has been an exhilarating journey to say the least! 🧵1/3

What a roller coaster the past few months have been! I'm excited and grateful for the opportunity to collaborate with such a badass team.

Stay tuned for more updates - blog posts, model improvements, and more...

1/4 Have you wondered how to optimize sys-perf for training Arctic-like models (MoE arch)? Let’s dive in! Our first technique: custom fused kernels. By crafting these kernels, we streamline irregular and sparse operators, boosting efficiency. #SnowflakeArctic #SystemOptimization

Snowflake Arctic's architecture vs. Grok, Mixtral, and Llama 3

ICYMI, this is a gigantic 480 billion parameter model made of 128 tiny experts, each not even 4B parameters

fully #opensource (Apache 2.0), link to detailed thread in comments

This is honestly more exciting than yet another LLM trained on internet data alone, and to boot, made by some very awesome people. Congratulations to the Neeva/Snowflake team