Songting Chen

Worth reading: How the Apache Arrow Format Accelerates Query Result Transfer by Ian Cook One of the hidden problems of intersystem communication is transcription of data formats. Ian does a wonderful job of explaining the "why" of Apache Arrow as well as the "how" of its advantages. Read on...

8

Join PyTorch at Snowflake's #DataCloudDevDay on June 6 in San Francisco to learn how to accelerate GenAI performance with Native PyTorch. Register: https://hubs.la/Q02xQStt0

18

Seismic’s AI enablement assistant, Aura Copilot, is now available for early access. Teams can now get faster access to answers and content, automate tedious tasks, and gain insights for better business decisions. Aura also now integrates with Microsoft Teams and Microsoft 365 Copilot to give teams in-the-moment answers and content recommendation. https://lnkd.in/d_mSbDXK

44

1 Comment

📣 At Databricks, we want to help customers build more #inference-friendly #llms. Our MixAttention architecture demonstrates that reductions in KV cache size make it possible to maintain model quality while improving inference speed and reducing memory footprint. Read our latest blog post to learn more: https://lnkd.in/gcHxM4DX

133

3 Comments

Follow along John Chun Hong Leung's new tutorial to learn how you can ingest, extract, and visualize time-series data by using PySpark on Databricks (implementation code included).

24

Meta's Llama models have set a new standard for open large language models (LLMs) in the industry. With a remarkable 10-fold increase in downloads from last year, Llama models are leading the way in openness, modifiability, and cost efficiency. This advancement empowers customers to create innovative solutions tailored to their specific needs. Excitingly, Oracle Cloud Infrastructure (OCI) Data Science now fully supports Llama 2, 3, and 3.1 models, even on CPUs. #Meta #LlamaModels #OracleCloud #DataScience

10

"The new Columnar capabilities enable rapid real-time intelligence for data analysis and quick insights, along with further opportunities for traditional ML and AI" ⚡ Check out this write up on Capella columnar from private preview user Will Thayer, MBA Technologist 📝 https://bit.ly/3JGo8UR

24

1 Comment

Exciting news for customers: Fine-tuning for Anthropic’s Claude 3 Haiku model is now available in Amazon Bedrock—currently the only managed service to offer this capability. This means our customers can now customize Claude 3 Haiku with their own data to build GenAI applications tailored to their specific needs, while keeping their data secure. Since we launched the preview a few months ago, we've already seen some great use cases. One of our customers, Robin AI, a legal AI company, is using this feature to optimize legal workflows and help lawyers speed up contract processes. With a fine-tuned Claude 3 Haiku in Bedrock, they are helping lawyers finish contract review processes 90% faster. Will be great to see how other customers use this new capability. https://lnkd.in/g2jU8Kfv

324

5 Comments

For those who missed the Snowflake Data Summit keynote; here’s my GeneAI summarization :-) 1) Sridhar the new Snowflake CEO is smart as a pup and amazing academic and career pedigree, but has very little charisma or public speaking skills. Hope he improves on the latter over time. 2) Jensen zoomed in for his Keynote because he is still in Taiwan for Compudex. Jensen has labeled Taiwan the heart of GenAI Compute, all of Nvidia (and AMD)’s major suppliers are in Taiwan. It’s amazing this little island the size of California holds the future to AI. 3) Snowflake and Nvidia announced their partnership. Basically it’s combining Snowflake LLM Datastore with Nvidia LLM training and inference/generate. This is revolutionary. Snowflake pioneered the separation of compute from storage in massive EDWs, and now the circle is completely with bringing Nvidia compute training and inference/generation close to storage, eliminating bandwidth cost, latency and data policy/privacy issues. And as someone with Information Retrieval background, I see GenAI has come full circle with RAG - it’s just Precision/Recall, attributes vectorization as “embedding”, K-Nearest Neighbor, User Search Intent, DAG, ML algorithms, etc all over again. Except this time with supercharged GPU SoC that is improving at 16x Moore’s Law. We live in exciting times.

29

8 Comments

🚀 KAWA Analytics is the easiest way to build enterprise-grade data apps on Snowflake From risk management to dynamic pricing, ESG to market abuse surveillance, KAWA AI Co-builder makes it effortless. 🎥 Watch our product manager Narjisse Bouri demonstrate how to build a market abuse monitoring system in no time.

5

1 Comment

Confluent's Field CTO, Kai Waehner, shares insights on the future of streaming analytics and data management, highlighting the importance of reliable data flow, collaboration, and security. Discover how Confluent's cutting-edge platform revolutionizes streaming analytics. Read the full interview: https://lnkd.in/gSMVCmvy #StreamingAnalytics #DataManagement #FutureOfData #BusinessIntelligence

80

Back in March, #Databricks launched #DBRX—a super-advanced state-of-the-art open language model that's truly next-level. This new model marks a major step forward for #OpenSource #LLMs. What’s really cool is that DBRX outperforms nearly all established open source models and even rivals #ClosedSource giants like #GPT4, #Gemini, and #Mistral across various benchmarks. Want to know what makes #DatabricksDBRX so awesome? We've got you covered! Check out the full article below to learn everything you need to know about it. https://lnkd.in/g3w845G7

2

We just launched a native Snowflake connector that cuts processing speeds AND costs in half by integrating SQream Blue's GPU-accelerated technology. Benchmarked by the TPCx-BB analysis, SQream Blue outperformed Snowflake, handling 30 TB of data at 2X the speed and 1/2 the cost. This connector allows users to seamlessly integrate SQream’s platform into existing Snowflake workflows, reducing data transfer needs and optimizing cloud spend. Experience efficiency like never before with SQream Blue.

21

Have you tried running the DBRX models from Databricks? In our latest blog post we provide a tutorial to get you started with the model on Ori’s public cloud. We were very impressed with DBRX’s performance on the Hugging Face Open LLM Leaderboard before the updates last week. While DBRX Instruct did not score as highly under the new scoring system, it's still one of the best open models for Language Understanding (MMLU), Programming (HumanEval), and Math (GSM8K). We also used BeFOri, the benchmarking framework from Ori to determine DBRX Base performs inference most efficiently between 40 and 256 concurrent requests on one of Ori's 8 x H100 PCIe bare metal servers. Check out the blog to learn how you can use BeFOri to optimise self-hosted LLM performance on any infrastructure! https://lnkd.in/gi55WEks

43

Much has been made since last week's announcement of Apple Intelligence at Apple's WWDC. Many have been pontificating on what Apple can do from a training perspective with the vast amount of data being collected on iPhones, and this incredibly comprehensive view they have of an individual consumer. But it's very important to remember that *most* of the valuable data on an iPhone is in third-party apps, the most important of which are already owned by Google, Meta, and other giants. While Apple sanctions all apps that can be installed on iPhones through the App Store's monopoly, it does not have the right to use the application specific or consumer specific data being collected by those apps. Evidence of breaching this confidence would quickly become a massive problem. Much like Google's acquisition of Hubspot in order to easily incorporate all of an enterprises' historical content for each of its customers to build customer-specific AI models (eg think an Accenture-specific model trained on all historical hubspot-housed Accenture marketing data, and all other Google data Google about Accenture, available with the click of a button), it will be interesting to see if Apple begins to methodically buy up mobile app companies to acquire their underlying data sets for use in training. Given Apple's penchant for vertical integration, it would seem a more relevant approach than attempting to do licensing partnerships as OpenAI has to get at underlying training data. If Apple wants to own the consumer from a GenAI and personal assistant perspective, they'll need to get at the data inside the myriad apps on the iPhone. https://lnkd.in/eH-BcfNe

6

🎩 US #ELECTIONS are like 🐘 #HADOOP Every elections run reminds us how well the Hadoop principles fit the vote. if you don't see the point from the get-go, here is the detailing the analogy: https://lnkd.in/eh-3grt6

2

Did you miss this awesome demo Hugo Lu did with Youssef Mrini over at Databricks 🏘 In Orchestra you can trigger and monitor Databricks Workbooks, Workflows, data quality tests, and lots more. Our users also leverage Orchestra's overarching monitoring and alerting capabilities to link upstream jobs like data ingestion (e.g. ADF) through Databricks into downstream jobs like Reverse ELT with Hightouch or Power BI dataset refreshes. Check it out here 🔗 https://lnkd.in/dU2grpum #dataengineering #databricks

7

1 Comment