Generative AI in Production: Best Practices and Lessons Learned
Meryem Arik: Meryem is a recovering physicist, co-founder, and CEO of TitanML.TitanML is an NLP development platform that focuses on the deployability of LLMs - allowing businesses to build smaller and cheaper language model deployments easily.
- Greg Loughnane is a Generative AI and LLM educator, community builder, CEO, and Founder of AI Makerspace, where he’s building useful onramps for people looking to join the unautomatable workforce of the future.
- Alessya Visnjic is the CEO of WhyLabs, the market leader in AI observability. AI teams at Square and Glassdoor use the WhyLabs platform to ensure AI models, from classification to LLMs, generate accurate, fair, and safe customer experiences.
- Chris Alexiuk, aka The LLM Wizard and educator extraordinaire. Chris was previously an Instructor and curriculum developer at FourthBrain and is currently a founding ML Engineer at Ox and CTO at AI Makerspace.
- Hannes Hapke is a principal machine learning engineer at Digits . He implemented deep learning systems from inception to production and has authored two books: building ml pipelines and NLP in action.
- Ville Tuulos is the co-founder and CEO of Outerbounds, a human-centric platform for data, ML, and AI projects based on metaflow.org, an open-source project he started while working on ML Infra and Architecture at Netflix.
Questions
What advice would you give someone looking to build a new generative AI-powered product or feature? What steps would you advice someone to take to understand its feasibility for real-world applications? (Panelists to ask: Ville, Hannes, Chris, Greg)
What are the most common challenges when translating a generative AI product/feature prototype into a production-ready product? (Panelists to ask: Ville, Chris, Greg, Alessya)
Can you share a war story where translating a generative AI-powered prototype to production posed unexpected challenges and how you overcame them? (Panelists to ask: Hannes)
Are specific tools, platforms, or infrastructure indispensable when moving from prototype to production, specifically for generative AI-powered products or features? (Panelists to ask: Ville, Chris, Alessya)
How do you address scalability when implementing generative AI models in production, especially when the original research might have been conducted on smaller datasets? (Panelists to ask: Hannes)
What measures do you take to ensure that a generative AI model, which works well in a controlled research environment, is robust and reliable when deployed in real-world scenarios? (Panelists to ask: Hannes, Alessya)
With the rapid advancements in generative AI research, where do you see the future of its implementation in production heading? (Panelists to ask: Chris, Greg)
What advice would you give aspiring practitioners looking to bridge the skill gap between building generative AI toy examples in a notebook to production implementation? (Panelists to ask: Ville, Greg)
社群媒體聆聽
#225 The Full Stack Data Scientist | Savin Goyal, Co-Founder & CTO at Outerbounds
The role of the data scientist is changing. Some organizations are splitting the role into more narrowly focused jobs, while others are broadening it. The latter approach, known as the Full Stack Data Scientist, is derived from the concept of a full stack software engineer, with this role often including software engineering tasks. In particular, one of the key functions of a full stack data scientist is to take machine learning models and get them into production inside software. So, what separates projects from production? Savin Goyal is the Co-Founder & CTO at Outerbounds. In addition to his work at Outerbounds, Savin is the creator of the open source machine learning management platform Metaflow. Previously Savin has worked as a Software Engineer at Netflix and LinkedIn. In the episode, Richie and Savin explore the definition of production in data science, steps to move from internal projects to production, the lifecycle of a machine learning project, success stories in data science, challenges in quality control, Metaflow, scalability and robustness in production, AI and MLOps, advice for organizations and much more. Find DataFramed on DataCamp https://www.datacamp.com/podcast and on your preferred podcast streaming platform: Apple Podcasts: https://podcasts.apple.com/us/podcast/dataframed/id1336150688 Spotify: https://open.spotify.com/show/02yJXEJAJiQ0Vm2AO9Xj6X?si=d08431f59edc4ccd Links Mentioned in the Show: Outerbounds - https://outerbounds.com/ Metaflow - https://metaflow.org/ Connect with Savin on Linkedin - https://www.linkedin.com/in/savingoyal/ [Course] Developing Machine Learning Models for Production - https://www.datacamp.com/courses/developing-machine-learning-models-for-production Related Episode: Why ML Projects Fail, and How to Ensure Success with Eric Siegel, Founder of Machine Learning Week, Former Columbia Professor, and Bestselling Author - https://www.datacamp.com/podcast/why-ml-projects-fail-and-how-to-ensure-success-with-eric-siegel-founder-of-machine-learning-week Rewatch sessions from RADAR: AI Edition - https://www.datacamp.com/radar-ai-2024 New to DataCamp? Learn on the go using the DataCamp mobile app - https://www.datacamp.com/mobile Empower your business with world-class data and AI skills with DataCamp for business - https://www.datacamp.com/business
Generative AI in Production: Best Practices and Lessons Learned
Meryem Arik: Meryem is a recovering physicist, co-founder, and CEO of TitanML.TitanML is an NLP development platform that focuses on the deployability of LLMs - allowing businesses to build smaller and cheaper language model deployments easily. - Greg Loughnane is a Generative AI and LLM educator, community builder, CEO, and Founder of AI Makerspace, where he’s building useful onramps for people looking to join the unautomatable workforce of the future. - Alessya Visnjic is the CEO of WhyLabs, the market leader in AI observability. AI teams at Square and Glassdoor use the WhyLabs platform to ensure AI models, from classification to LLMs, generate accurate, fair, and safe customer experiences. - Chris Alexiuk, aka The LLM Wizard and educator extraordinaire. Chris was previously an Instructor and curriculum developer at FourthBrain and is currently a founding ML Engineer at Ox and CTO at AI Makerspace. - Hannes Hapke is a principal machine learning engineer at Digits . He implemented deep learning systems from inception to production and has authored two books: building ml pipelines and NLP in action. - Ville Tuulos is the co-founder and CEO of Outerbounds, a human-centric platform for data, ML, and AI projects based on metaflow.org, an open-source project he started while working on ML Infra and Architecture at Netflix. Questions What advice would you give someone looking to build a new generative AI-powered product or feature? What steps would you advice someone to take to understand its feasibility for real-world applications? (Panelists to ask: Ville, Hannes, Chris, Greg) What are the most common challenges when translating a generative AI product/feature prototype into a production-ready product? (Panelists to ask: Ville, Chris, Greg, Alessya) Can you share a war story where translating a generative AI-powered prototype to production posed unexpected challenges and how you overcame them? (Panelists to ask: Hannes) Are specific tools, platforms, or infrastructure indispensable when moving from prototype to production, specifically for generative AI-powered products or features? (Panelists to ask: Ville, Chris, Alessya) How do you address scalability when implementing generative AI models in production, especially when the original research might have been conducted on smaller datasets? (Panelists to ask: Hannes) What measures do you take to ensure that a generative AI model, which works well in a controlled research environment, is robust and reliable when deployed in real-world scenarios? (Panelists to ask: Hannes, Alessya) With the rapid advancements in generative AI research, where do you see the future of its implementation in production heading? (Panelists to ask: Chris, Greg) What advice would you give aspiring practitioners looking to bridge the skill gap between building generative AI toy examples in a notebook to production implementation? (Panelists to ask: Ville, Greg)
Metaflow: Supercharging our data scientist productivity
Speakers: Ravi Kiran Chirravuri - Machine Learning Infrastructure, Netflix Jan Florjanczyk - Machine Learning Infrastructure, Netflix Abstract: Netflix's unique culture affords its data scientists an extraordinary amount of freedom. They are expected to build, deploy, and operate large machine learning workflows autonomously without the need to be significantly experienced with systems or data engineering. Metaflow, our ML framework (now open-source at metaflow.org), provides them with delightful abstractions to manage their project's lifecycle end-to-end, leveraging the strengths of the cloud: elastic compute and high-throughput storage. In this talk, we present our human-centric design principles that you can now adopt with ease. I don't think the talk will be very sales-y and focuses more on the life of a data scientist at Netflix and how the infrastructure is designed for a freedom & responsibility centric-culture. I think it is also expected to echo a lot with small-mid size companies who might not be able to afford a MLI team themselves and yet want to design/leverage infrastructure to improve DS productivity. We have also introduced a new scheduler integration (with AWS SFN) which we haven't publicly spoken about and it'll help dispel the belief that "Metaflow is not yet another DAG scheduler" but in fact interplays well with existing schedulers.
總共有 4 筆社群媒體資料需要解鎖才能查看