- Видео 54
- Просмотров 25 596
deepsense
США
Добавлен 5 мар 2020
deepsense.ai helps companies implement AI-powered solutions, with the main focus on AI Guidance and AI Implementation Services.
Our commitment and know-how have been appreciated by global clients including Nielsen, L’Oréal, Intel, Nvidia, United Nations, BNP Paribas, Santander, Hitachi, and Brainly.
Wherever you are on your AI journey, we can guide you and help implement projects in Generative AI, Natural Language Processing, Computer Vision, Predictive Analytics, MLOps and Data Engineering. We also deliver training programs to support companies in building AI capabilities in-house.
Our commitment and know-how have been appreciated by global clients including Nielsen, L’Oréal, Intel, Nvidia, United Nations, BNP Paribas, Santander, Hitachi, and Brainly.
Wherever you are on your AI journey, we can guide you and help implement projects in Generative AI, Natural Language Processing, Computer Vision, Predictive Analytics, MLOps and Data Engineering. We also deliver training programs to support companies in building AI capabilities in-house.
Speed up Deep Learning Models with Minimal Effort
Speaker: Michał Kustosz
Machine Learning Engineer
Discover how to significantly speed up PyTorch models on GPU with minimal effort by tackling compute, memory, and CPU overhead. This video explores the inner workings of GPUs, identifies bottlenecks using profiling tools, and shares techniques like leveraging Tensor Cores, using the Torch Compiler, applying quantization, and enabling Automatic Mixed Precision (AMP). Boost performance and reduce memory usage while maintaining model accuracy.
00:00 Intro
02:40 Compute bound
12:35 Memory bound
19:32 Overhead bound
21:35 Profiling GPU code
23:50 Solutions
Check our website: deepsense.ai/
Linkedin: www.linkedin.com/company/deepsense-ai
Facebook: facebook....
Machine Learning Engineer
Discover how to significantly speed up PyTorch models on GPU with minimal effort by tackling compute, memory, and CPU overhead. This video explores the inner workings of GPUs, identifies bottlenecks using profiling tools, and shares techniques like leveraging Tensor Cores, using the Torch Compiler, applying quantization, and enabling Automatic Mixed Precision (AMP). Boost performance and reduce memory usage while maintaining model accuracy.
00:00 Intro
02:40 Compute bound
12:35 Memory bound
19:32 Overhead bound
21:35 Profiling GPU code
23:50 Solutions
Check our website: deepsense.ai/
Linkedin: www.linkedin.com/company/deepsense-ai
Facebook: facebook....
Просмотров: 82
Видео
Programming foundation models. Leveraging DSPy for automated prompt engineering
Просмотров 14721 день назад
Speaker: Michał Pstrąg Machine Learning Engineer DSPy is one of the most popular libraries for automated prompt engineering, allowing developers to write better prompts faster. This session provides a concise overview of DSPy's features, including LM prompting, pipeline construction, and prompt optimization. Discover its ecosystem and explore the potential for developing automated, self-improvi...
Connecting the dots from docs. GraphRAG introduction
Просмотров 271Месяц назад
Speaker: Patryk Wyżgowski Machine Learning Engineer Join our speaker on a journey through an entire network of concepts such as Knowledge Graphs and GraphRAG. In this session, we'll explore the basic ideas, architecture, and characteristics of the GraphRAG method, including examples based on Neo4j Langchain as well as Microsoft's GraphRAG implementation. As a case study, we will delve into LLM-...
LLM-First: Where Tech Possibilities Meet User Value in Products and Services
Просмотров 239Месяц назад
In this exclusive interview, Tom Bianculli, CTO of @ZebraTechnologiesGlobal shares groundbreaking insights into how AI and UX are transforming industries. We explore the concept of LLM-first thinking and its impact on the frontline workforce, decision-making in AI, and the balance between tech innovation and user value. Key topics covered: Tech Possibility vs. User Value: How to align AI innov...
Revolutionizing Learning. Building an AI Product for 15 Million Users Daily
Просмотров 1792 месяца назад
💡 Join us as we dive into the world of AI in education with Bill Salak, CTO/COO of Brainly. Learn how he led the development of an AI-powered platform that supports 15 million students every day. 🎓 In this interview, Bill shares the challenges and breakthroughs in building an AI product at such scale, how Brainly ensures reliable and accurate learning experiences, and what the future holds for ...
Intro to Snowflake: More than SQL DBMS
Просмотров 962 месяца назад
Speaker: Grzegorz Rybak Data Engineer Explore the power of Snowflake for data-first projects and discover how this versatile platform can potentially become an end-to-end data layer in your company. This high-level overview session covers Snowflake's suite of features and capabilities in data storage, processing, notebooks-powered analysis, and sharing the results as a dataset or a fully-fledge...
Polar bears not on the verge of extinction. Pandas vs Polars comparison
Просмотров 1803 месяца назад
Speaker: Jakub Cierocki Senior Machine Learning Engineer UICN classifies Polar Bears as VU, that is 'vulnerable' (of extinction) species. But that's not really the case in the Python world. Polars DataFrame library, being a fresh competitor, challenges an old and infirm king - Pandas, quickly gaining popularity over time. A Rust-Python library whose development started in 2020 and 1.0 version w...
Understand your Data with Minimal Effort in 2024
Просмотров 1203 месяца назад
Speaker: Michał Grochecki Machine Learning Engineer Do you get overwhelmed when dealing with a new dataset? Are you struggling to find ways to analyze and visualize it? Not anymore! This talk covers methods that will help you to understand your data with little to no effort. 00:00 Intro & Agenda 1:06 Exploratory Data Analysis 1:37 Key factors for effective visualization 2:14 Understand your dat...
Disrupt or Be Disrupted: Pushing the Boundaries of AI in Niche Markets
Просмотров 4723 месяца назад
We had an insightful and engaging discussion with Dr. Burkhard Boeckem, CTO of Hexagon AB. Here are the highlights: ✨ 00:49 Hexagon's Mission & Focus: Dr. Boeckem introduced Hexagon, a leader in the geospatial industry, and explained their mission to drive innovation through AI. 03:07 AI Development Strategy: We explored Hexagon's strategic approach to AI, balancing cutting-edge research with ...
Maintaining Kubernetes applications with GitOps and ArgoCD
Просмотров 714 месяца назад
Speaker: Szymon Klimowicz MLOps Engineer Discover GitOps and its power for automated infrastructure management! In this video, we'll install ArgoCD on a Kubernetes cluster and deploy a JupyterHub service using GitOps principles. Join us for this hands-on tutorial and see how GitOps can transform your infrastructure management process! Check our website: deepsense.ai/ Linkedin: www.linkedin.com/...
Is your LLM behaving? Monitoring LLM applications
Просмотров 1865 месяцев назад
Speaker: Alan Konarski Senior Machine Learning Engineer Discover how to ensure LLM applications perform optimally. This talk covers common issues, the importance of monitoring, and key metrics. Gain insights into the tools and techniques used to enhance visibility and effectively monitor LLM applications. #llms #largelanguagemodels #rag 00:00 Intro 00:24 LLM Applications 05:20 LLM Applications ...
Bridging LLMs and databases. Lessons learned in production
Просмотров 2056 месяцев назад
Speaker: Mateusz Hordyński Technical Leader Prepare to dive into the exciting world of Large Language Models (LLMs) and structured data sources. In this session, we'll shed light on how to link LLMs to relational databases. Through exploration of case studies, you'll understand the benefits, the obstacles, and what's on the horizon for integrating LLMs and structured data. Last but not least we...
LLM is not listening to you? Do this instead! A Prompt Engineering Guide
Просмотров 2836 месяцев назад
Speaker: Sebastian Chwilczyński Machine Learning Engineer at deepsense.ai Have you ever been disappointed with GPT's response? If yes, join us for a deep dive into the world of Prompt Engineering. We will take you through motivation, state-of-the-art prompt engineering techniques and current problems alleviated with a very powerful concept of agentic systems. Warning: a lot of practical tips ar...
Gaussian Splatting: Crafting 3D Renders
Просмотров 3116 месяцев назад
Speaker: Natalia Czerep Machine Learning Engineer Join us for a deep dive into the world of Gaussian Splatting, where we explore the fundamentals and practical applications in 3D rendering. From scene rendering and editing to creating renders from sparse data, we will uncover the potential of the Gaussian splats. 00:00 Intro & Agenda 1:42 Context of the project - data generation of urban scenes...
The untapped potential of AI in large enterprises. It’s more than you think!
Просмотров 3407 месяцев назад
The untapped potential of AI in large enterprises. It’s more than you think!
The Rise of Small LMs. How we integrated RAG with SLMs into embedded devices
Просмотров 2867 месяцев назад
The Rise of Small LMs. How we integrated RAG with SLMs into embedded devices
Computer Vision for Mobile and Edge Devices
Просмотров 3028 месяцев назад
Computer Vision for Mobile and Edge Devices
Let's talk about R in RAG. Practical tips for improving your retrieval results
Просмотров 4178 месяцев назад
Let's talk about R in RAG. Practical tips for improving your retrieval results
React / TypeScript. How to start in 2024?
Просмотров 4579 месяцев назад
React / TypeScript. How to start in 2024?
OpenAI’s Assistants API. New OpenAI’s LLM tool
Просмотров 7259 месяцев назад
OpenAI’s Assistants API. New OpenAI’s LLM tool
LLMs deployment. Hugging Face Text Generation Inference and alternatives
Просмотров 2,4 тыс.9 месяцев назад
LLMs deployment. Hugging Face Text Generation Inference and alternatives
LLMs and WinRAR. What do they have in common?
Просмотров 39110 месяцев назад
LLMs and WinRAR. What do they have in common?
Identifying high-impact AI use cases in modern enterprises. A roadmap to success
Просмотров 1,8 тыс.10 месяцев назад
Identifying high-impact AI use cases in modern enterprises. A roadmap to success
Zero-shot object detection with Grounding DINO
Просмотров 2,3 тыс.11 месяцев назад
Zero-shot object detection with Grounding DINO
Kubeflow Pipelines on GCP (Vertex AI)
Просмотров 3 тыс.11 месяцев назад
Kubeflow Pipelines on GCP (Vertex AI)
Context extension challenges in Large Language Models
Просмотров 238Год назад
Context extension challenges in Large Language Models
Coding agents for Data Science. Challenges of complex code generation
Просмотров 194Год назад
Coding agents for Data Science. Challenges of complex code generation
Forecasting with deep learning . The revolution that didn't happen
Просмотров 180Год назад
Forecasting with deep learning . The revolution that didn't happen
You're doing a fantastic job! I have a quick question: My OKX wallet holds some USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). How can I transfer them to Binance?
First !
What would you recommend for building simple pipelines in Vertex AI that includes loading data from BQ, training and deploying model ( if KFP is an overkill) ?
Hey! Sorry for late response. Actually the use case you're asking about was the same exact thing I was working on. In my opinion it CAN be an overkill. If you want your engineers/users/stakeholders to see a beautiful graph, that "the data flows" and you have it all automated - then it's awesome. From my experience, there was too much time spent on learning/exploring the framework itself and shaping the solution in an optimal way. Besides that, I observed a significant slowdown in performance (mentioned in the presentation) - each component being a separate container is just an additional overhead. So I'd recommend to start researching/implementing with a simple set of scripts and once you're confident about your solution and ready to productionize/automate, then migrate everything to KFP. There are plenty examples of Vertex pipelines made by Google, I'd start with that, but those are usually much simpler than the real-world scenarios.
ollama is good alternative
Now I'm highly motivated to use LIDA in my upcoming project!
Great job Mateusz
Is there a version in polish?
Great job. Thanks :)
Well done
How can we avoid short-sightedness when seeking productivity through AI? Listen to Ambica Rajagopal, PHD from MICHELIN IT & Digital, in the full interview called "The untapped potential of AI in large enterprises" on our channel. #aiimpact #aibusiness #aitrends #aishorts #shorts #aiproject #michelin #deepsense #deeptalk #businesstalk #enterprises #aisolutions #tranding
Watching after a few months 🥲Hopefully not that different now
Nicely explained. Thank you.
Great video. I have a question. I have a input list, and my component can run on each item independently. So I am using dsl.parallel for this. But the thing is i have to use a big file from gcs in each of those instance of the component. How should i go about it?
Great overview!!! Another alternative to consider would be Nvidia Triton Inference Server
hi there. Are these models able to differentiate and label if e.g. a car is broken and what type of damage it has ? Does the complexity of the recognition solely depends on the datasets ?
Thanks Marianna, awesome video!
Nice and comprehensive slide on GroundingDINO!
how can we get the code to these algorithms?
Nice work. I was wondering if Ollama could be another alternative for TGI?
Insightful. Thanks for the efforts.
Thanks for a nice video!
codee??
Nice one Piotr. Well presented. Keep posting.
Dziękuje za film! Bardzo pomocny. Brakuje mi wspomnienia o barcie i bingu 😅 probowalam porównywać ich działania na jakiś samplowych taskach lecz ja jestem początkująca - czyli jako developer nie mam co korzystać z AI wersje goglowa i microsoftowa? 😅
hey so where is the software I need to use it
Looks great! Congrats! :)
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Learn more about us: deepsense.ai/
Where can we get it's source code?
of course its closed dumb dumb
does anyone know the links none of these sites or videos are giving a link just replaying that lady posing in different passions
Very inovative for people like me that would like to start a clothing business
✌️ P r o m o S M!!