POPULARITY
4 episodes with V100
2 episodes with V100
11 episodes with V100
7 episodes with V100
2 episodes with V100
2 episodes with V100
2 episodes with V100
3 episodes with V100
2 episodes with V100
Step into the future of policing where data is a mission-critical asset. Cindi Howson talks with Aimee Smith, Director of Data at the Metropolitan Police, about the Met's bold data transformation—from digitizing records dating back to 1829 to using AI and cloud tech for smarter, faster decisions. Hear how initiatives like the V100 program and real-time analytics help improve city safety. Key Moments: Genesis of the Met's Data Strategy (03:35) - The Met's data strategy's origin is traced to former Commissioner Cressida Dick's leadership, who envisioned leveraging data to transform policing, leading to a program building data capabilities and broadening analytics use beyond traditional intelligence and performance applications. Mission with Data and AI (13:34): The Met's overarching mission to use data and AI for precise decision-making is articulated, acknowledging the complexity of policing's multiple goals: crime prevention, incident response, organized crime intervention, victim service, and custody safety. Infrastructure Evolution (15:18): The transformation of the Met's data infrastructure over 5 years, from 8 separate operational systems to an integrated one with cloud technology adoption, is described, enhancing analytics and data science capabilities. V100 Initiative (19:58): The V100 initiative, a data and analytics effort to reduce violence against women and girls by prioritizing individuals with a history of harm, is explained. Concert Security Powered by Analytics (27:50): The use of ThoughtSpot by frontline officers is illustrated with a sergeant's innovative application for analyzing crime data around events like the Taylor Swift Eras tour to improve policing plans. AI Agent Development (36:37): An innovative project to build an AI agent that assists frontline officers at crime scenes by providing real-time guidance is outlined, aiming to improve public protection and investigative outcomes. Key Quotes:“So if an officer wants to start being able to do their own searches, creating their own sort of planners, thinking about doing their own trend analysis essentially, of crime data, which is great, isn't it? I mean, that's just exactly how you want ThoughtSpot to be used. Every officer has access to that.” - Aimee Smith"I like to think of it as a utility belt—you know how cops wear their utility belt? Well, hanging on there is this ThoughtSpot tool. A sergeant invented a way to use it for planning major events, concerts, to make sure our presence is right. And now that's replicable by other people who want to do the same thing." - Aimee Smith"One of the 5 principles of our business strategy for London to keep it safe is to be more precise in the use of data for decision making. So it's a high-level principle of our strategy. That makes data and analytics much harder, because there aren't enough data specialists and too many data parts to point at all those missions in one go.” - Aimee SmithMentionsMet Police's V100 InitiativeMet Police Develops an Open Data Strategy with the Open Data InstituteMet Police's Concert Preparation for Taylor Swift's Eras Tour Cressida Dick Reflects on Public Trust in the Digital Age The Data Protection ActGuest Bio Aimee Smith's distinguished career in the Metropolitan Police Service (MPS) spans almost a quarter-century, truly a testament to her profound dedication to integrating robust data into the very core of police decision-making. She embarked on her journey in 2001 as an Intelligence Analyst, steadily rising through the ranks. By 2014, her leadership capabilities led her to head UK Policing's largest Confidential Intelligence Unit. A pivotal "light-bulb moment" crystallized for her the critical importance of effective data management in driving operational outcomes, inspiring her to passionately spearhead the comprehensive MPS data transformation program. In a landmark achievement, Aimee was appointed as the first Director of Data for the MPS, where in 2019, she successfully established the inaugural Data Office within law enforcement, fundamentally reshaping how the service leverages its information. Hear more from Cindi Howson here. Sponsored by ThoughtSpot.
Happy Friday! Angie Gillenwater joins us with Apple TV, our “Adopt Me Please” Pet of the Week from the Kanawha Charleston Humane Association, plus Jim Strawn talking Live on the Levee, Logan Scott from V100 on his Camping For Kids and Hailey Morgan on tomorrow's Paws and Pistons Car Show in Winfield.
rWotD Episode 2877: WVIB Welcome to Random Wiki of the Day, your journey through Wikipedia’s vast and varied content, one random article at a time.The random article for Thursday, 20 March 2025 is WVIB.WVIB (100.1 FM, "V100") is a radio station broadcasting an urban adult contemporary format fed via satellite from Westwood One (known as "The Touch" or "Today's R&B and Old School"). The station is licensed to Holton, Michigan and serves the Muskegon market. It can be heard as far south as Allendale, Michigan, as far east as Lakeview, Michigan, and as far north as Ludington, Michigan. However, its range is limited by WBCH to the southeast and WSJP-FM to the west.This recording reflects the Wikipedia text as of 00:17 UTC on Thursday, 20 March 2025.For the full current version of the article, see WVIB on Wikipedia.This podcast uses content from Wikipedia under the Creative Commons Attribution-ShareAlike License.Visit our archives at wikioftheday.com and subscribe to stay updated on new episodes.Follow us on Mastodon at @wikioftheday@masto.ai.Also check out Curmudgeon's Corner, a current events podcast.Until next time, I'm generative Matthew.
9:10 Brian Connell of the Leukemia and Lymphonia Society/Keep American Covered on concerns with funding from the Trump Administration9:20 Rachelle Beatty on this weekend's WV Fishing Hunting and Outdoors Sports Show in Morgantown9:40 Logan Scott of V100 on our flood materials drive Wednesday
Happy Turkey Drop Day! The second annual WVRC Media-Charleston Turkey Drop is underway on our parking lot, 1111 Virginia Street East in Charleston, now til 6pm. Drop off your frozen turkeys and cash donations for Mountain Mission today. We will talk about it in detail this morning, plus Dr. Casey Sacks from BridgeValley is here and since we are only one week away from Thanksgiving Logan Scott from V100 will stop by to talk about the V100 Thanksgiving recipe contest.
AI Engineering is expanding! Join the first
WE SPEAK WITH ROBERT BECKMAN, CEO OF WICAB, INC. A COMPANY THAT HAS DEVELOPED THE “BRAINPORT V100” A TECHNOLOGY WHICH ALLOWS THE BLIND TO “TASTE THE LIGHT”.
Récitation du Coran lors de la prière du Tarawih pendant le mois de Ramadan 2023 à la mosquée Al Hikma, la sagesse, à Bruxelles --- Send in a voice message: https://podcasters.spotify.com/pod/show/belligh/message
Oracle has been actively focusing on bringing AI to the enterprise at every layer of its tech stack, be it SaaS apps, AI services, infrastructure, or data. In this episode, hosts Lois Houston and Nikita Abraham, along with senior instructors Hemant Gahankari and Himanshu Raj, discuss OCI AI and Machine Learning services. They also go over some key OCI Data Science concepts and responsible AI principles. Oracle MyLearn: https://mylearn.oracle.com/ou/learning-path/become-an-oci-ai-foundations-associate-2023/127177 Oracle University Learning Community: https://education.oracle.com/ou-community LinkedIn: https://www.linkedin.com/showcase/oracle-university/ X (formerly Twitter): https://twitter.com/Oracle_Edu Special thanks to Arijit Ghosh, David Wright, Himanshu Raj, and the OU Studio Team for helping us create this episode. -------------------------------------------------------- Episode Transcript: 00:00 The world of artificial intelligence is vast and everchanging. And with all the buzz around it lately, we figured it was the perfect time to revisit our AI Made Easy series. Join us over the next few weeks as we chat about all things AI, helping you to discover its endless possibilities. Ready to dive in? Let's go! 00:33 Welcome to the Oracle University Podcast, the first stop on your cloud journey. During this series of informative podcasts, we'll bring you foundational training on the most popular Oracle technologies. Let's get started! 00:46 Lois: Welcome to the Oracle University Podcast! I'm Lois Houston, Director of Innovation Programs with Oracle University, and with me is Nikita Abraham, Principal Technical Editor. Nikita: Hey everyone! In our last episode, we dove into Generative AI and Language Learning Models. Lois: Yeah, that was an interesting one. But today, we're going to discuss the AI and machine learning services offered by Oracle Cloud Infrastructure, and we'll look at the OCI AI infrastructure. Nikita: I'm also going to try and squeeze in a couple of questions on a topic I'm really keen about, which is responsible AI. To take us through all of this, we have two of our colleagues, Hemant Gahankari and Himanshu Raj. Hemant is a Senior Principal OCI Instructor and Himanshu is a Senior Instructor on AI/ML. So, let's get started! 01:36 Lois: Hi Hemant! We're so excited to have you here! We know that Oracle has really been focusing on bringing AI to the enterprise at every layer of our stack. Hemant: It all begins with data and infrastructure layers. OCI AI services consume data, and AI services, in turn, are consumed by applications. This approach involves extensive investment from infrastructure to SaaS applications. Generative AI and massive scale models are the more recent steps. Oracle AI is the portfolio of cloud services for helping organizations use the data they may have for the business-specific uses. Business applications consume AI and ML services. The foundation of AI services and ML services is data. AI services contain pre-built models for specific uses. Some of the AI services are pre-trained, and some can be additionally trained by the customer with their own data. AI services can be consumed by calling the API for the service, passing in the data to be processed, and the service returns a result. There is no infrastructure to be managed for using AI services. 02:58 Nikita: How do I access OCI AI services? Hemant: OCI AI services provide multiple methods for access. The most common method is the OCI Console. The OCI Console provides an easy to use, browser-based interface that enables access to notebook sessions and all the features of all the data science, as well as AI services. The REST API provides access to service functionality but requires programming expertise. And API reference is provided in the product documentation. OCI also provides programming language SDKs for Java, Python, TypeScript, JavaScript, .Net, Go, and Ruby. The command line interface provides both quick access and full functionality without the need for scripting. 03:52 Lois: Hemant, what are the types of OCI AI services that are available? Hemant: OCI AI services is a collection of services with pre-built machine learning models that make it easier for developers to build a variety of business applications. The models can also be custom trained for more accurate business results. The different services provided are digital assistant, language, vision, speech, document understanding, anomaly detection. 04:24 Lois: I know we're going to talk about them in more detail in the next episode, but can you introduce us to OCI Language, Vision, and Speech? Hemant: OCI Language allows you to perform sophisticated text analysis at scale. Using the pre-trained and custom models, you can process unstructured text to extract insights without data science expertise. Pre-trained models include language detection, sentiment analysis, key phrase extraction, text classification, named entity recognition, and personal identifiable information detection. Custom models can be trained for named entity recognition and text classification with domain-specific data sets. In text translation, natural machine translation is used to translate text across numerous languages. Using OCI Vision, you can upload images to detect and classify objects in them. Pre-trained models and custom models are supported. In image analysis, pre-trained models perform object detection, image classification, and optical character recognition. In image analysis, custom models can perform custom object detection by detecting the location of custom objects in an image and providing a bounding box. The OCI Speech service is used to convert media files to readable texts that's stored in JSON and SRT format. Speech enables you to easily convert media files containing human speech into highly exact text transcriptions. 06:12 Nikita: That's great. And what about document understanding and anomaly detection? Hemant: Using OCI document understanding, you can upload documents to detect and classify text and objects in them. You can process individual files or batches of documents. In OCR, document understanding can detect and recognize text in a document. In text extraction, document understanding provides the word level and line level text, and the bounding box, coordinates of where the text is found. In key value extraction, document understanding extracts a predefined list of key value pairs of information from receipts, invoices, passports, and driver IDs. In table extraction, document understanding extracts content in tabular format, maintaining the row and column relationship of cells. In document classification, the document understanding classifies documents into different types. The OCI Anomaly Detection service is a service that analyzes large volume of multivariate or univariate time series data. The Anomaly Detection service increases the reliability of businesses by monitoring their critical assets and detecting anomalies early with high precision. Anomaly Detection is the identification of rare items, events, or observations in data that differ significantly from the expectation. 07:55 Nikita: Where is Anomaly Detection most useful? Hemant: The Anomaly Detection service is designed to help with analyzing large amounts of data and identifying the anomalies at the earliest possible time with maximum accuracy. Different sectors, such as utility, oil and gas, transportation, manufacturing, telecommunications, banking, and insurance use Anomaly Detection service for their day-to-day activities. 08:23 Lois: Ok…and the first OCI AI service you mentioned was digital assistant… Hemant: Oracle Digital Assistant is a platform that allows you to create and deploy digital assistants, which are AI driven interfaces that help users accomplish a variety of tasks with natural language conversations. When a user engages with the Digital Assistant, the Digital Assistant evaluates the user input and routes the conversation to and from the appropriate skills. Digital Assistant greets the user upon access. Upon user requests, list what it can do and provide entry points into the given skills. It routes explicit user requests to the appropriate skills. And it also handles interruptions to flows and disambiguation. It also handles requests to exit the bot. 09:21 Nikita: Excellent! Let's bring Himanshu in to tell us about machine learning services. Hi Himanshu! Let's talk about OCI Data Science. Can you tell us a bit about it? Himanshu: OCI Data Science is the cloud service focused on serving the data scientist throughout the full machine learning life cycle with support for Python and open source. The service has many features, such as model catalog, projects, JupyterLab notebook, model deployment, model training, management, model explanation, open source libraries, and AutoML. 09:56 Lois: Himanshu, what are the core principles of OCI Data Science? Himanshu: There are three core principles of OCI Data Science. The first one, accelerated. The first principle is about accelerating the work of the individual data scientist. OCI Data Science provides data scientists with open source libraries along with easy access to a range of compute power without having to manage any infrastructure. It also includes Oracle's own library to help streamline many aspects of their work. The second principle is collaborative. It goes beyond an individual data scientist's productivity to enable data science teams to work together. This is done through the sharing of assets, reducing duplicative work, and putting reproducibility and auditability of models for collaboration and risk management. Third is enterprise grade. That means it's integrated with all the OCI Security and access protocols. The underlying infrastructure is fully managed. The customer does not have to think about provisioning compute and storage. And the service handles all the maintenance, patching, and upgrades so user can focus on solving business problems with data science. 11:11 Nikita: Let's drill down into the specifics of OCI Data Science. So far, we know it's cloud service to rapidly build, train, deploy, and manage machine learning models. But who can use it? Where is it? And how is it used? Himanshu: It serves data scientists and data science teams throughout the full machine learning life cycle. Users work in a familiar JupyterLab notebook interface, where they write Python code. And how it is used? So users preserve their models in the model catalog and deploy their models to a managed infrastructure. 11:46 Lois: Walk us through some of the key terminology that's used. Himanshu: Some of the important product terminology of OCI Data Science are projects. The projects are containers that enable data science teams to organize their work. They represent collaborative work spaces for organizing and documenting data science assets, such as notebook sessions and models. Note that tenancy can have as many projects as needed without limits. Now, this notebook session is where the data scientists work. Notebook sessions provide a JupyterLab environment with pre-installed open source libraries and the ability to add others. Notebook sessions are interactive coding environment for building and training models. Notebook sessions run in a managed infrastructure and the user can select CPU or GPU, the compute shape, and amount of storage without having to do any manual provisioning. The other important feature is Conda environment. It's an open source environment and package management system and was created for Python programs. 12:53 Nikita: What is a Conda environment used for? Himanshu: It is used in the service to quickly install, run, and update packages and their dependencies. Conda easily creates, saves, loads, and switches between environments in your notebooks sessions. 13:07 Nikita: Earlier, you spoke about the support for Python in OCI Data Science. Is there a dedicated library? Himanshu: Oracle's Accelerated Data Science ADS SDK is a Python library that is included as part of OCI Data Science. ADS has many functions and objects that automate or simplify the steps in the data science workflow, including connecting to data, exploring, and visualizing data. Training a model with AutoML, evaluating models, and explaining models. In addition, ADS provides a simple interface to access the data science service mode model catalog and other OCI services, including object storage. 13:45 Lois: I also hear a lot about models. What are models? Himanshu: Models define a mathematical representation of your data and business process. You create models in notebooks, sessions, inside projects. 13:57 Lois: What are some other important terminologies related to models? Himanshu: The next terminology is model catalog. The model catalog is a place to store, track, share, and manage models. The model catalog is a centralized and managed repository of model artifacts. A stored model includes metadata about the provenance of the model, including Git-related information and the script. Our notebook used to push the model to the catalog. Models stored in the model catalog can be shared across members of a team, and they can be loaded back into a notebook session. The next one is model deployments. Model deployments allow you to deploy models stored in the model catalog as HTTP endpoints on managed infrastructure. 14:45 Lois: So, how do you operationalize these models? Himanshu: Deploying machine learning models as web applications, HTTP API endpoints, serving predictions in real time is the most common way to operationalize models. HTTP endpoints or the API endpoints are flexible and can serve requests for the model predictions. Data science jobs enable you to define and run a repeatable machine learning tasks on fully managed infrastructure. Nikita: Thanks for that, Himanshu. 15:18 Did you know that Oracle University offers free courses on Oracle Cloud Infrastructure? You'll find training on everything from cloud computing, database, and security, artificial intelligence, and machine learning, all free to subscribers. So, what are you waiting for? Pick a topic, leverage the Oracle University Learning Community to ask questions, and then sit for your certification. Visit mylearn.oracle.com to get started. 15:46 Nikita: Welcome back! The Oracle AI Stack consists of AI services and machine learning services, and these services are built using AI infrastructure. So, let's move on to that. Hemant, what are the components of OCI AI Infrastructure? Hemant: OCI AI Infrastructure is mainly composed of GPU-based instances. Instances can be virtual machines or bare metal machines. High performance cluster networking that allows instances to communicate to each other. Super clusters are a massive network of GPU instances with multiple petabytes per second of bandwidth. And a variety of fully managed storage options from a single byte to exabytes without upfront provisioning are also available. 16:35 Lois: Can we explore each of these components a little more? First, tell us, why do we need GPUs? Hemant: ML and AI needs lots of repetitive computations to be made on huge amounts of data. Parallel computing on GPUs is designed for many processes at the same time. A GPU is a piece of hardware that is incredibly good in performing computations. GPU has thousands of lightweight cores, all working on their share of data in parallel. This gives them the ability to crunch through extremely large data set at tremendous speed. 17:14 Nikita: And what are the GPU instances offered by OCI? Hemant: GPU instances are ideally suited for model training and inference. Bare metal and virtual machine compute instances powered by NVIDIA GPUs H100, A100, A10, and V100 are made available by OCI. 17:35 Nikita: So how do we choose what to train from these different GPU options? Hemant: For large scale AI training, data analytics, and high performance computing, bare metal instances BM 8 X NVIDIA H100 and BM 8 X NVIDIA A100 can be used. These provide up to nine times faster AI training and 30 times higher acceleration for AI inferencing. The other bare metal and virtual machines are used for small AI training, inference, streaming, gaming, and virtual desktop infrastructure. 18:14 Lois: And why would someone choose the OCI AI stack over its counterparts? Hemant: Oracle offers all the features and is the most cost effective option when compared to its counterparts. For example, BM GPU 4.8 version 2 instance costs just $4 per hour and is used by many customers. Superclusters are a massive network with multiple petabytes per second of bandwidth. It can scale up to 4,096 OCI bare metal instances with 32,768 GPUs. We also have a choice of bare metal A100 or H100 GPU instances, and we can select a variety of storage options, like object store, or block store, or even file system. For networking speeds, we can reach 1,600 GB per second with A100 GPUs and 3,200 GB per second with H100 GPUs. With OCI storage, we can select local SSD up to four NVMe drives, block storage up to 32 terabytes per volume, object storage up to 10 terabytes per object, file systems up to eight exabyte per file system. OCI File system employs five replicated storage located in different fault domains to provide redundancy for resilient data protection. HPC file systems, such as BeeGFS and many others are also offered. OCI HPC file systems are available on Oracle Cloud Marketplace and make it easy to deploy a variety of high performance file servers. 20:11 Lois: I think a discussion on AI would be incomplete if we don't talk about responsible AI. We're using AI more and more every day, but can we actually trust it? Hemant: For us to trust AI, it must be driven by ethics that guide us as well. Nikita: And do we have some principles that guide the use of AI? Hemant: AI should be lawful, complying with all applicable laws and regulations. AI should be ethical, that is it should ensure adherence to ethical principles and values that we uphold as humans. And AI should be robust, both from a technical and social perspective. Because even with the good intentions, AI systems can cause unintentional harm. AI systems do not operate in a lawless world. A number of legally binding rules at national and international level apply or are relevant to the development, deployment, and use of AI systems today. The law not only prohibits certain actions but also enables others, like protecting rights of minorities or protecting environment. Besides horizontally applicable rules, various domain-specific rules exist that apply to particular AI applications. For instance, the medical device regulation in the health care sector. In AI context, equality entails that the systems' operations cannot generate unfairly biased outputs. And while we adopt AI, citizens right should also be protected. 21:50 Lois: Ok, but how do we derive AI ethics from these? Hemant: There are three main principles. AI should be used to help humans and allow for oversight. It should never cause physical or social harm. Decisions taken by AI should be transparent and fair, and also should be explainable. AI that follows the AI ethical principles is responsible AI. So if we map the AI ethical principles to responsible AI requirements, these will be like, AI systems should follow human-centric design principles and leave meaningful opportunity for human choice. This means securing human oversight. AI systems and environments in which they operate must be safe and secure, they must be technically robust, and should not be open to malicious use. The development, and deployment, and use of AI systems must be fair, ensuring equal and just distribution of both benefits and costs. AI should be free from unfair bias and discrimination. Decisions taken by AI to the extent possible should be explainable to those directly and indirectly affected. 23:21 Nikita: This is all great, but what does a typical responsible AI implementation process look like? Hemant: First, a governance needs to be put in place. Second, develop a set of policies and procedures to be followed. And once implemented, ensure compliance by regular monitoring and evaluation. Lois: And this is all managed by developers? Hemant: Typical roles that are involved in the implementation cycles are developers, deployers, and end users of the AI. 23:56 Nikita: Can we talk about AI specifically in health care? How do we ensure that there is fairness and no bias? Hemant: AI systems are only as good as the data that they are trained on. If that data is predominantly from one gender or racial group, the AI systems might not perform as well on data from other groups. 24:21 Lois: Yeah, and there's also the issue of ensuring transparency, right? Hemant: AI systems often make decisions based on complex algorithms that are difficult for humans to understand. As a result, patients and health care providers can have difficulty trusting the decisions made by the AI. AI systems must be regularly evaluated to ensure that they are performing as intended and not causing harm to patients. 24:49 Nikita: Thank you, Hemant and Himanshu, for this really insightful session. If you're interested in learning more about the topics we discussed today, head on over to mylearn.oracle.com and search for the Oracle Cloud Infrastructure AI Foundations course. Lois: That's right, Niki. You'll find demos that you watch as well as skill checks that you can attempt to better your understanding. In our next episode, we'll get into the OCI AI Services we discussed today and talk about them in more detail. Until then, this is Lois Houston… Nikita: And Nikita Abraham, signing off! 25:25 That's all for this episode of the Oracle University Podcast. If you enjoyed listening, please click Subscribe to get all the latest episodes. We'd also love it if you would take a moment to rate and review us on your podcast app. See you again on the next episode of the Oracle University Podcast.
As the sun sets on this wonderful series, I'm met with a cocktail of joy and sadness. I do wish I could do this forever and continue to grow but then I'd never make another YouTube episode. My love life would continue to bore me to tears. These things take up my whole weekend every week pretty much. The idea is to get the Youtube fired back up, test the waters and go from there. Who knows maybe in a week I'll scratch my head, run to the record store and be right back here breaking you off a piece. Come back like Jordan wearing the 45. In all seriousness, BIG thank you to any and everyone who clicks on this thing. Although I'm retiring the series, I'll keep them right here just incase someone wants to retroactively listen. My only goal for this episode in particular was to have as much fun as possible doing this, and to make it longer than the extended cut Lord of the Rings.. Which both of those things were easily accomplished. ALERT ALERT ALERT. DJ Witwicky Tee Shirts available at www.coffeestainclothing.com right now. Follow tremendousopinions on IG for future plans and updates. Thank you for being a part of this, I really appreciate it.Your Host with the Most,DJ Witwicky
Récitation du Coran lors de la prière du Tarawih pendant le mois de Ramadan 2023 à la mosquée Al Hikma, la sagesse, à Bruxelles --- Send in a voice message: https://podcasters.spotify.com/pod/show/belligh/message
Langsamfahrt: #61 - Freunde der V100, Metronom, 218 in Bayern
Langsamfahrt: #61 - Freunde der V100, Metronom, 218 in Bayern
Oracle has been actively focusing on bringing AI to the enterprise at every layer of its tech stack, be it SaaS apps, AI services, infrastructure, or data. In this episode, hosts Lois Houston and Nikita Abraham, along with senior instructors Hemant Gahankari and Himanshu Raj, discuss OCI AI and Machine Learning services. They also go over some key OCI Data Science concepts and responsible AI principles. Oracle MyLearn: https://mylearn.oracle.com/ou/learning-path/become-an-oci-ai-foundations-associate-2023/127177 Oracle University Learning Community: https://education.oracle.com/ou-community LinkedIn: https://www.linkedin.com/showcase/oracle-university/ X (formerly Twitter): https://twitter.com/Oracle_Edu Special thanks to Arijit Ghosh, David Wright, Himanshu Raj, and the OU Studio Team for helping us create this episode. ------------------------------------------------------- Episode Transcript: 00:00 Welcome to the Oracle University Podcast, the first stop on your cloud journey. During this series of informative podcasts, we'll bring you foundational training on the most popular Oracle technologies. Let's get started! 00:26 Lois: Welcome to the Oracle University Podcast! I'm Lois Houston, Director of Innovation Programs with Oracle University, and with me is Nikita Abraham, Principal Technical Editor. Nikita: Hey everyone! In our last episode, we dove into Generative AI and Language Learning Models. Lois: Yeah, that was an interesting one. But today, we're going to discuss the AI and machine learning services offered by Oracle Cloud Infrastructure, and we'll look at the OCI AI infrastructure. Nikita: I'm also going to try and squeeze in a couple of questions on a topic I'm really keen about, which is responsible AI. To take us through all of this, we have two of our colleagues, Hemant Gahankari and Himanshu Raj. Hemant is a Senior Principal OCI Instructor and Himanshu is a Senior Instructor on AI/ML. So, let's get started! 01:16 Lois: Hi Hemant! We're so excited to have you here! We know that Oracle has really been focusing on bringing AI to the enterprise at every layer of our stack. Hemant: It all begins with data and infrastructure layers. OCI AI services consume data, and AI services, in turn, are consumed by applications. This approach involves extensive investment from infrastructure to SaaS applications. Generative AI and massive scale models are the more recent steps. Oracle AI is the portfolio of cloud services for helping organizations use the data they may have for the business-specific uses. Business applications consume AI and ML services. The foundation of AI services and ML services is data. AI services contain pre-built models for specific uses. Some of the AI services are pre-trained, and some can be additionally trained by the customer with their own data. AI services can be consumed by calling the API for the service, passing in the data to be processed, and the service returns a result. There is no infrastructure to be managed for using AI services. 02:37 Nikita: How do I access OCI AI services? Hemant: OCI AI services provide multiple methods for access. The most common method is the OCI Console. The OCI Console provides an easy to use, browser-based interface that enables access to notebook sessions and all the features of all the data science, as well as AI services. The REST API provides access to service functionality but requires programming expertise. And API reference is provided in the product documentation. OCI also provides programming language SDKs for Java, Python, TypeScript, JavaScript, .Net, Go, and Ruby. The command line interface provides both quick access and full functionality without the need for scripting. 03:31 Lois: Hemant, what are the types of OCI AI services that are available? Hemant: OCI AI services is a collection of services with pre-built machine learning models that make it easier for developers to build a variety of business applications. The models can also be custom trained for more accurate business results. The different services provided are digital assistant, language, vision, speech, document understanding, anomaly detection. 04:03 Lois: I know we're going to talk about them in more detail in the next episode, but can you introduce us to OCI Language, Vision, and Speech? Hemant: OCI Language allows you to perform sophisticated text analysis at scale. Using the pre-trained and custom models, you can process unstructured text to extract insights without data science expertise. Pre-trained models include language detection, sentiment analysis, key phrase extraction, text classification, named entity recognition, and personal identifiable information detection. Custom models can be trained for named entity recognition and text classification with domain-specific data sets. In text translation, natural machine translation is used to translate text across numerous languages. Using OCI Vision, you can upload images to detect and classify objects in them. Pre-trained models and custom models are supported. In image analysis, pre-trained models perform object detection, image classification, and optical character recognition. In image analysis, custom models can perform custom object detection by detecting the location of custom objects in an image and providing a bounding box. The OCI Speech service is used to convert media files to readable texts that's stored in JSON and SRT format. Speech enables you to easily convert media files containing human speech into highly exact text transcriptions. 05:52 Nikita: That's great. And what about document understanding and anomaly detection? Hemant: Using OCI document understanding, you can upload documents to detect and classify text and objects in them. You can process individual files or batches of documents. In OCR, document understanding can detect and recognize text in a document. In text extraction, document understanding provides the word level and line level text, and the bounding box, coordinates of where the text is found. In key value extraction, document understanding extracts a predefined list of key value pairs of information from receipts, invoices, passports, and driver IDs. In table extraction, document understanding extracts content in tabular format, maintaining the row and column relationship of cells. In document classification, the document understanding classifies documents into different types. The OCI Anomaly Detection service is a service that analyzes large volume of multivariate or univariate time series data. The Anomaly Detection service increases the reliability of businesses by monitoring their critical assets and detecting anomalies early with high precision. Anomaly Detection is the identification of rare items, events, or observations in data that differ significantly from the expectation. 07:34 Nikita: Where is Anomaly Detection most useful? Hemant: The Anomaly Detection service is designed to help with analyzing large amounts of data and identifying the anomalies at the earliest possible time with maximum accuracy. Different sectors, such as utility, oil and gas, transportation, manufacturing, telecommunications, banking, and insurance use Anomaly Detection service for their day-to-day activities. 08:02 Lois: Ok.. and the first OCI AI service you mentioned was digital assistant… Hemant: Oracle Digital Assistant is a platform that allows you to create and deploy digital assistants, which are AI driven interfaces that help users accomplish a variety of tasks with natural language conversations. When a user engages with the Digital Assistant, the Digital Assistant evaluates the user input and routes the conversation to and from the appropriate skills. Digital Assistant greets the user upon access. Upon user requests, list what it can do and provide entry points into the given skills. It routes explicit user requests to the appropriate skills. And it also handles interruptions to flows and disambiguation. It also handles requests to exit the bot. 09:00 Nikita: Excellent! Let's bring Himanshu in to tell us about machine learning services. Hi Himanshu! Let's talk about OCI Data Science. Can you tell us a bit about it? Himanshu: OCI Data Science is the cloud service focused on serving the data scientist throughout the full machine learning life cycle with support for Python and open source. The service has many features, such as model catalog, projects, JupyterLab notebook, model deployment, model training, management, model explanation, open source libraries, and AutoML. 09:35 Lois: Himanshu, what are the core principles of OCI Data Science? Himanshu: There are three core principles of OCI Data Science. The first one, accelerated. The first principle is about accelerating the work of the individual data scientist. OCI Data Science provides data scientists with open source libraries along with easy access to a range of compute power without having to manage any infrastructure. It also includes Oracle's own library to help streamline many aspects of their work. The second principle is collaborative. It goes beyond an individual data scientist's productivity to enable data science teams to work together. This is done through the sharing of assets, reducing duplicative work, and putting reproducibility and auditability of models for collaboration and risk management. Third is enterprise grade. That means it's integrated with all the OCI Security and access protocols. The underlying infrastructure is fully managed. The customer does not have to think about provisioning compute and storage. And the service handles all the maintenance, patching, and upgrades so user can focus on solving business problems with data science. 10:50 Nikita: Let's drill down into the specifics of OCI Data Science. So far, we know it's cloud service to rapidly build, train, deploy, and manage machine learning models. But who can use it? Where is it? And how is it used? Himanshu: It serves data scientists and data science teams throughout the full machine learning life cycle. Users work in a familiar JupyterLab notebook interface, where they write Python code. And how it is used? So users preserve their models in the model catalog and deploy their models to a managed infrastructure. 11:25 Lois: Walk us through some of the key terminology that's used. Himanshu: Some of the important product terminology of OCI Data Science are projects. The projects are containers that enable data science teams to organize their work. They represent collaborative work spaces for organizing and documenting data science assets, such as notebook sessions and models. Note that tenancy can have as many projects as needed without limits. Now, this notebook session is where the data scientists work. Notebook sessions provide a JupyterLab environment with pre-installed open source libraries and the ability to add others. Notebook sessions are interactive coding environment for building and training models. Notebook sessions run in a managed infrastructure and the user can select CPU or GPU, the compute shape, and amount of storage without having to do any manual provisioning. The other important feature is Conda environment. It's an open source environment and package management system and was created for Python programs. 12:33 Nikita: What is a Conda environment used for? Himanshu: It is used in the service to quickly install, run, and update packages and their dependencies. Conda easily creates, saves, loads, and switches between environments in your notebooks sessions. 12:46 Nikita: Earlier, you spoke about the support for Python in OCI Data Science. Is there a dedicated library? Himanshu: Oracle's Accelerated Data Science ADS SDK is a Python library that is included as part of OCI Data Science. ADS has many functions and objects that automate or simplify the steps in the data science workflow, including connecting to data, exploring, and visualizing data. Training a model with AutoML, evaluating models, and explaining models. In addition, ADS provides a simple interface to access the data science service mode model catalog and other OCI services, including object storage. 13:24 Lois: I also hear a lot about models. What are models? Himanshu: Models define a mathematical representation of your data and business process. You create models in notebooks, sessions, inside projects. 13:36 Lois: What are some other important terminologies related to models? Himanshu: The next terminology is model catalog. The model catalog is a place to store, track, share, and manage models. The model catalog is a centralized and managed repository of model artifacts. A stored model includes metadata about the provenance of the model, including Git-related information and the script. Our notebook used to push the model to the catalog. Models stored in the model catalog can be shared across members of a team, and they can be loaded back into a notebook session. The next one is model deployments. Model deployments allow you to deploy models stored in the model catalog as HTTP endpoints on managed infrastructure. 14:24 Lois: So, how do you operationalize these models? Himanshu: Deploying machine learning models as web applications, HTTP API endpoints, serving predictions in real time is the most common way to operationalize models. HTTP endpoints or the API endpoints are flexible and can serve requests for the model predictions. Data science jobs enable you to define and run a repeatable machine learning tasks on fully managed infrastructure. Nikita: Thanks for that, Himanshu. 14:57 Did you know that Oracle University offers free courses on Oracle Cloud Infrastructure? You'll find training on everything from cloud computing, database, and security, artificial intelligence, and machine learning, all free to subscribers. So, what are you waiting for? Pick a topic, leverage the Oracle University Learning Community to ask questions, and then sit for your certification. Visit mylearn.oracle.com to get started. 15:25 Nikita: Welcome back! The Oracle AI Stack consists of AI services and machine learning services, and these services are built using AI infrastructure. So, let's move on to that. Hemant, what are the components of OCI AI Infrastructure? Hemant: OCI AI Infrastructure is mainly composed of GPU-based instances. Instances can be virtual machines or bare metal machines. High performance cluster networking that allows instances to communicate to each other. Super clusters are a massive network of GPU instances with multiple petabytes per second of bandwidth. And a variety of fully managed storage options from a single byte to exabytes without upfront provisioning are also available. 16:14 Lois: Can we explore each of these components a little more? First, tell us, why do we need GPUs? Hemant: ML and AI needs lots of repetitive computations to be made on huge amounts of data. Parallel computing on GPUs is designed for many processes at the same time. A GPU is a piece of hardware that is incredibly good in performing computations. GPU has thousands of lightweight cores, all working on their share of data in parallel. This gives them the ability to crunch through extremely large data set at tremendous speed. 16:54 Nikita: And what are the GPU instances offered by OCI? Hemant: GPU instances are ideally suited for model training and inference. Bare metal and virtual machine compute instances powered by NVIDIA GPUs H100, A100, A10, and V100 are made available by OCI. 17:14 Nikita: So how do we choose what to train from these different GPU options? Hemant: For large scale AI training, data analytics, and high performance computing, bare metal instances BM 8 X NVIDIA H100 and BM 8 X NVIDIA A100 can be used. These provide up to nine times faster AI training and 30 times higher acceleration for AI inferencing. The other bare metal and virtual machines are used for small AI training, inference, streaming, gaming, and virtual desktop infrastructure. 17:53 Lois: And why would someone choose the OCI AI stack over its counterparts? Hemant: Oracle offers all the features and is the most cost effective option when compared to its counterparts. For example, BM GPU 4.8 version 2 instance costs just $4 per hour and is used by many customers. Superclusters are a massive network with multiple petabytes per second of bandwidth. It can scale up to 4,096 OCI bare metal instances with 32,768 GPUs. We also have a choice of bare metal A100 or H100 GPU instances, and we can select a variety of storage options, like object store, or block store, or even file system. For networking speeds, we can reach 1,600 GB per second with A100 GPUs and 3,200 GB per second with H100 GPUs. With OCI storage, we can select local SSD up to four NVMe drives, block storage up to 32 terabytes per volume, object storage up to 10 terabytes per object, file systems up to eight exabyte per file system. OCI File system employs five replicated storage located in different fault domains to provide redundancy for resilient data protection. HPC file systems, such as BeeGFS and many others are also offered. OCI HPC file systems are available on Oracle Cloud Marketplace and make it easy to deploy a variety of high performance file servers. 19:50 Lois: I think a discussion on AI would be incomplete if we don't talk about responsible AI. We're using AI more and more every day, but can we actually trust it? Hemant: For us to trust AI, it must be driven by ethics that guide us as well. Nikita: And do we have some principles that guide the use of AI? Hemant: AI should be lawful, complying with all applicable laws and regulations. AI should be ethical, that is it should ensure adherence to ethical principles and values that we uphold as humans. And AI should be robust, both from a technical and social perspective. Because even with the good intentions, AI systems can cause unintentional harm. AI systems do not operate in a lawless world. A number of legally binding rules at national and international level apply or are relevant to the development, deployment, and use of AI systems today. The law not only prohibits certain actions but also enables others, like protecting rights of minorities or protecting environment. Besides horizontally applicable rules, various domain-specific rules exist that apply to particular AI applications. For instance, the medical device regulation in the health care sector. In AI context, equality entails that the systems' operations cannot generate unfairly biased outputs. And while we adopt AI, citizens right should also be protected. 21:30 Lois: Ok, but how do we derive AI ethics from these? Hemant: There are three main principles. AI should be used to help humans and allow for oversight. It should never cause physical or social harm. Decisions taken by AI should be transparent and fair, and also should be explainable. AI that follows the AI ethical principles is responsible AI. So if we map the AI ethical principles to responsible AI requirements, these will be like, AI systems should follow human-centric design principles and leave meaningful opportunity for human choice. This means securing human oversight. AI systems and environments in which they operate must be safe and secure, they must be technically robust, and should not be open to malicious use. The development, and deployment, and use of AI systems must be fair, ensuring equal and just distribution of both benefits and costs. AI should be free from unfair bias and discrimination. Decisions taken by AI to the extent possible should be explainable to those directly and indirectly affected. 23:01 Nikita: This is all great, but what does a typical responsible AI implementation process look like? Hemant: First, a governance needs to be put in place. Second, develop a set of policies and procedures to be followed. And once implemented, ensure compliance by regular monitoring and evaluation. Lois: And this is all managed by developers? Hemant: Typical roles that are involved in the implementation cycles are developers, deployers, and end users of the AI. 23:35 Nikita: Can we talk about AI specifically in health care? How do we ensure that there is fairness and no bias? Hemant: AI systems are only as good as the data that they are trained on. If that data is predominantly from one gender or racial group, the AI systems might not perform as well on data from other groups. 24:00 Lois: Yeah, and there's also the issue of ensuring transparency, right? Hemant: AI systems often make decisions based on complex algorithms that are difficult for humans to understand. As a result, patients and health care providers can have difficulty trusting the decisions made by the AI. AI systems must be regularly evaluated to ensure that they are performing as intended and not causing harm to patients. 24:29 Nikita: Thank you, Hemant and Himanshu, for this really insightful session. If you're interested in learning more about the topics we discussed today, head on over to mylearn.oracle.com and search for the Oracle Cloud Infrastructure AI Foundations course. Lois: That's right, Niki. You'll find demos that you watch as well as skill checks that you can attempt to better your understanding. In our next episode, we'll get into the OCI AI Services we discussed today and talk about them in more detail. Until then, this is Lois Houston… Nikita: And Nikita Abraham, signing off! 25:05 That's all for this episode of the Oracle University Podcast. If you enjoyed listening, please click Subscribe to get all the latest episodes. We'd also love it if you would take a moment to rate and review us on your podcast app. See you again on the next episode of the Oracle University Podcast.
Récitation du Coran lors de la prière du Tarawih pendant le mois de Ramadan 2023 à la mosquée Al Hikma, la sagesse, à Bruxelles --- Send in a voice message: https://podcasters.spotify.com/pod/show/belligh/message
Description Récitation du Coran lors de la prière du Tarawih pendant le mois de Ramadan 2023 à la mosquée Al Hikma, la sagesse, à Bruxelles --- Send in a voice message: https://podcasters.spotify.com/pod/show/belligh/message
Stefanie Myr is the head route setter, manager, and team coach at Climb Tacoma in WA. We met up in Leavenworth and talked about our similar upbringings in Christianity, why we both moved away from religion, finding “church” in the climbing community, confidence and self-belief, unique challenges as a short climber, being less certain and more curious, Stef's polyamorous relationship with her husband and partner, compersion, honest communication, doing what you can to make the world a little better, and much more!Listen to the Patron Show on Spotify!Check out Rhino Skin Solutions!rhinoskinsolutions.comUse code “NUGGET” at checkout for 20% off your next order!And check out EP 22 with Justin Brown to learn more about how to use Rhino products!Check out PhysiVantage!physivantage.com (link includes 15% off coupon)Use code "NUGGET15" at checkout for 15% off your next order!Check out Wonderful Pistachios!WonderfulPistachios.com to learn more!Check out Rumpl!rumpl.com/nuggetUse code "NUGGET" at checkout for 10% off your first order!Check out Chalk Cartel!chalkcartel.comUse code "NUGGET" at checkout for 20% off your next order!We are supported by these amazing BIG GIVERS:Leo Franchi, Michael Roy, David Lahaie, Robert Freehill, Jeremiah Johnson, Scott Donahue, Eli Conlee, Skyler Maxwell, Craig Lee, Mark and Julie Calhoun, Yinan Liu, Renzollama, Zach Emery, and Brandt MickolasBecome a Patron:patreon.com/thenuggetclimbingShow Notes: thenuggetclimbing.com/episodes/stefanie-myrNuggets:0:04:05 – Miniature carpentry, and a gift for me0:11:49 – Stef role at Climb Tacoma, and the gym as a core part of her community1:17:19 – How she first got a job at Climb Tacoma, and having a diverse group of people on the team0:20:16 – Stef's unique financial situation, and the sustainability of working at a climbing gym0:22:38 – Moving away from her Christian upbringing, and finding “church” at the climbing gym0:25:09 – Our parallels with church music and moving away from organized religion0:30:31 – Bullet point lists, and how the gym functions like a “church”0:32:07 – Meeting people who changed the way we thought about organized religion0:35:16 – Stef's upbringing and programming, and not realizing she was bisexual until college0:38:39 – Stef's husband Julien0:40:34 – Stef's confidence, choosing Giant Man as an objective, and gaining something from every climb0:46:17 – Frustration with height and grades, and needing to be V12 strong to climb V100:48:02 – My conversation with Nic Rummel about grades, and an announcement about the Patron show being on Spotify0:51:19 – Sending Pimpsqueak V8/9 in Leavenworth, ABR (always be rolling), and why recording send videos feels important0:56:02 – The power of seeing someone like you climb hard things1:00:42 – More about confidence, embracing powerful moves, learning from coaching little kids, and unlocking pieces of the puzzle1:08:18 – The moments that show you that you've gotten better at climbing, and Stef's mountain1:10:23 – Stef's dream boulder in Goldbar WA, and other goals in Leavenworth1:12:01 – How Stef got on The Nugget1:21:52 – Self-belief, wanting to be good, and the power of affirmation1:26:17 – Our brain's reaction to negative comments, and how Stef deals with trolls1:38:02 – The lawyer in my brain, and making people think differently by asking better questions1:39:51 – Less certain more curious1:42:47 – Stef's polyamorous relationship, and why she wanted to talk about it1:50:50 – Normalizing sex, and redefining cheating in an open relationship1:56:18 – Equating sex and intimacy to love, how cool her husband is, and the story of how she came to be with her partner2:03:22 – Jealousy, and honest communication2:05:32 – Why Stef wanted to talk about her open relationship, and why people who are poly are not free-for-alls2:09:53 – Bringing all of who you are to yourself as a climber2:11:53 – Compersion2:15:11 – The depth of her relationship with her husband before they opened their relationship, and the distinction between security and trust2:17:53 – Marriage, and Stef's future with her partner2:24:47 – Double the love and laundry, and Stef's life in a go bag2:26:50 – Check-ins, RADAR, smoking weed, and “ask don't assume”2:32:23 – Double-committing, and learning to communicate her plans2:39:18 – What Stef wishes people spent more time thinking about2:41:10 – Make the world a little better2:42:20 – My episode with Ethan Pringle about his dad, and the importance of airing out the messiness2:47:35 – Stef's sponsors2:51:08 – Pursuing the child-free life, and final thoughts
Invites are going out for AI Engineer Summit! In the meantime, we have just announced our first Actually Open AI event with Brev.dev and Langchain, Aug 26 in our SF HQ (we'll record talks for those remote). See you soon (and join the Discord)!Special thanks to @nearcyan for helping us arrange this with the Eleuther team.This post was on the HN frontpage for 15 hours.As startups and even VCs hoard GPUs to attract talent, the one thing more valuable than GPUs is knowing how to use them (aka, make GPUs go brrrr).There is an incredible amount of tacit knowledge in the NLP community around training, and until Eleuther.ai came along you pretty much had to work at Google or Meta to gain that knowledge. This makes it hard for non-insiders to even do simple estimations around costing out projects - it is well known how to trade $ for GPU hours, but trading “$ for size of model” or “$ for quality of model” is less known and more valuable and full of opaque “it depends”. This is why rules of thumb for training are incredibly useful, because they cut through the noise and give you the simple 20% of knowledge that determines 80% of the outcome derived from hard earned experience.Today's guest, Quentin Anthony from EleutherAI, is one of the top researchers in high-performance deep learning. He's one of the co-authors of Transformers Math 101, which was one of the clearest articulations of training rules of thumb. We can think of no better way to dive into training math than to have Quentin run us through a masterclass on model weights, optimizer states, gradients, activations, and how they all impact memory requirements.The core equation you will need to know is the following:Where C is the compute requirements to train a model, P is the number of parameters, and D is the size of the training dataset in tokens. This is also equal to τ, the throughput of your machine measured in FLOPs (Actual FLOPs/GPU * # of GPUs), multiplied by T, the amount of time spent training the model.Taking Chinchilla scaling at face value, you can simplify this equation to be `C = 120(P^2)`.These laws are only true when 1000 GPUs for 1 hour costs the same as 1 GPU for 1000 hours, so it's not always that easy to make these assumptions especially when it comes to communication overhead. There's a lot more math to dive into here between training and inference, which you can listen to in the episode or read in the articles. The other interesting concept we covered is distributed training and strategies such as ZeRO and 3D parallelism. As these models have scaled, it's become impossible to fit everything in a single GPU for training and inference. We leave these advanced concepts to the end, but there's a lot of innovation happening around sharding of params, gradients, and optimizer states that you must know is happening in modern LLM training. If you have questions, you can join the Eleuther AI Discord or follow Quentin on Twitter. Show Notes* Transformers Math 101 Article* Eleuther.ai* GPT-NeoX 20B* BLOOM* Turing NLG* Mosaic* Oak Ridge & Frontier Supercomputer* Summit Supercomputer * Lawrence Livermore Lab* RWKV* Flash Attention * Stas BekmanTimestamps* [00:00:00] Quentin's background and work at Eleuther.ai* [00:03:14] Motivation behind writing the Transformers Math 101 article* [00:05:58] Key equation for calculating compute requirements (tau x T = 6 x P x D)* [00:10:00] Difference between theoretical and actual FLOPs* [00:12:42] Applying the equation to estimate compute for GPT-3 training* [00:14:08] Expecting 115+ teraflops/sec per A100 GPU as a baseline* [00:15:10] Tradeoffs between Nvidia and AMD GPUs for training* [00:18:50] Model precision (FP32, FP16, BF16 etc.) and impact on memory* [00:22:00] Benefits of model quantization even with unlimited memory* [00:23:44] KV cache memory overhead during inference* [00:26:08] How optimizer memory usage is calculated* [00:32:03] Components of total training memory (model, optimizer, gradients, activations)* [00:33:47] Activation recomputation to reduce memory overhead* [00:38:25] Sharded optimizers like ZeRO to distribute across GPUs* [00:40:23] Communication operations like scatter and gather in ZeRO* [00:41:33] Advanced 3D parallelism techniques (data, tensor, pipeline)* [00:43:55] Combining 3D parallelism and sharded optimizers* [00:45:43] Challenges with heterogeneous clusters for distribution* [00:47:58] Lightning RoundTranscriptionAlessio: Hey everyone, welcome to the Latent Space podcast. This is Alessio, partner and CTO in Residence at Decibel Partners, and I'm joined by my co-host Swyx, writer and editor of Latent Space. [00:00:20]Swyx: Hey, today we have a very special guest, Quentin Anthony from Eleuther.ai. The context for this episode is that we've been looking to cover Transformers math for a long time. And then one day in April, there's this blog post that comes out that literally is called Transformers Math 101 from Eleuther. And this is one of the most authoritative posts that I've ever seen. And I think basically on this podcast, we're trying to give people an intuition around what are the rules of thumb that are important in thinking about AI and reasoning by AI. And I don't think there's anyone more credible than the people at Eleuther or the people training actual large language models, especially on limited resources. So welcome, Quentin. [00:00:59]Quentin: Thank you. A little bit about myself is that I'm a PhD student at Ohio State University, starting my fifth year now, almost done. I started with Eleuther during the GPT-NeoX20B model. So they were getting started training that, they were having some problems scaling it. As we'll talk about, I'm sure today a lot, is that communication costs and synchronization and how do you scale up a model to hundreds of GPUs and make sure that things progress quickly is really difficult. That was really similar to my PhD work. So I jumped in and helped them on the 20B, getting that running smoothly. And then ever since then, just as new systems challenges arise, and as they move to high performance computing systems and distributed systems, I just sort of kept finding myself falling into projects and helping out there. So I've been at Eleuther for a little bit now, head engineer there now, and then finishing up my PhD and then, well, who knows where I'll go next. [00:01:48]Alessio: Awesome. What was the inspiration behind writing the article? Was it taking some of those learnings? Obviously Eleuther is one of the most open research places out there. Is it just part of the DNA there or any fun stories there? [00:02:00]Quentin: For the motivation for writing, you very frequently see in like the DL training space, like these Twitter posts by like, for example, like Stas Bekman at Hugging Face, you'll see like a Twitter post that's like, oh, we just found this magic number and everything is like 20% faster. He's super excited, but doesn't really understand what's going on. And the same thing for us, we very frequently find that a lot of people understand the theory or maybe the fundamentals of why like AI training or inference works, but no one knows like the nitty gritty details of like, how do you get inference to actually run correctly on your machine split across two GPUs or something like that. So we sort of had all of these notes that we had accumulated and we're sort of sharing among engineers within Eleuther and we thought, well, this would really help a lot of other people. It's not really maybe appropriate for like a paper, but for something like a blog post or technical report, this would actually maybe squeeze a lot of performance out of people's hardware they're already running on. So I guess there are a lot of projects in Eleuther that we're sort of trying to share notes with people in a way that typical institutions don't. They sort of live within that institution and then you go to a different institution and they do something very similar, but without the lessons of the previous. And it's because everyone's trying to do their own special sauce with their own stack. Whereas Eleuther, we don't really have that constraint and we can just share everything to everybody. [00:03:14]Swyx: Yeah, this is a level of openness that basically very few people actually embrace. One, it's an extra effort to write things down, of course, but two, it is secret sauce and so that not many people do it. And therefore, oftentimes the only way to learn this stuff is to actually work in one of the large model labs. And so you guys are doing a lot. The only other instance where I can think of where people actually open sourced their process was Facebook's OPT. What else is similar, like sort of trade knowledge, but not formal research knowledge? [00:03:45]Quentin: I would say Bloom. So the Hugging Face Bloom project in big science and all of that, that was very open. I'd say it's the same caliber, if not more detailed than OPT. Other than that, I think there was like a doc from Microsoft on like their Turing NLG. Their paper is pretty relaxed in that it did talk about some of those challenges. Other than like OPT and Bloom and us, I can't think of any. It's a new thing. [00:04:10]Swyx: It matters that you are going for the sort of good enough rules of thumb, because I think a lot of people try to go for precision and being overly precise actually is not helpful. Right. Yes. [00:04:20]Quentin: You'll see some like statements in the blog posts that are just like, we think this is about 1.2 in our experience. And, you know, we don't go any further into detail and it would take maybe an extra month for us to chase down every single little piece of memory. But instead, like getting good enough is still helpful to people. [00:04:36]Alessio: Let's jump into it. The first part of the article, and we'll put this in the show notes so people will be following along with the post. So we don't need to read every single equation and every footnote for it. [00:04:46]Swyx: Okay. [00:04:46]Alessio: But the core equation here is that not the cost of compute, but the compute required to turn a transformer model is roughly equal to tau times T, where like T is the, where tau is the hardware setup throughput that you have. So number of GPUs times the actual flops per GPU. And then T is the time spent. I think people can visualize that pretty easily. It's basically like how many GPUs do you have and how much do you let them run for? And the things that come to it that people have read before in the Chinchilla paper in a way, and the OpenAI scaling law is that you can then equal this to 6PD, where P is the number of parameters in the model and D is the size of the, of the dataset in tokens. So talk a little bit about how people should think about the two. I think a lot of times the focus is on tokens parameter ratio in the training dataset and people don't think as much about the actual flops per GPU, which you're going to mention later in the blog post too, in terms of how much you can get out. So how should people think about this when they're building a model and where should they go to this equation as they're starting to think about training their own transformer-based [00:05:58]Swyx: model? [00:05:58]Quentin: You touched a little bit on the fact that people usually start with the dataset. So you have some dataset that you want to train a model on. And then from there, from the 6PD, you should see, okay, I should have about six tokens per parameter. So that determines my model size thereabouts for Chinchilla Optimal. So since then we've seen that need more something like 20 or more than that to get a good quality model. But the next question that should be on your mind in terms of a systems perspective is how long is it going to take for this model to train and what kind of budget should I expect? So let's say I want some cloud instance for some amount of time and each of them will have some price attached to it. So that's where the throughput comes in. So now that you have this model, this number of parameters, you should map that to a transformer architecture and you should benchmark what throughput you get on your software stack for that type of model. So now you have your flops per second on a single GPU. And then given whatever parallelism scheme, which I'm sure we'll get into, like data parallelism or tensor parallelism or whatever else, how is that flops number going to scale to whatever number of GPUs? And then from there, you're going to get a time. And if you have a time, you have a cost. Those are like the business answers that you'll be able to get using this formula. That's why we sort of split it into the T and the throughput terms so that you can solve for one of them, which is usually get throughput, need time, and from time you get cost. In a nutshell, that's the answer. [00:07:19]Alessio: One thing that I noticed, you mentioned some of these laws are only true when a thousand GPUs for one hour cost the same as one GPU for a thousand hours, given that we have a shortage of the biggest GPUs out there. Any thoughts there on how people should prioritize this? [00:07:36]Quentin: Yeah, so I would say you should find what the minimum number of GPUs is to just fit your model first. The memory bottleneck is your biggest problem if you have a sizable model. If it's a small model, nobody cares. But most models that people care about will need to be split across multiple GPUs. So find the minimum number of GPUs to just fit your one instance of your model and then calculate how long that's going to take. If it's a reasonable amount of time, then you're done. If it takes too long, then you need to start worrying about having multiple instances of that model. I always feel like you should go with the minimum number of GPUs because the more number of GPUs that you have, the more likely it is for things to break. So I would say just find out what time is reasonable for you and then fit the number of GPUs to that and no more. Because people get greedy and they say, if I have twice the GPUs, I can get this done in half the time. And then you end up taking three times the time because everything is breaking every day. And that's when I am up at midnight trying to fix your model that's broken. [00:08:34]Swyx: We had a previous guest which has invested a lot in their framework for training these things. Would there not be an equivalent open source framework you guys would have made that would help with scaling up GPUs linearly like that? Or is this an oversimplification? [00:08:50]Quentin: Okay, yeah. So maybe I should step back. Both Mosaic and us have our own sort of software stack recipe that scales well, theoretically. But I'll get to that in a minute. Mosaic is all based off optimizer sharding. So it's based off ZeRO. So you basically perfectly split your model optimizer and your parameters and your gradients across all of the different GPUs. So your aggregate memory is number of parameters divided by number of GPUs. Same thing for optimizer and so on. Whereas we at Eleuther use a Megatron deep speed based library. And for that, it's a bit more complex. So the efficiency can be a little higher, but it's more prone to failure at the same [00:09:30]Swyx: time. [00:09:30]Quentin: So you kind of have to tune it. In both cases, getting back to like the practical case, you should be able to get linear speed up by adding more GPUs. The problem is that there are hardware failures. You tend to have problems with like maybe loss will overflow if you have too many GPUs or maybe one GPU will hang. You might have software issues. You might have synchronization issues. And that's why I'm saying practically that you should take the minimum number of GPUs that you have because those are the easier cases to debug. That make sense? [00:10:00]Swyx: Yeah. [00:10:00]Quentin: Any more detail on any specific point? [00:10:02]Swyx: Not particularly, just because we haven't actually had to debug those things. But I imagine basically there's a lot of return towards encoding these knowledge into software and not repeating it again. So it makes a ton of sense. I think Alessio had more questions before we move too far into high level, more questions on just the equation itself. I think we want to spend time on essentially, this is the central equation of figuring out compute requirements. Yeah. [00:10:25]Alessio: Another thing in it is that the computer is like the forward pass and like the backwards pass and forward is 2PD, backward is 4PD. Why it's to the ratio between the two? Can you explain that? Why is it two and four? [00:10:39]Quentin: Yeah. [00:10:40]Alessio: Why is it twice the amount? [00:10:42]Quentin: Oh, okay. Intuitively for forward pass, you're just moving, you're propagating forward the inputs through the layer. And then in the backward pass, you're doing something a little more complex than that. You're doing back propagation. And I don't think I can explain it intuitively enough to go into more detail on the exact [00:10:58]Swyx: numbers. Yeah. [00:10:58]Quentin: That's okay. [00:10:59]Swyx: I feel like you want to get out a whiteboard and start drawing like, you know. [00:11:02]Quentin: That's what I would normally do. [00:11:03]Swyx: Tangents and gradients. It's actually surprisingly low to do the back propagation. Honestly, that's one of the fundamental things I love about the math of deep learning so far that as I've explored it, which is, it's surprisingly efficient as compared to other, I guess, numerical methods you might be exposed to and, you know, college calculus. Yeah. [00:11:22]Alessio: And I think the other thing is that things sound simple, you know, when people go on Twitter and say, Oh, 20 is like the optimal ratio. And it's like, then it's like, well, why is that the number? And the answer is usually much, much harder, like what we're seeing right now. So I think it's a, it's a good reminder that the numbers are simple, like all the best and most popular, like math equations are like, so elegant. Obviously the proof behind that is, it's not that easy. That's always a good reminder. [00:11:52]Swyx: I want to put this equation to the test a little bit. We can do this from either GPT-3's perspective or GPT-NeoX, whatever you're more comfortable with. You have this distinction of actual flops versus theoretical flops. And a lot of times when people report the flops it took to train a model, like we just saw one in Lama 2 where the estimate is something that the amount of flops and that's, that's what we go with. So GPT-3 took a 3.14 times 10 to the power 23 flops. That is the theoretical flops. I want to get to a point where I can sort of work out if a number passes the smell test. And I wonder how to do that because I should be able to plug in this equation, right? I know that GPT-3 was trained on 300 billion tokens. I know the parameter size of 175. Is it, is it just like a 6 times 175 times 300? Like I haven't done the math, but what are the nuances here that you might want to call out? [00:12:42]Quentin: Theoretical flops is usually given from, you have a given set of hardware and this is what you expect your hardware to get. The problem is that in practice, full utilization, that's the key word, right? Because in practice, there are a lot of cases where like you're spending time waiting on data movement from like the GPU to CPU. Or for example, you might be waiting to synchronize across the different GPUs. So there's a lot of idle time basically that you're going to be spending during training. [00:13:05]Swyx: Smell tests. [00:13:06]Quentin: I don't know if I have a smell test myself, to be honest, like maybe I'll look at like what sort of flops, what you would expect on like an A100. There's sort of just an expected flops for a given GPU that everyone sort of knows what you should expect. So like for an A100, that number is somewhere between 100 and 180. T flops is what you would expect to see on an A100. For a V100, like an older GPU, it's something more like 40 to 30. So people sort of know, given the kernels that we're running for a deep learning, what sort of flops you expect. And then you sort of compare that to the theory, to the theoretical flops that people are reporting and see if that matches your expectations. [00:13:47]Swyx: Yeah. [00:13:47]Alessio: And in the article you mentioned for the A100, like if you're seeing below 115 teraflops a second, there's something wrong with your model or hardware. How did you get to 115? Is it just, you know, production observability and like you've seen over months and months and months that like that's the baseline or how do you come up with the numbers like that? Yeah. [00:14:08]Quentin: For a number like that, we basically, we compared a lot of different frameworks. So like I mentioned before, Mosaic has their own framework and we have our own framework. They all have their own flop counters too, right? And we saw across a bunch of different hardware configurations that if you tune things correctly, you should be getting above 115 in pretty much all cases. So like there are some cases where things are tuned poorly or your system is a little weird, but we've never been able to get a new system and not been able to get above [00:14:35]Swyx: 115. [00:14:35]Quentin: If something is below 115, you have something really wrong in your software. But that's really all it is, is just comparing across software stacks and hardware systems. [00:14:44]Alessio: What about different GPUs? We had George Hotz on the podcast and he talked about AMD cards and how in theory their flops should be much better than some Nvidia cards, but the reality is like the CUDA runtime makes up for it. How should people think about improving that? You know, like do you see, okay, the A100 is like 115 teraflops. I'd rather just stick with this than try and figure out all the kinks of like a better AMD card or any thoughts there? [00:15:10]Swyx: Right. [00:15:10]Quentin: Well, that's sort of touching on developer time, right? And which ends up being more expensive because at the end of the day, the AMD and Rockham software stack has a long way to go. I would say most things run there, not particularly efficiently, but you're going to have weird bugs that no one has encountered before. One of the big pluses of going with the Nvidia and PyTorch stack is that there are thousands of GitHub issues with everyone facing the same problem as you and resolving them quickly and in an open source way is probably the biggest benefit of going with the Nvidia software stack right now. AMD has about the same hardware, software, not so much. And they haven't quite got the momentum in the open source realm, for example, to get close. Like something, for example, like Flash Attention, it's spread to more Nvidia GPU types than it has like to AMD at all. And waiting on those latest and greatest features to reach AMD is something that's prohibitive to a lot of people, but it's getting there. I'm running a lot of experiments on AMD right now because it's sort of reached the government lab supercomputers now. And so a lot of experiments are going there and it will catch up, I'd say within a few [00:16:14]Swyx: years. [00:16:14]Quentin: Awesome. [00:16:15]Swyx: Maybe just talk about what's available from the government labs and I heard the original, the origin of Eluther started with a grant for TPUs. Is that right? [00:16:24]Quentin: Yes, that was a little before me, but there was a lot of just like getting a grabbing a Google Cloud or TPU pod or something like that is a lot of the original TPU work on Mesh TensorFlow, which is like now like an ancient distributed deep learning library. [00:16:36]Quentin: Eluther got a grant, an insight grant with Oak Ridge last year, and we got quite a bit of Summit Compute. So Summit is a V100 based supercomputer. It's got some weirdness to it. So there's six V100 GPUs per node. And we did a lot of experiments there. It's a challenging system to scale to because your interconnect across nodes is kind of slow in comparison to within a node, which I think we'll get to later. But now Oak Ridge has moved to AMD. So the next grant that we're trying to work towards is on Frontier, which has four AMD GPUs per node and again has a slower interconnect across nodes. So we get all of those new challenges again to try and overlap things. But that's just like you have Oak Ridge, you have Lawrence Livermore. There's a lot of government supercomputers that you can apply for compute towards like open researchers too. It's sort of a new thing. I think we're one of the first like us and like Lion, for example, is another organization that's getting compute from government providers and such. They're all moving to AMD as well. And we look forward to exploring that with them. [00:17:42]Swyx: Yeah. [00:17:43]Alessio: The computing is definitely, it used to be easy to find the GPU. Now, not as much. So you got to find them anywhere. [00:17:49]Swyx: Yes. [00:17:49]Alessio: Let's talk about memory requirements a little bit. So you touched on this a little bit before and just before this, we had a trade out on the pockets from FlashAttention and memory speed was one of our main focuses, but this time we're being bound by actually memory size, like the VRAM itself, when it comes to model weights and parameters and optimizer states and all that fun stuff. Let's go through this and Sean, we can, we can take turns. There's a lot to cover here, but maybe we can start from model weights. So one topic we covered a lot in the past is precision and quantization. That's one of the obviously main driver of memory. You mentioned most of, in the article, most transformers are mixed precision, like FP16 plus FP32 or BF16 FP32, and they can be cast down. And you mentioned up to like INT8 without a lot of performance hit. So let's start there and maybe run people through some of the maths and like the byte per parameter ratio and different precision. [00:18:50]Swyx: Sure. [00:18:51]Quentin: So when I started deep learning, it was all FP32. You have 32 bits, four bytes per parameter. Things were pretty simple. You didn't have to do any loss scaling at all. But the problem was that you didn't get a whole lot of flops once NVIDIA moved to V100s and introduced Tensor cores. So Tensor cores do all of their computation at FP16 precision. So you're kind of throwing all of those away if you're doing things in FP32. So once the hardware moved to V100, the software moved to like mixed precision and APEX and AMP and such. And one counterintuitive part of mixed precision is that you actually require more memory when you're trained because you need an FP16 copy of the weights and an FP32 copy of the weights. The FP16 copy is where you're doing like your actual computation on the Tensor cores. So you get maybe it's not uncommon to get double the throughput that you would see before in FP32. And then you at each step update that FP32 copy with the FP16 update. So both need to be stored in memory. The problem with that is that FP16 is very precise but doesn't have a whole lot of range, [00:19:55]Swyx: dynamic range. [00:19:55]Quentin: So you have a really big mantissa if you're thinking in terms of like floating point representations, not a whole lot of exponent. So BF16 puts more of the bits from the mantissa back to the exponent. So you have a much higher range and a lower precision. And that gets rid of all of this instability problem and loss scaling and such that anyone familiar with debugging knows how unstable it can be, especially for large scale training. And BF16 does away with a lot of that, but it's only supported on A100s. So you see the back and forth between hardware and software. So every time NVIDIA introduces some new Tensor cores or BF16 support or something like that, the software adapts to support it and then training adapts. And then now you mentioned like Ind8 and such. Now we're seeing that you have some model that's been trained in FP16, FP32, whatever else. And then now you want to, with minimal loss and accuracy, quantize that model into a smaller representation like Ind8 and now like Ind4 and things like that and see what you can get away with. And then since deep learning is such like a stochastic problem that a lot of those last bits of precision don't really matter is what we're finding. And I expect that to continue. [00:21:06]Alessio: And so just to put some numbers to it, when you have a FP32, you need four bytes per parameter at inference time to load it in memory. If you have a eight bits model quantized down, you need one byte per parameter. So for example, in an H100, which is 80 gigabyte of memory, you could fit a 70 billion parameters in eight, you cannot fit a FP32 because you will need like 280 gigabytes of memory. So how much does that play into it? Like you mentioned it was all FP32 when you first started. Is it just like a development complexity thing, like going down to FP16 and then Ind8? Or if they could get a GPU with like a terabyte of VRAM, will people just load this memory as like FP32 weights or would they still want to quantize them to make them more efficient? Right. [00:22:00]Quentin: I would say even if you had infinite VRAM, you would still want a quantized model, just a bigger model that's quantized is what I would say. And that's because like I was mentioning there at the end, how like deep learning is very stochastic and a lot, you could have all the precision in the world, but ultimately it's meaningless when you still depend so much like on what the input is. And you depend so much on little variations and maybe a few more samples of training data would matter more. A lot of that precision in a nutshell doesn't really matter in deep learning. All that matters is the big picture. What is that neuron actually saying? And not the tiny details of what it might be thinking. Oh, I also wanted to mention that even if you have an A100, the actual model size is quite a bit smaller that you could load than what you mentioned. That's because of the KV cache. So the KV cache intuitively during inference, it only matters during inference and think intuitively if you're writing a paragraph, you want to remember every single previous word that you've written before you write the next word. So like what is autoregressive language modeling? It's filling in the next word, the next token. So if I say like the dog went to the, and I need to write the next word, I would say park or something. Before I write the next word, my memory is wiped and I have to read the whole thing again. That is life without a KV cache. And a KV cache says, remember everything that I've generated before, as well as all the context before what I've generated. But the memory overhead for a KV cache commonly is either comparable or larger than the model in some cases, if you have a really long context. And I think the exact equation is something like, oh, it's like two times the number of layers, times the number of heads, times the dimension of each head. And then there's two of those. You have one for K, one for V. But that was just a quick aside. Yeah. [00:23:44]Alessio: I know this is Transformers math, but do you think one of the interesting things about RNNs too, it's like moving away from this, like KV cache, the scales with the sequence length and having like a fixed sequence pass. I know those are some of the things that people are working on. [00:24:00]Swyx: Yeah. [00:24:00]Quentin: So there's a paper that I was involved with called RWKV that I would recommend people read. It is answering this exact question. So how do you get Transformers quality without this quadratic attention overhead that Transformers requires? So it is interesting. I don't know if I can really dive too deep into the technical details there. I'd recommend people read the paper. But yeah. [00:24:23]Swyx: Yeah. [00:24:23]Alessio: It's interesting to see if attention is all you need, or maybe attention is all we need, but we need better ways to make it infer in a good way. [00:24:33]Swyx: We've actually done an unreleased episode with one of the RWKV core members and they call it soft attention or light attention. I forget what they call it, but yeah, just ways to approximate it such that it's linear and not quadratic. That's great. Yeah. [00:24:47]Quentin: I didn't know that you were involved. [00:24:48]Swyx: That's great. How did you get involved? Is it just because like everyone just hangs out in Discord and talks about the future of Transformers? Oh yeah. [00:24:55]Quentin: I mean, the RWKV people specifically are in Eleuther all the time. Like they're very close collaboration with us. And my contribution was we have all of these experiments done by all of these people on RNNs and how they relate to Transformers and how do we turn that into a paper and disseminate that digestibly so that people don't have to read through like a Discord log from a year ago to understand what's going on. [00:25:16]Swyx: Oh my God. [00:25:16]Quentin: Just read this paper. So that took some work, but I wasn't a core contributor. So that's why I don't want to go into like the technical details. But yeah, that's how I did. [00:25:24]Swyx: We'll try to get that RWKV episode out. It seems like there's increasing mentions of it and they are doing pretty important work as far as scaling these models are concerned. Okay. So we discussed inference type quantization and memory requirements. And then you also had a section on training with a lot of stuff I think mentioned. I think we probably want to spend the most of our time on optimizer states and the Atom optimizer. Yeah. What are your takes on it and what should people keep in mind when they deal with these optimizers? Okay. [00:25:57]Quentin: I would say the Atom optimizer is good at what it does. It's sort of a broad question. So let me think. You have the copy of the weights and then you have your momentum and your variance that [00:26:08]Swyx: you store. [00:26:08]Quentin: And like, okay, maybe an intuitive explanation for momentum is that like, let's say you have a canyon and you're trying to get to the bottom. And if you're just doing basic SGD, then every step is going to be an equal size. Whereas if you're using something like Atom with the momentum term, then your steps should be progressively larger because you can see, oh, the general trend is we're heading downwards very quickly. But stepping back from that, since you have all of these extra terms in Atom, you require a lot more memory to store it. Like three times as much memory as SGD. And if you have all of this memory being spent on your optimizer states, then how do you distribute it across GPUs? Because you'll find that what ends up being your bottleneck more than just raw compute, raw flops on a given GPU is your parallelism. And that falls back onto how much model you can fit on a single GPU before you need to split it up across a bunch of GPUs. And then you end up spending time, more time with them talking to each other than actually making progress. So that's why all of this time in the blog post is spent on how do you distribute your model? What are all those different distributed strategies look like? Which ones are more efficient? And given that a lot of your memory is being spent optimizers, how do you distribute that optimizer specifically? Because a lot of people, when they talk about parallelism, they talk about model parallelism, the parameters themselves. In actuality, when you're training, a good portion of your memory is actually spent on optimizer states. So what specific part of that would you like to go into? Would you like to go into like zero or sharded optimizers? [00:27:36]Swyx: I think the sharded optimizer stuff is really interesting, but I think we're kind of leaving that towards the end, right? Because that's the maybe more advanced distributed sections. Here, I think we're just going for rough intuition for people who've maybe are familiar with the ideas of these optimizers, but haven't actually had to implement them yet. They read your code, but they don't really understand the intuition behind the code. I see. [00:28:00]Alessio: And Quentin, when you say in the blog post, it says, Adam is magic. How much of it is like actual magic, even to like people like you that are pretty close to the metal, so to speak? Are some of these things just come as gospel? It's like, I know this works, like I'm not touching it. I'm just leveraging it. How much of it are you actually thinking about improving on in your day-to-day work? I see. [00:28:22]Quentin: So I'm a systems guy. I'm an engineer. And a lot of these things come to me as magic. Adam comes to me as magic. I see it from the gods. I say, this is how a deep learning model is trained. And this is how the next step is calculated. And then I say, okay, how do I make that fast? I would say I do look at ways to improve upon it using things like second order optimizers. So there's a lot of research on there because they're hard to distribute. But the core contribution for me always comes down to someone else has done like some deep learning optimization and I need to make it run fast. So I can't really speak to the motivation of why Adam came about other than like simple, intuitive things like I mentioned with like the momentum. But what matters to me is that Adam takes more memory than SGD, specifically three times. And all of that memory needs to go somewhere and it needs to be split efficiently. [00:29:14]Swyx: Yeah. [00:29:14]Alessio: So when you add them all up, you got 12 bytes per parameter with vanilla Adam. [00:29:20]Swyx: Yeah. [00:29:20]Alessio: And then you still get the model parameters and memory too. So as you mentioned, you need to keep a copy of both for like a FB32, FB16 mixed, a copy of both quantization levels. So there's precision levels. So it's six bytes per parameter. Right. [00:29:36]Quentin: Taking a step back again, is that like, okay, most people think of your model getting big. So you need to split with model parallelism purely, something like tensor parallelism. But we can see that the model only takes like two bytes per parameter if we're doing FB16. Whereas the optimizer itself requires four bytes per parameter for the model states, four bytes for momentum, four bytes for variance. So what matters more is how do you split your optimizer efficiently and how do you store it efficiently? And something like bits and bytes, where the optimizer, you got like eight bit Adam, where those optimizer states is only one byte per parameter instead of four or something like that. That is going to give you a much better return on your model training and on your memory overhead required than if you were to, for example, quantize your pure like FB16 model weights down to int8 or something. So for training specifically, your optimizer memory matters a lot. The most in most cases. [00:30:31]Swyx: Well, yeah. [00:30:31]Alessio: And before we dive into zero, just to wrap up the items that you're going to shard later. So you have the parameters, you have the optimizer states, and then you have the gradients. Just maybe touch a little bit on that. And then we can talk about how to efficiently load them in GPUs. [00:30:48]Quentin: So the parameters are the FP32 copies of the parameters. We include them in the optimizer discussion. Some people don't, but just for clarity, it's 12 bytes per param for the optimizer states and four of them are for that FP32 copy of the weights. Four of them are for the momentum. I already went into why it's important to store momentum, but that's also per parameter. You need to store where that parameter is going and where it's been going in the past. You also need to know, okay, we know where it's going, but there's going to be bumps on this canyon that we're going down. So we need to store its variance. How often are those bumps? Should we be focusing more on the momentum? Or is this parameter just kind of jumping around everywhere? Those are all important answers that we need the optimizer to store, and it's per parameter. So that's where all three of those terms come from. And we also include some competing bits and bytes, for example, an SGD to show that depending on your optimizer, you may store all or none of these and in different representations. [00:31:50]Alessio: I'm looking at the total training memory. You essentially have model memory, optimizer memory, gradient memory, and activation memory. I think that's one of the last discussed things. So maybe just give people a little bit of a view. [00:32:03]Swyx: Yeah, this is completely new to me. [00:32:05]Alessio: Active, you know, recomputation, checkpointing, and all of that. [00:32:08]Swyx: Right. [00:32:09]Quentin: So, okay. So to summarize before activation checkpointing, which will be complicated, you have your model params, like I mentioned before, they used to be FP32. Now they're probably BF16, maybe FP16 if it's an older GPU. Then you have your optimizer. That's where a lot of the memory is going. And it's your high precision, usually FP32, copy of the weights. So that's four bytes per param. And then you have, optionally, a couple more terms like we just discussed, like momentum or variance or whatever else, depending on what your optimizer is. Then you have your gradients. So your gradients is what is the gradient update that we get after running the forward pass on the model. And that's going to be whatever your low precision copy of the weights is. So like two bytes per param, if you're using FP16 or BF16. And all of those are sort of set in stone. And that overhead is not going to go away for the duration of training. Your gradients might get cleared after you back propagate them, but your optimizer states and your model states aren't going away. That memory overhead will be there. Activation recomputation and activation memory is dynamic. So some people will come and have this problem where the model loads fine for training. But then when you actually run your first iteration, or you run some future iteration or something like that, you run out of memory, seemingly at random. And it's because of these activations that you're computing on the fly. Good summary, or do you want to get into activation recomputation now, or do you want me to touch on anything else? [00:33:35]Alessio: Yeah, I was going to say, when is the recomputation happening? How does it decide between recomputing versus storing? And talk a bit more about that, maybe. [00:33:47]Quentin: Yeah, okay. So there's a lot of different ways to do this, but I would say there are a few main ones. First is a very simple scheme. You recompute everything. Every single activation that you calculate is just going to be either used or thrown away until the end. So in that case, you care very much about memory. You care very little about compute. Maybe this would be a case where you have to distribute across a lot of different GPUs, for example. And your communication speed is really low. Then that might be a good case for you to just recompute everything. It happens rarely, but it happens. Next up would be something like selective recomputation. So in selective recomputation, which Megatron has a good paper on, and I believe the figure that we have in our blog post is from, in that case, you sort of do a weighted decision for each activation. So for really big activation tensors, you decide, is this going to be more expensive to save in terms of memory or to recompute in terms of compute? So that's sort of the smart scheme that Megatron implements. And there's a lot of different heuristics they use. It's probably not worth mentioning off this super long equation on a pod, but you should go and read that paper if you're interested on selective recomputation. And then a really stupid scheme that most people go with, including NeoX, would be something like, instead of doing all of these heuristics, you just say, if my tensor is bigger than X, I throw it away. And you set X to some static number, and that's it. And that is good enough for a lot of cases. [00:35:18]Swyx: Why is it good enough? [00:35:20]Quentin: You don't want to store more than, you know, X-sized tensor. And some fall above that, some fall below it. And you're not trying to squeeze. You care more about getting something close enough to what the actual heuristic should be without actually computing the heuristic because you don't want to spend the time writing that heuristic code. [00:35:37]Swyx: Cool. I think that does take us on a grand tour of the memory math. Is there any sort of high-level takeaway before we go into the distributed stuff? Zero and all that. Perhaps more detail than most people have ever encountered. And so I'll repeat the equation that Alessio mentioned again, which is total training memory now has all these components that you've mapped out for the first time as far as we're concerned. Model memory, optimizer memory, activation memory, gradient memory. We covered quite a few algorithms as to the choices you can make there. Anything else that you want to mention about just memory math? I don't think so. [00:36:11]Quentin: I think that about covers it. I will say that it's a very different scheme for training and inference. It's common for people to say, oh, BF16 is the best. Done. Whereas a more correct take is that during training, precision matters a bit more. So BF16 will be around longer for training than it will for inference, in which case your model is sort of already baked. And it definitely doesn't need some of those last bits of precision so you can get away much easier with going to int8 for inference rather than training. So everything that you learn for training has to be relearned for inference and vice versa. [00:36:44]Swyx: There's a third category. You're talking about training versus inference. This third category is emerging with regards to fine-tuning and perhaps parameter-efficient methods of fine-tuning. The naive way to implement fine-tuning is just to do more training. But I don't know if you've developed any intuitions over fine-tuning that's worth inserting here. Any intuitions? If you were to write fine-tuning math, what would go in there? That might be an interesting diff to training math. [00:37:10]Quentin: I think there's a lot of questions that are unanswered for fine-tuning. For example, we know scaling laws for training. And some people have done scaling laws for fine-tuning. But how does a model that's already been trained on one domain transfer to another in terms of fine-tuning size? How many tokens per parameter should you have for your fine-tuning dataset? Maybe I'm ignorant, but I feel like a lot of those sort of practical questions on how a model can transfer and how a model can learn or grok some new ability that wasn't in its original training dataset is something that I would definitely put inside a fine-tuning blog post. [00:37:45]Swyx: Something related to perplexity and, I guess, diversity of the tokens that you get. [00:37:49]Quentin: Yeah, sort of dataset transfer is something that I would be curious in. Learning rate transfer is another one. So your model has some decayed learning rate over the course of training. How does that change for fine-tuning? Things like that. [00:38:00]Swyx: All right, cool. Thanks for indulging that stuff. Sure. Yeah. [00:38:03]Alessio: I think after all of this, you can quickly do the math and see that training needs to be distributed to actually work because we just don't have hardware that can easily run this. So let's talk a bit about that. So zero is one of the first things that you mentioned here, which is focused on sharded optimizers. Maybe run people through that and how to think about it. [00:38:25]Swyx: Sure. [00:38:25]Quentin: So zero is centered around two communication operations. And the first is scatter. And people should be looking at the zero figure that I think we have. [00:38:35]Swyx: Yeah. [00:38:36]Quentin: So there's a figure in the paper with parameters, gradients, and optimizer states that people should be looking at when I'm talking about this. Every GPU is going to get its own equal portion of the slice. And if we're doing... There are different stages of zero, but let's just start off with assuming that it's an equal slice of the optimizer states, gradients, and parameters. That would be zero three, stage three in that case. And we do that with a scatter. And the scatter takes, say, one over end GPUs, plus this offset of that slice goes to that GPU. Now all of the GPUs have an equal slice that's in its rank order. And then during each training step, that GPU is going to wait for all of the other slices to communicate so that we now have a whole pie on that GPU, that single GPU. Once we have that whole pie, we do the forward pass on it. And then we distribute that forward pass to all of the others using a gather. So it's a scatter, reduced scatter specifically, and then a gather back to all the others. And you do that each step. So the point of it is that you're sharding these states across GPUs. And with the different stages, you'll see in that figure that the optimizer state is taking the most proportion, which is because of what I mentioned before. We're including the FP32 copy and we're doing atom. So we need those four bytes per param for momentum and for variance. And then zero stage one, which is the most common one, is just optimizer. Zero stage two is optimizer plus gradients. And zero stage three is optimizer gradients and model parameters. But it all comes back to this splitting up and then gathering together back and forth over and over. So you get a lot of communication overhead from zero. But the plus part of that is that you can overlap a lot of that movement with computation. [00:40:23]Alessio: How do you get the optimal number of GPUs to do this on? Is there a way to shard too much as well and put too much overhead? [00:40:31]Quentin: It depends more on what your interconnect is. Taking a step back, there is synchronization that's required, a lot of it, across all of these GPUs. And those tend to be cumulative. So if you go to too many GPUs on an interconnect that's too slow, then you're going to end up spending more time synchronizing. And that magic number where you spend more time synchronizing is going to be different depending on what your fabric is and what your GPU memory is specifically. Just how small of a slice is each GPU getting? I can't, for example, for Summit, that number comes out to be about 20 billion parameters. Now you have 20 billion parameters, and then your magic number of GPUs for that is going to be something like 100 to 200 scale. Beyond that, you're just going to end up spending more time communicating. And the actual flops dipping below some predetermined number by you is going to be whatever your sweet spot ends up being. [00:41:24]Alessio: And then, so this one was like hard for me to go through, so I'm excited to have you run through it, which is a 3D parallelism. [00:41:33]Swyx: It's fancy, it's cutting edge. [00:41:35]Alessio: Yeah, let's talk a bit more about that and some of the work. [00:41:38]Quentin: Okay, 3D parallelism. So what is each dimension? First is the really basic one. That's data parallelism. And data parallelism is you have a copy of the model. Let's say for simplicity, one copy fits on one GPU perfectly. Data parallelism is that now you have two GPUs, so you have one copy on GPU one, one copy on GPU two. Both of them do the forward and backward pass and then synchronize and average the gradients. And then that's a step. Data parallelism for 3D parallelism is actually zero. So it's, you're sharding the optimizer states across all of your different GPUs. Next up is tensor parallelism. Tensor parallelism is you split your model. Like say, if you have two GPUs, you split your model down the middle and each GPU on its tensor specifically is going to do its forward or backward operation on its tensor. And then only when necessary, it'll synchronize that tensor operation with the other GPU. It's a bit more complex than something like pipeline parallelism, which is the third dimension. In pipeline parallelism, let's say you have four layers in your model. And you have four GPUs. You put one layer on each GPU and then GPU one does the forward pass and then sends the output of its activations to GPU two. It does the forward pass, sends activations to three, and you're just moving down a line. That is a naive scheme in that all of the other GPUs are doing nothing while a single GPU is doing its forward or backward pass. So the reason it's called pipeline parallelism is because you're splitting your mini batch into micro batches. So GPU one will do the forward pass on micro batch one and then send to GPU two. And then while GPU two is running on that first micro batch, GPU one is working on the next micro batch. And so you're sort of pipelining the movement and computation of each micro batch. The problem with that is that you need a really big batch size in order to split it up into both mini batches and micro batches. So combining all three of those together, you get a 3D mesh of where each parameter and optimizer state and so on maps to each GPU. And that's 3D parallelism. So let's start diving into details on what have that made sense, what should I jump into more on? [00:43:55]Alessio: I think the main question is, do you need all of the GPUs to be the same to do this? Or can you have mismatching GPUs as well? [00:44:03]Quentin: Okay, two things matter. If there's a difference in VRAM for the two different kinds of GPUs, then you're going to be bottlenecked by whichever GPU has the lower amount of VRAM because it's going to run out of memory. And then you can't like whatever's left on the larger GPUs is going to be empty. As far as I'm aware, there's no like GPU single GPU aware memory overhead scheme that would account for that. The second problem is that let's say all of your GPUs have the same amount of VRAM, but half of them are really slow. And the problem with that is that those synchronizations that I mentioned earlier are going to kill you. So you're going to move as quickly as your slowest GPU in that case. So in both cases, you end up regressing to your slowest or smallest GPU. So you might as well have the same GPUs for all of them. Otherwise, you're wasting the nicer ones. And that also goes to your CPUs and your interconnect. So going back to the 20 billion parameter model that Eleuther was training, that was on a cluster that was sort of Frankenstein made during COVID when there was all of that shortage of network switches and such like that. So every node had a different network switch. And so you ended up moving at the speed of the slowest switch and getting everything tuned properly so that it's not worse than the slowest switch was challenging and is like a real world problem that sometimes comes up. [00:45:28]Alessio: Is this work widely accepted? Like I hadn't learned about this before studying for this episode. Is this something that people are still trying and researching? Or is everybody just aware of this and running this in production? [00:45:43]Quentin: What is this specifically? [00:45:44]Alessio: Like the sharded optimizers plus the 3D parallelism, bringing the two things together and having this kind of mesh strategy. [00:45:51]Quentin: I would say that a lot of major GPT-based models use this scheme. A lot of them now are sort of going with just a pure zero scheme. So just a pure sharded. You just shard everything. And then since that's so easy, everyone gets an equal slice. There's no such thing as a pipeline stage. There's no such thing as what tensor should go on which GPU. Instead, we shard everything equally and treat everything equally. It's a much easier problem to debug, to checkpoint, to run training on than it is with this 3D parallel scheme. I say 3D parallel gives you the most control and also the most ways to go wrong. And depending on whether you have more engineers or whether you have more GPUs, that should decide which of these you go with. [00:46:35]Swyx: It's also not too hard, right? You've basically outlined the five or six different numbers that you need to keep in your head. And it doesn't feel impossible that if you need to achieve that level of control, you've given everybody the main levers to do it with. And that's wonderful. Definitely. [00:46:51]Quentin: The problem that comes up is like, say, like, okay, GPT-4 came out. Now we have VLLMs. [00:46:57]Swyx: Whoa, what are VLLMs? Oh, okay. Virtual LLMs, like the Metro of Expert things? No, like visual. [00:47:03]Quentin: So now you have like multimodal models and such. How do you distribute that? Do you distribute it in a pipeline stage? And do you just shard it? Do you split the tensor and make a tensor parallel? It's sort of hard to change your model and add new features and such when you have this 3D parallel scheme. That's when I say hard. I mean, it's hard to sort of adapt and modify it to new features. [00:47:26]Alessio: I know we're at the hour mark, and I think we put our listeners through a very intense class today. So this was great, Quentin. And we're going to definitely link the article so that people can read it and follow along. Any other research that you're working on in this space that you want to shout out? I know one of our usual, I mean, wrong question is, what's the most interesting unsolved question in AI? So curious to hear if you think it's still on the training inference, math optimization, or are there more areas that people should pay attention to? [00:47:58]Quentin: I think in my area of research, there are two things that I think people should really care about. And the first is multimodal parallelism and RLHF. You were seeing more and more reinforcement learning and coming into the training loop. And so how do you split that some model or some GPUs are working on inference and some GPUs are working on training? And like I mentioned before, you have to relearn everything and they have very unique challenges. How do you split up a KV cache during training, for example? Those are challenges that are not well studied, I don't think. And then multimodal, you have like maybe a vision transformer and a text transformer. How do you split those up? Do you split them up equally? Do you put them on separate GPUs or do you just shard everything? And just maybe one GPU will have some vision, some text parameters. And then the second case I would say is that communication is very often a bottleneck. So we talk about 3D parallelism, but a lot of those like, for example, tensor parallelism, you can't go across nodes with. You'll just get killed in communication. So what I'm getting to is how should you compress your communication before it happens? So on the fly compression, you have some buffer that needs to be communicated. You compress it with a GPU kernel, then you send it across the network and then you decompress it, something like that. Making people spend less money on communication fabrics and more on GPUs as intended is sort of a thing that people need to explore. I think those are my two. [00:49:26]Alessio: Sean, you went over the other half of the lightning round before we wrap it up. [00:49:30]Swyx: That's a good brain dump. Cool. Yeah, I have so many more questions on the multimodal stuff, but that should be for another time. Acceleration, what has already happened in AI that you thought would take much longer? [00:49:42]Quentin: I would say flash attention. Guys, just talk to Tree. And flash attention is just sort of a really great set of kernels that I thought would take a while to get to us. [00:49:51]Alessio: Well, Quentin, thank you very much, man. This was super informative and I think hopefully helps demystify a little bit the blog post. I think people open it and it's like a lot of math on it. And I think you walking them through it was super helpful. So thank you so much for coming on. [00:50:07]Swyx: Of course. [00:50:08]Quentin: And I'm happy to answer any questions that people have offline if they have them. I do read my email. [00:50:13]Swyx: Email and Discord. Of course, yeah. [00:50:15]Quentin: Discord I'm even faster on. [00:50:16]Alessio: Thank you, everyone. [00:50:18]Swyx: Thanks, Quentin. [00:50:19] Get full access to Latent Space at www.latent.space/subscribe
Récitation du Coran lors de la prière du Tarawih pendant le mois de Ramadan 2023 à la mosquée Al Hikma, la sagesse, à Bruxelles --- Send in a voice message: https://podcasters.spotify.com/pod/show/belligh/message
Esta semana Mariano Urdín ha probado la Moto Guzzi V100, una marca que le trae grandes recuerdos de su época de piloto y que ha sabido evolucionar con una moto que incorpora un nuevo motor en V Transversal con toda la tecnología. También tiene algunas innovaciones curiosas en su carenado. Hablamos también del Mundial SBK, de las últimas victorias de Alvaro Bautista, y del resto de los pilotos de este otro gran mundial de motociclismo. ¿Podría ser la última temporada de Bautista? Hay rumores que hablan sobre ello. También te descubrimos una revolucionaria patente de Honda, orientada hacia motos off road y que tiene un control de los saltos y de la aterrizaje posterior a la moto. Algo que podría ser de mucha utilidad para los pilotos del Rally Dakar, puesto que se ha probado en una Honda CRF 300. Estamos en la semana del Gran Premio de Jerez y tanto si vas a viajar a la cita deportiva como si no, hemos preparado un final de podcast con muchos consejos para iniciar una ruta en moto, desde la preparación de la moto, el equipamiento, rodar en grupo o la planificación del viaje. Recuerda enviarnos tus dudas o sugerencias a redaccion@moto1pro.com.
On 580 Live from the Par Mar Stores Studio, Millie Snyder from The Shape Shop, Delegate Dana Ferrell and Matt Murphy with more on this evening's Working Women's Wednesday at the Red Carpet Lounge with V100. 580 Live is presented by Thornhill Auto Group.
On 580 Live from the Par Mar Stores Studio, nutrition expert Millie Snyder, Delegate and State Democratic Party Chair Mike Pushkin and more on V100's “Galentine's” Working Women's Wednesday after work at The Red Carpet. 580 Live is presented by Thornhill Auto Group.
Pastor Ben talks about Kingdom Vision and generosity. A look into the future of First Assembly and how you can partner with us. From the series Kingdom GenerositySupport the show
Hello everyone and welcome once again to the Ultimate Motorcycling weekly Podcast—Motos and Friends. My name is Arthur Coldwells. Moto Guzzi just launched the new V100 Mandello S and Senior Editor Nic de Sena went to the event in Italy. The hallowed Italian marque has radically redesigned its flagship motor—while managing to keep its iconic v-twin look. There are some big changes though, and not just to the motor; the new Guzzi looks fast and sporting. Nic gives us his thoughts and tells us whether the new Guzzi actually delivers on its considerable promise. In our second segment, Associate Editor Teejay Adams chats with artist Karl Hoffman. We recently met Karl at his Art Gallery H in the small, historic town of Tubac, Arizona. Karl's life journey has been nothing short of extraordinary, and of course it has involved motorcycles pretty much all the way. As a painter and jewelry designer, Karl's fine art and spectacular jewelry is absolutely spellbinding—as an artist herself, Teejay spent a long time talking to him and admiring the fruit of his considerable talents. His life story was so compelling she decided on the spot that she'd like to share it with you. This is the first part of two. So from all of us here at Ultimate Motorcycling, we hope you enjoy this episode! Art Gallery H in Tubac, AZ (video) Artist Karl Hoffman with Teejay Adams Moto Guzzi Mandello S
L'hebdo Du Repaire #32 - BMW S 1000 RR, Ducati Multistrada V4 Rally, QJ Motor GS550, Guzzi V100 Aviazione Navale, V-Twin Moto Morini, Rappel Honda Semaine du 26 septembre au 01 octobre Un résumé de l'actualité du monde moto en 15 mn maximum ! N'hésitez pas à parler de notre émission autour de vous ! Suivez-nous : http://www.lerepairedesmotards.com/ https://www.facebook.com/LeRepairedesMotards https://twitter.com/lerepaire https://www.instagram.com/lerepairedesmotards/ Podcast: https://anchor.fm/le-repaire-des-motards
Aerogeneradores con dos hélices / Caída en picado de los precios de GPU y RAM / En Japón se vuelven locos por las casetes / Película sobre las acciones meme / El bug de emergencias de los Pixel sigue
The Ladies100Too Slow To Disco·鸭鸭摇V100
İntikam ve Duygusal Kontrol - (V100)
Dr Mouhammad Ahmad LO | Les appels D'Allah aux gens de foi 16 | 100 Sourate Ali 'Imran نداءات الرحمن
P.Y.O.N Media present SpeakLife Nation Show. Hosted by LaShawnda Wilkins, SpeakLife Nation Show highlight the journey of professionals, artist, creatives, and other entrepreneurs in hopes to inspire others to chase their wildest dreams. On today's show LaShawnda features Credit Guy MKE, Lamont Smith! But it's more to him than his expertise in financial solutions. Tune in to SpeakLife Nation Show today. You just been Put On Notice!
P.Y.O.N Media presents Set The Tone featuring Crysy.B. Crysy.B is a young up and coming artist putting on for Milwaukee, WI. Her day view album Bipolar Love premiers February 25th on all streaming platforms. Tune in to Set The Tone to learn more about Crysy.B and her motivation for Bipolar Love. You just been Put On Notice!
P.Y.O.N Media presents A Heartfelt Story, hosted by Shannon King. A Heartfelt Story highlight individuals dedicated to impacting their community and empowering those around them. On today's episode Shannon King feature Laneice McGee. She's a boss on every level. Co- Founder of Big, Beautiful & Blessed & Founder of FEMA- Future Entrepreneurs Moving Ahead, Laneice McGee is putting her best foot forward everyday. Tune in to hear her Heartfelt Story right here on P.Y.O.N Media. You just been Put On Notice!
P.Y.O.N Media present SpeakLife Nation Show. Hosted by LaShawnda Wilkins, SpeakLife Nation Show highlight the journey of professionals, artist, creatives and other entrepreneurs in hopes to inspire others to chase their widest dreams. On today's show LaShawnda features Eboni P. Jones, The Cerebral Palsy Queen. She'll be discussing Cerebral Palsy Awareness Month and why she's perfectly able and not disabled! Tune in now and subscribe to P.Y.O.N Media on all podcasting platforms. You just been Put On Notice!
P.Y.O.N Media presents Set The Tone featuring Tarsha Wiggins!Tarsha has been "setting the tone" right here in Milwaukee, Wisconsin using urban/ hip hop music to drive meaningful conversations about mental health in the black & brown communities. Her mission is to reduce the stigma around mental illness. Tune in to hear how she's impacting her community and those around her with Speak Wellness & Trap Therapy. You just been put on notice!
In this week's episode, I cover more details on the LAPSU$ a on cyber gang, info on the recently disclosed Spring4Shell Vulnerability, Citrix DaaS announcement and more! Reference Links: https://www.rorymon.com/blog/episode-224-citrix-daas-announcement-spring4shell-chrome-v100-citrix-issues-more/
In this week's episode, I cover more detail on the LAPSU$ attacks on Okta, info on an actively exploited Zero Day Vulnerability in Chrome and more! Reference Links: https://www.rorymon.com/blog/episode-223-okta-security-breach-update-citrix-adc-bug-chrome-v100-more/
Reggie and JaQuan give all the latest, from their perspective with the latest on Wendy Williams WNBA Star Brittney Griner, V100.7's New TikTok Page and more!
Dans cet incroyable épisode, il est question : de la CNIL et de Google Analytics ; de la faille #log4shell ; d'euthentification, de la version 100 de Chrome et Firefox ; d'un nouveau container pour les requètes ; d'un incident poilu sur PostgreSQL ; de migration de métadata ; d'éxécution de code arbitraire et d'extension de cockpit JSON… pour finir, comme il se doit en musique.
P.Y.O.N Media presents Set the Tone featuring Brett Andrews, from The Brett Andrews Show. You can hear him on over 30 different radio stations and he's also the Program Director of a few. Brett Andrews have been setting the tone for over 25 years and counting. Tune in to hear how his passion for broadcasting has always been his calling. You just been put on notice!
P.Y.O.N Media presents A Heartfelt Story, hosted by Shannon King. A Heartfelt Story highlight individuals dedicated to impacting their community and empowering those around them. Shannon King help build up her community through the power of storytelling. Today's episode feature Kenneth Ginlack, Sr. Tune in to hear how he's been able to turn his life around after battling addictions, to obtaining two degrees and spreading his knowledge with the world. You just been Put On Notice!
P.Y.O.N Media presents Set The Tone featuring Cam Wallace. Putting on for Houston, TX. Cam Wallace has worked with some of the BIGGEST artist in the game. He's a song writer, producer and recording artist! Inspired by Chicago's very own Kanye (Ye), Cam can do it all! Tune in to Set The Tone and get put on notice about Cam Wallace today.You just been Put On Notice!
You On with fitness expert Shannon All Around. Tune in to get tips on maintaining a healthy body and mind with All Around Sports. Shannon has made an impact in her community and she's been an inspiration to so many. Tune in to Put You On Notice Podcast and get put on notice today.P.Y.O.N MediaOur Platform Your Opportunity
P.Y.O.N Media features Set The Tone, the show that speaks for it's self. Set the Tone highlight individuals from all walks of life, giving them recognition for their hard work and dedication in their respective fields. Today's episode feature Capitol Record's Regional Promotions Manager, Miriah Mark. She's "dope and do dope sh*t!" Tune in now to hear how she setting the tone everyday and putting herself in position to WIN!You just been Put On Notice!
Chris Lawrence and Carrie Houdesek host along with Steve Bishop and Jenny Murray from V100.The festivities translate great to the radio. There's performances, information on the floats, interviews, special guests, and tons of fun with Carrie and Chris.
29° Pillola. EICMA 2021. Il salone della moto di Milano, una tappa annuale da vedere e vivere in prima persona. Boicottata da BMW, KTM e DUCATI per i loro validi o non Motivi ma non da Triumph, MV, Gruppo piaggio, Le case giapponesi, Fantic, Benelli ed altre 800 presenze... Vi racconto le moto che mi hanno colpito in primis la Xcape 650, la Benelli trk 800, la V100 della guzzi per poi passare ad Aprilia Tuareg 660 ed un pò di elettriche... Insomma una bella fiera che vede i prodotti CINESI avanzare sempre più come al solito ma sempre più affidabili, sicuri e performanti tanto da essere scelti dalle case più blasonate.... Buon Ascolto Vi ricordo che il sito internet è già operativo e potete ascoltare tutte le pillole dalla prima all'ultima e vedere anche un pò di foto e video relativi alle mie puntate. Se avete voglia potete curiosare nelle restanti pagine del mio sito personale che si sta aggiornando piano piano... http://www.ivanbeghelli.it/pillole-di-mt.html Nel canale telegram del Podcast potrete trovare altri contenuti che riguardano la puntata. Per contattarmi, interagire con il Podcast M.T.& o lasciare commenti: canon.ice73@gmail.com www.ivanbeghelli.it #eicma #2021 #moto morini #xcape 650 #benelli trk 800 #Triumph #MV augusta #Piaggio #vespa #Fantic #benelli #guzzi #v100 #mandello #zero motorcycle #zero sr #cake #makka range
Molti si sono fermati all'estetica, che obiettivamente non esprime granché. Ma sotto le forme c'è un gran motore da 115 cavalli e una piattaforma completamente nuova. Prevedo che dopo la Mandello arriveranno moto entusiasmanti
Thanks to everyone who contributed through our Facebook group - you can join the fun here: facebook.com/groups/fulltankpodcast/ You can also let us know what you think of the show by messaging us on Instagram: instagram.com/moto.bob instagram.com/rarefiedroad And our YouTube channels are here too: youtube.com/c/motobob youtube.com/user/timaitkensmith
You On with Bailey Coleman, Senior VP of Programming at iHeartMedia and the voice of V100.7. Tune in to hear about her journey in radio and find out what was the absolute worst interview she's ever had. You'd never guess who made her feel uncomfortable! Put You On Notice Podcast, Our Platform Your Opportunity
You On wit Bailey Coleman, Senior VP of Programing at iHeartMedia and the voice of V100.7! Tune in to hear about her journey in radio and find out what was the absolute worst interview she's ever had. You'd never guess who it was with! Put You On Notice Podcast, Our Platform Your Opportunity --- Support this podcast: https://anchor.fm/putyouon-notice/support
Berkeley Packet Filters (BPF) — это технология ядра Linux, которая не сходит с первых полос англоязычных технических изданий вот уже несколько лет подряд. Конференции забиты докладами про использование и разработку BPF. В 100-м выпуске продолжаем разговоры о linux networking.Url podcast:https://dts.podtrac.com/redirect.mp3/fs.linkmeup.ru/podcasts/telecom/linkmeup-V100(2021-06).mp3
It's been nearly a month since the last podcast. The decision to add a video podcast component may prove to slow the process down too much for the quick "while my batteries charge" aspect of recording the show. I'm weighing whether or not to continue the video podcast. I'll still do it for interviews, but not the updates. What are your thoughts on that? Here are some of the topics discussed Losi LMT Updates *New* Losi V100 On-road car *New* Arrma Vorteks Monster Truck Racing Feeling disconnected from the community Upcoming Schedule Axialfest Badlands
Set The Bar... in the studio? That's right we back in the studio for the first time this year! Huge shout out to V100.7 very own Bailey Coleman for this opportunity. Tune in to Set The Bar with The Put On Squad. On today's episode we pay homage to DMX, discuss Kanye West deal with Netflix and put you on notice about some stock you should be buying right now. Tune in to Put You On Notice Podcast, Our Platform Your Opportunity. --- Support this podcast: https://anchor.fm/putyouon-notice/support
Foreign Language Learning. Universal Teaching For Foreign Language Learning. V100. English/French. Aboussouan International For Foreign Language Learning. The Stairs For Continuing Education For All ages. --- This episode is sponsored by · Anchor: The easiest way to make a podcast. https://anchor.fm/app
This episode we sat with Radio Personality Promise of 105.7. One of the Biggest Radio voices in the city of Milwaukee! It was dope to hear his story of coming up including his time at the historic v100 here in Milwaukee WI to now being the morning VOICE OF THE NEW 105.7! That and celeb stories!!! Vibe with us!
In this episode, AC & CJ catch up on the Microsoft cloud news, including updates to Microsoft Surface, Azure, Microsoft Teams and Skype!News Parler Is Gone, But Hackers Say They Downloaded Everything First Surface Pro 7+ brings next level performance and versatility to the enterprise Microsoft is building a new Outlook app for Windows and Mac powered by the web Microsoft’s Azure Modular Datacenter: What’s ‘special clouds’ got to do with it? Microsoft Designing Its Own Chips for Servers, Surface PCs Introducing Microsoft Cloud for Retail Release Notes for Skype 8.67 Picks AC’s Pick Can Apple’s M1 help you train models faster & cheaper than NVIDIA’s V100? CJ’s Pick Now how many USB-C™ to USB-C™ cables are there? (USB4™ Update, September 12, 2019)
The Platform 293 features a forty five minutes hip-hop and R&B mix by Milwaukee's own DJ O! Heard on Milwaukee's air waves frequently working with one of the most popular stations V100.7! He's now booking future gigs in Milwaukee and the Atlanta, Georgia area. Reach out to him on his socials by clicking the links below and enjoy his latest mix on The Platform! FB: https://www.facebook.com/djomilwaukee IG: https://www.instagram.com/djo414/ Mixcloud: https://www.mixcloud.com/djo414/ Twitter: https://twitter.com/djo414
My Daily Thought is more random than it is focused..lol Thank you for watching and listening!! Please like, share, comment, and subscribe!! Be blessed!! Website / KIRWKC Anchor Podcast Site: www.kirwkc.com iHeartRADIO: https://www.iheart.com/podcast/269-keepin-it-real-with-kc-73615909/ Pandora: https://www.pandora.com/podcast/keepin-it-real-with-kc/PC:46195 Apple Podcast: https://podcasts.apple.com/us/podcast/keepin-it-real-with-k-c/id1494499465 Daily Motion: https://www.dailymotion.com/KIRWKC YouTube: https://www.youtube.com/channel/UCyCEtmZUJo3tbrqrJLWRxZw Twitter: https://twitter.com/kirwkc @kirwkc Instagram: https://www.instagram.com/kirwkc Facebook: https://www.facebook.com/kirwkc --- This episode is sponsored by · Anchor: The easiest way to make a podcast. https://anchor.fm/app Support this podcast: https://anchor.fm/kirwkc/support
Another non-factor on the political landscape. Thank you for watching and listening!! Please like share, comment, and subscribe!! Be blessed!! Website / TPES Anchor Podcast Site: www.thepurpleelephantshow.com Apple Podcast: https://podcasts.apple.com/us/podcast/the-purple-elephant-show/id1373583160 Daily Motion: https://www.dailymotion.com/ThePurpleElephantShow Rumble: https://rumble.com/user/ThePurpleElephantShow Parler: https://parler.com/profile/Thepurpleelephantshow/posts @Thepurpleelephantshow Instagram: https://www.instagram.com/thepurpleelephantshow BitChute: https://www.bitchute.com/channel/o2yy59QDShg3 --- This episode is sponsored by · Anchor: The easiest way to make a podcast. https://anchor.fm/app Support this podcast: https://anchor.fm/the-purple-elephant-show/support
Kickin' It with Tyree is a talk show with a twist. Reaching out to artists with an objective to display their talents and artistry. Not only does it allow these artists to display their art, but also gives them a platform to provide insight, depth and background information about the projects that dedicate their time to, and It also gives us all a chance to have fun. Tyree and Bri sit down for some good talks as talks about her plans for 2021, pitching to V100 host Reggie Brown, what it's like being a woman in this industry, becoming a DJ and what it has felt like to grind her way to the top. --- Support this podcast: https://anchor.fm/tyree-pope/support
The final week of 2020 has finally arrived. I can't thank all you listeners enough and a special thank you goes out to all of the guest DJs on the show. As we head into 2021 we'll have aired 275 episodes starting with number 272 today by Milwaukee's DJ Roc. His third time being heard on the show and this time around he gives listeners 45 minutes of latin, house, hip-hop and trap music. During the pandemic he's remained a bi-weekly resident at District on Water every other Saturday and he'll also be ringing in the New Year there so hit him up for details if you're going out in Milwaukee. He's heard on V100 as an on air DJ as well as being an official DJ for the Milwaukee Bucks. Be sure to follow him on his socials by clicking the links below and enjoy Roc's latest mix, on The Platform. FB - https://www.facebook.com/DJROCMKE/ IG - https://www.instagram.com/djroc13/ Mixcloud - https://www.mixcloud.com/djrocmke/ Twitter - https://twitter.com/ROCMKE
Randy Raley, veteran St. Louis disc jockey and current program director at V100 in Topeka, joins host Mark Reardon to talk music including Tom Petty’s Wildflowers boxset and Bruce Springsteen. Movie reviewer Dan Buffa takes a look at The Croods 2 and more recommendations. See omnystudio.com/listener for privacy information.
V100.7 Reggie Smooth Az Butta about to put you on notice! Check out the newest episode of Put You On Notice Podcast featuring a new segment, Caught in The Web. Is a man only loved for what he can provide? You Just Been Put On Notice --- This episode is sponsored by · Anchor: The easiest way to make a podcast. https://anchor.fm/app
Country music legend Charlie Daniels has died at the age of 83 after suffering a stroke near Nashville. Longtime KSHE Disc Jockey Randy Raley, now the Program Director for V100 in Topeka, joins host Mark Reardon to reminisce about the musician.
In this episode, you are going to get real talk about real issues going on right now. It was a breath of fresh air to have an open conversation with Dres. Dres is a globally-inspired artist and mental health advocate. For over ten years, he has served his community by providing support to inner-city kids and young adults who struggle with their mental health (trauma, depression, psychosis, autism, Aspergers, ADHD etc.). In 2017, he launched his entertainment & lifestyle company “Urban & Arrogant". Through his company, he releases music, merchandise, and delivers visual content. His latest single "Countin' Racks" has been played on Iheart Radio Station; V100.7 FM, and Radio Milwaukee (88.9 F.M.). In 2020, Dres was awarded the Shepherd Express "Best Rap Hip-Hop Artist" award. Dres is scheduled to release his debut pop-rap album which will be available on all streaming platforms later on this year. Your feedback is always welcome. If you want to share your story on Local First Podcast all you need to do to get started go to ScheduleMyPodcast.com (http://ScheduleMyPodcast.com) This is episode is brought to you by: EXACTA Corp (https://wp2.myexactamundo.com/) EXACTA Corporate Organizer CRM&More brings order to chaos in your business by helping you Plan |Execute | Analyze and Close more business. RARE Leaders (http://rareleaders.com) Connect - Collaborate - Contribute through RARE Conversations Support this podcast
The Platform episode forty-seven features a mix from Roc! Our guest for the second time on the show and today on Cinco De Mayo! Being a mexican American, he takes pride in his heritage and loves working latin music into his regular sets. He gets to show some of that off with this mix today. You can hear him around the city of Milwaukee at every major venue, on the radio as a member of the V100 mix squad or at Milwaukee Bucks games as an official DJ. While sports are postponed and venues are closed, be sure to follow him on his social media pages to tune in for future live streams! FB - https://www.facebook.com/DJROCMKE/ IG - https://www.instagram.com/djroc13/ MixCloud - https://www.mixcloud.com/djrocmke/ Twitter - https://twitter.com/ROCMKE
The Platform episode forty features a mix from DMatic! We also have to wish him a happy birthday! One of Milwaukee's favorites, you can listen to him all around the city but more specifically at his residencies at District on Water and Points View. He's an avid Office lover and he's one of the radio DJ's for V100. Today he gets into it with some trap, moombah, reggaeton and more. Be sure to follow him on his social media pages below, wish him a happy birthday and enjoy! FB - https://www.facebook.com/djDMaticMKE/ IG - https://www.instagram.com/djdmatic/ MixCloud - https://www.mixcloud.com/DMaticMKE/ Twitter - https://twitter.com/djDMatic
This week on YBO...IT'S REESE'S BIRTHDAY on the 23rd.. We get into What’s Going On and give our thoughts on Reggie Brown being pushed from V100, Tyler Perry’s new movie “A Fall From Grace” (wigs included), and Fox Soul. For Random Sh*t of the Week we talk about listening to old episodes, the best work shift for us and the Netflix show “The Circle.” When it comes to Dating, Relationships & Sex for you nasties ask are there no more rules in dating and being toxic. We close the show with our weekly anon and inspiration of the week. Find more at ybopodcastmke.com
Welcome back to Teezy Talks Episode 130! This weeks guest was V100.7, Reggie “Smooth Az Butta” Brown We talked about everything and shoutout to Reggie for being an open book; Tune in, he was pulling out receipts and all! We discussed How He got started How he got his Dj name the Young MA situation Love life His pics with celebrities And much more! We were unable to play our artist this week but shoutout to the winner of the poll Ag Da Gif ft Duch- No Excuses @teezytalks-Instagram @teezytalks-twitter Teezy Talks Podcast- Facebook
УРААА!!! 100!!! Так! обязательно его лайкаем и репостим, маме и папе покажи, друзьям обязательно! Всем расскажите, что есть такой подкаст, который желает вам с утра доброго утра) Треклист как обычно: AFFKT feat. Sutja Gutierrez - The Show (Original Mix) Cosmo's Midnight feat. Age.Sex.Location - Have It All Dotan - Home II Jamie N Commons - Won't Let Go JMSN - Drinkin' Kali feat. Скриптонит, Truwer - На пределе KEYMONO - Morning Kindness feat. Jazmine Sullivan - Hard To Believe Kyle Watson feat. Kylah Jasmine - You Boy Linda Lyndell - What A Man Metallica - Enter Sandman Paul Woolford, Karen Harding - You Already Know RÜFÜS DU SOL - Underwater (Adam Port Remix) Sabb - Jeopardized Tove Lo - Glad He's Gone (Major Lazer Remix) Tuxedo - The Tuxedo Way Us3 feat. Rahsaan, Gerard Presencer - Cantaloop (Flip Fantasia) План Ломоносова - Шаляпин _______________________ Послушать подкаст можно еще тут: https://t.me/paladinfm https://vk.com/paladinfm _______________________
УРААА!!! 100!!! Так! обязательно его лайкаем и репостим, маме и папе покажи, друзьям обязательно! Всем расскажите, что есть такой подкаст, который желает вам с утра доброго утра) Треклист как обычно: AFFKT feat. Sutja Gutierrez - The Show (Original Mix) Cosmo's Midnight feat. Age.Sex.Location - Have It All Dotan - Home II Jamie N Commons - Won't Let Go JMSN - Drinkin' Kali feat. Скриптонит, Truwer - На пределе KEYMONO - Morning Kindness feat. Jazmine Sullivan - Hard To Believe Kyle Watson feat. Kylah Jasmine - You Boy Linda Lyndell - What A Man Metallica - Enter Sandman Paul Woolford, Karen Harding - You Already Know RÜFÜS DU SOL - Underwater (Adam Port Remix) Sabb - Jeopardized Tove Lo - Glad He's Gone (Major Lazer Remix) Tuxedo - The Tuxedo Way Us3 feat. Rahsaan, Gerard Presencer - Cantaloop (Flip Fantasia) План Ломоносова - Шаляпин _______________________ Послушать подкаст можно еще тут: https://t.me/paladinfm https://vk.com/paladinfm _______________________
УРААА!!! 100!!! Так! обязательно его лайкаем и репостим, маме и папе покажи, друзьям обязательно! Всем расскажите, что есть такой подкаст, который желает вам с утра доброго утра) Треклист как обычно: AFFKT feat. Sutja Gutierrez - The Show (Original Mix) Cosmo's Midnight feat. Age.Sex.Location - Have It All Dotan - Home II Jamie N Commons - Won't Let Go JMSN - Drinkin' Kali feat. Скриптонит, Truwer - На пределе KEYMONO - Morning Kindness feat. Jazmine Sullivan - Hard To Believe Kyle Watson feat. Kylah Jasmine - You Boy Linda Lyndell - What A Man Metallica - Enter Sandman Paul Woolford, Karen Harding - You Already Know RÜFÜS DU SOL - Underwater (Adam Port Remix) Sabb - Jeopardized Tove Lo - Glad He's Gone (Major Lazer Remix) Tuxedo - The Tuxedo Way Us3 feat. Rahsaan, Gerard Presencer - Cantaloop (Flip Fantasia) План Ломоносова - Шаляпин _______________________ Послушать подкаст можно еще тут: https://t.me/paladinfm https://vk.com/paladinfm _______________________
The FLYEst online radio show returns for Season 4 (bonus ep) ... You dont want to miss this weeks show playing new music and topics! Topic: "The State of Milwaukee" Segments: 1. Studies Show 2. Live Call Ins 3. FLYE (top) Five 4. Message of the Day ft . "The PASTOR" Tune in to hear conversation on current events and the hottest underground music from Milwaukee & other cities. This week we have music from Milwaukee, Chicago, Miami, St. Louis, ect. Got Music you want heard on FLYE Radio, email us @ FlyeRadio414@gmail.com --- The live link ---> http://tobtr.com/s/10376099 Live call ins will be available from 8:15pm - 9:30pm -- toll free - (760) 454-8829 Join the group FLYE Radio Topics for more dialogue and follow us on Twitter and Instagram @FlyeRadio
This week Eric asked why u was masterbate with in house and well phone calls happened and we asked tough questions! VIBE WITH US! Sponsor this week is DJ HEATHEN of V100
This week Eric asked why u was masterbate with in house and well phone calls happened and we asked tough questions! VIBE WITH US!Sponsor this week is DJ HEATHEN of V100
Today is the first day of July, the first day of Q3, and the official halfway point of the year. Since we only have six months left in 2019, if you've had time to reflect on the six months that have passed, you may be lamenting on how far you are from the goals you set in January. (Or you may have crushed your goals completely, to which I say, yay you!) But if you are nowhere near where you thought you'd be, consider this your pep talk - it's time to get back in the game. I'm here to let you know, all is not lost. Even if you were asleep from January - June, you can make up the difference with the months we have left. You simply need to revisit your goals, tweak your strategy, and decide what you're going to do differently. But first, realize you're not alone. According to U.S. News & World Report, the failure rate for New Year's resolutions is said to be about 80 percent, and most lose their resolve by mid-February. I for one am guilty of hopping off the consistency train. At the end of May, I looked at my annual goals and felt a little demoralized; I simply was not on track with many of the goals I'd set. But in true Amanda fashion, I had some surprise "sleeper" wins that helped me close out the month, the quarter, and the first half of the year in spectacular fashion. Here's what's been going on: One of my goals for the year? More international travel. But for whatever reason when I set the goal in January, I failed to book any tickets or plan any trips. Fast forward to May: a friend asked if I could join her in Italy on the heels of a big conference she was attending for work. I said yes, spent an epic week in Italy in June, and just like that, my travel and adventure goals are back on track for the year. Another goal? Double my podcast downloads in 2019. As June approached, I could see I wasn't on track to meet that goal. I had exceeded last year's numbers up to this point, but doubled them? Not quite. Then by chance I went looking for parenting podcasts to help me navigate a few issues with my kids as puberty is setting in, we're switching schools in the fall, and I need a more effective way to communicate with my soon to be teenaged son. I randomly met a parenting podcaster at a media event in mid-May. She's doing a challenge to record an episode every day this year, which inspired me to challenge myself to publish an ambitious 30 episodes in 30 days. While I set out to do the challenge because I was inspired by the parenting podcast, I inadvertently boosted my podcast downloads due to the sheer volume of new episodes. And just like that, we're back on track and back on schedule to reach the 2x goal for the year. Sidebar: I learned a lot about epic productivity, motivation, and building habits during that challenge and will share the lessons in a future missive, but for now, here are a few of my favorite episodes: PYG 108 - You have to sell it PYG 100 - What are you willing to create? PYG 99 - Muscle Memory PYG 97 - Making yourself at home on a new level PYG 92 - Build a brand library so people can find your ideas Another big goal? Land media exposure. June was an exceptional month for our PR clients in Maximum Exposure. After two months of crafting ideas and pitching, I was honestly getting a little worried that we hadn't had more media wins. So at the beginning of June, I decided to further examine our process to figure out what might have been missing. We focused more on breaking news, increased our time spent on research, and changed the order of the weekly program touch points. The result? Our changes worked like magic. In the month of June alone, our students were interviewed or featured by Fast Company, the Washington Post, the Washington Post Express, ABC Milwaukee, V100.7 FM in Milwaukee, ABC 7 DC and Let's Talk Live. While media exposure wasn't tied to a specific goal I set at the beginning of the year, it was tied to a goal I set when we opened our doors to the program in April. And it was a great example of how sometimes you may plan for a trickle of sustained success over time, but end up with a flood of it all at once - if you can hone in on the right new strategy. At the end of the day - by trickle or by flood - you can still reach your 2019 goals. It may not happen out of the gate, but when you're honest about what isn't working and you decide to win, success can come at you fast. If you're looking at July 1 with a side-eye because you are nowhere near where you thought you'd be by now, remember, you can change your results if you change your strategy. You can make big things happen fast. You can make up the difference in the months we have left. You simply need to revisit your goals, tweak your strategy, and decide what you're going to do differently. If your incremental strategy hasn't worked so far, maybe it's time to do something absurdly different so you can get absurdly different results. Even if you're not blown away by what you've accomplished in the months behind us, get excited. You can turn this year around! *** We have one more spot to join us next month in Maximum Exposure. This is a high touch group PR program designed to get you media exposure - interviews, segments and features to pontificate about what you know best. If you're ready for PR results, and want a PR team that won't be happy until your face is splashed all over tv screens and newspaper pages, apply here.
he FLYEst online radio show returns for Season 4 (bonus ep) ... You dont want to miss this weeks show playing new music and topics! Topic: "PILOT Show" Segments: 1. Studies Show 2. Live Call Ins 3. FLYE (top) Five 4. Message of the Day ft . "The PASTOR" Tune in to hear conversation on current events and the hottest underground music from Milwaukee & other cities. This week we have music from Milwaukee, Chicago, Miami, St. Louis, ect. Got Music you want heard on FLYE Radio, email us @ FlyeRadio414@gmail.com --- The live link ---> http://tobtr.com/s/10376099 Live call ins will be available from 8:15pm - 9:30pm -- toll free - (760) 454-8829 Join the group FLYE Radio Topics for more dialogue and follow us on Twitter and Instagram @FlyeRadio
Welcome to #NoWahala with @theycallmeTUNE x @justbawo! Tune into Episode 39, with special guest Poizon Ivy (The DJ)! In this episode, the group discuss a variety of topics with Global Spin Award winning DJ - Poizon Ivy, including her upbringing, her career and working for the NBA (Dallas Mavericks organization), dating, her vision for Kenya's future within the African music industry, collaborations with African artists and recent joint venture with Ghanaian producer/DJ GuiltyBeatz. More on Ivy below! ___________________ “When it’s all said and done, the only thing that should matter is how well you do the job; not your gender,” says DJ Poizon Ivy. Born Ivy Awino in the Lang’ata suburb of Nairobi, Kenya, and raised in Dallas, Texas, Ivy mastered both the piano and cello before discovering her love for yet another instrument: turntables. A bonafide sports aficionado, Poizon Ivy has managed to not only create waves for herself in the music world but also in the sports realm. She’s currently the official deejay for Skylar Diggins’ Shoot 4 The Sky Basketball Camp Tour, and the team DJ for both the Dallas Wings (WNBA) and Dallas Mavericks (NBA). With her spirit of transcending boundaries and creating history, Poizon Ivy became the first female mixer to grace the airwaves at WKKV-FM as a V100.7 Mix Squad DJ in 2013. She’s also a contributing mixer on Nairobi’s 103.5 Homeboyz Radio. A frequent guest mixer, she has also been featured on Sirius XM’s Sway In the Morning on Shade 45 and DASH Radio to name a few. As one of the most in-demand talents in the country, she has been called on by major brands, corporations, and non-profits including Atlantic Records, Red Bull, Adidas, Mountain Dew, Ashley Stewart, and the Milwaukee Brewers. Poizon Ivy does not only aspire to impact those who listen to her mix, she impacts young women in various endeavors. When she’s not honing her skills, Ivy is a devoted mother, she’s advocating for various causes, working with nonprofit organizations that enable young women pursuing careers in music, mentoring young mothers, and working with organizations that aide health issues affecting young women. ___________________ Featured Songs of The Week: IVY: Arrow Bwoy ft. Demarco - Love Doctor BAWO: Musmah ft. David Meli - Spend Time TUNE: Burna Boy ft. DJDS - Innocent Man Listen to this week's music selections on the #NoWahalaRadio playlists below! Spotify: nowahalapod.com/radioS Apple Music: nowahalapod.com/radioA _______ Also, as mentioned in the episode, feel free to submit messages and show us support here! -Thanks for Listening!- nowahalapod.com/spotify | instagram.com/NoWahalaPod | facebook.com/NoWahalaPod | twitter.com/NoWahalaPod #NoWahala [No Wahala] --- This episode is sponsored by · Anchor: The easiest way to make a podcast. https://anchor.fm/app
DJ Nu Stylez is a mixer on Milwaukee’s V100.7 FM See acast.com/privacy for privacy and opt-out information.
The nightlife was behind me, and my future was unsure. I knew working at my grandparent's construction company wasn't going to be my career. I had no idea what I was going to do in life. All I knew is that I had a fresh start and some time to get it figured out. When I got the call from Jeff Peterson at Topeka's Rock Radio V100, everything changed again for me. I was getting my shot and on a path that I dreamed of when I was a kid making little radio shows on an old reel-to-reel recorder my dad gave me. --- This episode is sponsored by · Anchor: The easiest way to make a podcast. https://anchor.fm/app Support this podcast: https://anchor.fm/everyoneneedsalittle/support
Jantzen Fugate, a serial entrepreneur who's had many successes and some epic failures that we'll get into a little bit later. He was named to the V100 in Utah, a list of the top 100 entrepreneurs. In 2015 he built his business loan brokerage to be Top 20 best companies to work for, and in 2016 they were named in the top 15 best companies to work for in the State of Utah. He's the founder of PulNoMor a patent pending dog leash, Streamlined Record Retrieval which is the first business I started at the age of 21. It's still going strong today, growing by about 15% year-over-year and it's ran by my dad and brother. He and his wife own a pretty successful downtown Thai Restaurant. And the founder of the Business Loan Broker Academy and Business Loan Broker Conference. During This Show We Discuss… Mr. Jantzen Fugate's background in the business credit and financing space Types of business financing that are working well right now The best types of financing to look at for someone who has credit issues Suggested types of financing for startup businesses The current difficulty to getting finance for a business Best types of financing for real estate investors Some of the best credit lines for business owners Loan programs that offer the best interest rates Factors that determine how much money a borrower may get Factors that determine the rates a borrower may pay Mr. Fugate's thoughts on the importance of business credit More about what Mr. Fugate is doing with Business Loan Broker Academy The type of training needed to become a business loan broker Type of ongoing training is needed Type of education and events available for business loan brokers One thing you every entrepreneur should be doing but aren't One action item everyone should do immediately after hearing this show! And much more…
Bailey Coleman is one of the only female Program Directors I've met in the radio industry, and she's a total boss in that role and out. She tells it how it is, and she has a huge heart to match her big personality. We talked about how long she's been in the biz, her ginormous family, and Sister Strut coming up this weekend in Milwaukee!
The FLYEst online radio show returns for Season 3 ep 4... You dont want to miss this weeks show playing new music and topics! Topic: "LEVELS OF LOYALTY" Segments: 1. Studies Show 2. Live Call Ins 3. FLYE (top) Five 4. Message of the Day ft . "The PASTOR" Tune in to hear conversation on current events and the hottest underground music from Milwaukee & other cities. This week we have music from Milwaukee, Chicago, Miami, St. Louis. Got Music you want heard on FLYE Radio, email us @ FlyeRadio414@gmail.com --- The live link ---> http://tobtr.com/s/10376099 Live call ins will be available from 8:15pm - 9:30pm -- toll free - (760) 454-8829 Join the group FLYE Radio Topics for more dialogue
The FLYEst online radio show returns for Season 3 ep 2... You dont want to miss this weeks show playing new music and topics! Topic: "DEFINITION OF HATING"? Segments: 1. Studies Show 2. Live Call Ins 3. FLYE (top) Five 4. Message of the Day ft . "The PASTOR" Tune in to hear conversation on current events and the hottest underground music from Milwaukee & other cities. This week we have music from Milwaukee, Chicago, Miami, St. Louis. Got Music you want heard on FLYE Radio, email us @ FlyeRadio414@gmail.com --- The live link will be posted on CoolsCorner.com Live call ins will be available from 8:15pm - 9:30pm -- toll free - (760) 454-8829 Join the group FLYE Radio Topics for more dialogue
Followup AR Kit in Apple Maps Apple’s Machine Learning “Journal” Intel Skylake, Kaby Lake Hyperthreading errata Debian Announcement Detection story Amazon EC2 G3 Instances IBM Z14 - Buzzword driven computing Nvidia V100 in the wild Nvidia DGX-1 V100 pricing Neural Network hardware for Microsoft HoloLens Google Glass for Industry Adobe Flash EOL Adobe announcement Google Chrome no longer runs Flash by default - December 2016 How does the Internet Work? What is “Webscale”? Rails can’t Scale Ruby Ruby on Rails CRUD Vint Cerf MVC - Model-View-Controller Framework Song NodeJS Express Global Interpreter Lock JRuby Elixir Phoenix web framework Getting to Facebook DNS BGP Anycast Content Delivery Networks Fastly Varnish SSL Handshake Seconds count in web performance Aftershow Japanese Sumo Robots - Youtube highlights compilation Rules
On Monday, July 31, a standing-room-only crowd at Club Garibaldi enjoyed a ton of jokes, plenty of jabs, and the deft hosting skills of FOX6 anchor Ted Perry at the 3rd annual Roast Of Milwaukee. Comedians Ryan Lowe, Greg Bach, Addie Blanchard, David Louis, Rita Landis, AJ Grill, and Carly Malison all lambasted the city (and each other), and Perry and V100.7 DJ Promise joined in on the biting-but-loving roasting action. Milwaukee Record's own Tyler Maas did a set, too, and was unfavorably compared to a sickly Dave Grohl at least once (Matt Wild took a "Sheldon from The Big Bang Theory" hit). Oh, and we raised a bunch of cash and supplies for the Milwaukee Women's Center. Static Eyes, meanwhile, blew the house down before and after the show.
Host Tyler Maas recently met with up-and-coming musician Vincent VanGreat at the Cactus Club. They were also joined by Promise, a V100.7 radio personality who highlights the best in Milwaukee-made hip-hop every weekend on his “Heat From The Street” segment. Together, the trio talked up the history of Milwaukee hip-hop, its ups and downs, struggles rappers face in this market, and who is helping to bolster the city’s rap rep.
[[Topic]] "Pure Deception" - a Trick or scheme used to get what you want; Deciet. LIVE Call in toll free #(760)-454-8829 [save it in your phone] Tune in to hear Unfiltered talk and new Music with @Martin2Cool , J-E-Double F, and Rachel Luis . LIVE call ins and intervews every show. You never know who's going to be in the building... Are you an Unsigned Artist? Do YOU want your music heard in new markets?Welcome to Thee FLYEst Online Radio Show in the woooooooorld! We are the MidWest Connect for all of your musical needs! Wait til you hear the new Interviews and Topic content. BIGGER THAN EVER!!! Send Your music and videos to FLYEradio414@gmail.com in MP3 Format. If You haven't already, please send us your Twitter Names and Follow us @FlyeRadio.
Part two of Lone Star Sports & Entertainment General Manager & Advocare V100 Texas Bowl Executive Director, David Fletcher's D1.ticker interview focuses on Fletcher's unique role and other high profile events Lone Star plays a hand in.
Today's 1.Q features Lone Start Sports & Entertainment General Manager & Advocare V100 Texas Bowl Executive Director, David Fletcher. Listen in as he describes the early season process his team executes to start planning for the bowl and what nuances go into selling title sponsorship for the event .
Feel Like You're Everything ~ FLYEDating 101 Only playing the HOTTEST music in Milwaukee!FLYE Radio show is brought to you on CoolsCorner.com. Your Midwest connect for music, entertainment and current events!{Hosts} @Martin2Cool @FLYE_Jeff @Qui_lo_lo @LoveQueenStephGo to CoolsCorner.com to see how you can get your music sent to FLYE Radio@FLYERadio
Thank you for tuning in to FLYE Radio presented by COOLSCORNER.com ~ New weekly topics and breaking music from the hottest underground artist around the world. Todays show "New Year , New Me" || FLYE Radio Show || January 13th, 2016 will air @ 6:00pm CST. You can listen from your cell phone, computer, or any smart device by clicking this page at said time. Guest Host: J-E-Double and Queen Steph. Visit CoolsCorner.com to see additional content (Videos, Music from the show, Photo's, Blogs and More). Want your music or content submitted to be heard on FLYE Radio or posted on COOLSCORNER.com? Email CoolsCorner414@gmail.com with your Bio attached to the content. Please enjoy the show! This is for YOU! -F.L.Y.E. [Feel Like You're Everything]
Tune in on New Year's Eve as we get you set for the kickoff between Boston College and Arizona! A special 45-minute pregame show will highlight the matchups between the Eagles and the Wildcats, and we'll break down the game as we get set for the opening kick from Shreveport, Louisiana. We'll be answering your questions and talking exclusively about BC football as we get to celebrate the team one more time in 2013!
© 2020-2025 Ivy Podcast Discovery LLC
