Best PyTorch Podcasts (2025)

1
Compiler collectives 16:33

1y ago16:33

16:33

Compiler collectives are a PT2 feature where by compiler instances across multiple ranks use NCCL collectives to communicate information to other instances. This is used to ensure we consistently decide if inputs or static or dynamic across all ranks. See also PR at https://github.com/pytorch/pytorch/pull/130935…

1
TORCH_TRACE and tlparse 15:28

1+ y ago15:28

15:28

TORCH_TRACE and tlparse are a structured log and log parser for PyTorch 2. It gives useful information about what code was compiled and what the intermediate build products look like.By PyTorch

1
Higher order operators 17:10

1+ y ago17:10

17:10

Higher order operators are a special form of operators in torch.ops which have relaxed input argument requirements: in particular, they can accept any form of argument, including Python callables. Their name is based off of their most common use case, which is to represent higher order functions like control flow operators. However, they are also u…

1
Inductor - Post-grad FX passes 24:07

1+ y ago24:07

24:07

The post-grad FX passes in Inductor run after AOTAutograd has functionalized and normalized the input program into separate forward/backward graphs. As such, they generally can assume that the graph in question is functionalized, except for some mutations to inputs at the end of the graph. At the end of post-grad passes, there are special passes th…

1
CUDA graph trees 20:50

1+ y ago20:50

20:50

CUDA graph trees are the internal implementation of CUDA graphs used in PT2 when you say mode="reduce-overhead". Their primary innovation is that they allow the reuse of memory across multiple CUDA graphs, as long as they form a tree structure of potential paths you can go down with the CUDA graph. This greatly reduced the memory usage of CUDA grap…

1
Min-cut partitioner 15:56

1+ y ago15:56

15:56

The min-cut partitioner makes decisions about what to save for backwards when splitting the forward and backwards graph from the joint graph traced by AOTAutograd. Crucially, it doesn't actually do a "split"; instead, it is deciding how much of the joint graph should be used for backwards. I also talk about the backward retracing problem.…

1
AOTInductor 17:30

1+ y ago17:30

17:30

AOTInductor is a feature in PyTorch that lets you export an inference model into a self-contained dynamic library, which can subsequently be loaded and used to run optimized inference. It is aimed primarily at CUDA and CPU inference applications, for situations when your model export once to be exported once while your runtime may still get continu…

1
Tensor subclasses and PT2 13:25

1+ y ago13:25

13:25

Tensor subclasses allow you to add extend PyTorch with new types of tensors without having to write any C++. They have been used to implement DTensor, FP8, Nested Jagged Tensor and Complex Tensor. Recent work by Brian Hirsh means that we can compile tensor subclasses in PT2, eliminating their overhead. The basic mechanism by which this compilation …

1
Compiled autograd 18:07

1+ y ago18:07

18:07

Compiled autograd is an extension to PT2 that permits compiling the entirety of a backward() call in PyTorch. This allows us to fuse accumulate grad nodes as well as trace through arbitrarily complicated Python backward hooks. Compiled autograd is an important part of our plans for compiled DDP/FSDP as well as for whole-graph compilation.…

1
PT2 extension points 15:54

1+ y ago15:54

15:54

We discuss some extension points for customizing PT2 behavior across Dynamo, AOTAutograd and Inductor.By PyTorch

1
Inductor - Define-by-run IR 12:06

1+ y ago12:06

12:06

Define-by-run IR is how Inductor defines the internal compute of a pointwise/reduction operation. It is characterized by a function that calls a number of functions in the 'ops' namespace, where these ops can be overridden by different handlers depending on what kind of semantic analysis you need to do. The ops Inductor supports include regular ari…

1
Unsigned integers 13:07

1+ y ago13:07

13:07

Traditionally, unsigned integer support in PyTorch was not great; we only support uint8. Recently, we added support for uint16, uint32 and uint64. Bare bones functionality works, but I'm entreating the community to help us build out the rest. In particular, for most operations, we plan to use PT2 to build anything else. But if you have an eager ker…

1
Inductor - IR 18:00

1+ y ago18:00

18:00

Inductor IR is an intermediate representation that lives between ATen FX graphs and the final Triton code generated by Inductor. It was designed to faithfully represent PyTorch semantics and accordingly models views, mutation and striding. When you write a lowering from ATen operators to Inductor IR, you get a TensorBox for each Tensor argument whi…

1
Dynamo - VariableTracker 15:55

1+ y ago15:55

15:55

I talk about VariableTracker in Dynamo. VariableTracker is Dynamo's representation of the Python. I talk about some recent changes, namely eager guards and mutable VT. I also tell you how to find the functionality you care about in VariableTracker (https://docs.google.com/document/d/1XDPNK3iNNShg07jRXDOrMk2V_i66u1hEbPltcsxE-3E/edit#heading=h.i6v7gq…

1
AI Drives the Software-Defined Heterogeneous Computing Era 8:08

2y ago8:08

8:08

By Dr. Sakyasingha Dasgupta. The rapid development of artificial intelligence (AI) applications has created enormous demand for high-performance and energy-efficient computing systems. However, traditional homogeneous architectures based on Von Neumann processors face challenges in meeting the requirements of AI workloads, which often involve massi…

1
Multimodal Generative AI on Energy-Efficient Edge Processors 6:30

2y ago6:30

6:30

By Dr. Sakyasingha Dasgupta. Edge computing will grow exponentially in the next few years, as more and more devices and applications demand low-latency, high-performance, and privacy-preserving computation at the edge of the network. However, one of the biggest challenges facing edge computing is handling the increasing complexity and diversity of …

1
Efficient Edge AI Chips with Reconfigurable Accelerators 7:56

2y ago7:56

7:56

By Nikolay Nez. DNA IP and MERA are a potent combination, reducing the hardware-specific knowledge OEMs would otherwise need to power AI applications efficiently. Much of the AI inference conversation focuses on delivering as many operations as quickly as possible. Massive GPU-based implementations can find homes in data centers with practically un…

1
再構成可能なアクセラレータを搭載した効率的なエッジAIチップ 10:30

2y ago10:30

10:30

DNA IPとMERAの強力なコンビネーションで、ユーザーがAIアプリケーションを効率的に動かすためのハードウエア固有の特別な知識習得の必要性を減らすことができます。 AI推論の話題の多くは、できるだけ早く多くの処理を提供することに焦点が当てられています。データセンターのように実質的に無制限の電力と冷却を備えた施設では、GPUベースの大きなシステムを実装することが可能です。しかし、データセンター以外の場所（広義では「エッジ」）にある組み込みシステムの制約が加わると、サイズ、重量、電力、利用率を考慮してスケーリングされた、より効率的なエッジAIチップが不可欠となります。EdgeCortix Dynamic Neural Accelerator (DNA) プロセッサアーキテクチャは、カスタムA…

1
Connecting Edge AI Software with PyTorch, TensorFlow Lite, and ONNX Models 6:31

2y ago6:31

6:31

By Antonio Nevado. Data scientists and others may have concerns moving PyTorch, TensorFlow, and ONNX models to edge AI software applications – MERA makes it easy and is model-agnostic. PyTorch, TensorFlow, and ONNX are familiar tools for many data scientists and AI software developers. These frameworks run models natively on a CPU or accelerated on…

1
What is edge AI inference doing for more devices? 7:05

2y ago7:05

7:05

By Jeffrey Grosman. AI inference is a common term - but what is edge AI inference? EdgeCortix provides an answer in terms of workloads, efficiency, and applications. Artificial intelligence (AI) is changing the rules for many applications. Teams train AI models to recognize objects or patterns, then run AI inference using those models against incom…

1
Opensource Licensing for LLMs 24:00

2+ y ago24:00

24:00

In this episode we discuss licensing and commercial use for large language models. How does open source licensing affect you as an entrepreneur or decision maker? How do you navigate the line that separates commercial vs non-commercial use? Join our Discord to participate in the discussion: https://discord.gg/94xbKyTX…

1
GPT4, AI Transforming Business, the Future of Applications 13:49

2+ y ago13:49

13:49

Should AI already be transforming your business? And what does all of this mean for the future of applications? In this episode, we discuss GPT-4 and how the emergence of an increasingly closed-source AI ecosystem can be damaging for the entire industry.By Lightning AI

1
A new wave of AI-based products and the resurgence of personal applications 35:57

2+ y ago35:57

35:57

In this episode, Luca and I talk about Sarah Guo's advice to AI Entrepreneurs, Aligning models to customer needs, Luca's predictions about the future of AI and Programing without Programming, or Automation for Everyone. Also, if you want to learn more, check out our Read Log: https://lightningai.notion.site/The-AI-Buzz-with-Luca-and-Josh-Episode-5-…

1
ChatGPT + Bing and How to start an AI company in 3 easy steps. 35:26

2+ y ago35:26

35:26

In this episode, Luca and I talk about ChatGPT + Bing, Google vs Microsoft, Artificially learning high order logic, and how to start an AI company in 3 easy steps. Also, if you want to learn more, check out some of our sources here: https://lightningai.notion.site/Readlog-21-Feb-2023-eb0f44e895ce4c81b5777e6360f1324e…

1
Unbacked SymInts 21:31

2+ y ago21:31

21:31

This podcast goes over the basics of unbacked SymInts. You might want to listen to this one before listening to https://pytorch-dev-podcast.simplecast.com/episodes/zero-one-specialization Some questions we answer (h/t from Gregory Chanan): - Are unbacked symints only for export? Because otherwise I could just break / wait for the actual size. But m…

1
Zero-one specialization 21:07

2+ y ago21:07

21:07

Mikey Dagistes joins me to ask some questions about the recent recent composability sync https://www.youtube.com/watch?v=NJV7YFbtoR4 where we discussed 0/1 specialization and its implications on export in PT2. What's the fuss all about? What do I need to understand about PT2 to understand why 0/1 specialization is a thing?…

1
Constitutional AI, Emergent Abilities and Foundation Models 33:43

2+ y ago33:43

33:43

In this episode, Luca and Josh talk about Constitutional AI, a way to get models to train themselves not to say bad words, Emergent Abilities and Foundation Models.By Lightning AI

1
Big Data, Reinforcement Learning and Aligning Models 36:54

2+ y ago36:54

36:54

In this episode, Luca and Josh chat about Big Data and The Pile, an open source resource for training models, Reinforcement Learning and Aligning Models.By Lightning AI

1
ChatGPT, Transformers and Attention 38:30

2+ y ago38:30

38:30

In the first episode of The AI Buzz with Luca and Josh we talk about ChatGPT, Transformers and Attention.By Lightning AI

1
torchdynamo 25:35

2+ y ago25:35

25:35

What is torchdynamo? From a bird's eye view, what exactly does it do? What are some important things to know about it? How does it differ from other graph capture mechanisms? For more reading, check out https://docs.google.com/document/d/13K03JN4gkbr40UMiW4nbZYtsw8NngQwrTRnL3knetGM/edit#By PyTorch

1
PyTorch 2.0 17:51

2+ y ago17:51

17:51

Soumith's keynote on PT2.0: https://youtu.be/vbtGZL7IrAw?t=1037 PT2 Manifesto: https://docs.google.com/document/d/1tlgPcR2YmC3PcQuYDPUORFmEaBPQEmo8dsh4eUjnlyI/edit# PT2 Architecture: https://docs.google.com/document/d/1wpv8D2iwGkKjWyKof9gFdTf8ISszKbq1tsMVm-3hSuU/edit#By PyTorch

1
History of functorch 19:10

3y ago19:10

19:10

Join me with Richard Zou to talk about the history of functorch. What was the thought process behind the creation of functorch? How did it get started? JAX’s API and model is fairly different from PyTorch’s, how did we validate that it would work in PyTorch? Where did functorch go after the early user studies? Where is it going next?…

1
The Story Behind PyTorch and the Community Who Maintains It, with Soumith Chintala 33:21

3y ago33:21

33:21

There’s no need to bury the lead here. Soumith Chintala is the central figure in a major transition in the world of Artificial Intelligence and Machine Learning. He works at Meta where he’s the manager of PyTorch, an open source machine learning framework that was recently transferred to the Linux Foundation. PyTorch enables ML engineers to deploy …

1
Managing an Open Source Program Office, with Ashley Wolf, GitHub 30:10

3y ago30:10

30:10

It’s a consistent pattern at most companies: High-value data and corporate memory are stored in isolated channels on disparate systems. Old processes are protected by those who have been there the longest. The problem is, the DNA of the company becomes lost as long-time employees depart, making it difficult for new hires to find what is available, …

1
Is the finance industry using open source? Yes. Yes it is!, with Gabriele Columbro 30:43

3y ago30:43

30:43

With major software vulnerabilities popping up on what seems like a weekly basis and government regulation imminent when it comes to providing a software bill of materials for any application sold to the United States government, collaboration on open source security is no longer optional. Large enterprises have come to realize that it's better to …

1
Become a Hybrid in the Open Source Community, with Ana Jiménez 25:45

3y ago25:45

25:45

“I usually say that I’m a hybrid,” Ana Jiménez says. In this context what does that even mean, what is a hybrid? According to the Oxford Language Dictionary, a hybrid is “a word formed from elements taken from different languages, for example television ( tele- from Greek, vision from Latin).” If we use that as our definition, Ana Jiménez Santamari…

1
Waiting for the SBOM to Drop, with Allan Friedman 33:11

3y ago33:11

33:11

Allan Friedman was one of the first, if not THE first person to talk with me about the need for a mandatory software bill of materials to be attached to all software back in 2017 when he was Director of Cybersecurity Initiatives for the US Department of National Telecommunications and Information Administration (NTIA). In today’s show we’ll do a de…

1
LFNetworking and Edge Computing, with Arpit Joshipura 27:00

3y ago27:00

27:00

Ahmedabad is the largest city in the state of Gujarat (goo jer raht) in western India. It has a population of over eight million people. This is where Arpit Joshipura, GM of LFNetworking at the Linux Foundation, was born and raised. The city of Ahmedabad is divided into two major sections, dissected by the Sabarmati River. The east side is what’s c…

1
The Call for Code Project, with Daniel Krook 26:14

3y ago26:14

26:14

Technology influences every aspect of our life. It's hard to remember a time when analog was separate from the digital. How do we balance the pace of innovation with its social impact when everything is changing so quickly? For Daniel Krook, these two threads converged in 1995. Dan went to Trinity College, a small liberal arts school in Hartford, C…

1
Games and Digital Media in Open Source, with Royal O'Brien 31:27

3y ago31:27

31:27

The Unreal gaming engine launched in 1998. It was a fun time. It was like, “Oh my God, we can build our own games and gaming maps!” But those earlier in the gaming cycle thought there was a better alternative already on the market: the launch of Quake in 1996. Royal O’Brien, currently GM of Digital Media and Games at the Linux Foundation was one of…

1
Building Your Open Source Strategy, with Nithya Ruff 44:10

3y ago44:10

44:10

"A lot of the companies I've worked for were trying to transition from proprietary or very siloed products, very black box products to more standards based products. And that seems to be a common thread that I've gone through even in proprietary companies. Avaya also was trying to move away from a very, very monolithic black box voice messaging sys…

1
Management and Diversity in the Open Source Community with Sara Chipps 36:37

3y ago36:37

36:37

As Sara Chipps delved deeper into the world of open source and moved up into senior engineering and management roles, there was a specific skill she learned about how to work with and manage engineers. "This is something you never get taught in school. This is something we don't stress to junior developers. This is something that senior developers …

1
A New Model for Technical Training, with Clyde Seepersad 35:04

3y ago35:04

35:04

Clyde Seepersad is the Senior Vice President & General Manager, of the Training & Certification Project at The Linux Foundation. He carries the idea with him that failure is temporary. Knowing that can help you get through some pretty intense situations. On the flip side, knowing success is temporary gives you a chance to store away some of those g…

1
Learning rate schedulers 19:35

3y ago19:35

19:35

What’s a learning rate? Why might you want to schedule it? How does the LR scheduler API in PyTorch work? What the heck is up with the formula implementation? Why is everything terrible?By PyTorch

1
Weak references 16:46

3y ago16:46

16:46

What are they good for? (Caches. Private fields.) C++ side support, how it’s implemented / release resources. Python side support, how it’s implemented. Weak ref tensor hazard due to resurrection. Downsides of weak references in C++. Scott Wolchok’s release resources optimization. Other episodes to listen to first: https://pytorch-dev-podcast.simpl…

1
Strides 20:31

3y ago20:31

20:31

Mike Ruberry has an RFC about stride-agnostic operator semantics (https://github.com/pytorch/pytorch/issues/78050), so let's talk about strides. What are they? How are they used to implement views and memory format? How do you handle them properly when writing kernels? In what sense are strides overspecified, and therefore, not worth slavishly reim…

1
The Business Side of Open Source, with Patrick Debois 30:25

3y ago30:25

30:25

The first time Patrick Debois came into contact with Open Source was in the early stages of development of the Linux kernel, compiling it on floppies on his 486 machine. To tell you how long ago that was, the Intel 486 was introduced in 1989, It was the first chip in the line to include a built-in math coprocessor. Patrick was an early adopter of c…

1
A Life in Open Source, with Brian Behlendorf 51:54

3y ago51:54

51:54

Brian Behlendorf came from a science and technology background. In fact, his parents met at IBM where his father was a Cobol programmer. During the 1980s, Brian was comfortable in front of a TRS 80 and a PC junior doing basic programming and term reports. He quickly found his way onto Usenet and participation on mailing lists around the band REM or…

1
Balancing Priorities at the CNCF, with Priyanka Sharma 29:18

3+ y ago29:18

29:18

Priyanka Sharma has had a long career in tech. After graduating from Stanford in 2009, she worked at Google in the Online Partnerships Group, was a technical consultant where she onboarded new DoubleClick clients, and acted as interim Product Manager for internal insights tools. From there she moved to OutRight, leading the promotional launch for t…

1
The Untold Stories of Open Source - An Introduction to the new Podcast from the Linux Foundation 8:49

3+ y ago8:49

8:49

Open Source is embedded in every software application you touch today. It’s impossible to build a large scale application without it. The real question is, what’s the story behind that component, application, or framework you just downloaded? Not the specs. Not the functionality. The real story: “Who wrote the code? What is their backstory? What le…

Podcasts Worth a Listen

PyTorch Podcasts

Podcasts Worth a Listen

Quick Reference Guide