Hey, everyone. Welcome to the Latent Space Podcast. This is Alessio, partner and CTO in Residence and Decibel Partners, and I'm joined by my co-host, Swiggs, founder of Small.ai. And today we're back with Jeremy Howard, I think your third appearance on Latent Space. Welcome. Wait, third? Second? - Well, I grabbed you in Europe.

- I see. - Very fun standing outside streets. - I never heard that, by the way. You've got to send me a link. I've got to hear what it sounded like. - Yeah, yeah. - I think the two episodes are six hours, so there's plenty to listen. We'll make sure to send it over.

Yeah, we're trying this thing where the major ML conferences, we, you know, do a little audio tour of the conference and give people a sense of what it's like. But the last time you were on, you declared the end of fine-tuning. I hope that... I know that, you know, I sort of editorialized the title a little bit, and I know you were slightly uncomfortable with it, but you just own it anyway.

I think you're very good at the hot takes. And we were just discussing in our pre-show that things have... It's really happening, that the continued pre-training is really happening. Yeah, absolutely. I think people are starting to understand that treating the three ULM fit steps of, like, pre-training, you know, and then the kind of, like, what people would now call "instruction tuning," and then, I don't know if we've got a general term for this, DPO, RLHFE step, you know, but, you know, the task training, they're not actually as separate as we originally suggested they were in our paper.

And when you treat it more as a continuum, and that you make sure that you have, you know, more of kind of the original data set incorporated into the later stages, and that, you know, we've also seen with, like, LLAMA3, this idea that those later stages can be done for a lot longer.

These are all of the things I was kind of trying to describe there. It wasn't, like, yeah, wasn't the end of pre-training. Sorry, it wasn't the end of fine-tuning, but more that we should treat it as a continuum, and we should have much higher expectations of how much you can do with an already trained model.

You can really add a lot of behavior to it. You can change its behavior. You can, you know, you can do a lot. So a lot of our research has been around trying to figure out how to modify the model by a larger amount rather than starting from random weights, because I get very offended at the idea of starting from random weights.

Yeah, I saw that in iClear in Vienna, there was an outstanding paper about starting transformers from data-driven pyres. I don't know if you saw that one. They called it sort of never trained from scratch, and I think it was kind of rebelling against, like, the sort of random initialization of it.

Yeah, I've, you know, that's been our kind of continuous message since we started Fast.ai, is if you're training from random weights, you better have a really good reason, you know, because it seems so unlikely to me that nobody has ever trained on data that has any similarity whatsoever to the general class of data you're working with, and that's the only situation in which I think starting from random weights makes sense.

Yeah, the other trends since our last pod that I would point people to is I'm seeing a rise in multi-phase pre-training. So Snowflake released a large model called Snowflake Arctic, where they detailed three phases of training, where they had, like, a different mixture of, like, there was, like, 75% web in the first instance, and then they reduced the percentage of the web text by 10% each time and increased the amount of code in each phase.

And I feel like multi-phase is being called out in papers more. I feel like it's always been a thing, like, changing data mix is not something new, but calling it a distinct phase is new, and I wonder if there's something that you're seeing on your end. Well, so they're getting there, right?

So the point at which they're doing proper continued pre-training is the point at which that becomes a continuum rather than a phase. So the only difference with what I was describing last time is to say, like, oh, there should, you know, there's a function or whatever which is happening every batch.

And it doesn't, like, it's not a huge difference, but it's like back, you know, I always used to get offended when people had learning rates that, like, jumped. And so one of the things I started doing early on in Fast.ai was to say to people, like, no, you should actually have, your learning rate schedule should be a function, not a list of numbers.

So now I'm trying to give the same idea about training mix. There's been pretty public work from Meta on schedule-free optimizers. I don't know if you've been following Aaron DeFazio and what he's doing. Just because you mentioned learning rate schedules, you know, what if you didn't have a schedule?

I mean, I don't care very much, honestly. Like, I don't think that schedule-free optimizer's that exciting. It's fine. We've had non-scheduled optimizers for ages, like, Les Wright, who's now at Meta, who was part of the Fast.ai community there, created something called the Ranger optimizer. You know, I actually like having more hyperparameters, you know, as soon as you say schedule-free, then, like, well, now I don't get to choose.

And there isn't really a mathematically correct way of, like, I actually try to schedule more parameters rather than less. So, like, I like scheduling my epsilon in my atom, for example. I schedule all the things. So, but then the other thing we always did with the Fast.ai library was make it so you don't have to set any schedules.

So Fast.ai always supported, like, not, you didn't even have to pass a learning rate. Like, it would always just try to have good defaults and do the right thing. But to me, I like to have more parameters I can play with if I want to, but that you don't have to.

And then the more, less technical side, I guess, of your issue, I guess, with the market was some of the large research labs taking all this innovation kind of behind closed doors and whether or not that's good, which it isn't. And now we could maybe make it more available to people.

And then after a month, a month after we released the episode, there was the whole Sam Altman drama and, like, all the OpenAI governance issues. And maybe people started to think more, okay, what happens if some of these kind of labs, you know, start to break from within, so to speak, and the alignment of the humans is probably going to fall before the alignment of the models.

So I'm curious, like, if you have any new thoughts, and maybe we can also tie in some of the way that we've been building Answer as, like, a public benefit corp and some of those aspects. Sure. So, yeah, I mean, it was kind of uncomfortable because two days before Altman got fired, I did a small public video interview in which I said I'm quite sure that OpenAI's current governance structure can't continue and that it was definitely going to fall apart.

And it fell apart two days later. And a bunch of people were like, "What did you know, Jeremy?" - What did Jeremy see? - I didn't see anything. It's just obviously true. And so, yeah, so my friend Eric Ries and I spoke a lot before that about, you know, Eric's, I think, probably most people would agree, the top expert in the world on, kind of, start-up and AI governance.

And, you know, we could both clearly see that this didn't make sense to have, like, a so-called non-profit where then there are people working at a commercial company that's owned by or controlled nominally by the non-profit where the people in the company are being given the equivalent of stock options.

Like, everybody there was working there with expecting to make money largely from their equity. So the idea that then a board could exercise control by saying, like, "Oh, we're worried about safety issues and so we're going to do something that decreases the profit of the company," when every stakeholder in the company, their remuneration pretty much is tied to their profit, it obviously couldn't work.

So, I mean, that was a huge oversight there by someone. And I guess it's, like, I guess part of the problem is that the kind of people who work at non-profits, you know, and in this case the board, you know, who are kind of academics and, you know, people who are kind of true believers, I think it's hard for them to realize that 99.999% of the world is driven very heavily by money, especially huge amounts of money.

So, yeah, Eric and I had been talking for a long time before that about, like, well, what could be done differently? Because also companies are sociopathic, like, by design. And so the alignment problem, as it relates to companies, has not been solved. Like, companies become huge, they devour their founders, they devour their communities, and they do things where even the CEOs, you know, often of big companies tell me, like, "I wish our company didn't do that thing." But, you know, I know that if I didn't do it, then I would just get fired, and the board would put in somebody else.

And the board knows if they don't do it, then their shareholders can sue them, because they're not maximizing profitability or whatever. So, what Eric's spent a lot of time doing is trying to think about, like, how do we make companies less sociopathic, you know? Or maybe a better way to think of it is, like, how do we make it so that the founders of companies can ensure that their companies continue to actually do the things they want them to do?

So, you know, when we started a company, you know, like, well, A, we very explicitly decided we're going to start a company, not a academic lab, not a non-profit, you know. We created a Delaware C Corp, you know, the most company kind of company. But when we did so, we told everybody, you know, including our first investors, which was you, Alessio.

They sound great. We are going to run this company on the basis of maximizing long-term value. You know? So, you know, in fact, so when we did our second round, which is an angel round, we had everybody invest through a long-term SPV, which we set up, where everybody had to agree to vote in line with long-term value principles.

So, like, it's not just, it's never enough just to say to people, like, okay, we're trying to create long-term value here for society as well as for ourselves, and everybody's like, oh, yeah, yeah, I totally agree with that. But when it comes to like, okay, well, here's a specific decision we have to make, which will not maximize short-term value, people suddenly change their mind.

So, you know, it has to be written into the legal documents of everybody, so that there's no question that that's the way the company has to be managed. So, then you mentioned the PBC aspect, Public Benefit Corporation, which I never quite understood previously. And it turns out it's incredibly simple.

Like, it took, you know, like one paragraph added to our corporate documents to become a PBC. It was cheap, it was easy, but it's got this huge benefit, which is, if you're not a Public Benefit Corporation, then somebody can come along and offer to buy you, with a stated description of, like, turning your company into the thing you most hate, right?

And if they offer you more than the market value of your company and you don't accept it, then you are not necessarily meeting the, kind of, your fiduciary responsibilities. So, the way, like, Eric always described it to me, you know, is like, if Philip Morris came along and said that you've got great technology for marketing cigarettes to children, so we're going to pivot your company to do that entirely, and we're going to pay you 50% more than the market value, you're going to have to say yes.

If you have a PBC, then you are more than welcome to say no, if that offer is not in line with your stated public benefit. So, our stated public benefit is to maximize, you know, the benefit to society through using AI. So, given that more children smoking doesn't do that, then we can say, like, no, we're not selling to you.

Yeah, and I was looking back at some of our emails. You sent me an email on November 13th about talking, and then on the 14th, I sent you an email working together to free AI, was the subject line. And then that was, kind of, the start of the seed round.

And then two days later, someone got fired. So, this was, like, not even, you know, you were having these thoughts even before. We had, like, a public example of, like, why some of the current structures didn't work. So, yeah, you were very ahead of the curve, so to speak.

I would love just to, you know, people can read your awesome introduction, blog, and answer, and the idea of having an R&D lab versus our lab, and then a D-Lab somewhere else. I think, to me, the most interesting thing has been hiring, and some of the awesome people that you've been bringing on that maybe don't fit the central casting of Silicon Valley, so to speak.

Like, sometimes I go there, like, playing baseball cards, you know. People are like, oh, what teams was this person on? Where did they work? Versus focusing on ability. So, I would love for you to give a shout out to some of the awesome folks on the team. So, you know, there's, like, a graphic going around describing, like, the people at XAI, you know, the Elon Musk thing, and, like, they're all connected to, like, you know, multiple of Stanford, Meta, DeepMind, OpenAI, Berkeley, Oxford.

It's just, look, these are all great institutions, and they have good people, and I'm definitely not at all against that, but, damn, there's so many other people. And one of the things I found really interesting is, kind of, anytime I, almost anytime I see something which I think, like, this is really high quality work, and it's, like, something I don't think would have been built if that person hadn't built the thing right now, I nearly always reach out to them and ask to chat.

And I tend to dig in to find out, like, okay, you know, why did you do that thing? Everybody else has done this other thing. Your thing's much better, but it's not what other people are working on. And, like, 80% of the time, I find out the person has a really unusual background.

So, like, often they'll have, like, either they, like, came from poverty, and, like, didn't get an opportunity to go to good school, or they, like, you know, had dyslexia and, you know, got kicked out of school in year 11, or, you know, or they had a health issue that meant they couldn't go to university, or something happened in their past, and they ended up out of the mainstream, and then they, kind of, succeeded anyway.

And those are the people that, throughout my career, I've tended to, kind of, accidentally hire more of. But, like, it's not exactly accidentally. It's, like, when I see somebody who's done, two people who have done extremely well. One of them did extremely well in exactly the normal way, from the background, entirely pointing in that direction, and they achieved all the hurdles to get there.

And, like, okay, that's quite impressive, you know. But another person who did just as well, despite lots of constraints, and doing things in really unusual ways, and came up with different approaches, like, that's normally the person I'm likely to find useful to work with, because they're often, like, risk-takers, they're often creative, they're often extremely tenacious, they're often very open-minded.

So, that's the kind of folks we, you know, I tend to find myself hiring. And I think, like, so now at Answer.ai, it's a group of people that are strong enough that nearly every one of them has independently come to me in the past few weeks and said, and told me that they have imposter syndrome, and they're not convinced that they're good enough to be here, you know.

And I kind of heard it at the point where I was like, okay, I don't think it's possible that all of you are so far behind your peers that you shouldn't get to be here. But I think part of the problem is, like, as an R&D lab, the great developers look at the great researchers and they're like, wow, these big-brained, crazy research people with all their math and shit, they're too cool for me, oh my god.

And then the researchers look at the developers and they're like, oh, they're killing it, making all this stuff with all these people using it, and talking on Twitter about how great it is. And I think they're both a bit intimidated by each other, you know. And so I have to kind of remind them, like, okay, there are lots of things in this world where you suck compared to lots of other people in this company, but also vice versa, you know, for all things.

And the reason you came here is because you wanted to learn about those other things from those other people and have an opportunity to, like, bring them all together into a single unit. So, you know, it's not reasonable to expect you're going to be better at everything than everybody else.

Even though, like, I guess the other part of it is for nearly all of the people in the company, to be honest, they have nearly always been better than everybody else at nearly everything they're doing, nearly everywhere they've been. So it's kind of weird to be in this situation now where it's like, gee, I can clearly see that I suck at this thing that I'm meant to be able to do compared to these other people, where I'm like the worst in the company at this thing for some things.

So I think that's a healthy place to be, you know, as long as you keep reminding each other about that's actually why we're here. And it's been really nice to see, like, it's all a bit of an experiment, like, we don't have any managers. We don't have any hierarchy from that point of view.

So, for example, I'm not a manager, which means I don't get to tell people what to do or how to do it or when to do it. And it's been a bit of an experiment to see how that would work out. And it's been great, like, so, for instance, Ben Clavier, who you might have come across, he's the author of Ragatouille.

He's the author of Rerankers, super strong information retrieval guy. And a few weeks ago, he, you know, this additional channel appeared on Discord, on our private Discord called Bert24. Like, these people started appearing, as in our collab sections. We have a collab section for, like, collaborating with outsiders. And these people started appearing.

There are all these names that I recognize, like Bert24. And they're all talking about, like, the next generation of Bert. And I start following along. It's like, okay, Ben decided that I think, quite rightly, we need a new Bert. Because everybody, like, so many people are still using Bert.

And it's still the best at so many things. But it actually doesn't take advantage of lots of best practices. And so, he just went out and found basically everybody who's created better Berts in the last four or five years, brought them all together. Suddenly, there's this huge collaboration going on.

So, yeah, I didn't tell him to do that. He didn't ask my permission to do that. And then, like, Benjamin Warner dived in. And he's like, oh, I created a whole Transformers from scratch implementation designed to be maximally hackable. He originally did it largely as a teaching exercise to show other people.

But he was like, I could, you know, use that to create a really hackable Bert implementation. In fact, he didn't say that. He said, I just did do that. You know, and I created a repo. And then everybody's like, starts using it. They're like, oh, my God, this is amazing.

I can now implement all these other Bert things, you know. And it's not just answer AI guys. There, you know, there's lots of folks, you know, who have, like, contributed new data set mixes and blah, blah, blah. So, I mean, I can help in the same way that other people can help.

So, like, then Ben Clavier reached out to me at one point and said, like, okay, can you help me, like, what have you learned over time about how to manage, you know, intimidatingly capable and large groups of people who you're nominally meant to be leading? And so, you know, like, I try to help, but I don't direct.

Another great example was Kerim, who, after our FSTP QLORA work, decided quite correctly that it didn't really make sense to use LORA in today's world. You want to use the normalized version, which is called DORA. And like, two or three weeks after we did FSTP QLORA, he just popped up and said, okay, I've just converted the whole thing to DORA, and I've also created these VLLM extensions, and I've got all these benchmarks, and, you know, now I've got training of quantized models with adapters that are as fast as LORA and as, actually, better than, weirdly, fine-tuning.

I was just like, okay, that's great, you know? And yeah, so, the things we've done to try to help make these things happen as well is like, we have, so we don't have any required meetings, you know, but we do have a meeting for each pair of major time zones that everybody's invited to, and, you know, people see their colleagues doing stuff that looks really cool, and say like, oh, how can I help, you know, or how can I learn, or whatever.

So another example is Austin, who, you know, amazing background, he ran AI at Fidelity, he ran AI at Pfizer, he ran browsing and retrieval for Google's DeepMind stuff, created Gemma.cpp, and he's been working on a new system to make it easier to do WebGPU programming, because again, he quite correctly identified, like, you know, this is a way that not everybody has to use CUDA, not everybody has to use NVIDIA, you can do stuff on your own computer, optionally through the browser, we need to make this easier to do.

And so I, yeah, so I said to him, like, okay, I want to learn about that, not an area that I have much expertise in, so, you know, he's going to show me what he's working on and teach me a bit about it, and hopefully I can help contribute.

I think one of the key things that's happened in all of these is everybody understands what Eric Gilliam, who wrote the second blog post in our series, the R&D historian, describes as everybody has total flexibility to do what they want, but we all understand, like, kind of roughly why we're here, you know, we all have the same, you know, we agree with the premises around, like, you know, everything's too expensive, everything's too complicated, you know, people are building too many vanity foundation models rather than taking better advantage of fine-tuning, like, there's this kind of general, like, sense of, like, we're all on the same wavelength about, you know, all the ways in which current research is fucked up and, you know, all the ways in which, you know, we kind of try, you know, worried about centralization and we, you know, we all care a lot about not just research for the point of citations, but research that actually wouldn't have happened otherwise and actually is going to lead to real-world outcomes and so, yeah, with this kind of like shared vision, people understand, like, you know, so when I say, like, oh, well, you know, tell me, Ben, about BERT 24, what's that about, and he's like, you know, like, oh, well, you know, you can see from an accessibility point of view or you can see from a kind of a actual practical impact point of view, there's far too much focus on decoder-only models and, you know, like, BERT's used in all of these different places and industry and so I can see, like, in terms of our basic principles, what we're trying to achieve, this seems like something important and so I think that's, like, a really helpful that we have that kind of shared perspective, you know.

Yeah, and before we maybe talk about some of the specific research, when you're, like, reaching out to people, interviewing them, what are some of the traits, like, how do these things come out, you know, usually? Is it working on side projects that, you know, you're already familiar with? Is there anything, like, in the interview process that, like, helps you screen for people that are more, less pragmatic and more research-driven versus some of these folks that are, like, are just going to do it, you know, they're not waiting for, like, the perfect process?

Anybody who comes through the recruiting is interviewed by everybody in the company. You know, our goal is 12 people, so it's not an unreasonable amount and, like, the way I, so the other thing to say is everybody so far who's come into the recruiting pipeline, everybody bar one, has been hired, so, which is to say our original curation has been good.

And that's actually pretty easy because nearly everybody who's come in through the recruiting pipeline are people I know pretty well, so, you know, Jono Whittaker and I, you know, he worked on the stable diffusion course we did, he's outrageously creative and talented and he's super, like, enthusiastic tinkerer, just likes making things and, you know, Benjamin was one of the strongest parts of the fast.ai community, which is now the alumni, it's like hundreds of thousands of people and, you know, again, like, they're not people who a normal interview process would pick up, right?

So Benjamin doesn't have any qualifications in math or computer science, Jono was living in Zimbabwe, he was not, you know, he was working on, like, helping some African startups, you know, but not FANG kind of credentials, but yeah, I mean, when you actually see people doing real work and they stand out above, you know, we've got lots of Stanford graduates and OpenAI people and whatever in our alumni community as well, you know, when you stand out above all of those people, anyway, obviously you've got something going for you, you know, him and I worked together on the masks study we did in the proceeding at the National Academy of Science.

So, you know, we had worked together and, again, that was a group of, like, basically the 18 or 19 top experts in the world on public health and epidemiology and research design and so forth, and Austin was, you know, one of the strongest people in that collaboration. So yeah, you know, like, I've been lucky enough to have had opportunities to work with some people who are great and, you know, I'm a very open-minded person, so I kind of am always happy to try working with pretty much anybody and some people stand out.

You know, there have been some exceptions, people I haven't previously known, like Ben Clavier actually I didn't know before, but, you know, with him, like, I just read his code and I'm like, oh, that's really well-written code, like I, and like it's not written exactly the same way as everybody else's code, and it's not written to do exactly the same thing as everybody else's code.

So yeah, and then when I chatted to him, it's just like, I don't know, I felt like we'd known each other for years, like we just were on the same wavelength, and, but I could pretty much tell that was going to happen just by reading his code. I think you express a lot in the code you choose to write and how you choose to write it, I guess, you know, or another example, this guy named Vic, who was previously the CEO of DataQuest, and like, in that case, like, he's, you know, he's created a really successful startup, he's like, he won the first, basically, Kaggle NLP competition, which was automatic essay grading.

He's got the current state-of-the-art OCR system, Syria, again, he's just a guy who obviously just builds stuff, you know, he doesn't ask for permission, he doesn't need any, like, external resources. Actually, Karim's another great example of this, I mean, I already knew Karim very well because he was my best ever master's student, but it wasn't a surprise to me, then, when he then went off to create the world's state-of-the-art language model in Turkish on his own, in his spare time, with no budget, you know, from scratch, this is not fine-tuning or whatever, he like, went back to Common Crawl and did everything, so, yeah, it's kind of, I don't know what I'd describe that process as, but it's not at all based on credentials.

Assemble based on talent, yeah. We wanted to dive in a little bit more on, you know, turning from the people side of things into the technical bets that you're making. Also a little bit more on Bert, I was actually, we just did an interview with Yitay from Rekka, I don't know if you're familiar with his work, but also another encoder-decoder bet, and one of his arguments was actually people kind of over-index on the decoder-only GPT-3 type paradigm, I wonder if you have thoughts there that is maybe non-consensus as well.

Yeah, no, absolutely, so I think it's a great example, so one of the people we're collaborating with a little bit with Bert24 is Colin Raffle, who is the guy behind, yeah, most of that stuff. You know, between that and UL2, there's a lot of really interesting work, and so one of the things I've been encouraging the Bert group to do, and Colin has as well, is to consider using a T5 pre-trained encoder backbone as a thing you fine-tune, which I think would be really cool.

But he was saying, you know, Colin was also saying actually just use encoder-decoder as your Bert, you know, why don't you use that as a baseline, which I also think is a good idea. Yeah, look, you know, what technical arguments are, you know, are people underweighting? I mean, Colin would be able to describe this much better than I can, but I'll give my slightly non-expert attempt.

Look, I mean, think about like diffusion models, right, like in stable diffusion, like we use things like UNet, we, you know, you have this kind of downward path and then in the upward path you have the cross connections, which you, it's not a tension, but it's like a similar idea, right?

You're inputting the original encoding path into your decoding path. It's critical to make it work, right, because otherwise in the decoding part, the model has to like do so much kind of from scratch, right? So like if you're doing translation, like that's a classic kind of encoder-decoder example. If it's decoder only, you never get the opportunity to find the right, you know, feature engineering, that feature encoding for the original sentence.

And it kind of means then on every token that you generate, you have to recreate the whole, the whole thing, you know. So if you have an encoder, it's basically saying like, okay, this is your opportunity model to create a really useful feature representation for your, for your input information.

So I think there's really strong arguments for encoder-decoder models anywhere that there is this kind of like context or source thing, you know. And then why encoder only, well because like so much of the time what we actually care about is like, you know, a classification. You know, it's like an output.

It's like we're not generating an arbitrary length sequence of tokens. So anytime you're not generating an arbitrary length sequence of tokens, decoder models don't seem to make much sense to me. Now the interesting thing is, you see on like Kaggle competitions, that decoder models still are at least competitive with things like Deberta v3.

But they have to be way bigger to be competitive with things like Deberta v3, and the only reason they are competitive is because people have put a lot more time and money and effort into training the decoder only once, you know. There isn't a recent Deberta, there isn't a recent Bert.

So yeah, it's a whole part of the world that people have slept on a little bit, and this is just what happens. This is how trends happen, rather than like, to me everybody should be like, oh let's look at the thing that has shown signs of being useful in the past but nobody really followed up with properly.

That's the more interesting path, you know, but people tend to be like, oh I need to get citations. So what's everybody else doing? Can I make it 0.1% better, you know, or 0.1% faster? That's what everybody tends to do. Yeah, so I think it's like, ETA's work commercially now is interesting because here's like a whole, here's a whole model that's been trained in a different way, so there's probably a whole lot of tasks it's probably better at than, you know, GPT and Gemini and Claude.

So that should be a good commercial opportunity for them if they can figure out what those tasks are. Well, if rumors are to be believed, and he didn't comment on this, but, you know, Snowflake may figure out the commercialization for them, so we'll see. Let's talk about FSDP, Qlora, Qdora and all of that awesome stuff.

One of the things we talked about last time, some of these models are meant to run on systems that nobody can really own, no single person. And then you were like, well, what if you could fine tune a 70B model on like a 4090? And I was like, no, that sounds great, Jeremy, but like, can we actually do it?

And then obviously, you all figured it out. Can you maybe tell us some of the worst stories behind that, like the idea behind FSDP, which is kind of taking, you know, sharped data parallel computation, then Qlora, which is do not touch all the weights, just go quantize some of the model, and then within the quantized model only do certain layers, instead of doing everything.

Well, to the adapters. Yeah, yeah. To the adapters. Yeah, I will leave the floor to you. I think before you published it, nobody thought this was like a short term thing that we're just going to have. And now it's like, oh, obviously you can do it, but it's not that easy.

Yeah. I mean, to be honest, it was extremely unpleasant work to do. This is like, not at all enjoyable. So I kind of did version 0.1 of it myself before we had launched the company, or at least the kind of like the pieces, which is, they're all pieces that are difficult to work with, right?

So for the quantization, you know, I chatted to Tim Detmers quite a bit, and, you know, he very much encouraged me by saying like, yeah, it's possible. He actually thought it'd be easy, it probably would be easy for him, but I'm not Tim Detmers. You know, so he wrote Bits and Bytes, which is his quantization library, and, you know, he wrote that for a paper.

He didn't write that to be production like code. It's now like he's using it. He wrote it in one night, apparently. Yeah. Yeah. So, you know, like, it's not particularly well structured. There's lots of code paths that never get used. There's lots of, you know, multiple versions of the same thing.

You have to try to figure it out. So trying to get my head around that was hard, and, you know, because it's like, the interesting bits are all written in CUDA, it's hard to like to step through it and see what's happening. And then, you know, FSTP is this very complicated library in PyTorch, which not particularly well documented.

So the only really way to understand it properly is, again, just read the code and step through the code. And then, like, Bits and Bytes doesn't really work in practice unless it's used with PEFT, the Hugging Face library, and PEFT doesn't really work in practice unless you use it with other things.

And there's a lot of coupling in the Hugging Face ecosystem where, like, none of it works separately. They all work together, which I don't love. So yeah, trying to just get a minimal example that I can play with was really hard. And so I ended up having to rewrite a lot of it myself, to kind of create this minimal script.

One thing that helped a lot was Medec had this Llama Recipes repo that came out just a little bit before I started working on that. And like, they had a kind of role model example of, like, here's how to train FSDP Laura. Didn't work with QLaura on Llama. Actually, a lot of that had been put together, like, a lot of the stuff I discovered, the interesting stuff, had been put together by Les Wright, who's, he was actually the guy in the Fast.ai community I mentioned who created the Ranger Optimizer.

So he's doing a lot of great stuff at Meta now. So yeah, I kind of, that helped get some minimum stuff going, and then it was great once Benjamin and Jono joined full-time. And so we basically hacked at that together, and then Kerim joined, like, a month later or something.

But gee, it was just a lot of, like, fiddly detailed engineering on, like, barely documented bits of obscure internals. So my focus was to see if it kind of could work, and I kind of got a bit of a proof of concept working, and then the rest of the guys actually did all the work to make it work properly.

And you know, every time we thought we had something, you know, we needed to have good benchmarks, right? So we'd, like, it's very easy to convince yourself you've done the work when you haven't, you know, so then we'd actually try lots of things and be like, oh, in these, like, really important cases, the memory uses higher, you know, or it's actually slower.

And we'd go in and we'd just find, like, all these things that were nothing to do with our library that just didn't work properly. And nobody had noticed they hadn't worked properly because nobody had really benchmarked it properly. So we ended up, you know, trying to fix a whole lot of different things.

And even as we did so, new regressions were appearing in, like, Transformers and stuff that Benjamin then had to go away and figure out, like, oh, how come FlashAttention doesn't work in this version of Transformers anymore with this set of models, and, like, oh, it turns out they accidentally changed this thing so it doesn't work.

You know, there's just, there's not a lot of really good performance type evals going on in the Open Source ecosystem. So there's an extraordinary amount of, like, things where people say, like, oh, we built this thing and it has this result, and when you actually check it, it doesn't.

So yeah, there's a shitload of war stories from getting that thing to work. And it did require a particularly, like, tenacious group of people and a group of people who don't mind doing a whole lot of, kind of, like, really janitorial work, to be honest, to get the details right, to check them.

Yeah. Yeah, we had the tree DAO on the podcast, and we talked about how a lot of it is, like, systems work to make some of these things work. It's not just, like, beautiful pure math that you do on a blackboard. It's, like, how do you get into the nitty-gritty of it.

I mean, FlashAttention is a great example of that. Like, it's, it basically is just, like, oh, let's just take the attention and just do the tailed version of it, which sounds simple enough, you know. But then implementing that is challenging at lots of levels. Yeah. What about inference? You know, obviously, you've done all this amazing work on fine-tuning.

Do you have any research you've been doing on the inference side, how to make local inference really fast on these models, too? We're doing quite a bit on that at the moment. We haven't released too much there yet, but one of the things I've been trying to do is also just to help other people.

And one of the nice things that's happened is that a couple of folks at Meta, including Mark Seraphim, have done a nice job of creating this CUDA mode community of people working on, like, CUDA kernels or learning about that, and I tried to help get that going well as well and did some lessons to help people get into it.

So there's a lot going on in both inference and fine-tuning performance and a lot of it's actually happening kind of related to that. Also the PyTorch team have created this Torch AO project on quantization. And so there's a big overlap now between kind of the FastAI and AnswerAI and CUDA mode communities of people working on stuff about inference and fine-tuning, but we're getting close now.

You know, our goal is that nobody should be merging models, nobody should be downloading merged models, everybody should be using basically quantized plus adapters for almost everything, and just downloading the adapters, and that should be much faster. So that's kind of the place we're trying to get to. It's difficult, you know, because, like, Kerem's been doing a lot of work with VLM, for example.

These inference engines are pretty complex bits of code. They have a whole lot of custom kernel stuff going on as well, as do the quantization libraries. So we've been working on that with also quite a bit of collaborating with the folks who do HQQ, which is a really great quantization library and works super well.

So yeah, there's a lot of other people outside AnswerAI that we're working with a lot who are really helping on all this performance optimization stuff, open source. Just to follow up on merging models, I picked up there that you said nobody should be merging models. I think that's interesting because, you know, obviously a lot of people are experimenting with this and finding interesting results.

I would say, in defense of merging models, you can do it without data. That's probably the only thing that's going for it. To explain, it's not that you shouldn't merge models, it's that you shouldn't be distributing a merged model. You should distribute a merged adapter, 99% of the time, and actually often one of the best things happening in the model merging world is actually that often merging adapters works better.

The point is, Sean, that once you've got your new model, if you distribute it as an adapter that sits on top of a quantized model that somebody's already downloaded, then it's a much smaller download for them, and also the inference should be much faster, because you're not having to transfer FB16 weights from FB, from HBM memory at all, or ever load them off disk, you know, all the main weights are quantized, and the only floating point weights are in the adapters, so that should make both inference and fine-tuning faster.

Got it, got it, okay, perfect. We're moving on a little bit to the rest of the Fast universe. I would have thought that, you know, once you started Answer.ai, that the sort of Fast universe would be kind of on hold, and then today you just dropped FastLight, and it looks like, you know, there's more activity going on in sort of FastLand.

Yeah, so FastLand and AnswerLand are not really distinct things, AnswerLand is kind of like the FastLand grown up and funded, they both have the same mission, which is to maximize the societal benefit of AI broadly. We want to create thousands of commercially successful products at Answer.ai, and we want to do that with like 12 people, so that means we need a pretty efficient stack, you know, like quite a few orders of magnitude more efficient, not just for creation, but for deployment and maintenance than anything that currently exists.

People often forget about the 'D' part of our R&D firm, so we've got to be extremely good at, you know, creating, deploying, and maintaining applications, not just models. Much to my, you know, horror, the story around creating web applications is much worse now than it was 10 or 15 years ago, in terms of like, if I say to a data scientist, here's how to create and deploy a web application, you know, either you have to learn JavaScript or TypeScript, and about all the complex, like, libraries like React and stuff, and all the complex, like, details around security and web protocol stuff, around how you then talk to a back-end, and then all the details about creating the back-end.

You know, if that's your job, you know, and you're, you know, you have specialists who work in just one of those areas, it is possible to, for that to all work, but compared to like, oh, write a PHP script and put it in the home directory that you get when you sign up to this shell provider, which is what it was like in the 90s, you know, here are those 25 lines of code, you're done, and now you can pass that URL around to all your friends, you know, or put this, you know, .pl file inside the CGI bin directory that you got when you signed up to this web host.

So yeah, the thing I've been mainly working on the last few weeks is fixing all that, and I think I fixed it. I'll go to this thing called fastHTML. I don't know if this is an announcement, but I can tell you guys. So yeah, there's this thing called fastHTML, which basically lets you create a complete web application in a single Python file.

Unlike excellent projects like Streamlit and Gradio, you're not working on top of a highly abstracted thing that's got nothing to do with web foundations, you're working with web foundations directly, but you're able to do it by using pure Python. There's no template, there's no ginger, there's no separate like CSS and JavaScript files.

It looks and behaves like a modern SPA web application. And you can create components for like Daisy UI, or Bootstrap, or Shoelace, or whatever fancy JavaScript and/or CSS, Tailwind, etc. library you like, but you can write it all in Python. You can pip install somebody else's set of components and use them entirely from Python.

You can develop and prototype it all in a Jupyter Notebook if you want to. It all displays correctly, so you can like interactively do that. And then you mentioned Fastlight, so specifically now if you're using SQLite in particular, it's like ridiculously easy to have that persistence, you know, and you can basically, all of your handlers will be passed database-ready objects automatically that you can just call .delete, .update, .insert on.

Yeah, you get session, you get security, you get all that. So it's, again, like with most of everything I do, it's very little code. It's mainly tying together really cool stuff that other people have written, so. You don't have to use it, but a lot of the best stuff comes from its incorporation of HTMX, which to me is basically the thing that changes your browser to make it work the way it always should have.

So it's a, it just does four small things, but those four small things are the things that are basically unnecessary constraints that HTML should never have had. So it removes the constraints. It sits on top of Starlet, which is a very nice, you know, kind of lower level platform for building these kind of web applications.

The actual interface matches as closely as possible to FastAPI, which is a really nice system for creating the kind of classic JavaScript type applications. And Sebastian, who wrote FastAPI, has been kind enough to help me think through some of these design decisions and so forth. I mean, everybody involved has been super helpful.

Actually, I chatted to Carson, who created HTMX, you know, also about it, chatted to some of the folks involved in Django. Like, everybody in the community I've spoken to definitely realizes there's a big gap to be filled around, like, highly scalable web foundation based, you know, pure Python framework with a minimum of fuss.

So yeah, I'm getting a lot of support and trying to make sure that FastHTML works well for people. Yeah, I would say, when I heard about this, I just texted Alexio, I think this is going to be pretty huge. You know, like, people consider Streamlit and Gradio to be the state of the art, but I think there's so much to improve in, you know, having sort of, what do you say, what do you call web foundations and web fundamentals at the core of it, I think would be really helpful.

Yeah, it's based on 25 years of thinking and work for me. So like, FastML was built on a system much like this one, but that was of hell. And so I spent, you know, 10 years working on that. We had millions of people using that every day, really pushing it hard.

And I really always enjoyed working in that. So you know, and obviously lots of other people have done, like, great stuff and particularly HTMX, you know. So I've been thinking about like, yeah, how do I pull together the best of the web framework I created for FastML with HTMX?

There's also things like Pico CSS, which is the CSS system, which by default, FastHTML comes with. Although as I say, you can pip install anything you want to, but it makes it like, super easy to, you know, so we're trying to make it so that just out of the box, you don't have any choices to make, you know, if you don't want to.

You can make choices, but for most people, you just, you know, it's like the PHP in your home directory thing. You just start typing and just by default, you'll get something which looks and feels, you know, pretty okay. And if you want to then write a version of Gradio or Streamlit on top of that, you totally can.

And then the nice thing is if you then write it in kind of the Gradio equivalent, which will be, you know, I mentioned we'll create some kind of pip installable thing for that. Once you've outgrown, or if you outgrow that, it's not like, okay, throw that all away and start again in this like whole separate language, but it's like this kind of smooth, gentle path that you can take step-by-step because it's all just standard web foundations all the way, you know.

Yeah. Got it. So, you know, just to wrap up the sort of open source work that you're doing, you know, you're aiming to create thousands of projects with a very, very small team. And I haven't heard you mention once AI agents or AI developer tooling or AI code maintenance, you know, I know you're very productive, but you know, what is the role of AI in your own work?

So I'm making something. I'm not sure how much I want to say just yet. Okay. Give us a nibble. All right, I'll give you the key thing. So I've created a new approach. It's not called prompt engineering. It's called dialogue engineering. And I'm creating a system for doing dialogue engineering.

It's currently called AI magic. I'm doing most of my work in this system and it's making me much more productive than I was before I used it. So I always just build stuff for myself and hope that it'll be useful for somebody else. Think about chatGPT with Code Interpreter, right?

The basic UX is the same as a 1970s teletype, right? So if you wrote APL on a teletype in the 1970s, you typed onto a thing, your words appeared at the bottom of a sheet of paper and you'd like hit enter and it would scroll up. And then the answer from APL would be printed out, scroll up, and then you would type the next thing, which is also the way, for example, a shell works, like bash or ZSH or whatever.

It's not terrible, you know, like we all get a lot done in these like very, very basic teletype style REPL environments, but I've never felt like it's optimal, you know, and to me, you know, so, and everybody else has just copied chatGPT. So it's also the way BART and Gemini work.

It's also the way the Claude web app works. And then you add Code Interpreter and the most you can do is to like plead with chatGPT to write the kind of code I want. It's pretty good for very, very, very beginner users who like can't code at all, like by default now the code's even hidden away, so you never even have to see it ever happened.

But for somebody who's like wanting to learn to code or who already knows a bit of code or whatever, it's, it seems really not ideal. So okay, that's one end of the spectrum. The other end of the spectrum, which is where Sean's work comes in, is, oh, you want to do more than chatGPT, no worries.

Here is Visual Studio Code. I run it. There's an empty screen with a flashing cursor. Okay, start coding, you know. And it's like, okay, you can use systems like Sean's or like Cursor or whatever to be like, okay, Apple K in cursors, like create a form that blah, blah, blah, but it's, in the end, it's like a convenience over the top of this incredibly complicated system that full-time sophisticated software engineers have designed over the past few decades in a totally different environment as a way to build software, you know.

And so we're trying to like shoehorn in AI into that. And it's, it's not easy to do, and I think there are like much better ways of thinking about the craft of software development in a language model world to be much more interactive, you know. So the thing that I'm building is, is neither of those things.

It's something between the two. And it's built around this idea of crafting a dialogue, you know, where the outcome of the dialogue is, you know, the artifacts that you want, whether it be a piece of analysis or whether it be a Python library or whether it be a technical blog post or whatever.

So as part of building that, I've created something called Claudette, which is a library for Claude. I've created something called Cosette, which is a library for OpenAI. There are libraries which are designed to make those APIs much more usable, much easier to use, much more concise. And then I've written AI magic on top of those.

And that's been an interesting exercise because I did Claudette first, and rather than try to like, I was looking at what Simon Willison did with his fantastic LLM library, and his library is designed around like, let's make something that supports all the LLM inference engines and commercial providers. I thought, okay, what if I did something different, which is like make something that says Claude friendly as possible and forget everything else.

So that's what Claudette was. So for example, one of the really nice things in Claude is pre-fill. So by telling the assistant that this is what your response started with, there's a lot of powerful things you can take advantage of. So yeah, I created Claudette to be as Claude friendly as possible.

And then after I did that, and then with Claude, particularly with GPT 4.0 coming out, I kind of thought, okay, now let's create something that's as open AI friendly as possible. And then I tried to look to see, well, where are the similarities and where are the differences? And now can I make them compatible in places where it makes sense for them to be compatible without losing out on the things that make each one special for what they are.

So yeah, those are some of the things I've been working on in that space. And I'm thinking we might launch AI magic via a course called how to solve it with code. The name is based on the classic Polya book, if you know how to solve it, which is, you know, one of the classic math books of all time, where we're basically going to try to show people how to solve challenging problems that they didn't think they could solve without doing a full computer science course, by taking advantage of a bit of AI and a bit of, like, practical skills.

And it's particularly for this, like, whole generation of people who are learning to code with and because of ChatGPT. Like, I know a lot of people who didn't really know how to code, but they've created things because they use ChatGPT, but they don't really know how to maintain them or fix them or add things to them that ChatGPT can't do, because they don't really know how to code.

So this course will be designed to show you how you can, like, you know, either become a developer who can, like, supercharge their capabilities by using language models, or become a language model first developer who can supercharge their capabilities by understanding a bit about process and fundamentals, so, yeah.

Nice. That's a great spoiler, you know. I guess the fourth time you're going to be on Learning Space, we're going to talk about AI magic. Jeremy, before we wrap, this was just a great run through everything. What are the things that when you next come on the podcast in nine, 12 months, we're going to be like, "Man, Jeremy was, like, really ahead of it." Like, is there anything that you see in this space that maybe people are not talking enough?

You know, what's the next company that's going to fall, like, in drama internally? Anything? You know, hopefully we'll be talking a lot about fast HTML and hopefully the international community that at that point has come up around that, and also about AI magic and about dialogue engineering. Hopefully dialogue engineering catches on, because I think it's the right way to think about a lot of this stuff.

What else? I'm just trying to think about more on the research side. Yeah, I think, you know, I mean, we've talked about a lot of it. Like, I think encoder-decoder architectures, encoder-only architectures, hopefully we'll be talking about, like, the whole re-interest in BERT that BERT 24 stimulated. There's a state-space model that came out today that might be interesting for just general discussion.

One thing that stood out to me with Cartesia's blog post was that they were talking about real-time ingestion of billions and trillions of tokens, and keeping that context, obviously, in the state-space that they have. I'm wondering what your thoughts are, because you've been entirely transformers the whole time. Yeah, no, so obviously my background is RNNs and LSTMs, and I'm still a believer in the idea that state is something you can update, you know.

So obviously Sepp Hochreiter came up, came out with XLSTM recently. Oh my god, okay, another whole thing we haven't talked about, just somewhat related. I've been going crazy for, like, a long time about, like, why can I not pay anybody to save my KV cache, you know, for, like, I just ingested the Great Gatsby or the documentation for Starlet or whatever, you know, I'm sending it as my prompt context.

Why are you redoing it every time, you know? So Gemini is about to finally come out with KV caching, and this is something that Austin actually in Gemma.cpp had had on his roadmap for years, well, not years, months, long time, is that the idea that the KV cache is, like, a thing that, like, it's a third thing, right?

So there's RAG, you know, there's in-context learning, you know, and prompt engineering, and there's KV cache creation. I think it creates, like, a whole new class, almost, of applications or of techniques where, you know, for me, for example, I very often, like, I very often work with, like, really new libraries, or I've created my own library that I'm now writing with, rather than on.

So I want all the docs in my new library to be there all the time. So yeah, I want to upload them once, and then all of, have a whole discussion about building this application using FastHTML, well, nobody's got FastHTML in their, in their language model yet. I don't want to send all the FastHTML docs across every time.

So one of the things I'm looking at doing in AI Magic, actually, is taking advantage of some of these ideas, so that you can have the documentation of the libraries you're working on be kind of always available. So there'll be ways to, you know, something over the next 12 months people will be spending time thinking about is how to, like, where to use RAG, where to use fine-tuning, where to use KV cache storage, you know, and, and how to use state, because in state models and XLSTM, again, state is something you, you update.

So how do we combine the best of all of these worlds? >> And Jeremy, I know before you talked about how some of the autoregressive models are not maybe a great fit for agents. Any other thoughts on like JEPA, diffusion for text, any interesting thing that you've seen pop up?

>> In the same way that, like, we probably ought to have state that you can update, i.e. XLSTM and state models, in the same way, a lot of things probably should have an encoder, JEPA and diffusion both seem like the right conceptual mapping for a lot of things we probably want to do.

So the idea of, like, there, there should be a, a piece of the generative pipeline, which is like thinking about the answer and coming up with a sketch of what the answer looks like before you start outputting tokens. That's where it kind of feels like diffusion ought to fit, you know, and diffusion is, because it's not autoregressive, it's like, let's try to, like, gradually de-blur the picture of how to solve this.

So this is also where dialogue engineering fits in, by the way. So with dialogue engineering, one of the reasons it's working so well for me is I use it to kind of, like, craft the thought process before I generate the code, you know. So yeah, there's a lot of different pieces here, and I don't know how they'll all kind of exactly fit together.

I don't know if JEPA is going to actually end up working in the text world, I don't know if diffusion will end up working in the text world, but they seem to be, like, trying to solve a class of problem which is currently unsolved. Awesome, Jeremy, this was great, as usual.

Thanks again for coming back on the pod. Thank you, Lysia. Thank you, Sean. And thank you all for listening. Yeah, that was fantastic.