Mark Huang: evaluation

Transcript context

…One more question that I had was on LoRa and taking some of these capabilities out and bringing them to other model. You mentioned Weng's work. He tweeted about we're going to take this LoRa adapter for the Gradient 1 million context extension, and you're going to be able to apply that to other model. Can you just generally explain to people how these things work with language models? I think people understand that with stable diffusion, you have these LoRa patches for different types of styles. Does that work similarly with LLMs? And is it about functionality? Can you do LoRa patches with specific knowledge? What's the state of the art there? Yeah, I think there's a huge resurgence in what I would call model alchemy to a certain extent, because you're taking all of these LoRa's and you're mixing them together. And then that's a lot of the model merging stuff that I think Charles Goddard does and a lot of others in the open community, right? Because it's a really easy way. You don't need training, and you can test and evaluate models and take the best skills and mix and match. I don't think there has been as much empirical study, like you're saying, for how shows the same type of... It's not as interpretable as stable diffusion to a certain extent. Because even we have experimented with taking deltas in the same methodology as Wing, where we'll take a delta of an already trained model, try to see how that has created, in a sense, an ROHF layer, right? Taking the LLAMA instruct layer, subtracting the base model from that, and then trying to apply that LoRa adapter to another model and seeing what it does to it. It does seem to have an effect, though. I will not lie to say I'm really surprised how effective it is sometimes. But I do notice that for more complex abilities, other than more stylistic stuff, it kind of falls through. Because maybe it requires a much deeper path in the neural network, right? All these things, these weights are just huge trees of paths that the interesting stuff is the road less traveled, to a certain extent. And when you're just merging things brute force together that way, you don't quite know what you'll get out all the time. There's a lot of other research that you have merged ties and you have all these different types of techniques to effectively just apply a singular value decomposition on top of weights and just get the most important ones and prevent interference across all the other layers. But I think that that is extremely interesting from developer community. And I want to see more of it, except it is to a certain extent, kind of polluting the leaderboards these days because it's so targeted. And now you can kind of game the metric by just finding all the best models and then just merging them together to do that. And I'll just add one last bit is basically the most interesting part about all that actually to me is when people are trying to take the lowers as a way of like, short circuiting the training process. So they take the lowers, they merge it in, and then they'll fine tune afterwards. So like the fine tuning and the reinitialization of a little bit of noise into all the new merged models provides like kind of a learning tactic for you to get to that capability a little bit faster. There's a lot there. I really like the comparison of ties merging to singular value decomposition. I looked at the paper and I don't really think I understood it on that high level until you just said it. We have to move on to benchmarking. This is a very fun topic. Needle in a haystack. What are your thoughts and feelings? And then we can discuss the other benchmarks first, but needle in a haystack.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.