audio audioduration (s) 3.66 30 | text stringlengths 85 608 | source stringclasses 1
value |
|---|---|---|
I'm going to explain how to fine tune the latest open source models, all the way from Gemma 3, Quen 3, Lama 4, Phi 4, and Mistral Small. I'll explain the pros and cons of using Unsloth versus using Transformers. I'll explain also how to do fast evaluations using VLLM separate to the fine tuning. And then I'll go throug... | 0 | |
For a video overview, I'll briefly describe why you might want to fine tune. Hopefully you've got a sense already if you're watching this video, but I'll recap on that. I'll very briefly describe how to prepare data. You probably should be spending 90 % of your time on data preparation, and I have a lot of videos cover... | 0 | |
But I'll talk about the pros and cons of using Unsloth versus Transformers for fine tuning. Then I'll talk a little about running fast evaluations. You should be evaluating your performance before you fine tune and afterwards and during as well. And I'll show you how you can speed that up a lot by using VLLM in the sam... | 0 | |
I'll give a few general fine tuning tips before then going to a live demo of fine tuning using Unsloth and Transformer notebooks that I've put together in the advanced fine tuning repo. So why should you fine tune? Generally, this is ideally a last resort. You have tried out doing prompt engineering. You have included ... | 0 | |
You need to improve your answer structure and format. For example, you want the model to take a certain approach by maybe recapping on some of the information, processing it, and maybe reasoning over it, and then giving a final structured answer. Sometimes that means you have tool calling. So you want to have a structu... | 0 | |
Now, there are two others you might consider as well. One is if you want to improve accuracy beyond just using a retrieval method. Back in one of my earlier videos, I show how you can combine retrieval with fine tuning and get the best possible approach. So you can improve performance beyond just a retrieval approach o... | 0 | |
That's like GRPO, Group Relative Policy Optimization, and other related techniques. I have a series of videos from earlier in 2025 that cover that, and I'll let you try them out there. For today, though, we're going to focus not so much on reasoning, although I think I will make another reasoning video soon. We're goin... | 0 | |
But a lot of the same principles apply if you wanted to fine tune on structure, although I'll point you for more details on fine tuning for JSON responses or function calling to this video right here. I'm not going to talk too much about data preparation, but I will point you to, well, I'll give you a few tips and poin... | 0 | |
This is where you take, say, magazine content, newsletter articles, or books, and you feed that in without much pre -processing. I mean, you'll clean it up, but you're not going to change it into Q &A pairs. This continued pre -training here, typically, this is difficult to do on top of an existing model because it can... | 0 | |
So often, unless you have a very large amount of data and you're willing to do continued pre -training followed by post -training, then it's probably not recommended to try and do the continued pre -training. Instead, what generally is recommended, particularly with smaller amounts of data, like, say, up to maybe 100 ,... | 0 | |
And these are often data sets you synthetically create using documents and using LLMs to create questions and answers from those documents. Now, to prepare a synthetic data set, it's recommended to use a large language model. You want to generate not just questions and answers, but ideally questions, evaluation criteri... | 0 | |
If you further want the answers to involve reasoning or to involve chain of thought, you probably need to augment them further. And I do have a video showing how with these augmented answers, you can get up to very high levels of accuracy with fine -tuned models. And that's true whether you're fine -tuning open source ... | 0 | |
Now, one caveat, I don't say this with 100 % confidence, maybe 75 % confidence. My sense is that as models get stronger, it's harder to fine -tune. And that's because the model starts off from a very good point. And in fine -tuning, you are risking damaging the model in some ways. I'll give you a simple example. If you... | 0 | |
And if you train it without reasoning data, you could just drag the performance below what the reasoning performance was, even though you're adding the right content to it. So I think in some ways, fine -tuning is getting a bit trickier. And if you are going to try and fine -tune for reasoning, and I mean reasoning not... | 0 | |
But if you're trying to use reasoning in kind of a verbal domain, that's going to be tricky. It's something I want to cover maybe in a later video, but just kind of watch out. Because if you don't have proper reasoning data sets developed, you do risk maybe regressing performance. Okay, so I'm going to talk about the l... | 0 | |
Those are two other examples. Maybe I'll go through those at some point. But the one that I've been using most often on this channel is either Transformers or unsloth. Unsloth is effectively a wrapper. I mean, I don't mean that in a negative way. It's brought a lot of improvements to the Transformers library in terms o... | 0 | |
So unsloth generally is two times faster than Transformers. And that's because of a variety of tricks. It's not just one thing. It's like the accumulation of about five or six tricks that result in faster fine tuning. Also, unsloth provides a unified function for loading multimodal models. So if you have a model like J... | 0 | |
Whereas with unsloth, it's just unified how you load that model. So that makes it a lot easier to have one script that supports a lot of different models. Unsloth for now is single GPU only. So if you have a very large model, you might need to use Transformers instead, which does support multi GPU. By default, Transfor... | 0 | |
So it will chunk up the model by layers, which means your GPUs are not all active when you're fine tuning because you will use one GPU to process these layers, then the next GPU, then the next. That might sound inefficient and it is, but it's quite simple and robust. So I wouldn't rule that out if you need to tune a la... | 0 | |
That's where you split the model across GPUs, but you split the matrices essentially so that all GPUs are being used more or less at once. And you can find a video on fully sharded data parallel on this channel if you just look up FSDP. Something else about Unsloth, because it's a wrapper, sometimes there are issues wh... | 0 | |
Also, if you're trying to use AI to help you code, there will tend to be more documentation on the advanced features in Transformers. And because everything is being wrapped by Unsloth, it's sometimes harder to access the features by working through Unsloth. So basically what I'm saying is if you're trying to use some ... | 0 | |
For now also, JAMA 3 is still broken in the sense that the configuration file won't allow you to run inference on VLLM. So if you do fine -tune a JAMA 3 model on Unsloth, I'll point you to an issue where there's potentially a fix. I expect that it will be fixed at some point, but something to keep a note on. And also, ... | 0 | |
It's 192 gigabytes in VRAM. So if you have a model, you can probably fit a model that's up to 150 billion parameters in 8 -bits. Unsloth allows you to fine -tune in 8 -bits now, not just in 4 -bits. I don't recommend 4 -bit. I know it's very popular doing Qlora, but I have found, particularly because of merging back ad... | 0 | |
8 -bit is probably quite good as well. So yeah, if you have a model that's 16 -bits, you can train up to probably 80 or 70 billion parameters on a B200 with Unsloth, or in 8 -bits, probably even something larger. So this is not even that much of a limitation. You're not going to be able to fine -tune DeepSeq using Unsl... | 0 | |
If you wanted to fine -tune, let's say, Lama 4 and the Maverick version, you're not going to fit that on a single GPU. But you could fit the Scout version in a single GPU in 8 -bits, because it's 100 billion parameters, roughly. And it's about one byte per parameter, so in 8 -bits. So that would be fitting onto that GP... | 0 | |
You can just run inference using Transformers or Unsloth. But neither of these libraries are designed for inference, so they don't do what's called continuous batching. You can send in a batch of tokens, so you have to size the batch so you don't run out of memory. You can slightly automate that, actually, through Tran... | 0 | |
But the inference is just not optimal in terms of the back -end, and it will be significantly slower than using something like VLLM or SG -Line. So I recommend, and that's why the scripts I'll show you today, they have two parts. They've got an inference part or an evaluation part that's run with VLLM, and then they've... | 0 | |
If you have a very big model, it can take a bit of time to load. But again, if you're only on one GPU, this is probably not going to be a big constraint, and reloading the model should be fairly fast, given it's already going to be on your disk from the fine -tuning. Now for some questions on which model to fine -tune.... | 0 | |
It's an Apache 2 license, and it tends to be strong in evaluations. Just I've heard from customers as well when I see results across the different models they've tried. So it would be one of my top recommendations. Gemma 3 is a very strong model as well, but the license is custom. So if you're at a bigger company where... | 0 | |
You can just click on these links to get more info on the licenses. 5 .4 from Microsoft is a permissive license. It also allows for reasoning. These top two models don't. So if you want to fine -tune for reasoning, maybe 5 .4 is a good option. Lama 4 is a custom license, and it's also very large. The Scout model's 100 ... | 0 | |
The quality is probably not much better than Gemma 27B, Gemma 3, or potentially even Mistral. So I recommend probably just using Mistral or Gemma 3 over Lama 4. Quen 3 is a very strong model. I think it is probably stronger than all of these models here, and it's Apache 2. But you do have the issues that come with usin... | 0 | |
And also there is a backdoor risk with any language model here, but you have to weigh that in the context of where and how it was developed. These models are increasingly being used to control agents, and that provides an extra angle by which you can have danger if the model is controlling your agent in a malicious way... | 0 | |
So Quen 3, very strong, perhaps very good choice if you want to fine -tune for reasoning, but you do have to be careful of the censorship and the backdoor issues. Now, just a few general fine -tuning tips before I move to a demo. Spend 80 or 90 % plus of your time on data preparation. That probably means watching some ... | 0 | |
I explained in my data prep video how you can rephrase certain questions in order to make sure they're not verbatim in your training and your eval set. But I do recommend including a verbatim copy of some of your training data set, because by including a verbatim copy and also a version, a data set that's not in your t... | 0 | |
Measuring overfitting is another reason to use the eval set during training, so you should be calculating training and eval losses. Make sure to evaluate before and after fine -tuning. And then one kind of random tip here is do inspect the chat template being used when you're fine -tuning. Some chat templates have got ... | 0 | |
So you may want to remove that when you're doing the fine -tuning. So other things that can appear unexpectedly may also adversely affect the quality. Okay, so that's it for the theory portion. I'm going to move now and show you how to fine -tune. And if you want to find the scripts I'm going to show you, they're avail... | 0 | |
And I've actually refreshed the repository. Let me show you here. I've just cloned it over using Windsurf. Historically, the fine -tuning repo was organized according to branches. So different branches would have different content. And those branches are still there. For example, if you want synthetic scripts, scripts ... | 0 | |
There are a large variety of scripts, and they're still there in different branches. But going forward, I'm going to leave all of the scripts within the main branch here. And I've started to create a clean folder for data prep. That was the most recent video. And now I've got a clean folder here for fine -tuning. And t... | 0 | |
One is VLLM unsloth. It uses VLLM for evaluation and then unsloth for fine -tuning. This one here is VLLM transformers, so it uses VLLM for eval and transformers for fine -tuning. And then this one here is pure transformers, so it will just use transformers both for evaluation and for fine -tuning. And it will automati... | 0 | |
So I'm going to show you a fine -tune using the unsloth script here. The transformer script more or less follows it. And when I go through this, I'll highlight a few points where there's a difference between using unsloth and the fine -tuning. Now, to get going on a GPU, I'm going to use RunPod and this one -click temp... | 0 | |
Now, here, I said 192. That was wrong. Sorry. I meant, I guess, what I should have said was 180. So if you run a B200 for $7, it looks like, yeah, $8 an hour. Maybe I said $7. You can fit 180 gigabytes of VRAM. We're just going to run with a H100, which is 80 gigabytes. We won't be running quite as big a model. So I'll... | 0 | |
And you can see here that we've got a CUDA 12 .1 and PyTorch 2 .2 template. So I'll get this going. Okay. So once the pod has started, we're going to connect and open up Jupyter. And then I'm going to upload my notebook, the unsloth one. Now, just when I'm uploading unsloth here, notice that I can also upload a require... | 0 | |
So you can either install the latest version of the dependencies, or you can install from the requirements file if you want to ensure reproducibility. Okay. So I apologize for hurting the eyes of those who are sensitive to white light. I have swapped this over to dark mode. And we're going to start off with an evaluati... | 0 | |
Well, it's a trellis data set called Touch Rugby and Comprehensive. It's the QA data set I generated in a recent video, and it consists of a series of questions and answers, and then evaluation criteria for marking a given answer correct. And these were all generated using, I think, the Gemini Pro 2 .5 model based on a... | 0 | |
So the first thing I'm going to do is I'm going to run some installations here. Sometimes if you've run the fine tuning first with unsloth, you might need to uninstall unsloth. But since we're starting fresh here, we don't need to uninstall unsloth. We'll just run ahead and do the installs, the most important of which ... | 0 | |
It's a new model I haven't trained, so I'm kind of curious what the results will be and see if we hit any issues. But I will give some commentary as we go about some of the other models. So my first comment here is if you're using the QEN model, if you want to fine tune and disable reasoning using VLLM, you do need the... | 0 | |
So just a heads up, if you're going to use QEN 3 and you want to disable reasoning, you need to install VLLM from source for now. All right. So while that's installing, we'll move on here. Do restart the kernel after the installs are done. This is where I was saving my requirements to a TXT file. You could, if you wish... | 0 | |
VLLM on slot and then dash system. Now, the dash system means that we're installing onto the system. We're not in a virtual environment. And this makes sense because when we started the Docker image on this GPU, we already had CUDA and PyTorch installed and we want to make use of those. We don't want everything to be p... | 0 | |
And now I'm going to log into Hugging Face. So I'm going to, I won't add it as a Git credential. I will get a token. And I've got a token and we're logged in now. I can save that. At least close it. And the model that we're going to evaluate. So we'll evaluate the base model. Then we'll do fine tuning. And then we'll e... | 0 | |
And I'm going to set model slug equal to the phi model. And I've got the phi model up over here. And yeah, the dark mode is pretty poor on Safari. So I'll paste it in. And the data set is the touch rugby data set. The training split is called train. The eval split is called eval. There is a mirror eval split. This is l... | 0 | |
And within this data set, there's a column for a question. There's a column for evaluation criteria, which we'll use for grading. And there's a column for answer, which we'll use for fine tuning. So the answer is for fine tuning. The evaluation criteria is for grading. And for setting up the judge, we're going to use a... | 0 | |
And this makes sure that we download the weights into the disk. So we want, when we're on RunPod, we want the weights to go on the volume. We don't want it to go on the container. The volume here is about 500 gigabytes. The container is only, I think, maybe 10. So I'll run this. That just sets those variables. And now ... | 0 | |
So I'm just going to go over to AI Studio. I'll do this off screen. And I'm going to create an API key. So I'll paste in that API key. And that's set now. And just to briefly show you the judge, this is not yet the judge. It's just setting the API key. It's actually saving it to a local environment variables file so th... | 0 | |
Here we're setting up an OpenAI client. So we're hitting Gemini using an OpenAI client. And here we've just set up this function called chat that allows us to send messages in with a single message into Gemini. Okay, next we're going to prepare the data set. And when we prepare it, we can turn a test mode on. | 0 | |
You'll notice that back here earlier when I defined the data set, there's a flag to set test equals to true. If you just want to look at a small sample of the data set, you can set test equals to true. And it will also print a lot more debug logs down below in the script. So I've got that set to false. And we're going ... | 0 | |
And yeah, if we're just inspecting, we're going to slice the data so that we're only going to take a few rows. So here we're potentially sampling if test is true. Otherwise, we're loading the full data set and we're just printing out to make sure everything is in order here. So when I print the data and eval set, the t... | 0 | |
This is how you load with VLLM. You pass in the model slug, the GPU maximum memory utilization, the data type and the max sequence length, which we set up earlier, and our sampling parameters. So here, when we're generating, we're going to use a temperature that's defined by temperature here, which I think I've set to ... | 0 | |
You don't want tokens that are very low probability because they can throw your prompt or they can throw your completion off in an unexpected direction. So at this point, the model is going to be loaded. It'll be downloaded from Hugging Face first, and then the shards will be loaded onto the GPU. So this is typically w... | 0 | |
So you can see there's about seven gigabytes. And notice here that we are using FlashInfer. FlashInfer is a relatively recent library. It's faster than FlashAttention, and if you have it installed, it will be used by VLLM. If you don't install it, I think it won't be used by default. So this model, Phi4mini, relatively... | 0 | |
It's then going to create CUDA graphs, so it's going to calculate forward some paths for optimally doing the computations. This takes a bit of time, but then it makes the inference faster because you've pre -computed what's called the graph. Okay, so when the model's loaded, we can run evaluation. To run evaluation, we... | 0 | |
Here we have an expert evaluator tasked with determining if the answer satisfies the specified evaluation criteria. We're telling Gemini that it will receive a question, the evaluation criteria, and then the model's answer that needs to evaluate it. And it's just going to score a one or zero, so it's here right or wron... | 0 | |
Okay, so this is pretty much it. We are going to define a helper. This is the evaluation result. We want the evaluator to first give a reason and then indicate whether the model is correct or not. We'll then extract, using a regular expression, the score, whether it's one or zero, to determine whether it's marked corre... | 0 | |
And so if you include all of the thinking along with the answer, it's going to be harder to evaluate. And for that reason, if there is any thinking, we're going to strip the thinking here. Then we have a function to evaluate. And you can see it generates an answer. Sorry, this evaluate answer takes in a generated answe... | 0 | |
Then we'll call the judge LLM and we'll get the response and parse that. And when we want to evaluate a model, here's where we need to create the messages with the problem, generate an answer, and then pass that answer in for evaluation. So that's the full loop. And just a note here that if you want to disable thinking... | 0 | |
So we've defined that evaluation function and we're just going to run it on a row. This is the fifth row of evaluation data. And it looks like my API key is invalid. So I'm going to go back up and I'm going to reset that API key. And to do that, I think I just need to search for the word reset. Okay. | 0 | |
So by the way, you can also select OpenAI for grading as well, if you wish, but I'm going to use Gemini. And let's now put in my API key. I'm just going to get a new key here. I must have pasted in a wrong key. Okay. Let's try that. And we'll continue on down and see if that evaluates the model for us. And here, VLLM h... | 0 | |
I shouldn't have rerun that. I should have just gone straight to the evaluation because the model was already loaded. So I hit a CUDA out of memory. See, VLLM will often use up the full memory for when it loads a model. It will pre -allocate that memory. So it's recommended don't rerun the model loading. Okay. So it is... | 0 | |
So yeah, before I restart the kernel, I'm just going to set that reset back to false because I should have fixed my API key. And now I can run all the cells. So yeah, the model is reloading. And I think sometimes the graph can be cached. So that improves. It improves the speed for loading as well. And you can see for t... | 0 | |
So it will automate batching for us like that. And here we go. It's now evaluated. The question, what's the regulation about touch rugby participants covering the ball with their clothing? And here is the generated answer. And here is the evaluation criteria. And the judge has marked it correct because it's saying that... | 0 | |
But what we want to do is we want to evaluate a batch of questions. So we're going to use batching. We're going to have VLLM answer multiple questions in parallel. And then we're going to use threading to make parallel calls to Gemini's API. So that's what's happening here. We're batch evaluating the model by building ... | 0 | |
Again, if you want to disable thinking, that's the line to include. And then we're going to judge using a thread pool executor that's going to make multiple parallel requests. And we'll get back the final score here. So you can now run a short test. You can set test to true. It'll just run two rows of the data set. And... | 0 | |
And then it's going to judge those. So here's a sample generation. In fact, it's just giving us the five generations first. And then it's giving us the judging of those five. And you can see the results coming out here. And it looks like we've gotten 40 % correct. Now we're going to evaluate the full evaluation data se... | 0 | |
So actually, when you evaluate, you want to evaluate multiple times on that same eval data set. Or maybe if you had a very large eval data set, you wouldn't have to do this. But because my data set is 32, you'll find some variance if you just run it once. So I recommend running it at least three times. And for that, I'... | 0 | |
And I'm going to just copy this here because I've run it previously on Gemma 3. Let's just create a new cell. And it's going to print out the data set name, the eval split name. So we're running the comprehensive data set, the eval split, and we're running the phi model. And we should be ready to go. Now, one thing I d... | 0 | |
So actually, what I need to do is set test equals false and run it again. And that will just suppress all of the detail logging. You can see here we're going to run on all of the prompts. And then it's going to evaluate those using the judge. Now, actually, you could increase the number of threads here. Gemini's API is... | 0 | |
And I think I have an issue with the kernel because I can see my GPU memory is not working. That may be because I just stopped it during processing. Sometimes if you stop VLLM midway, you just run into these issues. So maybe I should have let it run out. But I didn't feel like that because it was printing too much debu... | 0 | |
And just while we're waiting for that running, I'm actually running the comprehensive dataset as opposed to a manual dataset that I curated. So this should actually be moved down. You can see evaluation is starting there, by the way. And I can probably just move it in this way here just by clicking the down button. And... | 0 | |
You can see here I ran it previously on the Mistral model. And if I just paste in a copy of this, I'm adding in test equals false to make sure we don't have too much debug logs. And you can see now we're processing the prompts. So 32 prompts. And I think I actually should have put that with a lowercase test. | 0 | |
Either way, we can probably wait for it to complete and then rerun it so it doesn't print out all of the logs here. So it's going to print out 32 of the answers. And here you can see it making those parallel calls to Gemini. And we're running basically 96 different prompts here because we're running this three times. S... | 0 | |
Now, I'm just going to run this again so it doesn't print the verbose logs by setting test equals false. And in the meantime, we can take a look at some of the results. So here with Mistral Small, when I ran this three times, I got an average of 13 answers correct. Somewhere around 40 % overall. I can show you also, I ... | 0 | |
So the QEN 1 .7B, I scored five. So about half the amount, five instead of 13. And I can show you also, let's see, do I have any other results down here? JAMA 3 .4B. And JAMA 3 .4B scores nine. So, yeah, the Small QEN model scoring about four. That's including reasoning, by the way. Sorry, five. JAMA 4B scoring about n... | 0 | |
And now let's go up and see how we score with the PHY model. I expect, I don't know, maybe something like the JAMA 4B. Somewhere around nine. Okay, we got six. Then the next one, we got four. So that just shows you the variance. And that's why it's valuable to run multiple times. And then the last one here, we've score... | 0 | |
So actually, this model, the PHY mini, is not much better than the QEN 1 .7B. So what this does is it gives us a baseline. We've got 5 .3 correct. We're now going to run fine tuning. And we'll come back and run this again. And we're going to see if we get an improvement in the performance. Now, improving performance is... | 0 | |
It's a raw set of answers that were generated by Gemini Pro, which may not match, probably doesn't match the kind of logits or the probability pattern of the model that we're training here. So I could definitely do a better job of improving this data. So I'm not sure we're going to improve performance here by fine tuni... | 0 | |
Now, for running fine tuning, we're going to need to use unsloth. And we're going to uninstall VLLM to make sure we don't have conflicts. If you want to speed up the fine tuning, you can install flash attention. But it does give issues with QEN. So that's just a little warning. And if you want to use it, when you load ... | 0 | |
If I'd restarted the kernel, it would have gotten rid of these warnings. That's OK, though. It's going to still correctly install. And I'm going to now restart my kernel. And we should have unslots installed. Now, just two troubleshooting things. If you find that there's a conflict with TorchVision, you do not need Tor... | 0 | |
Unslots has a custom cross -entropy that saves on memory. I think maybe on compute. You can disable it if there are issues by running this line here. OK, so I've restarted the kernel. And I should still be logged in. Actually, it looks like maybe I'm not fully logged in. So I'm going to go across and get a token. And t... | 0 | |
Now, the model that we want to train is going to be the PHY model again. So I need to populate that. And, yeah, we can set the max sequence length to 8000. We're going to fine -tune in 16 bits, which I recommend. But 8 -bit, I think, actually is not bad. So if you wanted to save memory, you could do that. We're going t... | 0 | |
And if you wanted, you could use a different data set for eval by setting this here to the load data set right here. OK, so we're going to set these variables. I've got this little helper function. This is just a function to clear CUDA. If you've loaded the model and you want to reload a model, you can clear out the mo... | 0 | |
And now we're going to load the model. So I'll run this here. And I'm going to print out the padding side. Typically, for fine -tuning, you will want to pad on the right -hand side. Or you'll want to use whatever the model's default is. But for inference, using VLLM, you typically want left padding. So just a note ther... | 0 | |
Now, unsloth here is downloading the model, which I do not want because the model should actually already be downloaded. So what I need to do is potentially set this cache directory and retry. And again, it's downloading the model here. So I'm going to restart the kernel. And let's just check that we have the name of t... | 0 | |
5 -4. Yeah, it could be that unsloth is downloading it from its own repo. Because unsloth has got a version of all these models. So I'm not entirely sure, but I suspect that may be what's happening. If I look at the 5 -4 mini model here, this is the original model. But if I copy this, there's probably an unsloth versio... | 0 | |
Yeah, unsloth has got this version here. And it may just be defaulting to this. So that's why it's actually downloading the model. Even though I've downloaded it already for VLLM. That's okay, though. We can keep going. Allow it to download. And then we'll see what padding side is default. And we'll also print out the ... | 0 | |
So actually, unsloth is automatically setting it to the end of text token. And it's using the left -hand side for padding, which should be fine. And you can see the model architecture here. 31 layers, the MLPs, the attention. And actually, the attention is fused. So the Q, K, and V are fused together. Unsloth might dec... | 0 | |
So this here is a function just for me to inspect the size of these different modules. This is relevant because larger matrices you should train more slowly. And this affects how we put on the adapters. And, yeah, we actually need to adjust this code to work for phi because the layout is not the same. | 0 | |
So if I go to chat GPT and I create a new conversation, I can say update this code to support the phi architecture. And sorry, my screen is small there. I'm just going to paste in the phi architecture now. So I'll paste this in and we'll print. Now, why am I going to this trouble to see the dimensions? | 0 | |
I'm doing it because I want to know what I should set my LoRa Alpha to. And the LoRa Alpha should be the square root of the smallest matrix dimension. Basically, LoRa Alpha is kind of setting a bar for how you think about the relative training rate of the adapters, the LoRa adapters versus the main matrices. We're usin... | 0 | |
We're just going to put these little adapters that clip on to the main model and we're going to fine -tune those instead. But because they're smaller, you need to train them faster. The size of the adapters is determined by the rank. The larger the rank, the slower you want to train it. And that's actually scaled autom... | 0 | |
Because it's all about the relative size of the adapters compared to that original matrix that we're going to freeze. Okay, so it's given us this unwrap function. And we will see if it works. So let's see here. Is it giving us both functions? Yes. So I think I can just paste that in. And if I run it. Yeah. So now we ca... | 0 | |
We've got this fused layer, which has got the smaller dimension of 3000. And we've got the MLP, which has got 3000. So actually what we want to do here is we want square root of 3000. And I think that's going to work out. 32 is about square root of 1000. So it should be about 1 .7 times this. So something like 50 is go... | 0 |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 14