3 ms·
7B LLaMA is a terrible general purpose model, but the finetunes are pretty good at very specific roles, like dialogue/roleplay, a dungeon master bot or even cod
by brucethemoose2 3y ago
7B LLaMA is a terrible general purpose model, but the finetunes are pretty good at very specific roles, like dialogue/roleplay, a dungeon master bot or even code completion.
The metrics are good though, perhaps placing this closer to 13B.
And 8K context is huge. When you can stuff that much example text in, it gives the model more to "latch onto," and its also the point where you would start worrying about RAM/VRAM consumption for a ~13B model.
- aryamaan 3y agowould appreciate resources on fine tuning
- brucethemoose2 3y agoHonestly, I dunno. I think most people are using lit-llama or EasyLM (on TPUs) for finetuning? QLORA is the gold standard for more affordable training. As for datasets, just look at the open datasets the best-in-class models are using, like Vicuna or https://huggingface.co/NousResearch/Nous-Hermes-13b https://huggingface.co/NousResearch/Nous-Hermes-13b Some model datasets like Manticore, Chronos or the infamous Pygmalion are more "secretive," but you can find the dataset gathering scripts on Github or in community chats.
- SparkyMcUnicorn 3y agoThis and/or text-generation-webui training doc are a good place to start. https://github.com/zetavg/LLaMA-LoRA-Tuner https://github.com/zetavg/LLaMA-LoRA-Tuner
- ttt3ts 3y agohttps://huggingface.co/blog/stackllama https://huggingface.co/blog/stackllama You can easily finetune 7B or 15B LORA model with that on consumer GPUs.
- minimaxir 3y agoThat blog post demonstrates that it's not "easily" finetuneable, just possible to finetune. There's many technical considerations even beyond hardware (dataset formatting, training hyperparameter nuances) that do not make it accessible to the newbie experimenting with LLMs. It's a rabbithole, and unfortunately there's no good shortcuts.
- Solvency 3y agoWhy have there been thousands of overnight AI/GPT startups and products in the last few months and NOT a single simple intuitive "fine tuning wizard" app? That seems like such an obvious glaring gap.
- minimaxir 3y agoBecause the ChatGPT API (and analogous competitors) is cheap enough that it's both faster and more cost effective to just use it instead instead of using your own model, with maybe some shenanigans to handle its shortcomings without increasing cost much if at all.. And that was before gpt-3.5-turbo-0613, which dropped the price more and is about 2-3x faster. There are startups that do finetuning on your own data, but with zero hints on how to preprocess your data and absurd costs (both upfront training and GPUs for serving inference) that's it's extremely difficult to argue from a customer business perspective compared to just using an API.
- hyperliner 3y ago[dead]
- shapefrog 3y ago> Why have there been thousands of overnight AI/GPT startups and products in the last few months and NOT a single simple intuitive "fine tuning wizard" app? Vapourware GPT startup inc is valued at $2bn the afternoon after you form the company and buy your first macbook. Actual usage of Ai, fine tuning etc. I can offer you $100,000 for 30% of your company if you can demonstrate a fully working product.
- lhl 3y agoHere's a good writeup that goes into more depth than most of the READMEs on fine-tuning: https://erichartford.com/uncensored-models https://erichartford.com/uncensored-models
- ttt3ts 3y agoDo you know a dataset for fine-tuning roleplaying?
- Tostino 3y agoYou must have missed the memo... It's now super easy to extend the context of 2k llama models to 8k, 16k, or even 32k with just a small fine tune and a tweak to the code. You still need the memory to be able to go that high, but it's totally doable.
- thomasahle 3y agoWhat memo/paper does this?
- Tostino 3y agohttps://arxiv.org/abs/2306.15595 https://arxiv.org/abs/2306.15595
- imtemplain 3y ago[dead]
- brucethemoose2 3y agoI saw the SuperHOT LORAs as well. But I assumed full training would give better perplexity for large contexts, and perhaps this method would be more effective at 16K+ with an 8K model to start with.
- Tostino 3y agoPossibly, but the perplexity has shown to decrease while fine-tuning a 2048 model on larger context sizes for outputs within it's original context limit...so, more research needed.