23 ms·
Gorilla: Large Language Model connected with massive APIs
- billconan 3y agothe thing is that, many tasks require a workflow, rather than a single api call. can this model figure out a workflow?
- jfarina 3y agoIt sounds like it can expedite the interpretation of the APIs documentation.
- toomuchtodo 3y agoBuilds integrations to APIs vs no code/low code providers doing it it seems. The logical next step is to make LLM workflow running robust and maintainable.
- visarga 3y agoThe dataset is good though, covers many model APIs. Datasets are portable, can be used in many ways.
- sandGorgon 3y agowould this outperform the api<>instruction pairs stuffed into a vector db and running an retrieval augmented chain of thought on openai this is how openai plugins work right ?
- aledalgrande 3y ago> Gorilla, a finetuned LLaMA-based model that surpasses the performance of GPT-4 on writing API calls. When combined with a document retriever, Gorilla demonstrates a strong capability to adapt to test-time document changes, enabling flexible user updates or version changes keywords "finetuned" and "document retriever" So it's fine tuned to create API calls but still uses a db to fetch the API shape. They also say that they're doing a "retriever aware" training, meaning passing examples in the dataset that point out that a document with the API shape is gonna follow the question, in the hope that the model learns to always check on that before answering.
- kodah 3y agoHow feasible would it be for me to run things like this and a ChatGPT/CoPilot style LLM on my local servers?
- vs4vijay 3y agoWelcome to https://old.reddit.com/r/LocalGPT/ https://old.reddit.com/r/LocalGPT/
- nacs 3y agohttps://old.reddit.com/r/LocalLLaMA/ https://old.reddit.com/r/LocalLLaMA/ < much larger/active (and its not just for the meta-llama)
- seydor 3y agocheck out oobabooga, you can run most llama derivatives with a browser UI
- behnamoh 3y agocan people stop using animal names for their models and methods? At least with LlaMA, the name was an acronym, but Gorilla is clearly just marketing. flagged.
- anaganisk 3y agoWhy should they, and it seems very silly to flag a post for their preference of name. Are users on Twitter birds? Or is Mastodon filled with elephants? Name is merely a symbol in today's world Google is synonymous to search but Google is not just search anymore. I'm honestly curious about your gripe with it.
- behnamoh 3y agoSome people confuse research with production. Imagine if Einstein named his theory the Theory of Chimps, just to get attention.
- sebastiansm 3y agoImagine working in the Manhattan project outside Manhattan.
- sebzim4500 3y agoThere's loads of stuff in physics with silly sounding names. quarks, anyons, etc.
- EddieEngineers 3y agoIf Gorilla upsets you, wait till you hear about Cummingtonite!
- anaganisk 3y agoWhat's the point of research if it can't gain attention, how many people really understood or even know about some space magic, before interstellar kind of managed to ELI5. Research is meant for everyone, and in today's world you bring attention to it with catchy names. I wonder if Google was named PageRank Algorithm runner, how many would've even used it.
- zellyn 3y agoNote: if you read the paper carefully, this is LlaMA-derived, so suffers from the same use restrictions. We really need the open LLMs to catch up!
- jimsimmons 3y agoRealistically you can assume there’ll be one soon and develop with these models and switch over
- quickthrower2 3y agoThere are some already (one of the openllamas I think, and that discord group made one with the linear attention recently although trained on fewer params I think) Maybe not easy to switch to a non-meta-llama ?
- operator-name 3y agoAddress not found?
- sudoapps 3y agoI can see this being useful as a specialized (fine-tuned) LLM in a chain of LLMs for full autonomy
- ethanbond 3y agoDon't worry folks, the years and years of "of course we'd never connect experimental AI systems to the open internet" assurances were never necessary for safe development and deployment. This reassessment was founded on the last several months that showed us these systems are close to 100% dependable, they never deceive humans, they are entirely aligned with human morals (you know, all those morals we all agree on), they are advancing at a totally predictable rate, and we never see unexpected behaviors from them. Besides, they can only be wielded by people who we trust have good intentions. No, this re-assessment has nothing to do with market dynamics incentivizing people to deploy experimental systems further into the infrastructure we depend on.
- andrewmutz 3y agoMany of us have not been persuaded by the pervasive, kneejerk fear of AI that floods social media. If social media had been around in the past, we'd have never developed trains or electricity or cars because of the crippling fear that something bad might happen along the way.
- moritzwarhier 3y agoI don't see cars as being comparable to the universal usage of electrical energy as an intermediate of converting physical forms of energy. Not even close. And also, the "deployment" of cars if you will, has brought and still brings malignant effecfs on society, not only positives. If by cars you mean cars for transporting individual people in non-critical situations and the corresponding industrial and political developments, especially around urban planning. Nor arguing about the significance of the effects, but the discovery of electricity and how to use it seem more useful to me than the discovery of using fossil fuel cumbustion for transporting people. Road transit in total is a different thing of course.
- andrewmutz 3y agoIf a technology has positives and negatives consequences for the world, but the positives outweigh the negatives, then we want those technologies to be developed. For cars, the positives outweigh the negatives.
- ManuelKiessling 3y agoPlug: I’ve built something vaguely similar, but of course a trillion times less brilliant and powerful, yet nonetheless useful for some straightforward cases where really you just want GPT et al. to call your API at just the right moment of the conversation, with JSON that is guaranteed to be valid: https://github.com/manuelkiessling/php-ai-tool-bridge https://github.com/manuelkiessling/php-ai-tool-bridge
- danmaw 3y agoVery cool and thanks for sharing, would you say this is similar to langchains tool concept?
- ManuelKiessling 3y agoYes, absolutely — probably in the sense that it is the MVP of the "Agents" concept from LangChain, built with duct tape. But it works surprisingly well, because the underlying idea (again, from LangChain) is simply clever.
- EddieEngineers 3y agoCool work @shishirpatil - I'll check it out later & certainly join the discord! Saying this, have you considered you may be responsible for opening the AI floodgates to the internet & are starting the apocalypse of Humankind? I'm only half '/s' here, the speed everything is moving at is insane.
- worik 3y agoThere sign in uses a google form, and is insisting I am logged in to google. What is this? I mean, what is the date? 2005?
- seydor 3y agoIf people are surpassing GPT4 with finetuned Llama-7B, this is the end for NotOpenAI's parade
- jondwillis 3y agoCan’t wait for this to land in langchain!
- fruktmix 3y agoHarambe tried to warn us
- mluo 3y agoAlpaca Llama Vicuna -> Gorilla Chad move
- aledalgrande 3y agoPaper if their site isn't loading https://arxiv.org/pdf/2305.15334.pdf https://arxiv.org/pdf/2305.15334.pdf
- krayush 3y agoThe world works on APIs and this unlocks LLMs to interact with the rest of the world. Great work.
- elexhobby 3y agoNot sure if this is obvious. But its incorrect to ditch on GPT-4. The paper uses self-instruct on GPT-4 to generate the training data on which it is fine-tuned. This paper would not exist without GPT-4. Although they claim GPT-4 can be replaced by any LLM, I'm sure the results would be nowhere as good, and so they stuck with GPT-4.
- fnordpiglet 3y agoIt does make me wonder what the converged fixed point on this technique is. If I fine tune with GPT4 to make model A, which then performs better than GPT4, then fine tune model B with A, at what point does either artifacting or diminishing returns set in?
- elexhobby 3y agoGPT-4 is powerful over a diverse set of tasks. They use it to build a model which is better for a narrow sub-task. Pretty sure the model is sub-optimal to GPT-4 for everything else.
- fnordpiglet 3y agoYeah, but there have been papers being published in general LLMs as well finetuned off of GPT4 instead of humans. Even in the narrow space the question remains. If I build a superior model for task X using gpt4 and it’s superior at X, can I think use my new model to train another model in X and continue to see benefits ?
- elexhobby 3y agoFor another possibility, see https://arxiv.org/abs/2305.15717 https://arxiv.org/abs/2305.15717. The new models may not actually be better - the evaluation may be broken.