Persimmon-8B(adept.ai) |
Persimmon-8B(adept.ai) |
The success of ChatGPT and my current work has had me thinking a lot about the "product" applications of large language models. I work at Pulumi on www.pulumi.com/ai; it's a GPT-3.5 and GPT-4 interface using retrieval augmented generation to generate Pulumi programs, and user experience is top of mind for me.
(Fingers crossed this doesn't hug our site to death here for the reasons I'm about to explain.)
To be blunt: I have found it surprisingly difficult to find the right tools to host models without dramatically worsening the UX. In theory we should be able to fine-tune a model against our own SDKs and synthetically generated code to improve the model's output and to guard against hallucination when retrieval fails. In practice, self-hosted model serving APIs have really poor time-to-first-token or even completely lack streaming behavior. It's a non-starter to build a product on something where a user has to sit and watch a spinner for a minute or more. I've been looking at the vLLM project with great interest, but haven't found much else.
---
For folks in MLops, deploying models with streaming APIs:
1. Is it mostly accurate that none of the model serving tools created prior to ChatGPT are great for streaming, interactive use cases?
2. How are you currently serving these models as an API and what upcoming tools are you exploring?
For the authors: How does your inference optimization compare to vLLM, or other tools using techniques such as continuous batching and paged attention?
IME the streaming API in text-generation-inference works fine in production. (Though some of the other solutions may be better). I've used it with Starcoder (15B) and the time-to-first-token / tokens per second all seem quite reasonable out of the box.
* scored 18.9 on HumanEval (coding) where Llama2 7B scored 12.2
* was trained from the beginning with a 16k context using a modified RoPe where many models are simply fine-tuned using RoPe to gain longer context windows after the base model has been trained at 4k.
Can anyone share ideas on how important the 2nd one is? Do LLMs benefit from large context windows using RoPe during pretraining?
Weights haven't been released, though.
https://twitter.com/suchenzang/status/1699926157028897078?s=... notes some issues directly comparing the 16k context number. the odd choice of tokenizer means its effectively like a 10-12k model (? ballpark, not calculated)
The article claims 18.9 for the base model, but also claims 20.7 for the fine tuned model.
I'm concerned about the current download's availability - its two URLs to some object storage. I find that these go dark rather quickly for many different reasons (accidentally moving it, bandwidth limits, deleting it later, etc).
I'm curious if there's a reason it's not also hosted on huggingface? I'm not saying they're the best place, but redundancy is good, most models have entries there, they have a very good cdn, and isn't as likely to go dark accidentally.
1) In the results table, Llama2 base is being compared to Persimmon base and finetuned, and only the latter performs better. Would a comparison to Llama2-chat be possible/fair?
2) The Llama-2 numbers for MMLU in that table seem different from those in the HF leaderboard and the Llama-2 webpage presentation. Is it the 1-shot variant that is different or are these measurements not 100% standard and reproducible?
The numbers are different because the measurement is different. The blog post explains that we sample from the models and expect answers rather than relying on perplexity measurements.
The inference code is shared as a proof of concept, it is not meant to be a production ready deploy. Also worth noting that not all LLMs are used to produce text which is read by humans.
It’s funny you say production, because all of the errors I ran into suggest the container is expecting your production architecture.
My advice is stream first then make synchronous convenience wrappers on top of that. Also, lean on community standards for PoC. I’m guessing your investors are interested in making this scale as cheaply as possible, but that is probably the least important feature for people evaluating your model’s quality locally.
I am an AI novice but why can't they automated this with AI? I thought the whole point of these tools was to automated tasks that are error prone and require lots of attention to details. Computers are great at that kind of stuff so it's surprising they haven't applied AI techniques to automate parts of the AI pipeline like converting code from Python to C++.
Could someone briefly explain what this means? multimodal as in picture, but if unused then presumably that part is somehow untrained...so it wouldn't know what to do with the picture?
The embeddings form the vocabulary of the model. The vocabulary "namespace" has 70k empty slots so you could introduce your own tokens and train on top of that, where token = some patch of multimodal data.
The AI race to zero must be accelerated with $0 free models and less control from gatekeepers such as ClosedAI
From my understanding, you'd have to repeat the experiment isolating each variable to see what difference each one actually makes, no?
It's safe to assume they're worse at every task than larger models, so I wouldn't look at use cases in terms of what tasks they can do compared to larger models.
But what's good about them is they're smaller so they can run on smaller and cheaper hardware. So an example would be to fine-tune and then run on some sort of local user device rather than in the cloud. This might become more practical in the future as hardware improves.
Perhaps for basic code completion and simple writing tasks?
edit: not sure why op is getting downvotes, this is a very reasonable question imo; maybe the characterization of kernel compilation as "AI" vs. just "software"?
The whole thing seems obviously amenable to gradient based optimization and data augmentation with synthetic code generators. It is surprising that no one is pursuing such approaches to improving the optimization pipeline in kernel compilation/fusion/optimization because it is just another symbol game with much better defined metrics than natural language models.
If it was necessary for some reason... Running a language model to keep something like this is sync over long term training and iteration would likely be more expensive than a developer's time AND block the researcher in a verification loop on the output which still probably needs to be checked by the developer (they could be the same person which will just deepen the frustration they experience).
The use of a lot of garbage accounts in this thread and lack of model details also looks pretty shady...
Back to the question at hand. How exactly do AI companies plan to build AGI if they can not optimize the AI development loop with their tools and techniques?
you are confusing (syntactic) validation from verification. verifying code is an incredibly hard problem.
You can get a lot of value out of a models even if they are not capable of AI development because most people aren't doing things that are as complicated as AI development.
it quite literally is useful for consumer usecases though.
For example, one consumer usecase that is being used by a lot of students right now is cheating on their homework.
It is right now being used for all sorts of consumer things like that.
Also, if you have an opinion you can just say what your opinion is. You don't have to hide it behind questions and alt accounts.
You can just have an opinion and say it.