Has anyone taken the plunge on fine tuning any of the LLaMA 2 models? Working my way through that using the Huggingface “autotrainer” … and have many questions.

Welcome!

Kevin Wang

joined #06-technical-discussion. Also,

babat00nday

joined.

Folks my Vercel Edge-> LangChainStream -> ChatOpenAI chatbot setup, streaming is bugging out and losing conversation mid-stream after I add a system message with 8k tokens mid chat. Has anyone seen this streaming issue with LangChainStream -> ChatOpenAI packages before? My current hypotheses are: 1. There might be ChatOpenAI issues with heavily loaded context? 2. Is it possible that free tier Vercel is causing this weird behavior? 3. Is adding context via system message mid-chat the issue? Would it be better to start a new chat altogether? Any suggestions/intuitions would be helpful. cc: @Hao Sun

Hello friends, Our product launch is due in a few weeks and I need to better understand what frameworks ML engineers use in computer vision ie. Tensoflow, Pytorch, etc. to ensure we support all frameworks for a clean experience. What frameworks do your computer vision models use? We may have to add another layer of abstraction, this will help in our development, thank you!! PS. We (leap-labs.com) are building a dashboard that interprets ML models, extracting what a model has learned directly from the model itself independent of its datasets. Starting with computer vision. DM me for more info!

Any backend people here familiar with DBT?

Curious if anyone's done much hacking with language model prompt frameworks like Microsoft Guidance, and what use cases you've seen work well with those frameworks vs. raw text prompts.

Hey everyone, So we've been digging into ways to reduce hallucinations and make LLM's output more robust and deterministic. We're seeing some crazy good early results in our test runs 🤞 If you're running into any issues with prompt/output quality, just DM and we can chat!

Hi all, we are working on a project for a major AI lab to make safer AI. Our team is establishing a diverse group of researchers passionate about rigorously testing and defining how AI can be safely used. We are adding 10 members (low time, *very high impact*). If this is you, please consider applying here by Wednesday. Thanks.

anyone have a great modern stack relative to AI-gen and/or no-code for standing up iOS and Android apps quickly from mockups to code, etc?

Welcome!

John C

joined #06-technical-discussion.

Hello Everyone, I am looking for simple elegant way of fine tuning diffuser using dreambooth on small sample dataset. Any example you shared would be appreciated. I have no experience with diffuser so the examples on google kind of confusing me. I want to do a simple task on google colab T4 GPU.

Is there anything folks are using to ask ChatGPT (or a similar tool) stuff with a later than 2021 cutoff date? Specifically I'm thinking of trying to ask prompts around VisonOS development for the new Apple headset

Hi @Alex, are you willing to implement something on your end? Or are you looking for some already built tool?

I want to improve my retrieval with my vector database (in my case, PGVector). Sometimes, it comes back to some unrelated articles . My plan is too have an LLM look at the results and decide if its actually a good result but I'm afraid it will hurt speed. Has anyone tried to do something like this - AI filtered retrieval results?

Has anyone played around with Nvidia Nemo Guardrail? Would love to chat about experience &/or if anyone knows similar products out there?

Welcome!

Vladi Delsoglio

joined #06-technical-discussion. Also,

hsyun

and

Allen

joined.

Per user monthly cost estimates for OpenAI model usage. Ouch! GPT-4 is a no-go.

Doing some research and made this comparison of hosted model providers such as beam.cloud banana.dev fal.ai together.ai replicate.com brev.dev gradient.ai, etc.! Anyone tried some of these and has some thoughts or opinions? https://jamesie.notion.site/List-of-hosted-inference-providers-a7129c3825274d1d8128ebc80a56bfec?pvs=4

This message was deleted.

I’ve noticed that the browser version of chatgpt4 is alot better than the playground version which must have something to do with the default system prompt. Does anyone have an idea bout what the initial settings / prompt is for the chatgpt4?

I think they try to keep that a secret.

Welcome!

walkerx

joined #06-technical-discussion. Also,

Msaritzky

and

TV Feeder

joined.

Where do you find the browser in the OpenAI playground. I dont have it

https://twitter.com/trsohmers/status/1707146728783093802 (fascinating. from my co-founder)

Sup all! Has anyone made an embeddable chatbot before? I'm porting my Chrome extension (ShopAdvisor screenshot attached) to an embeddable widget (example screenshot attached from MattressFirm). I've never made an embeddable widget before - any tips/gotchas to consider? Anyone I can ping sparingly if I get stuck? Thanks! 🙏

Welcome!

Ali Mirzaei

joined #06-technical-discussion.

This message was deleted.

Welcome!

Avishaan

joined #06-technical-discussion.

Hello, I am trying to build an Q/A chatbot that can read pdfs along with the ability to read hand written comments. Has anyone tried to do this? Any tips to consider?

Does OpenAI have any API access to the image input feature?