- Видео 24
- Просмотров 132 060
Tosh Velaga
США
Добавлен 7 ноя 2018
Self taught developer. I've been programming for 6+ years.
On this channel we cover the intersection of full stack development and advances in AI, whether they be transformer or diffuser based.
Feel free to reach out to me if you have any questions :)
On this channel we cover the intersection of full stack development and advances in AI, whether they be transformer or diffuser based.
Feel free to reach out to me if you have any questions :)
DBRX is the best open source model
Paper: www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm
GPT4 Architecture Leak: www.linkedin.com/pulse/supposed-leak-gpt4-architecture-alvaro-duran-tovar/
hosted inference: huggingface.co/spaces/databricks/dbrx-instruct
To run inference you can use diffusers or vllm, but you're gonna need 4 A100's.
GPT4 Architecture Leak: www.linkedin.com/pulse/supposed-leak-gpt4-architecture-alvaro-duran-tovar/
hosted inference: huggingface.co/spaces/databricks/dbrx-instruct
To run inference you can use diffusers or vllm, but you're gonna need 4 A100's.
Просмотров: 563
Видео
The problem with the GPT Store...
Просмотров 8779 месяцев назад
The GPT Store just launched publicly a few days ago. Already there are 3 million GPTs, a majority of which are either spam or have less than 10 conversations. In this video we talk about why the GPT store is not a great platforms for developers/entrepreneurs to build on. We also cover some simple steps OpenAI can take to improve the experience.
Recreating Google's Viral Gemini Demo using OpenAI GPT-4V
Просмотров 2,5 тыс.10 месяцев назад
I recreated part of the recent Gemini demo, specifically the part where the AI identifies which cup is hiding the ball. Here's all the code: github.com/toshvelaga/gemini-cup-demo-gpt4 Gemini Demo: sundarpichai/status/1732433036929589301 Fast forward to the 2:30 minute part of the demo to see which part I am recreating here.
Run Stable Diffusion as an API on AWS SageMaker
Просмотров 4,5 тыс.11 месяцев назад
Run Stable Diffusion as an API using AWS s3 and AWS SageMaker. Here's the article that this video is based on: www.philschmid.de/sagemaker-stable-diffusion Code: colab.research.google.com/drive/15T52oXb1huqUD8sGpzo5aoqiYzuG4INk?usp=sharing Interested in building an application on top of Stable Diffusion? Feel free to book a free consultation here: cal.com/toshvelaga/stablediffusionai
Running Stable Diffusion on Google Colab
Просмотров 940Год назад
Tutorial on using the diffusers library from hugging face to set up common pipelines including text to image, image to image, upscaling, and controlnet. Everything is free to use on Google Colab. Be sure to change the runtime from CPU to GPU. Colab link: colab.research.google.com/drive/16Yrf8YzHl_7zaiFgtQsD-rfIkOLum3-L
Running Automatic1111 Stable Diffusion Web UI on a GPU for Free
Просмотров 14 тыс.Год назад
Tutorial on how to use Automatic1111 Stable Diffusion Web UI using Sagemaker Studio. You get 4 hours of free GPU per day and 8 hours of CPU. Code to run: git clone github.com/AUTOMATIC1111/stable-diffusion-webui.git conda install glib=2.51.0 -y python launch.py listen xformers enable-insecure-extension-access ngrok NGROK_AUTHTOKEN
Claude API on AWS Bedrock
Просмотров 6 тыс.Год назад
If you don't have access to Claude, using AWS Bedrock is the fastest way to get access to Claude 2 and Claude Instant. AWS bedrock is an offering from Amazon that lets you run several foundation models including Claude, Stable Diffusion, and Cohere. Running Claude on AWS Bedrock is a good option because it's all serverless meaning that you only pay for what you use. Collab with code: colab.rese...
Cloudflare AI Inference & AI Gateway Tutorial
Просмотров 3,7 тыс.Год назад
Cloudflare just released their AI inference offering along with a partnership with Hugging Face. In this tutorial I show you how to use their REST API for inference as well as how to use the AI Gateway product, which I am most excited for. The AI Gateway product lets you cache, rate limit, and log errors/responses/tokens for inference endpoints from Hugging Face, OpenAI, Replicate, and Cloudfla...
Google Palm 2 Overview and Maker Studio Examples
Просмотров 200Год назад
You can apply for access to use Google's LLM PALM 2 which is on par with GPT-3.5: makersuite.google.com There are different model sizes: Gecko, Otter, Bison, Unicorn. You only get access to Bison. Palm is 75 tokens / second and 8000 token context window for bison-text and 4000 token context window for bison-chat - A fine tuned version of PaLM is also being used for medical questions, called Med...
Devs are switching from VS Code to Cursor and you should to
Просмотров 14 тыс.Год назад
Cursor is a fork of VSCode that lets you use GPT-3.5 and GPT-4 directly within your editor to write code, chat with files, and create documentation. Demo of how to use Cursor.so. Use ChatGPT directly within your codebase to write and analyze code. Probably going to be using this over VS code now since it's so damn good Edit: tried to go back to VS Code and can't. #openai #chatgpt #vscode
MongoDB Vector Search Tutorial
Просмотров 9 тыс.Год назад
MongoDB recently released their vector db offering. If you're trying to develop an AI based product with noSQL, their Atlas Vector Search is super convenient since you can use the DB to store user data alongside embeddings. Pros - Super easy to set up - If you're already using Atlas to store data, no need to configure an extra db just to store vector embeddings - Free tier will last for a while...
MetaGPT - Multi-Agent Framework with GPT-4
Просмотров 7 тыс.Год назад
MetaGPT is a framework that uses GPT-4 to automate multiple roles within a software company. For example product managers, software architects, project managers and software engineers. It writes code, documentation, user stories, competitive analysis, and creates diagrams. Github: github.com/geekan/MetaGPT #ai #gpt4
Llama 2 Beats Every Open Source AI Model
Просмотров 624Год назад
Meta just released their latest open source model Llama 2. The model with 70 billion parameters ranks the highest on several evals compared to other open source models including Falcon and MPT. Check out the benchmark results here: ai.meta.com/llama/ Check out a16z hosting the models on Replicate: replicate.com/a16z-infra Try Llama 2 7b and 13b on Typeblock: www.typeblock.co/resources/llama-2 U...
OpenAI Code Interpreter's Most Practical Use Case
Просмотров 1,5 тыс.Год назад
In this video we go over how to analyze data from a database in Code Interpreter. Note: I've also tried manipulating images, such as changing the colors and turning them into gifs. The quality of the results was not really good enough to actually use. Let me know if you been able to manipulate images using Code Interpreter for a legit use case. You can also upload videos and audio files, howeve...
Gorilla LLM: Write APIs using LLMs
Просмотров 2,7 тыс.Год назад
Gorilla LLM is an LLM trained on scraped data from Hugging Face, Torch Hub, and Tensorflow Hub. The model can write AI APIs based on this data. It's commercially usable and open source. In this video we talk about how it works, go over two examples including object detection, and talk about weaknesses. Github Link: github.com/ShishirPatil/gorilla Collab: colab.research.google.com/drive/1qWvsUt-...
Using GPT-Engineer to build entire apps [Demo + Explanation]
Просмотров 16 тыс.Год назад
Using GPT-Engineer to build entire apps [Demo Explanation]
OpenAI API GPT-3.5-TURBO-0613 Function Calls [LATEST VERSION]
Просмотров 3,4 тыс.Год назад
OpenAI API GPT-3.5-TURBO-0613 Function Calls [LATEST VERSION]
Streaming OpenAI Chat Completions Using React and Node JS
Просмотров 13 тыс.Год назад
Streaming OpenAI Chat Completions Using React and Node JS
How to use chatGPT directly in VS Code with Code GPT
Просмотров 14 тыс.Год назад
How to use chatGPT directly in VS Code with Code GPT
Next.JS 13.4 When to Use Client Components vs Server Components
Просмотров 2,2 тыс.Год назад
Next.JS 13.4 When to Use Client Components vs Server Components
Next.JS Get and Post Requests Using the App Directory
Просмотров 1,1 тыс.Год назад
Next.JS Get and Post Requests Using the App Directory
Connecting Langchain to any SQL Database Demo
Просмотров 12 тыс.Год назад
Connecting Langchain to any SQL Database Demo
🎯 Key points for quick navigation: 00:00 *Demonstrates how to set up readable streams using the OpenAI Chat Completion endpoint with React.js and Node.js.* 00:28 *The app streams responses word by word for improved UX, instead of waiting for the completion of the entire response.* 00:57 *On the server side, the OpenAI Node.js library imports and uses GPT-3.5 turbo to handle chat completions with streaming enabled.* 01:39 *The response data is processed in chunks, converted from a buffer to a string, split by new lines, and further handled to send to the client.* 02:32 *Chunks are parsed to remove excess data and extract content, which is sent to the client.* 03:27 *Client-side React code uses the Fetch API to post requests to the server and pipe responses through a TextDecoderStream for readability.* 04:08 *The setup requires a Node.js server endpoint and a React front end; links to the code are provided in the description for further reference.* Made with HARPA AI
Great, Thumbnail is completely misleading. Thought i would see how can we connect sqlserver from colab using langchain. You never touched it!!!
Is it safe to connect lang chain with our production database
Now, is it still available? I would like you to come back and review it again. Thank you very much.
Isn't this doing double work?
hey i made this video a while ago, since then openai has changed their node library and how it handles streaming. I would copy the code they have for streaming responses here: github.com/openai/openai-node
is it possible to add the controlnet extension as well because i can't seem to get it to work
How to make the same process with SadTalker?
ChatGpt subscription is not the same as Open IA subscription, to use CodeGpt you NEED an APIKEY this is from OpenIA.. do not confuse people.. thanks
Can we install mongodb vector locally? I have the mongodb community version installed locally, does that support this vector embedding?
Hey last time I checked a few months ago it was only available through their hosted atlas offering. Just did a quick search and wasn't able to find any updates so tbh not sure.
Thank you for the well explained video.
thanks, bro, I thought this was something complex and only Vercel could do this. Now I have also created my own streaming solution
Amazing
Min 1:00, give it some time to "THINK"? Tell me you're on your 40s without telling me you're on your 40s :D
Haha this was so long ago I have no idea how long it took
Hi Tosh, fantastic video, thank you. I was able to get up and running. I'm now trying to do this with lambda labs so I don't face the same time limitation, but the conda installation (step 2) doesn't work there. Says conda command not found. Any tips to get around this?
the code is outdated already, and not reproducible.
Hey, have you found any alternative to Ngrok? It frequently ran out of request limit lately and requires us to upgrade plan.
You may want to mention that you need a credit card
Hey Tosh, for some reason this code doesn't work for Azure open AI api. Can you tell me why?
can't open file '/home/studio-lab-user/launch.py': [Errno 2] No such file or directory. anything missing?
How to return SQLQuery from the chain just wanna get that :(( any idea ?
getting this error ERROR: Could not install packages due to an OSError: [Errno 28] No space left on device: '/home/studio-lab-user/.conda/envs/studiolab/lib/python3.9/site-packages/torch/distributed/_spmd/__init__.py'
does this still work???
Uptalk alert 😂
Awesome as usual @tosh so great to see content from you again partner.
Thanks so much Rich :).
hello amazing vid. I got stuck at the end of python launch.py command "Traceback (most recent call last): File "/home/studio-lab-user/stable-diffusion-webui/launch.py", line 48, in <module>" It´s sound any solution to you?
Thank you so much, Tosh, for the tutorial. It was perfect.
ngrok is no longer free. Could you explain how to do it with a similar free service like zrok? thank you
Hey I think ngrok is still free, I just checked their website. Where does it say they got rid of the free tier. If they did I gotta find a new tunneling service haha :)
@@toshvelaga Since January 2024, it has a 1GB/month limit, which for me runs out after using the notebook once or twice. And then it asks you to subscribe or wait for next month to be able to keep using it for free. I heard about zrok, but have not been able to make it work, and decided to ask you (:
ahhh i see thanks for clarifying. that sucks i guess ngrok is desperately trying to make money. I havend heard of zrok before but will check it out@@gianluca-aries
I am encountering an error "Configuration is not a constructor"
@Tosh, great video, but Im curious to know how are you setting the URL to the DB?
This doesn’t work anymore
Can you please make a new updated video on how to install stable diffusion because it doesn't work for me anymore, when I write "python launch.py --lis..." an error appears, I believe it's not just me that this error is happening to . This happened when I decided to reinstall it because I had some buggy Lora, so I did the same procedure as in the video and it didn't work anymore, I believe it was because of the stable diffusion version, I don't know :(
this is what appears to me: Traceback (most recent call last): File "/home/studio-lab-user/stable-diffusion-webui/launch.py", line 48, in <module> main() File "/home/studio-lab-user/stable-diffusion-webui/launch.py", line 44, in main start() File "/home/studio-lab-user/stable-diffusion-webui/modules/launch_utils.py", line 465, in start import webui File "/home/studio-lab-user/stable-diffusion-webui/webui.py", line 13, in <module> initialize.imports() File "/home/studio-lab-user/stable-diffusion-webui/modules/initialize.py", line 36, in imports shared_init.initialize() File "/home/studio-lab-user/stable-diffusion-webui/modules/shared_init.py", line 40, in initialize from modules import styles File "/home/studio-lab-user/stable-diffusion-webui/modules/styles.py", line 9, in <module> class PromptStyle(typing.NamedTuple): File "/home/studio-lab-user/stable-diffusion-webui/modules/styles.py", line 11, in PromptStyle prompt: str | None TypeError: unsupported operand type(s) for |: 'type' and 'NoneType' (studiolab) studio-lab-user@default:~/stable-diffusion-webui$
I get the error TypeError: unsupported operand type(s) for |: 'type' and 'NoneType' after launching, what should I do?
I got this as well. Any luck fixing it? @toshvelaga can you help please? it's possible it needs python 3.10 but sagemaker won't allow that...
I'm getting an out of memory error with ml.g4dn.xlarge when trying to make the call to the endpoint. The notebook instance is ml.t3.medium but I don't think there's any big loading to memory happening in the notebook during the call. I think the error comes from the endpoint instance itself and it's not enough for stable diffusion v2.1.
Awesome video, my friend. Thanks for this!! I made it all the way through but when I request the URL in a browser it showed me the 1-time "This is a free app... do this if you don't want your users to see this page, etc, etc, click here to go on to your tunneled app" page and when I clicked to continue (and when requesting the url on subsequent attempts), it just hangs and then eventually gives ERR_TIMED_OUT. Any ideas?
Nevermind I'm an idiot!! ngrok-free DOT app was being automatically blocked by my ISP - just had to whitelist it and all is operating smooth as can be. Awesome video!! Only complaint (of the whole setup, not your vid) is the tiny storage space available in SageMaker. I'm assuming it can be done with an s3 instance, but is there any way to mount a network drive I have running in my LAN and specify a111 look there for models... or loras... or at least store training datasets and/or output images there? Or what about a different cloud storage (other than s3) like Google Drive, Dropbox, etc?
Hi I wanted to know, can I create this similar index structure without using the cloud version of mongodb atlas .... And just using the plain desktop version of mongodb ?
Hey unfortunately last time I checked it only works for atlas
someone please tell me what theme this is and what icon pack
the gui is a dogs breakfast IMHO. It's very simple which is a good thing, but there is so much jumping around between screens on their web gui. if they just make the gui reactive and real time then you would not need to jump around so much and the changes would come to your screen without clicking around.
Hi @Tosh i am trying to connect mssql it is not working can you help me
what about controlnet?
Yes it is wonderful, and no, you should not if you are Python people, pylance not gonna work with Cursor due to license problems. This got me well.
Dude love your content and it was super helpful and most importantly worked. But I am having difficulties running restarting the same project again.
Whenevr I run this command python launch.py --listen --xformers --enable-insecure-extension-access --ngrok NGROK_AUTHTOKEN... It gives me the error python: can't open file '/home/studio-lab-user/launch.py': [Errno 2] No such file or directory
hey there thanks for the kind words and apologies for the late reply, would you be able to type the following into the terminal: ls ls will list all the files and directories. Can you check that there is a launch.py? If not you either don't have the files or are in the wrong working directory
@@toshvelaga Thanks man...... I was able to fix the issue. It was like you said I was in the wrong directory. Btw I am trying to run Comphy ui on Sagemaker Studio Labs can you please make a guide on that or guide me to some resources
@@rishabhsingh1406 Hi. I've been getting the same error. Think you can explain the steps to work around it in laymen terms?
I have this problem in the process of using, please help me to see what is going on. ERR_NGROK_725 Network bandwidth exceeded This ngrok account has exceeded its network bandwidth. If you’re the developer of this page, you'll need to upgrade your plan. If you're a visitor of this page, contact the developer for more information. Used to be fine. Did ngrok change the rules?Probably only 1G of network bandwidth is available..It used to be infinite??
ty tosh
I got ERROR: TypeError: Cannot read properties of undefined (reading 'on') when i return the response i dont get a .data object i jsut get this: Stream { iterator: [AsyncGeneratorFunction: iterator], controller: AbortController { signal: AbortSignal { aborted: false } } }
How expensive is it to set up a AWS cloud computer
im sure they will make changes but i already knew it would be spam central 😂...Also im deep in open source multi agent systems,so im not crying over open Ai
hell yeah, love the open source work! What are some of the open source projects? Would love to check it out
@@toshvelaga .Working with bot swarm teams and getting them to do things. Having them interact with other LLMS
thanks! will this solution work with a backend in node/express, a frontend in react? (no next.js)
Yes absolutely, you can think of Next js as being a wrapper built on top of React so it will work :)
Imagine being pissed cause nobody discovers your gpt because of too much competition. What did you expect? Obviously it's your responsability to promote your app, just as you'd do with any other kind of app. Ffs😂
ughhhhhhh, bro...no offense but i think you've missed the point of this video Just to remind you all OpenAI has to do to make the GPTStore a good experience for users/developers is to copy what works for other app stores, i.e. an approval process and user reviews
Definitely, I've message Logan from OpenAI about the spam they're looking into it. Have high hopes they can turn it into a great experience. They're super good at listening and implementing user feedback. Also thanks for the very thoughtful comment :)
Love it! Could you maybe next time do a video that talks about gemini(is there a store?) and/or update us on this matter of GPTstores. Thanks)
Sure, Bard/Gemini does have extensions atm but they are only limited to other Google products like Google Maps and RUclips. They haven't officially launched their own store for extensions yet :)
Quality! Great video! ❤
Thanks Rich! Appreciate the kind words!