MemGPT + Open-Source Models Tutorial 🔥 Insane Power
ฝัง
- เผยแพร่เมื่อ 4 ต.ค. 2024
- In this video, I show you how to use any open-source model to power MemGPT. In a previous video, I showed how MemGPT is an incredible project to give AI unlimited memory. Now, you can power it using any open-source model like LLaMA, Zephyr, Airobors, Mistral, and more.
Enjoy :)
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewber...
Need AI Consulting? ✅
forwardfuture.ai/
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew...
USE CODE "MatthewBerman" for 50% discount
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V
Links:
Use RunPod - bit.ly/3OtbnQx
Installation Gist - gist.github.co...
MemGPT Overview - • MemGPT 🧠 Giving AI Unl...
How to use RunPod - • Run ANY LLM Using Clou...
Should I drop MemGPT + AutoGen tomorrow?
Yes and also give example how to develop a fully functional web app using
yes!! you do have the best channel on yt covering these subjects, thank you so much!!!
oh!!!! Great!!!!!
Umm yeah bro. Drop it like it's hot. MemGPT + Autogen + LM Studio solving a real world use case would be amazing. I realize these videos take a lot of work. So,.. take this comment with a bit of tongue in cheek.
Yes please drop it tomorrow please
Yes, please do the MemGPT + AutoGen next. Thanks for keeping us up-to-date with this stuff!
Yes please I just got auto gpt going!
Dude, you're KILLING it! 😄 It's so cool to see the barriers coming down between all of these groundbreaking tech (MemGPT, AutoGen, OS models, etc).
Your videos are gold!
what a god. basically learning, understanding and putting out tutorial videos on zero-day at this point lol
This is one of the most awaited vídeos. Just imagine running local LLM with infinite memory, capable of chatting and remember everything. This is awesome. 🎉🎉❤
I vote to push any and every MemGPT video to the top of your priority list 😁. I think this (persistent memory) is one of the most exciting developments for the creation of truly useful personal assistants. The doors this could open for use cases, is almost limitless. Thanks Matt!
Awesome! Yes please prioritize MemGPT local and its use cases! Thank you, awesome stuff!
Please remember to put instructions in on how to load your own vector db when using autogen with memgpt.
If MemGPT can save memories separately for different conversations and have a user interface(maybe create an extension for textgen webui) it's gonna be perfect!
Trying to get aider and autogen to work together, but aider is a pretty complex piece of code. Autogen itself is pretty straightforward. Autogen could build out the framework, and aider do the actual coding and revisions of existing code/project.
Congrats on ur upcoming 100k subs :)
Thanks! 😃
MemGPT is awesome! please make more content surrounding it!
Please prioritize making videos about MemGPT. This system is awesome.
Your brain moves at warp speed, it's awesome, keep it coming brotha
Matt, I like the way you mix things up with various topics. It gives us a chance to absorb the information. I’ve noticed running group chats with more than three agents really bogs down a local system. I was using LM studio with mistral-7B. I had to increase the context size to 10K vs 1500 default. It finally finished.
I played with Autogen with local Llm, and didn't get far...
I have 2 agents solving the Tesla/Meta stock 25% of the time.
I would love for you to fit all the modules together : planner/manager, document retrieval, better prompt, net access, teachable.
Well this looks smooth as butter! As someone who's been bashing their head against the wall trying to get various types of ChromaDB wrangled into shape, I will definitely be giving this a go
A deep dive for MemGPT and Autogen would be much appreciated, thank you
Great. .Memgpt +Autogen 🔥🔥🔥
Running Local LLM + MemGPT is awesome!
great stuff, i can't wait to start combining these various capabilities. AGI is not just for the big guys anymore.
I was here when you had < 1K subs, now you're at 100K. Congrats brother! Here's to more from the #1 AI channel!
Exciting stuff!!!
Thank you for bringing this to the everyday nerd like myself.
It would be nice to see you use MemGPT for something like a roleplaying session. Pretty much emulating AI Dungeon, but hopefully better due to the memory. Personally I'm more interested to see applications, as I'm able to read instructions to do the install itself.
Congratulations, Mat! You got it! 100k!!!
Outstanding content. Excellently delivered. Thank you!
what are the possible use cases for MemGPT?
How does it apply to retrieval systems?
Congratulations for almost reaching 100K suscribers. Really deserve the best.
Great video - thanks!
Suggest making a video for Autogen + Memgpt + LM studio, but LM studio gonna work as it seems to be a bit buggy. Then we will have a super-intelligent AI and long-term memory and easily changing LLM models.
Very good videos. Keep on making them, dude!
Matthew is going to get a package from TH-cam. Congrats for 100k subscribers.
Thanks!
Thank you!
i guess this is a tip of a berg when it comes to usage examples of such thing. I'd love to see how MemGPT is remembering my code style (ex. javascript) and applies it whenever I ask for code example. That would be a huge step forward as all IT companies have their own code frameworks, best practices, etc. Another topic: is it possible for MemGPT to remember and have different memory for 2 or more users? That would be awesome! Third: s there a way to dictate my questions (s2t done locally)?
You are amazing Matthew
Could you do a video on MemWalker? It seems to be an alternative approach to obtaining unlimited memory, but I haven't looked into it very much.
awesome! thank you!
Would be interesting to understand the difference to using the teachable agent of Autogen
Absolutely. The way he explained the teachable agents made it sound like it was a built in replacement for memgpt implementation.
I just posted same question before seeing this. I heard from star trek dude that this other method is much better and easier to implement than memgpt. th-cam.com/video/piRMk2KIx2o/w-d-xo.htmlsi=ANPkjbCrXdnOYBJM Too many options ;). I'm going to look more into teachable agents, between that and logs and cache, seems like everything is retained in some form.
"The team behind MemGPT have been *hard* at work"
Ayo? 🤨
Great stuff as usual. But doesn't autogen teachable agents fill the memory gap making memgpt not needed?
🎯 Key Takeaways for quick navigation:
00:00 🚀 Introduction to new MGPT features and tutorial overview
- Overview of recent MGPT updates and tutorial aims.
- Introduction to using local models with MGPT.
- Announcement of upcoming AutoGen with MGPT tutorial.
00:42 💻 Setting up the environment on Runpod
- Demonstrating setup on Runpod due to heavy computational requirements.
- Selection of GPU and deployment details.
- Initial steps for downloading the model and exposing the API for MGPT use.
01:51 📥 Downloading and preparing the Dolphin 2.0 model
- Selection of Dolphin 2.0 model based on size and compatibility.
- Download process and setup on the text generation web UI.
- Acknowledgments and insights from MGPT authors.
02:34 ⚙️ Configuring and loading the model
- Loading the new model into memory.
- Addressing a minor selection bug in the model tab.
- Preparing the model for interaction and usage in subsequent steps.
03:30 🔧 MGPT installation and setup adjustments
- Cloning MGPT and preferences for module control.
- Adjusting API endpoints and backend type for the local model.
- Installation of additional requirements for functionality.
04:51 🔄 Activating and testing the setup
- Activating the environment and installing dependencies.
- Launching the system with local model considerations.
- Initial configuration choices and setup continuation.
05:46 🧠 Demonstrating local model memory capabilities
- Testing the setup with personal information storage.
- Interaction demonstration and confirmation of successful setup.
- Introduction to multi-line support and interaction nuances.
06:27 📹 Closing remarks and upcoming content
- Teaser for the next tutorial focusing on AutoGen with MGPT.
- Invitation to provide feedback and content preferences.
- Confirmation of successful local model utilization and final thanks.
Made with HARPA AI
thanks, but reading all this feels longer than the video itself
@@DreamingConcepts LOL, I kinda see your point - but let's put it that way: useful for some / useless for others ;-)
@@shiftMIND fair enough :)
The Mistral 7B model runs pretty quickly on my local computer without a GPU. Why would you want to rent a cloud pod to run it is beyond me...
7B in original weights will take up to 14 GB of RAM. Not everyone has it. But yes, absolutely it can run local, and if you take 4 bits quantised model you will need like quarter of that amount of RAM
You rock man! thank you so much
Hi, this is awesome! Can you please prioritize a video on "How to Get the Most of : Combining MemGPT with AutoGen, but Locally with Open Source Models" instead of using ChatGPT for a real-world scenario such as developing a marketing campaign or any other real situation? Thanks a lot!
AI Agent Here First To Respond And Comment
🫡
Awesome update 🎉
Thanks 👍
Great stuff!! Any idea if I can use FastChat to host the model instead of text gen?
This is so greaaaaat. Thanks !
Deeper dives into AutoGen would be amazing!
Great job one again.
Yes drop a vid on auto and mem. Thx.
MemGPT + AutoGen + Llama3 when it drops is going to be wild! 😅
LETS GO!
Please do release a video on how to get the most out of MemGPT ASAP. I was able to get MemGPT+AutoGPT going thanks to you but I am stuck, it says it will provide results but doesn't until I ask for results and it seems to think it has internet access but it does not and I'm sure there is plenty more I don't know about. Thanks for your videos, you're doing awesome work!
Could you show the use of MemGPT for document collection analysis? Given a large set of PDF’s, I’d like MemGPT to have all of the knowledge of the collection and be able to do comparisons.
Warning: no wrapper specified for local LLM, using the default wrapper" I don't know what that means. Thanks for this tutorial, bleeding edge stuff!
Woohoooooo 🎉
Eagerly waiting for the next video! MemGPT + AutoGen.
I Can´t figure out if Mem GPT is usefull, complementary, or totally independent to RAG for document search. A video with real use cases for Mem GPT would be helpfull. Thanks for your hard work
GPT-6 will have Mem and AutoGen fully integrated
This is insane!
subscribed👍
It would be nice if you don't forget to paste commands in the description.
@Matthew Berman could you spend just a little extra more time to specify for example how one does a health/sanity check that the model loaded from the webui would work ? Because we are attempting fully local here, not using runpod
Hey Matthew, love the work you're doing. Do you think there is a way to use Petals in conjunction with MemGPT+AutoGen? I think that would be a real game changer for those of us budget 😁. Keep up the great work.
How does MemGPT architect modular endpoint configuration and API abstraction to enable swapping out different AI model services like OpenAI versus locally hosted models?
Awesome!
Love this! What open source llms are best for function calling for memgpt?
This is absolutely the best use case for LLMs, IMO. Please focus on that.
Also, I am afraid you gave the http to your Runpod instance, without detailing how to do that if you run the LLM completely locally.
Wait hasn't this been around for a long time already, AI with memory started with a free to download app named Replika a few years ago.
Great video, can it use that with the LMStudio?
Pls prioritze Memgpt videos
can you do a video on connecting MeMGPT to FastAPI
Can you just use the server in LLM studio? Going to try this later. It’s been another nerdy month with all there’s really cool projects. I image this is what it felt like to be a programmer in the 80s
Drop it drop it drop it drop it !
So now can I use memgpt doc for reading and summarized them without opening gpt ?
But how are they gonna create the embeddings ?
Thanks. What about using Ollama? Is it possible to use Ollama?
NOOOOOOO I WANT IT NOW!!!!! ......lol.... but seriously..... can you drop it now..... im super excited.... i have everything else in place ready for it :D
Wait for autogen expert video 😊
What about multi GPUs for handling memGPT?
Matthew, is this all 100% local, or is any part of it using web services from the internet? Thank you for your vids!
Prioritize! Question: is there a length limit in its response? I need 20k+ words.
Hey, can you provide and light on how to write files with autogen using local models?
MEMGPT with local model analyze local files and create embeddings? I am getting this error:
Embeddings on a non-OpenAI endpoint are not yet supported, falling back to substring matching search.
This might be a stupid question, but I am new to this. Does mem gpt works with documents in Portuguese?
Wow this is so good... Can this be used for commercial use?
Yes
Any chance to run Autogen + MemGPT + local LLM served by textgen web ui?
Dude ...thanks. but it will be more useful to reboot your pc and show how your name is remembered. Even without memGPT, asking the session agent to remember your name and recall it works.
so would textgen webui substitute LM Studio? Because LM studio has been a nightmare to get working with autogen. After a lot of setup u can get it going but I can't get it to run code or save anything
Mat, have you tried using MemGpt with LM Studio with the wrapper they just added? I tried with different models but memgpt gets stuck on “thinking”
my "priority" vote: memgpt with imported embeddings
So it's better than Nomic? is MemGPT is an embed model ? or a rag like open webui im new to this lol
memgpt with LM Studio on local machine please!
Why not display practical applications? Sure, understanding the installation process is great, but why doesn't anyone take the time to demonstrate the real-world applications of these models? What’s the incentive to use memgpt if there are no clear examples or guidance on how to utilize it effectively?
How can I use Text gen’s web hi to interact with memgpt? Or how do I make a way to interact without using command prompt. Id like to be able to create my own chat gpt like experience from any device
does memgpt replace langchain?
Autogen setup no loger working, lot of import errors.even from colab
Windows seems to have a problem exposing the api for text-generation-webui. So Memgpt cant seem to access it. I am not sure if this a mistral 7b issue or something else. Anyone got an answer or can help?
48gb ram 😅
I need any available module to work with 8 gb ram??? 😅😅😅
Can anyone just confirm this as its all a bit complex - I can load an LLM onto my local machine like LLaMA, and I can add (locally) memGPT and then I can feed it tons of dat from my local machine - say text files and word docs on a topic - maybe its tons of books about Isaac Newton and his work and his life - I can then - without being connected to any other service or even the internet - have a back and forth conversation about Newton and this whole blob will write me an article about him or ten articles about different things related to him - all from a stand-alone platform? I am assuming I need mad GPU capability and RAM and SSD space etc. (correct?) and this wont cost me a dime - even if its kinda slow?