MemGPT: AI Agent 🤖 with Unlimited Memory 🧠 | Store and Load Converstations
ฝัง
- เผยแพร่เมื่อ 25 ก.ค. 2024
- In this video, we will look at the open source MemGPT project that is aiming towards solving the small context window problem that is impacting the current LLMs. We will install the repo on our local machine and customize the personas of both the AI assistant and the human agent.
Let's Connect:
☕ Buy me a Coffee: ko-fi.com/promptengineering
|🔴 Support my work on Patreon: Patreon.com/PromptEngineering
🦾 Discord: / discord
💼Consulting: calendly.com/engineerprompt/c...
Links:
MemGPT website: memgpt.ai
MemGPT Repo: github.com/cpacker/MemGPT
MemGPT paper: arxiv.org/abs/2310.08560
Timestamps:
[00:00] Intro
[01:10] MemGPT Paper
[02:40] How it works?
[04:07] Local installation
[06:12] How to use memgpt
[08:30] How to change Personas
[10:35] How memory is updated
[12:48] Save and Load Conversations - วิทยาศาสตร์และเทคโนโลยี
Want to connect?
💼Consulting: calendly.com/engineerprompt/consulting-call
🦾 Discord: discord.com/invite/t4eYQRUcXB
☕ Buy me a Coffee: ko-fi.com/promptengineering
|🔴 Join Patreon: Patreon.com/PromptEngineering
I just like this guy's voice tbh, not over inducing, not salesy, just genuinely explaining stuff just for the sake of explaining stuff. And without telling me to subscribe like 10x
Awesome vid!
First here much love to you and your content keep it up 💯
Awesome video! Would love another on using MemGPT to chat with documents
Yes
@@engineerprompt I think it would be especially interesting to see how MemGPT could be used to take on the persona of an author, ask and store answers to questions that pertain to the text and guide the user through the concepts in the text. MemGPT seems like the perfect solution for such an experience
Hello. What if i want to have a gpt locally to train it with my local data and not output anything outside in apis ? Also i would like to have the conversations stored locally. Important is that i want my localgpt to have access to my full app code so that it learns from it. What should I use ?
how can i use this chat bot with my local documents ?? can you do for us onther video please a,d man you are amazing thank you so much ! you are better than anyone else
Are we allowed to delete the venv or is it qilling someone ?
You can integrate that into your LocalGPT project, no?
I hope they will lower the costs for API-calls.
I toyed a bit with the open-source code interpreter model a while back using GPT-4 and the money evaporate in front of your eyes.
I can imagine that there is no different outcome when using MemGPT with GPT-4 right?
Yes, at the moment, it's not really feasible for larger projects before of the cost.
What is the barrier to adding this memGpt functionality to your localGPT project qnd specifically using llms like mistral 7b?
I think it’s going to be the model themselves. At the moment, smaller models are not good at following complex instructions and working with agents. Hopefully this will change soon
Thanks for everything. Your are deep tracks my man.
Very Interesting project. I am curious how it will manage in an multi user environment where u chat with a gui like gradio of streamlit. Not really interested in close source solution. Hopefully support for open source llm will come soon. Thanks for showing us in new dev in llm.👍
I think it will take a while for open source models to be used in projects like these. They still lack instruction following abilities. If you notice most of the projects that use agents are based on gpt4. Even ChatGPT can’t keep up with the complexity involved
@@engineerpromptnot so long there was a paper from Microsoft research about extending the context length to 1 million tokens or more. Autogen could be used for open source llm as well. Mistral has already a token length of 8k and 16k on some open source llm also. I think memgpt is Just extending the conversation and cannot for example summarize a book with 100k tokens or more. Its proven that size not always matter. Look at mistral. I think that a federative llms finetuned for multiple tasks working together can beat gpt4....
But how do models with larger context windows perform against the same model with the original smaller window?
MemGPT GPT Engineer combination could be fun 🤔
and really expensive as well 😀
Amazing feature. Could you please create a video with embeddings and seeing how it performs? Thanks in advance.
Yes, that’s coming soon
Thats what we want yet people can't seem to figure that out.
Can MenGPT be integrated with local LLM model?
Yes, watch the video being released in a couple of hours
How you generated the avatar
check this out: th-cam.com/video/V2efVSXSlqc/w-d-xo.html
Thanks, very interesting projects as usual.
Would be nice to be able to run a llm locally for this MemGPT. Maybe you already coverded such tweaks in one of your previous tutorials.
I think its going to take some time before local models are powerful enough
@@engineerprompt What if a more narrow purpose was the goal? Is it possible to reduce the knowledge areas so that only a specific area was needed? For instance, say only cooking knowledge was required, not knowledge of car repair, science, etc. Would this help an LLM running on a local GPU?
It would be very interesting to combine memgpt and Secfillings together🤣
agree :)
That AI generated Avatar was cool!!
(It was AI generated right !!!☠☠☠)
Im working on KEKgpt. It’ll troll all online gpt instances and tell them bad jokes. It’s to teach resilience. 😅
I love it
Hello, Can you please make full video session on How to Integrate MemGPT into Gradio interface.
Anyone work and have done this before please share me, Thanks!
A persistent pickle? Wouldn't you want to save in vector form? Or am I getting confused? I want what it remembers in vector form I think for ease of use. Am I missing something? Ah, it does incorporate llamaindex abilities.
You use pickle to save or keep FAISS vector store. Save_local.. ;-)
Make the waw avatar from the picture with lipsynic. head movements blink eye very realistic. what did you do it with?
I do not recall seeing the @engineerprompt AI avatar in previous videos, it must be new. The voice (or AI voice-over) is the same as in other videos - is it real or also AI generated?
u handsome devil. Your mouth moves almost perfectly.
Any idea when these models, open ai chat gpt 4 / API will become free for all? It feels like back in the day when the internet could only be used by a few rich people. Only bing offers people gpt4, any idea why that is?
Probably never. Once they’re old enough, OpenAI will just “discontinue” (deprecate) them in favor of their newer models. However, open-source models will always be trying to catch up, so I’m sure that by the time these models are deprecated, OSS (open source software) will have already caught up to and/or exceeded their capabilities. But for the time being, don’t hold your breath.
@@georgesanchez8051 OK, will try to master my patience to an expert level then. Keep up the good work. 👍🏻
Please create a UI like gradio or flask for chat bot
My only question: is it free?
No . For OpenAI api , it costs
Thx
ye thobda hta beech me se ye to ise side me lga chota krke
Can we stop saying "increasing the context window"?
None of these functionalities even come close to mimicking the context window of any LLM.
These approaches are simply context-manipulators.
They twist and turn the data that is being fed into the context-window of the LLM.
These guys in the Open-source world (no hard feelings, you are doing an amazing job and we as a community are growing with your contributions) are simply using terms like "unlimited context window" because that sounds fancy... grabs attention... and etc!
How many time do u have to make a vid like this lol.... I swear they never actually work
except if you watch the video, it works perfectly.
what not work ?
it worked for me, fwiw. MemGPT has updated some of their installation instructions, so I modified the installation from what Mohamed demonstrated, but I have it running all day now and with impressive results.
you can believe what you want but im telling you that mother f'er aint real!
I don't need an AI to do all this nice nice apologizing and such. I don't want an AI that does small talk either. I don't think many do. So why show this stuff?
MemGPT is cool but imho this is distracting.