An Introduction to LLM Agents | From OpenAI Function Calling to LangChain Agents
HTML-код
- Опубликовано: 25 авг 2024
- #automation #python #LLM #langchain #chatgpt
In this video, we’ll go through an introduction on LLM-based agents. We'll start with discussing some intuitions about agents in general and their implementation, covering from Python + API implementations, to OpenAI function calling and LangChain agentic implementations. Then we dive into LangChain as a framework to build interesting agents exploring some core features.
📚 Chapters:
00:00: Introduction to the video and the topic of agents.
00:36: Personal introduction and definition of agents as a combination of thought and action.
00:51: Tools and three complexity levels of agents.
01:03: Discussion on the OpenAI's Function API.
01:38: Defining an agent in simple terms and the decision-making process.
02:05: Example of the decision-making process applied to attending a live training.
02:58: Simplistic definition of an agent in the context of LLMs.
03:20: Introduction to LLMs and their basic function.
03:51: Example of LLM output and introduction to tools for real-world actions.
04:19: Discussion on seminal papers on combining LLMs with tools.
05:02: Python functions as tools for LLMs and system setup.
05:36: Introduction to the paper "React" and its contributions to agents.
06:41: Recap of fundamental papers on agents and LLMs.
07:00: Surge in popularity of LLM-based agents and applications.
08:04: Popular agent implementations and their features.
09:14: Discussion on GPT-based agents and their functionalities.
10:01: Complexity levels in building agents and setting up task executions.
11:02: Level one of agent implementation using Python functions.
12:27: Execution of Python functions and the limitations of this approach.
13:37: Introduction to OpenAI's Function API and its usage.
14:43: Detailed explanation of setting up and using OpenAI's Function API.
17:12: Introduction to LangChain as a framework for agents.
17:52: Cognitive architecture and its relevance to agents.
18:53: The routing process in agent implementation.
19:19: LangChain's framework features and core elements.
20:07: Use of LangChain for common tasks and integrations.
21:21: Prompt templating and dynamic prompts in LangChain.
22:25: Output parsing with LangChain and integration with Pantic.
23:09: LangChain expression language for building application chains.
24:21: The agent loop and its key components in LangChain.
25:31: Schema and structured interactions in LangChain.
26:28: Inputs to the agent and the loop structure.
27:19: Discussion on the agent loop code and runtime.
28:15: Tools in LangChain as functions for agents to invoke.
29:37: LangChain's focus on action and real-world applications.
30:38: The future of LangChain and its ease of use.
31:18: LangChain toolkits and integrations for LLMs.
31:31: References for the presentation and closing remarks.
🔗 Links:
- Subscribe!: / @automatalearninglab
- Join Medium: / membership
- Tiktok: www.tiktok.com...
- Twitter: / lucasenkrateia
- LinkedIn: / lucas-soares-969044167
Support the Channel!
- Buy me a cup of coffee: tr.ee/7tYsD-tUu2
- Paypal: paypal.me/luca... 
Наука
this video is a goldmine of knowledge, thank you
kk thanks!
I myself am fascinated by everything to do with this topic and I can’t wait to get started with my own projects. Subscribed.
Nice!
awesome video. I like the balance of concept explanation and high level code explanations! perfect for beginners without dumbing down! Thank you so much
Nice! Glad to hear! :)
Excellent video! Very educational and easy to understand!
thanks!!! :)
Very insightful content. Thanks a lot
Great video with deep insights and high levels. It will be good to combine a second video with runnable codes to visually see the results.
Oh. nice! Great idea! :)
Very good! Thanks.
You're very welcome!
Great video
Thanks! :)
Congratulations! Great!
Thanks! :)
Cannot find level 2 in this video. Level 1 is function in prompt, level 3 is about langchain. what should level 2 be? : )
level 2 was supposed to be openai function calling
At 12:12 in the timestamp, you indicated that you would include the repository link in the description, but I'm unable to locate it.
Check out this repo, the notebooks that start with 1..... should contain all the code you're looking for. github.com/EnkrateiaLucca/oreilly_live_training_agents/tree/main/notebooks
Nice explanation, thank you. I'm interested in your opinion of Autogen vs LangGraph.
I Think LangGraph is langchain's reaction to AutoGen, in the short term future I think we'll see both being used for agent orchestration, however for controlling outputs of agents, Langchain does a much better job, so if LangGraph can allow langchain specific agents to be even more useful, I think it could give it the edge, given that AutoGen gives you more higher level type control of how agents communicate with each other rather than something like chains and llm output control which is what you see with LangChain
@@automatalearninglab thank you for replying. I was wondering, for learning and developing a POC whether I should start with Autogen or LangGraph. Thanks for your input.
Do you have the repo or an example video where you make an agent with optional call functions?
I think my video on the 3 levels of complexity maybe would be interesting to you? I'll make more in the future just about function calling.
Hi Lucas, grazie for the nice content.... would love to hear you opining about in what case would be better to use LLAMA vs Langchain....
So they are not necessarily mutually exclusive, langchain is more about orchestration, llama 2 is more about using a good local model. I'd say that if you just need a simple llm running locally you can use something like ollama which is straight forward API to use llama2 (although langchain also has a pretty simple one). If you want to have like a more complex application that does more interesting stuff like chains running summarization others running like extraction and so on, then something like langchain starts to make more sense.
@@automatalearninglab thank you!
@@TheSardOz Oh! Thank youuu! :)
oh my god 😍
kkk thank youuu
nice vid bro, only one suggestion: the mic quality could be better
Yep noted
The audio is almost unbearable. Probably should fix that.
Yep I did! Thanks for the feedback! ( you are so very right)
im pretty sure u got the wrong mic plugged in 💀
kkkk, maybe. My latest videos now have the proper decent audio, hopefully I won't make this mistake again.