LangGraph: Agent Executor
Вставка
- Опубліковано 16 січ 2024
- In this video we will go over how to re-create the canonical LangChain "AgentExecutor" functionality in LangGraph. The benefits of doing it this way are that it will be much easier to modify parts of it to fit your own custom logic.
Notebook: github.com/langchain-ai/langg...
LangChain Agents: python.langchain.com/docs/mod...
This is fantastic!!! Thank you so much for putting this Playlist, very clear and well explained... This is really the best way to use Agents.... Graph is brilliant!
I really like the idea of looking at conversations, actions and functions as a branching graph neural network rather than as a linear based branch.
All sort of lightbulbs are popping in my head about modeling very complex interactions between agents, humans in the loop and data
This is awesome, thank you
Can you throw some light on how LangGraph is different from AutoGen, in terms of capabilities or use-cases it tries to solve for. AutoGen is also muti-agent workflow, which can be branched, cyclic and LangGraph also looks the same.
It will be amazing if the langchain team makes lcel and lang graph available in rust. This sort of continue and end are very intuitive with Rust Enums. The code will also be less clunky. The app will also be faster if smaller parts of chains are executed with gguf models in rust. I am not sure, but i guess we can do that with the ollama integration somehow but a native adaptor would be cool too.
Video sound is too low
Thanks for the video! Is there an advantage of using agents in langgraph vs RAG chains in langgraph?
Is it possible to use AgentExecutor instead, to encapsulate tool running logic into single node?
Is it just me noticing the latency for checking the weather in SF 8+ sec. Been fiddling with multi agents in OpenAI using multiple agents operating on the same thread. My issue is latency all the time. Simply writing a slightly more complex prompt yields fast response times (if that is an important thing). But maybe the whole point here is not to send everything off to a GPT4/GPT4o type model for every call?.
how can I add a function to a node that takes multiple arguments?
can anyone explain to me how does the agent predicts to continue or to finish, i watched the video twice but couldn't find an answer ?
When I try to pass the tools to ToolsExecutor, when langgraph tries to use them, i get the following error :
Attribute Error: 'list' object has no attribute 'tool'.
Has anyone faced this error on ToolsExecutor?
I’m really confused why you’re explaining every line of code with no context, like why not show what the program does first so it will make sense? Should I skip ahead to figure out what it does? Who chooses the format of these and why choose such a strange and unconventional format? Usually coding tutorials show what programs do first.
I agree, I left the video even more confused than when I came in from the blog post.
Also a visual representation of the abstractions, in special the graph itself, would be great
I don't undersdtand... It's all there; at the end, you see the outcome. You can even click on the description of the video to have all the code, cut and paste it and execute it... What more do you want?
This example doesn't work anymore!
why?
Having Graph in name is so misleading. It is rather an agents state machine with transition rules.
you guys are amazing , 95% of import libraries are langchain , sounds to me langchain invented the whole AI industry , wrapping others efforts to build your so called "ecosystem" and get developers locked in , which is not a decent AI startup should be doing .