Using LangChain with DuckDuckGO Wikipedia & PythonREPL Tools 

totally agree the issue with this is how to put it in a colab for people to play with it. I always try to release some code so people can try it out themselves. I need to look at it again.

@@samwitteveenai My thought was that perhaps for some of your vids, you have a little segment that translates the usage to langflow so that we can see how to begin to string graphical nodes together to get the same kind of outcomes that you're showing with your code bits? Thanks again for making all this more accessible for the 'common man' :)

LangChain works with several LLMs, including llama derivatives.

Yes, use Vicunia 7B, or would you wait for red pajamas?

@@jawadmansoor6064, I would say, play with several of them before you start training the model with your custom data.

I actually tested it on Gpt4all 7B 4bit, vicuna 7B 4 but (llama based) models, frankly it was not correctly working, i tested it on the python_repl by parsing a CSV file and asking it to find the rows of the CSV, but unfortunately it didn't pass the correct python code to do that, but it correctly identified that it should use the pandas library and dataframe method. It was looping again and again for ever with each time giving a wrong response. I think we should do further testing and i am sure that larger model like the vicuna 13B or the Openassistant 30B model will work correctly(i was not able to test the bigger model as am resource limited)

@@randomNunber thank you for tests and reporting results

Check out my 1st video on LangGraph it has some of these there

HERO

try pip install --upgrade langchain

Yes you can do that in the same way as the PDF vid, but use an external vector store like Pinecone etc

The actual prompt is shown at 7:27

i've tested it and getting errors for specfying the tools. probably some updates with the libraries : //

less than 2 cents

for this video it would be low , in cents (probably 5-10cents max)

There is a the paid GSERP tool for google searches. You could also write one reasonably easily by modifying something like DuckDuckScrape

yes if you code it up to do that.

@@samwitteveenaiThanks, Witteveen would you happen to know which commands I need for the API ?

I am not sure I understand what you are trying to do if you want to run your langchain app as an API that can be called via http etc you could use FastAPI or Flask and then expose a port.

@@samwitteveenai yes that's exactly it, I will use FastAPI thank you.

Chatsnacks looks interesting will check it out.

Yes the OpenAI model is what is deciding via text generation which tool it should use. the LangChain filters for that in the background.

Yeah they changed the API for this the day after I released the video. I will update the code over the next few days.

check out the video after this I did it with turbo and explained the issues and differences.

I just updated the the notebook now. Thanks to you and others who pointed this out.

:D not quite but it's pretty cool.

almost all don't work on these base prompts unfortunately.

@@samwitteveenai thank's for the feedback. Are there potential data sets that could be used for fine tuning for these prompting tasks?

@@adriangabriel3219 not that are public AFAIK, I am looking at releasing a model for this.

@@samwitteveenai could you elaborate a bit on how the data looks like that you are using for fine-tuning? So how do the instructions look like? Are you using open-source models to generate your training data? Have you used models like Dolly or OpenAssistant together with a sample seed to generate datasets? Could you make a video on that if so? I have tried Dolly extensively together with the approach from the Alpaca model (175 sample questions -> then generate more instructions based on those samples) and it failed miserably.

its actually a 3rd party that runs the API

i disagree, ran it 2 days straight around $15.00 usd

for this video it would be low , in cents (probably 5-10cents max)

best to reach out on Linkedin