ChatLlaMA fixes

[AAnirudh07]

Hey @diegofiori!
Thank you for making chatllama public! I was looking into how to create my own dataset and came across this issue:

• [Chatllama] Question: How to load the ChatLLaMA model weights from the directory for training? Also, can you provide a sample of the training data? #200

In the meantime, I decided to experiment with langchain agents to generate a dataset. During my experimentation, I encountered a few small bugs which I have fixed in this pull request:

1. I added openai to the dependencies
2. I updated the readme with instructions on how to add the openai api key as an environment variable
3. I called the main() function in generate_dataset.py (currently the script executes but nothing is called).

super excited for your training example!

[diegofiori]

Hi @AAnirudh07, thank you very much for the PR! With the training example we will provide also the script for converting the generated data in the correct format ;) Stay tuned :)

[AAnirudh07]

Awesome thanks!