winddude/stock_price_chat

Stock Price Chat Lora

GitHub | Blog

Stock Price Chat is an intent/action model. It is an experiment of a type of RaG(Retrevial augmented generation) for answer plain text querries for stock prices.

Usage

1. download the base Llama2 7b model
2. replace the tokenizer with the tokenizers in this repo
3. when loading a model, after loading the tokenizer make sure to call model.resize_token_embeddings(len(stokenizer))
4. with peft to load the adapter in this repo.

The model needs to be augmented with knowledge for yFinance, so use code found here: https://github.com/getorca/stock_price_chat. More details on the archetecture of the intent/action loop are also available here.

Basic Prompt Format

<|SYSTEM|>You are a bot that provides stock prices. From a user input first create an action with the ticker and date in a jsons string. If you are sent an action and knowledge create the response with the stock price from the provided knowledge for the date the user asks.<|END_SYSTEM|>
<|INPUT|>user input<|END_INPUT|>  
<|ACTION|>action string generated by the model<|END_ACTION|>
<|KNOWLEDGE|>knowledge string returned via the api call<|END_KNOWLEDGE|>
<|RESPONSE|>plain text response generated by the model<|END_RESPONSE|>