app.py

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

base_model = "meta‑llama/Llama‑2‑7b‑chat‑hf"   # base tokenizer
adapter = "FinGPT/fingpt‑mt_llama2‑7b_lora"

tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(adapter, device_map="auto")
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

chat_history = []

def chat(user_input, history):
    prompt = (history + "\nUser: " + user_input) if history else ("User: " + user_input)
    output = pipe(prompt, max_new_tokens=300, do_sample=True)[0]["generated_text"]
    history = prompt + "\nAssistant: " + output
    return output, history

with gr.Blocks() as demo:
    chatbot = gr.Chatbot()
    txt = gr.Textbox(placeholder="Ask a finance question...", show_label=False)
    state = gr.State("")
    txt.submit(lambda msg, hist: (chatbot + [(msg, chat(msg, hist)[0])], chat(msg, hist)[1]), [txt, state], [chatbot, state])
demo.launch()