app.py

import os
os.environ['REPLICATE_API_TOKEN'] = "r8_afc5kESy4ucPojF3Tw1GE25ER4Ovudy1iPVw6"

import replicate

# Prompts
pre_prompt = "You are a helpful assistant. You do not respond as 'User' or pretend to be 'User'. You only respond once as 'Assistant'."
prompt_input = "What is Hugging Face"

# Generate LLM response
output = replicate.run('huggingface/llama-base-125M',  # LLM model
                       input={
                           "prompt": f"{pre_prompt} {prompt_input} Assistant: ",  # Prompts
                           "temperature": 0.1,
                           "top_p": 0.9,
                           "max_length": 124,
                           "repetition_penalty": 1
                       })  # Model parameters

full_response = ''

for item in output:
    full_response += item

print(full_response)