Как разобраться в LLama (llama_cpp)

def get_answer(llm, prompt):
 
      output = llm(
            prompt, # Prompt
            max_tokens=32, # Generate up to 32 tokens, set to None to generate up to the end of the context window
            stop=['Q:', '\n'], # Stop generating just before the model would generate a new question
            echo=False # Echo the prompt back in the output
            #echo=True # Echo the prompt back in the output
      ) # Generate a completion, can also call create_completion
 
      return output
 
llm = Llama(
      model_path=model_path,
      n_gpu_layers=-1, # Uncomment to use GPU acceleration
)
 
 
questions = [
      'Q: Name the planets in the solar system? A: ',
]
 
 
for question in questions:
 
      output = get_answer(llm, question)
      print()
      print(Fore.GREEN + question)
      print(Fore.CYAN + str(output['choices'][0]['text']))

SYSTEM_PROMPT = 'Llama answer'
SYSTEM_TOKEN = 1788
USER_TOKEN = 1404
BOT_TOKEN = 9225
LINEBREAK_TOKEN = 13
 
ROLE_TOKENS = {
    "user": USER_TOKEN,
    "bot": BOT_TOKEN,
    "system": SYSTEM_TOKEN
}
 
#!##########################################################################################
def get_message_tokens(model, role, content):
    message_tokens = model.tokenize(content.encode('utf-8'))
    message_tokens.insert(1, ROLE_TOKENS[role])
    message_tokens.insert(2, LINEBREAK_TOKEN)
    message_tokens.append(model.token_eos())
    return message_tokens
 
#!##########################################################################################
def get_system_tokens(model):
    system_message = {
        "role": "system",
        "content": SYSTEM_PROMPT
    }
    return get_message_tokens(model, **system_message)
 
#!##########################################################################################
def interact(model_path, n_ctx=2000, top_k=30, top_p=0.9, temperature=0.2, repeat_penalty=1.1):
 
    model = Llama(model_path=model_path, n_ctx=n_ctx, n_parts=1,)
 
    system_tokens = get_system_tokens(model)
    tokens = system_tokens
    model.eval(tokens)
 
    while True:
 
        user_message = input('> ')
        message_tokens = get_message_tokens(model=model, role='user', content=user_message)
        role_tokens = [model.token_bos(), BOT_TOKEN, LINEBREAK_TOKEN]
        tokens += message_tokens + role_tokens
        generator = model.generate(tokens, top_k=top_k, top_p=top_p, temp=temperature, repeat_penalty=repeat_penalty)\
 
        for token in generator:
            token_str = model.detokenize([token]).decode('utf-8', errors='ignore')
            tokens.append(token)
            if token == model.token_eos():
                break
            print(token_str, end='', flush=True)
        print()
 
interact(model_path)