README.md

metadata

language:
  - ru
  - en

from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig

with ('generation_config.json').open('w') as fp:
    json.dump({
        "pad_token_id": 0,
        "bos_token_id": 1,
        "eos_token_id": 2,
        "temperature": 0.3,
        "top_p": 0.9,
        "top_k": 50,
        "do_sample": True,
        "max_new_tokens": 1536,
        "repetition_penalty": 1.1,
        "no_repeat_ngram_size": 15,
    }, fp, indent=4)

MODEL_NAME = "Vikhrmodels/Vikhr_instruct"
TEMPLATE = "<s>{role}\n{content}</s>\n"
SYSTEM_PROMPT = "Ты – полезный помощник по имени Вихрь. Ты разговариваешь с людьми и помогаешь им."

model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
model.to('cuda')
model.eval()

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=False)
generation_config = GenerationConfig.from_pretrained("generation_config.json")

class Conversation:
    def __init__(
        self,
        message_template=DEFAULT_MESSAGE_TEMPLATE,
        system_prompt=DEFAULT_SYSTEM_PROMPT,
    ):
        self.message_template = message_template
        self.messages = [{
            "role": "system",
            "content": system_prompt
        }]

    def add_user_message(self, message):
        self.messages.append({
            "role": "user",
            "content": message
        })

    def get_prompt(self, tokenizer):
        final_text = ""
        for message in self.messages:
            message_text = self.message_template.format(**message)
            final_text += message_text
        final_text += 'bot'
        return final_text.strip()


def generate(model, tokenizer, prompt, generation_config):
    data = tokenizer(prompt, return_tensors="pt")
    data = {k: v.to(model.device) for k, v in data.items()}
    output_ids = model.generate(
        **data,
        generation_config=generation_config
    )[0]
    output_ids = output_ids[len(data["input_ids"][0]):]
    output = tokenizer.decode(output_ids, skip_special_tokens=True)
    return output.strip()

inputs = ["Как тебя зовут?", "Кто такой Колмогоров?"]

for inp in inputs:
    conversation = Conversation()
    conversation.add_user_message(inp)
    prompt = conversation.get_prompt(tokenizer)

    output = generate(model, tokenizer, prompt, generation_config)
    print(inp)
    print(output)

wandb