|
from typing import Dict, Any |
|
import logging |
|
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
from peft import PeftConfig, PeftModel |
|
|
|
|
|
LOGGER = logging.getLogger(__name__) |
|
logging.basicConfig(level=logging.INFO) |
|
|
|
|
|
class EndpointHandler(): |
|
def __init__(self, path=""): |
|
config = PeftConfig.from_pretrained(path) |
|
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, load_in_8bit=True, device_map='auto') |
|
self.tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path) |
|
|
|
self.model = PeftModel.from_pretrained(model, path) |
|
|
|
def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]: |
|
""" |
|
Args: |
|
data (Dict): The payload with the text prompt and generation parameters. |
|
""" |
|
LOGGER.info(f"Received data: {data}") |
|
|
|
prompt = data.pop("prompt", data) |
|
parameters = data.pop("parameters", None) |
|
|
|
input = self.tokenizer(prompt, return_tensors="pt") |
|
|
|
LOGGER.info(f"Start generation.") |
|
if parameters is not None: |
|
output = self.model.generate(**input, **parameters) |
|
else: |
|
output = self.model.generate(**input) |
|
|
|
prediction = self.tokenizer.decode(output[0]) |
|
LOGGER.info(f"Generated text: {prediction}") |
|
return {"generated_text": prediction} |
|
|
