File size: 1,975 Bytes
6d612ea 7f041bb 6d612ea 7f041bb 6d612ea 43e5ff2 8cb3d7b 7f041bb 6d612ea 7f041bb 6d612ea 7f041bb 6d612ea |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 |
import logging
from typing import Any, Dict
import torch.cuda
from peft import PeftConfig, PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
LOGGER = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)
device = "cuda" if torch.cuda.is_available() else "cpu"
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,
trust_remote_code=True,
device_map="auto"
)
self.tokenizer = AutoTokenizer.from_pretrained(
config.base_model_name_or_path, trust_remote_code=True)
self.tokenizer.pad_token = self.tokenizer.eos_token
# Load the Lora model
self.model = PeftModel.from_pretrained(model, path, torch_dtype=model.dtype)
self.model.eos_token_id = self.tokenizer.eos_token_id
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}")
# Get inputs
prompt = data.pop("inputs", None)
parameters = data.pop("parameters", None)
if prompt is None:
raise ValueError("Missing prompt.")
# Preprocess
input_ids = self.tokenizer(
prompt, return_tensors="pt").input_ids.to(device)
# Forward
LOGGER.info(f"Start generation.")
if parameters is not None:
output = self.model.generate(input_ids=input_ids, **parameters)
else:
output = self.model.generate(input_ids=input_ids)
# Postprocess
prediction = self.tokenizer.decode(output[0])
LOGGER.info(f"Generated text: {prediction}")
return {"generated_text": prediction}
|