File size: 1,714 Bytes
a7ff2fc |
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 |
from typing import List, Any
import torch
from peft import PeftModel, PeftConfig
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
def preprocessing(data):
texts = list()
i = 0
if len(data) <= i+4000:
texts = data
else:
while len(data[i:]) != 0:
if len(data[i:]) > 4000:
string = str(data[i:i+4000])
texts.append(string)
i = i + 3800
else:
string = str(data[i:])
texts.append(string)
break
return texts
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
peft_model_id = "sooolee/flan-t5-base-cnn-samsum-lora"
config = PeftConfig.from_pretrained(peft_model_id)
class EndpointHandler:
def __init__(self, path=""):
# load model and tokenizer from path
self.model = AutoModelForSeq2SeqLM.from_pretrained(config.base_model_name_or_path, device_map='auto')
self.tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
self.model = PeftModel.from_pretrained(self.model, path, device_map='auto')
def __call__(self, data: Any) -> List[str]:
inputs = data.pop("inputs", data)
# process input
texts = preprocessing(inputs)
inputs = self.tokenizer(texts, return_tensors="pt", padding=True, ) # truncation=True
with torch.no_grad():
output = self.model.generate(input_ids=inputs["input_ids"].to(device), max_new_tokens=60, do_sample=True, top_p=0.9)
summary = self.tokenizer.batch_decode(output.detach().cpu().numpy(), skip_special_tokens=True)
return summary |