Spaces:
Paused
Paused
import os | |
import logging | |
from pathlib import Path | |
from typing import Optional | |
from pydantic import BaseModel | |
from transformers import AutoTokenizer, AutoModelForQuestionAnswering | |
import torch | |
# Set up logging | |
logger = logging.getLogger(__name__) | |
# Define the model directory | |
MODEL_DIR = Path(__file__).parent.parent / "_models" | |
MODEL_DIR.mkdir(parents=True, exist_ok=True) # Create the directory if it doesn't exist | |
# Hugging Face authentication token (from environment variable) | |
AUTH_TOKEN = os.getenv("auth_token") | |
if not AUTH_TOKEN: | |
raise ValueError("Hugging Face auth_token environment variable is not set.") | |
class DataLocation(BaseModel): | |
""" | |
Represents the location of a model (local path and optional cloud URI). | |
""" | |
local_path: str | |
cloud_uri: Optional[str] = None | |
def exists_or_download(self): | |
""" | |
Check if the model exists locally. If not, download it from Hugging Face. | |
""" | |
if not os.path.exists(self.local_path): | |
if self.cloud_uri is not None: | |
logger.warning(f"Downloading model from Hugging Face: {self.cloud_uri}") | |
# Download from Hugging Face | |
tokenizer = AutoTokenizer.from_pretrained( | |
self.cloud_uri, token=AUTH_TOKEN | |
) | |
model = AutoModelForQuestionAnswering.from_pretrained( | |
self.cloud_uri, token=AUTH_TOKEN | |
) | |
# Save the model and tokenizer locally | |
tokenizer.save_pretrained(self.local_path) | |
model.save_pretrained(self.local_path) | |
logger.info(f"Model saved to: {self.local_path}") | |
else: | |
raise ValueError(f"Model not found locally and no cloud URI provided: {self.local_path}") | |
return self.local_path | |
# Define the model location | |
class QAModel: | |
def __init__(self,model_name,model_locaton): | |
""" | |
Initialize the QA model and tokenizer. | |
""" | |
self.model_name = model_name | |
self.model_location = model_locaton | |
self.tokenizer = None | |
self.model = None | |
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
self._load_model() # Call the method to load the model and tokenizer | |
def _load_model(self): | |
""" | |
Load the tokenizer and model. | |
""" | |
# Ensure the model is downloaded | |
model_path = self.model_location.exists_or_download() | |
# Load the tokenizer and model from the local path | |
self.tokenizer = AutoTokenizer.from_pretrained(model_path) | |
self.model = AutoModelForQuestionAnswering.from_pretrained(model_path,return_dict=True).to(self.device) | |
logger.info(f"Loaded QA model: {self.model_name}") | |
def inference_qa(self, context: str, question: str): | |
""" | |
Perform question-answering inference. | |
Args: | |
context (str): The text passage or document. | |
question (str): The question to be answered. | |
Returns: | |
str: The predicted answer. | |
""" | |
if self.tokenizer is None or self.model is None: | |
raise ValueError("Model or tokenizer is not loaded.") | |
# Tokenize inputs | |
inputs = self.tokenizer(question, context, return_tensors="pt", truncation=True, padding=True) | |
inputs = {key: value.to(self.device) for key, value in inputs.items()} | |
# Perform inference | |
with torch.no_grad(): | |
outputs = self.model(**inputs) | |
# Extract answer | |
answer_start_index = outputs.start_logits.argmax() | |
answer_end_index = outputs.end_logits.argmax() | |
predict_answer_tokens = inputs["input_ids"][0, answer_start_index : answer_end_index + 1] | |
answer = self.tokenizer.decode(predict_answer_tokens, skip_special_tokens=True) | |
return answer | |
def load_qa_pipeline(model_name: str = "Gowtham122/albertqa"): | |
""" | |
Load the QA model and tokenizer. | |
""" | |
model_location = DataLocation( | |
local_path=str(MODEL_DIR / model_name.replace("/", "-")), | |
cloud_uri=model_name, # Hugging Face model ID | |
) | |
qa_model = QAModel(model_name,model_location) | |
return qa_model | |
def inference_qa(qa_pipeline, context: str, question: str): | |
""" | |
Perform QA inference using the loaded pipeline. | |
""" | |
return qa_pipeline.inference_qa(context, question) |