|
import subprocess |
|
import sys |
|
import torch |
|
import base64 |
|
from io import BytesIO |
|
from PIL import Image |
|
import requests |
|
from transformers import AutoModelForCausalLM, AutoProcessor |
|
from tokenizers import Tokenizer, pre_tokenizers |
|
import os |
|
|
|
def install(package): |
|
subprocess.check_call([sys.executable, "-m", "pip", "install", "--no-warn-script-location", package]) |
|
|
|
class EndpointHandler: |
|
def __init__(self, path=""): |
|
|
|
required_packages = ['timm', 'einops', 'flash-attn', 'Pillow','transformers==4.41.0.dev0'] |
|
for package in required_packages: |
|
try: |
|
install(package) |
|
print(f"Successfully installed {package}") |
|
except Exception as e: |
|
print(f"Failed to install {package}: {str(e)}") |
|
|
|
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
|
print(f"Using device: {self.device}") |
|
|
|
|
|
self.model_name = "arjunanand13/LADP_Florence-40e" |
|
self.model = AutoModelForCausalLM.from_pretrained( |
|
self.model_name, trust_remote_code=True |
|
).to(self.device) |
|
|
|
|
|
self.tokenizer = self.load_tokenizer() |
|
|
|
|
|
self.processor = AutoProcessor.from_pretrained(self.model_name, trust_remote_code=True) |
|
|
|
if torch.cuda.is_available(): |
|
torch.cuda.empty_cache() |
|
|
|
def load_tokenizer(self): |
|
"""Manually loads the tokenizer and adds a whitespace pre-tokenizer.""" |
|
try: |
|
tokenizer = Tokenizer.from_pretrained(self.model_name) |
|
tokenizer.pre_tokenizer = pre_tokenizers.Whitespace() |
|
print("[INFO] Whitespace pre-tokenizer added.") |
|
return tokenizer |
|
except Exception as e: |
|
print(f"[ERROR] Failed to load tokenizer: {str(e)}") |
|
return None |
|
|
|
def process_image(self, image_data): |
|
"""Processes image data from file path or base64-encoded string.""" |
|
print("[DEBUG] Attempting to process image") |
|
try: |
|
if isinstance(image_data, str) and len(image_data) < 256 and os.path.exists(image_data): |
|
with open(image_data, 'rb') as image_file: |
|
print("[DEBUG] File opened successfully") |
|
image = Image.open(image_file) |
|
else: |
|
print("[DEBUG] Decoding base64 image data") |
|
image_bytes = base64.b64decode(image_data) |
|
image = Image.open(BytesIO(image_bytes)) |
|
|
|
print("[DEBUG] Image opened:", image.format, image.size, image.mode) |
|
return image |
|
except Exception as e: |
|
print(f"[ERROR] Error processing image: {str(e)}") |
|
return None |
|
|
|
def __call__(self, data): |
|
"""Processes input and generates model output.""" |
|
try: |
|
inputs = data.pop("inputs", data) |
|
|
|
if isinstance(inputs, dict): |
|
image_path = inputs.get("image", None) |
|
text_input = inputs.get("text", "") |
|
else: |
|
image_path = inputs |
|
text_input = "What is in this image?" |
|
|
|
print("[INFO] Image path:", image_path, "| Text input:", text_input) |
|
|
|
image = self.process_image(image_path) if image_path else None |
|
|
|
model_inputs = self.processor( |
|
images=image if image else None, |
|
text=text_input, |
|
return_tensors="pt" |
|
) |
|
|
|
model_inputs = {k: v.to(self.device) if isinstance(v, torch.Tensor) else v |
|
for k, v in model_inputs.items()} |
|
|
|
with torch.no_grad(): |
|
outputs = self.model.generate(**model_inputs) |
|
|
|
decoded_outputs = self.processor.batch_decode(outputs, skip_special_tokens=True) |
|
print(f"[INFO] Generated text: {decoded_outputs[0]}") |
|
return {"generated_text": decoded_outputs[0]} |
|
|
|
except Exception as e: |
|
print(f"[ERROR] {str(e)}") |
|
return {"error": str(e)} |
|
|
|
|
