Spaces:
Running
on
Zero
Running
on
Zero
File size: 2,233 Bytes
52e44f1 18f33a7 52e44f1 18f33a7 52e44f1 18f33a7 52e44f1 18f33a7 d5f1e98 18f33a7 d5f1e98 18f33a7 52e44f1 18f33a7 52e44f1 18f33a7 52e44f1 18f33a7 52e44f1 |
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 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 |
import torch
from PIL import Image
from transformers import Florence2ForConditionalGeneration, Florence2Processor
MODEL_ID = "ducviet00/Florence-2-large-hf"
# Global variables for lazy loading
_model = None
_processor = None
_device = None
_torch_dtype = None
def _load_model():
"""Load model and processor lazily"""
global _model, _processor, _device, _torch_dtype
if _model is None:
_device = "cuda:0" if torch.cuda.is_available() else "cpu"
_torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
print(f"Loading model {MODEL_ID} on {_device} with dtype {_torch_dtype}...")
_model = Florence2ForConditionalGeneration.from_pretrained(
MODEL_ID, dtype=_torch_dtype, trust_remote_code=True
).to(_device) # type: ignore
_processor = Florence2Processor.from_pretrained(
MODEL_ID, trust_remote_code=True, use_fast=True
)
print("Model loaded successfully!")
return _model, _processor, _device, _torch_dtype
def get_task_response(task_prompt: str, image: Image.Image, text_input=None):
"""Return associated task response
Task can be:
'<MORE_DETAILED_CAPTION>'
'<DETAILED_CAPTION>'
'<CAPTION>'
"""
# Lazy load model only when needed
model, processor, device, torch_dtype = _load_model()
if text_input is None:
prompt = task_prompt
else:
prompt = task_prompt + text_input
# Ensure image is in RGB mode
if image.mode != "RGB":
image = image.convert("RGB")
if processor is None:
raise ValueError("processor is None")
inputs = processor(
text=prompt,
images=image,
return_tensors="pt", # type: ignore
).to(device, torch_dtype)
generated_ids = model.generate(
input_ids=inputs["input_ids"],
pixel_values=inputs["pixel_values"],
max_new_tokens=1024,
num_beams=3,
)
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
parsed_answer = processor.post_process_generation(
generated_text, task=task_prompt, image_size=(image.width, image.height)
)
return parsed_answer[task_prompt]
|