s444415/donut
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

testing

Browse Source
This commit is contained in:
mkozlowskiAzimuthe 2023-01-22 23:32:33 +01:00
parent 8f5b72c2fa
commit 6ba533aeda
1 changed files with 45 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

90
single_eval.py
View File

@ -7,21 +7,21 @@ from fastapi import FastAPI, UploadFile, File
import io import io
import os import os
print("Set up config") # print("Set up config")
config_vision = VisionEncoderDecoderConfig.from_pretrained("Zombely/plwiki-proto-fine-tuned-v3.2") # config_vision = VisionEncoderDecoderConfig.from_pretrained("Zombely/plwiki-proto-fine-tuned-v3.2")
config_vision.encoder.image_size = [1920, 2560] # (height, width) # config_vision.encoder.image_size = [1920, 2560] # (height, width)
config_vision.decoder.max_length = 768 # config_vision.decoder.max_length = 768
processor = DonutProcessor.from_pretrained("Zombely/plwiki-proto-fine-tuned-v3.2") # processor = DonutProcessor.from_pretrained("Zombely/plwiki-proto-fine-tuned-v3.2")
model = VisionEncoderDecoderModel.from_pretrained("Zombely/plwiki-proto-fine-tuned-v3.2", config=config_vision) # model = VisionEncoderDecoderModel.from_pretrained("Zombely/plwiki-proto-fine-tuned-v3.2", config=config_vision)
processor.image_processor.size = [1920, 2560][::-1] # should be (width, height) # processor.image_processor.size = [1920, 2560][::-1] # should be (width, height)
processor.image_processor.do_align_long_axis = False # processor.image_processor.do_align_long_axis = False
# dataset = load_dataset(config.validation_dataset_path, split=config.validation_dataset_split) # # dataset = load_dataset(config.validation_dataset_path, split=config.validation_dataset_split)
device = "cuda" if torch.cuda.is_available() else "cpu" # device = "cuda" if torch.cuda.is_available() else "cpu"
model.eval() # model.eval()
model.to(device) # model.to(device)
print("Print ipconfig") print("Print ipconfig")
os.system("ipconfig") os.system("ipconfig")
@ -33,40 +33,40 @@ app = FastAPI()
async def test(): async def test():
return {"message": "Test"} return {"message": "Test"}
@app.post("/process") # @app.post("/process")
async def process_image(file: UploadFile= File(...)): # async def process_image(file: UploadFile= File(...)):
request_object_content = await file.read() # request_object_content = await file.read()
input_image = Image.open(io.BytesIO(request_object_content)) # input_image = Image.open(io.BytesIO(request_object_content))
# prepare encoder inputs # # prepare encoder inputs
pixel_values = processor(input_image.convert("RGB"), return_tensors="pt").pixel_values # pixel_values = processor(input_image.convert("RGB"), return_tensors="pt").pixel_values
pixel_values = pixel_values.to(device) # pixel_values = pixel_values.to(device)
# prepare decoder inputs # # prepare decoder inputs
task_prompt = "<s_cord-v2>" # task_prompt = "<s_cord-v2>"
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids # decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
decoder_input_ids = decoder_input_ids.to(device) # decoder_input_ids = decoder_input_ids.to(device)
print("Start processing") # print("Start processing")
# autoregressively generate sequence # # autoregressively generate sequence
start_time = time.time() # start_time = time.time()
outputs = model.generate( # outputs = model.generate(
pixel_values, # pixel_values,
decoder_input_ids=decoder_input_ids, # decoder_input_ids=decoder_input_ids,
max_length=model.decoder.config.max_position_embeddings, # max_length=model.decoder.config.max_position_embeddings,
early_stopping=True, # early_stopping=True,
pad_token_id=processor.tokenizer.pad_token_id, # pad_token_id=processor.tokenizer.pad_token_id,
eos_token_id=processor.tokenizer.eos_token_id, # eos_token_id=processor.tokenizer.eos_token_id,
use_cache=True, # use_cache=True,
num_beams=1, # num_beams=1,
bad_words_ids=[[processor.tokenizer.unk_token_id]], # bad_words_ids=[[processor.tokenizer.unk_token_id]],
return_dict_in_generate=True, # return_dict_in_generate=True,
) # )
processing_time = (time.time() - start_time) # processing_time = (time.time() - start_time)
# turn into JSON # # turn into JSON
seq = processor.batch_decode(outputs.sequences)[0] # seq = processor.batch_decode(outputs.sequences)[0]
seq = seq.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "") # seq = seq.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
seq = re.sub(r"<.*?>", "", seq, count=1).strip() # remove first task start token # seq = re.sub(r"<.*?>", "", seq, count=1).strip() # remove first task start token
seq = processor.token2json(seq) # seq = processor.token2json(seq)
return {"data": seq['text_sequence'], "processing_time": f"{processing_time} seconds"} # return {"data": seq['text_sequence'], "processing_time": f"{processing_time} seconds"}