Numeric values are not properly decoded in PaddleOCR-VL outputs

[passing2961]

Hi,

When I call the PaddlePaddle/PaddleOCR-VL model using the code below, I encounter inaccurate outputs—specifically, numerical values are not decoded correctly.

For example, the model outputs:

“(e.g., Gemini-.-Pro gains only +.%, while DeepSeek-R declines by -.%)”

whereas the original text in the image is:

“(e.g., Gemini-2.5-Pro gains only +1.8%, while DeepSeek-R1 declines by -0.1%)”.

Am I missing some configuration or preprocessing step?

from PIL import Image
import torch
from transformers import PaddleOCRVLProcessor, PaddleOCRVLForConditionalGeneration
from transformers import AutoProcessor, AutoModelForImageTextToText

# ---- Settings ----
model_path = "PaddlePaddle/PaddleOCR-VL"
image_path = "../test.png"
task = "ocr" # Options: 'ocr' | 'table' | 'chart' | 'formula'
# ------------------

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

PROMPTS = {
    "ocr": "OCR:",
    "table": "Table Recognition:",
    "formula": "Formula Recognition:",
    "chart": "Chart Recognition:",
}

image = Image.open(image_path).convert("RGB")

model = AutoModelForImageTextToText.from_pretrained(
    model_path,
    # trust_remote_code=True,
    torch_dtype=torch.bfloat16,
    # attn_implementation="flash_attention_2", 
).to(dtype=torch.bfloat16, device=DEVICE).eval()

processor = AutoProcessor.from_pretrained(model_path, use_fast=False, trust_remote_code=True)

messages = [
    {
        "role": "user", 
        "content": [
            {"type": "image", "image": Image.open(image_path).convert("RGB")},
            {"type": "text",  "text": PROMPTS[task]}
        ]
    }
]

inputs = processor.apply_chat_template(
    messages, 
    tokenize=True, 
    add_generation_prompt=True, 	
    return_dict=True,
    return_tensors="pt"
).to(DEVICE)

with torch.inference_mode():
    generated_ids = model.generate(
        **inputs,
        max_new_tokens=1024,
        # do_sample=False,
        # use_cache=True
    )
generated_ids_trimmed = [out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
output_text = processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
print(output_text)