olmocr/tests/test_molmo.py

60 lines
2.0 KiB
Python
Raw Permalink Normal View History

2025-01-22 15:23:08 -08:00
import unittest
2025-02-14 20:42:19 +00:00
import pytest
2025-01-22 15:23:08 -08:00
import requests
2025-01-29 15:25:10 -08:00
from PIL import Image
from transformers import (
AutoModelForCausalLM,
AutoProcessor,
AutoTokenizer,
GenerationConfig,
)
2025-01-22 15:23:08 -08:00
2025-02-14 20:42:19 +00:00
@pytest.mark.nonci
2025-01-22 15:23:08 -08:00
class MolmoProcessorTest(unittest.TestCase):
def test_molmo_demo(self):
# load the processor
processor = AutoProcessor.from_pretrained(
2025-01-29 15:30:39 -08:00
"allenai/Molmo-7B-O-0924",
2025-01-22 15:23:08 -08:00
trust_remote_code=True,
2025-01-29 15:30:39 -08:00
torch_dtype="auto",
2025-01-22 15:23:08 -08:00
)
# load the model
model = AutoModelForCausalLM.from_pretrained(
2025-01-29 15:30:39 -08:00
"allenai/Molmo-7B-O-0924",
2025-01-22 15:23:08 -08:00
trust_remote_code=True,
2025-01-29 15:30:39 -08:00
torch_dtype="auto",
2025-01-22 15:23:08 -08:00
)
device = "cuda:0"
model = model.to(device)
# process the image and text
2025-01-29 15:30:39 -08:00
inputs = processor.process(images=[Image.open(requests.get("https://picsum.photos/id/237/536/354", stream=True).raw)], text="Describe this image.")
2025-01-22 15:23:08 -08:00
# move inputs to the correct device and make a batch of size 1
inputs = {k: v.to(model.device).unsqueeze(0) for k, v in inputs.items()}
print("Raw inputs")
print(inputs)
print("\nShapes")
2025-01-23 10:58:43 -08:00
# {('input_ids', torch.Size([1, 589])), ('images', torch.Size([1, 5, 576, 588])), ('image_masks', torch.Size([1, 5, 576])), ('image_input_idx', torch.Size([1, 5, 144]))}
2025-01-29 15:30:39 -08:00
print({(x, y.shape) for x, y in inputs.items()})
2025-01-22 15:23:08 -08:00
print("\nTokens")
print(processor.tokenizer.batch_decode(inputs["input_ids"]))
# generate output; maximum 200 new tokens; stop generation when <|endoftext|> is generated
2025-01-29 15:30:39 -08:00
output = model.generate_from_batch(inputs, GenerationConfig(max_new_tokens=200, stop_strings="<|endoftext|>"), tokenizer=processor.tokenizer)
2025-01-22 15:23:08 -08:00
# only get generated tokens; decode them to text
2025-01-29 15:30:39 -08:00
generated_tokens = output[0, inputs["input_ids"].size(1) :]
2025-01-22 15:23:08 -08:00
generated_text = processor.tokenizer.decode(generated_tokens, skip_special_tokens=True)
# print the generated text
2025-01-29 15:30:39 -08:00
print(generated_text)