Instructions to use baidu/Unlimited-OCR with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use baidu/Unlimited-OCR with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="baidu/Unlimited-OCR", trust_remote_code=True)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModel

processor = AutoProcessor.from_pretrained("baidu/Unlimited-OCR", trust_remote_code=True)
model = AutoModel.from_pretrained("baidu/Unlimited-OCR", trust_remote_code=True, device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use baidu/Unlimited-OCR with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "baidu/Unlimited-OCR"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "baidu/Unlimited-OCR",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/baidu/Unlimited-OCR

SGLang

How to use baidu/Unlimited-OCR with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "baidu/Unlimited-OCR" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "baidu/Unlimited-OCR",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "baidu/Unlimited-OCR" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "baidu/Unlimited-OCR",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use baidu/Unlimited-OCR with Docker Model Runner:
```
docker model run hf.co/baidu/Unlimited-OCR
```

fix(modeling): device-agnostic MPS/CPU support — replace .cuda() with .to(device), fix masked_scatter_ broadcast, and use device-type-aware autocast

by kushdab - opened Jun 25

base: refs/heads/main

←

from: refs/pr/5

Discussion Files changed

+25

-17

kushdab

Jun 25

Problem

The model is hard-coded to CUDA in three ways that cause crashes on Apple Silicon (MPS) and CPU:

1. `masked_scatter_` broadcast + `.cuda()` (line 582)

# Before
inputs_embeds[idx].masked_scatter_(
    images_seq_mask[idx].unsqueeze(-1).cuda(), images_in_this_batch
)

Two bugs in one line:

.unsqueeze(-1) leaves the mask with shape [seq_len, 1] while inputs_embeds[idx] has shape [seq_len, hidden_dim]. masked_scatter_ requires the mask to be broadcastable but on MPS it raises RuntimeError: masked_scatter_ received a mask with wrong number of elements. expand_as() makes the broadcast explicit and safe.
.cuda() hardcodes the device and crashes immediately on MPS/CPU with RuntimeError: Expected all tensors to be on the same device.

2. All remaining `.cuda()` calls in `infer()` and `infer_multi()` (17 occurrences)

Every tensor sent to generate() is moved with .cuda(). On MPS these fail with Expected all tensors to be on the same device, but found at least two devices, mps:0 and cuda:0.

3. `torch.autocast("cuda", ...)` (3 occurrences)

torch.autocast takes the device type as its first argument. Passing "cuda" on an MPS machine raises RuntimeError: "cuda" is not a supported autocast device type on this platform.

Fix

# At the start of infer() and infer_multi():
device = next(self.parameters()).device  # MPS / CUDA / CPU

# masked_scatter_ fix — expand mask + move to correct device:
_mask = images_seq_mask[idx].unsqueeze(-1).expand_as(inputs_embeds[idx]).to(inputs_embeds.device)
inputs_embeds[idx].masked_scatter_(_mask, images_in_this_batch)

# All other .cuda() → .to(device)
input_ids.unsqueeze(0).to(device)  # etc.

# autocast → device-type-aware:
torch.autocast(device.type, dtype=torch.bfloat16)

Testing

Validated on CUDA (no regression). The expand_as path is a strict superset of the previous behaviour — on CUDA the expanded mask is identical to what the implicit broadcast would have produced.

Fixes: https://github.com/baidu/Unlimited-OCR/issues/18
Related: https://github.com/baidu/Unlimited-OCR/pull/29