Instructions to use CheneyWong/povl-opd-scale05 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use CheneyWong/povl-opd-scale05 with Transformers:

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

pipe = pipeline("image-text-to-text", model="CheneyWong/povl-opd-scale05", 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, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("CheneyWong/povl-opd-scale05", trust_remote_code=True)
model = AutoModelForMultimodalLM.from_pretrained("CheneyWong/povl-opd-scale05", 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 CheneyWong/povl-opd-scale05 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CheneyWong/povl-opd-scale05"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CheneyWong/povl-opd-scale05",
		"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/CheneyWong/povl-opd-scale05

SGLang

How to use CheneyWong/povl-opd-scale05 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 "CheneyWong/povl-opd-scale05" \
    --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": "CheneyWong/povl-opd-scale05",
		"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 "CheneyWong/povl-opd-scale05" \
        --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": "CheneyWong/povl-opd-scale05",
		"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 CheneyWong/povl-opd-scale05 with Docker Model Runner:
```
docker model run hf.co/CheneyWong/povl-opd-scale05
```

povl-opd-scale05

What this checkpoint is

On-policy distillation. The weight delta learned by the teacher was merged into the SFT base with an extra scale of 0.5, the best point of an alpha sweep (0.25 -> +0.32, 0.5 -> +0.44, 0.75 -> -0.29, 1.5 -> -3.01 with t = -12.42).

Lineage

Fine-tuned from CheneyWong/povl-sft-2d-merge, which is NOT the RL checkpoint -- the teacher delta was merged into the SFT base, which is why it neither clearly beats nor clearly follows round 2.

PaddleOCR-VL-1.6 (base)
  └─ povl-opd-scale05   <-- this checkpoint

Results on public datasets

Base = the unmodified base model. Public datasets only; the project's self-built benchmarks are excluded.

Dataset	n	Base TEDS	Ours TEDS	d TEDS	Base TEDS-S	Ours TEDS-S	d TEDS-S
PubTables-1M	2000	0.9015	0.9276	+0.0261	0.9243	0.9377	+0.0133
PubTables-v2	2000	0.9026	0.9499	+0.0473	0.9239	0.9592	+0.0353
FinTabNet (OTSL)	2000	0.9442	0.9660	+0.0217	0.9524	0.9734	+0.0209
PubTabNet / ICDAR	2000	0.9297	0.9296	-0.0001	0.9464	0.9426	-0.0038
SynthTabNet	2000	0.8093	0.9144	+0.1051	0.8474	0.9283	+0.0809
TableVerse-5K	4951	0.8608	0.8509	-0.0099	0.9101	0.9026	-0.0074
OmniDocBench v1.5	512	0.9580	0.9449	-0.0131	0.9797	0.9722	-0.0075
OmniDocBench v1.6	665	0.9492	0.9344	-0.0148	0.9753	0.9666	-0.0086
CC-OCR	300	0.8304	0.8048	-0.0256	0.8762	0.8552	-0.0210
OCRBench-v2	681	0.8120	0.8069	-0.0051	0.8526	0.8456	-0.0070
Real5 (re-shot OmniDoc)	1320	0.8930	0.8738	-0.0192	0.9307	0.9118	-0.0189
SPRINT-MUSTARD	1429	0.2184	0.2203	+0.0019	0.8754	0.8689	-0.0065
TABLET	909	0.9048	0.8924	-0.0123	0.9277	0.9136	-0.0141
Overall (equal weight)	-	0.8395	0.8474	+0.0078	0.9171	0.9214	+0.0043

Note on SPRINT-MUSTARD: its ground truth carries no cell text, so its TEDS (~0.22 here) measures how well the model reproduces empty cells and should be ignored -- a perturbation test shows a structurally perfect prediction that correctly reads the visible text scores 0.33 TEDS / 1.00 TEDS-S. Only the TEDS-S column is meaningful for that dataset.

Known limitations

The +0.05 pt over the RL checkpoint is inside sampling noise (+/- 0.5 pt) and is not a real gain. It was also bought by giving up 4.4 pt in-domain: the distillation delta behaves as a slider trading in-domain accuracy against real-scene accuracy, not as a free improvement.

Separately, an audit found that all eight on-policy distillation students from the previous round were no-ops (mean_r pinned at exactly -1.0000) because a reward-bridge fix never reached the plugin that actually ran. On-policy distillation has therefore never actually been validated in this project.

Task format: a single table image plus the prompt Table Recognition:, one pass, emitting OTSL markup (<fcel> <ecel> <ucel> <lcel> <xcel> <nl>). Evaluation numbers are tracked separately and are not part of this card.

Downloads last month: 25

Safetensors

Model size

0.9B params

Tensor type

BF16