Instructions to use nvidia/Nemotron-3.5-Parse with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvidia/Nemotron-3.5-Parse with Transformers:

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

pipe = pipeline("image-text-to-text", model="nvidia/Nemotron-3.5-Parse", 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 AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("nvidia/Nemotron-3.5-Parse", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use nvidia/Nemotron-3.5-Parse with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/Nemotron-3.5-Parse"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nvidia/Nemotron-3.5-Parse",
		"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/nvidia/Nemotron-3.5-Parse

SGLang

How to use nvidia/Nemotron-3.5-Parse 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 "nvidia/Nemotron-3.5-Parse" \
    --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": "nvidia/Nemotron-3.5-Parse",
		"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 "nvidia/Nemotron-3.5-Parse" \
        --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": "nvidia/Nemotron-3.5-Parse",
		"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 nvidia/Nemotron-3.5-Parse with Docker Model Runner:
```
docker model run hf.co/nvidia/Nemotron-3.5-Parse
```

Nemotron-3.5-Parse / reference /multilingual /README.md

emelryan

sosahi

ronay-nv

Nemotron 3.5 Parse release

dca5b2d 15 days ago

preview code

Raw History Blame Contribute Delete

3.2 kB

Multilingual OCR evaluation

MOSCAR uses simple synthetic renderings of text in each language to test basic OCR capability. These results do not guarantee accuracy on complex scanned documents, handwriting, degraded images, or other real-world variations. Character bag F1 measures character overlap without considering character order.

These results compare the published Nemotron Parse 2.0 and Nemotron 3.5 Parse weights on identical evaluation pages.

Evaluation settings

MOSCAR uses 8,111 held-out synthetic document pages, with 45–50 pages per language/script entry. The 163 entries represent 162 distinct language codes; Simplified and Traditional Chinese are separate entries. IndicVisionBench OCR uses 876 test pages across 10 languages.

Both models use one pass with temperature 0, top-k 1, repetition penalty 1.1, and an output limit of 8,744 tokens. Each model uses its documented default task prompt: Parse 2.0 uses </s><s><predict_bbox><predict_classes><output_markdown><predict_no_text_in_pic>, and Parse 3.5 uses </s><s><predict_bbox><predict_classes><output_markdown><predict_text_in_pic>. A full shared-prompt control, using predict_text_in_pic for both models, is also recorded in the results JSON. Model-specific image preprocessing is preserved. Parse 2.0 boxes are mapped back using its published transform_bbox_to_original helper; Parse 3.5 boxes are scaled from content-normalized coordinates. The same frozen benchmark scorer is used for both models. No ParseBench multi-pass merging or postprocessing is applied to these OCR evaluations. All pages, including outputs reaching the token limit, remain in the results.

MOSCAR CER divides the total character edit distance by the total reference character count after Unicode NFKC normalization, whitespace normalization, and case folding. Script and overall CER are recomputed from those totals. Character bag F1 ignores character order and averages the per-page F1 scores. The CSV localization recall uses a bounding-box IoU threshold of 0.5; matched class accuracy is computed only on localized blocks. IndicVisionBench uses the benchmark adapter's character/word ANLS and corpus CER/WER with whitespace and parsing/semantic-markup normalization. ANLS averages pages; CER and WER aggregate edit and reference counts. Metrics are percentages in the tables and CSV files. CER and WER include insertion errors and are not clipped at 100%. CER and WER are lower-is-better; ANLS and character bag F1 are higher-is-better.

The results JSON records evaluated model revisions, model-weight digests, input/label manifest hashes, prediction digests, coverage, token-limit counts, and the scoring snapshot digest. These identify the evaluated artifacts without requiring internal filesystem paths.