Instructions to use KiteFishAI/Minnow-OCR-I1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use KiteFishAI/Minnow-OCR-I1 with Transformers:

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

pipe = pipeline("image-text-to-text", model="KiteFishAI/Minnow-OCR-I1")
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)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("KiteFishAI/Minnow-OCR-I1")
model = AutoModelForMultimodalLM.from_pretrained("KiteFishAI/Minnow-OCR-I1")
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 = tokenizer.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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use KiteFishAI/Minnow-OCR-I1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "KiteFishAI/Minnow-OCR-I1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KiteFishAI/Minnow-OCR-I1",
		"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/KiteFishAI/Minnow-OCR-I1

SGLang

How to use KiteFishAI/Minnow-OCR-I1 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 "KiteFishAI/Minnow-OCR-I1" \
    --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": "KiteFishAI/Minnow-OCR-I1",
		"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 "KiteFishAI/Minnow-OCR-I1" \
        --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": "KiteFishAI/Minnow-OCR-I1",
		"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 KiteFishAI/Minnow-OCR-I1 with Docker Model Runner:
```
docker model run hf.co/KiteFishAI/Minnow-OCR-I1
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Minnow-OCR-I1

Minnow-OCR-I1 is the baseline release in the Minnow OCR series from KiteFishAI — document intelligence models packaged for air-gapped, on-prem deployment in regulated enterprise environments (BFSI, healthcare, legal).

Domain-adapted variants fine-tuned on Indian enterprise documents — multilingual forms, insurance surveyor reports, clinical records, and regulatory filings — will follow in subsequent Minnow-OCR releases.

Why this exists

Regulated Indian enterprises frequently cannot pull models from external sources at deployment time: air-gapped infrastructure, data-residency mandates, and procurement controls require a vetted, internally distributed model registry. The Minnow-OCR series provides that registry layer — known-good checkpoints, reproducible serving configurations, and a clear upgrade path to domain-adapted versions.

Model description

The underlying model is a compact (~0.9B parameter) encoder-decoder vision-language model for complex document understanding, built on the GLM-V architecture with a CogViT visual encoder, a lightweight cross-modal connector, and a GLM-0.5B language decoder. It delivers state-of-the-art results for its size class on major document benchmarks, including OmniDocBench and formula/table recognition tasks.

Usage

from transformers import AutoModelForImageTextToText, AutoProcessor

model_id = "KiteFishAI/Minnow-OCR-I1"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(model_id, device_map="auto")

For high-throughput serving:

python -m sglang.launch_server --model KiteFishAI/Minnow-OCR-I1 --port 8080

License and attribution

The model weights are released under the MIT License, originally developed and released by Z.ai as GLM-OCR. This repository redistributes those weights with attribution, per the license terms. If you use the full OCR pipeline with PP-DocLayoutV3 layout analysis, note that component is licensed under Apache License 2.0; comply with both licenses.

About KiteFishAI

KiteFishAI builds deployable enterprise intelligence systems for regulated Indian enterprises — full on-prem stacks spanning custom models, fine-tuning on enterprise data, and air-gapped deployment. Learn more at kitefishai.com or huggingface.co/KiteFishAI.

Downloads last month: -

Safetensors

Model size

1B params

Tensor type

BF16

Model tree for KiteFishAI/Minnow-OCR-I1

Base model

zai-org/GLM-OCR

Finetuned

(26)

this model

Quantizations

1 model

Evaluation results

Overall on allenai/olmOCR-bench View evaluation results

source leaderboard

75.2 ^*
Arxiv Math on allenai/olmOCR-bench View evaluation results

source leaderboard

80.7
Old Scans Math on allenai/olmOCR-bench View evaluation results

source leaderboard

68.3
Table Tests on allenai/olmOCR-bench View evaluation results

source leaderboard

77.6
Old Scans on allenai/olmOCR-bench View evaluation results

source leaderboard

37.6
Multi Column on allenai/olmOCR-bench View evaluation results

source leaderboard

76.7
Long Tiny Text on allenai/olmOCR-bench View evaluation results

source leaderboard

86.9
Headers Footers on allenai/olmOCR-bench View evaluation results

source leaderboard

95.8