Instructions to use MuseMesh/sansar-ocr-2b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MuseMesh/sansar-ocr-2b with Transformers:

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

pipe = pipeline("image-text-to-text", model="MuseMesh/sansar-ocr-2b")
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("MuseMesh/sansar-ocr-2b")
model = AutoModelForMultimodalLM.from_pretrained("MuseMesh/sansar-ocr-2b", 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 MuseMesh/sansar-ocr-2b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MuseMesh/sansar-ocr-2b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MuseMesh/sansar-ocr-2b",
		"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/MuseMesh/sansar-ocr-2b

SGLang

How to use MuseMesh/sansar-ocr-2b 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 "MuseMesh/sansar-ocr-2b" \
    --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": "MuseMesh/sansar-ocr-2b",
		"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 "MuseMesh/sansar-ocr-2b" \
        --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": "MuseMesh/sansar-ocr-2b",
		"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 MuseMesh/sansar-ocr-2b with Docker Model Runner:
```
docker model run hf.co/MuseMesh/sansar-ocr-2b
```

sansar-ocr-2b / training /label_stats.json

ks2812

v0.1.0: OCR-VLM v1.2-clean (Qwen3.5-2B fine-tune), bench_v1 median letter CER 1.05% (transformers)

83299c4 verified 26 days ago

Raw History Blame Contribute Delete

2.17 kB

{
 "version": "v1.2-clean",
 "built_from": "/run/media/kushal/Kush/sansar/data/ocr_labels/v1",
 "rules": "(every train / val row of v1.1 is decided by them; test = the 54 bench_v1 pages, unchanged):\n  1. tiers silver_vision, llm_claude, llm_codex and sources textbooks_vision, air_vision, llm_claude, llm_codex: dropped.\n  2. any remaining row that carries an ocr_engine, a cer_vs_vision value or an id prefix of those tiers\n     (tbv_, airv_, llmc_, llmx_): dropped (none expected; a guard).\n  3. synthetic pages that typeset a paragraph whose corpus record lists `vagartha` (Sarvam's Vagartha dataset) among\n     its source_corpora: dropped. The paragraph ids come from --vagartha_ids (id<TAB>source_corpora, made by a\n     grep of train_slice_plus_all_v2 for the synthetic text cache's ids); a page is dropped when it shares one\n     16-letter window (hnorm, stride 1 on the page) with any of those paragraphs.\n  4. kept tiers must be gold_twin, gold_digital (textbooks: text layer, AIR: OCR-free decode) and synthetic (curated\n     e-text typeset): asserted.",
 "splits": {
  "train": {
   "pages": 19377,
   "books": 1057,
   "chars": 27080866,
   "by_tier": {
    "gold_twin": 11074,
    "gold_digital": 7206,
    "synthetic": 1097
   },
   "by_source": {
    "twin": 11074,
    "textbooks": 3367,
    "air": 3839,
    "synthetic": 1097
   }
  },
  "val": {
   "pages": 711,
   "books": 31,
   "chars": 1045871,
   "by_tier": {
    "gold_twin": 209,
    "gold_digital": 478,
    "synthetic": 24
   },
   "by_source": {
    "twin": 209,
    "textbooks": 36,
    "air": 442,
    "synthetic": 24
   }
  },
  "test": {
   "pages": 54,
   "books": 53,
   "chars": 86038,
   "by_tier": {
    "gold_bench": 54
   },
   "by_source": {
    "bench": 54
   }
  }
 },
 "dropped": {
  "train: synthetic page typesets a Vagartha (Sarvam) paragraph": 847,
  "train: tier llm_claude (llm_claude)": 420,
  "train: tier llm_codex (llm_codex)": 219,
  "train: tier silver_vision (air_vision)": 908,
  "train: tier silver_vision (textbooks_vision)": 2881,
  "val: synthetic page typesets a Vagartha (Sarvam) paragraph": 26,
  "val: tier silver_vision (air_vision)": 125
 }
}

	{
	"version": "v1.2-clean",
	"built_from": "/run/media/kushal/Kush/sansar/data/ocr_labels/v1",
	"rules": "(every train / val row of v1.1 is decided by them; test = the 54 bench_v1 pages, unchanged):\n 1. tiers silver_vision, llm_claude, llm_codex and sources textbooks_vision, air_vision, llm_claude, llm_codex: dropped.\n 2. any remaining row that carries an ocr_engine, a cer_vs_vision value or an id prefix of those tiers\n (tbv_, airv_, llmc_, llmx_): dropped (none expected; a guard).\n 3. synthetic pages that typeset a paragraph whose corpus record lists `vagartha` (Sarvam's Vagartha dataset) among\n its source_corpora: dropped. The paragraph ids come from --vagartha_ids (id<TAB>source_corpora, made by a\n grep of train_slice_plus_all_v2 for the synthetic text cache's ids); a page is dropped when it shares one\n 16-letter window (hnorm, stride 1 on the page) with any of those paragraphs.\n 4. kept tiers must be gold_twin, gold_digital (textbooks: text layer, AIR: OCR-free decode) and synthetic (curated\n e-text typeset): asserted.",
	"splits": {
	"train": {
	"pages": 19377,
	"books": 1057,
	"chars": 27080866,
	"by_tier": {
	"gold_twin": 11074,
	"gold_digital": 7206,
	"synthetic": 1097
	},
	"by_source": {
	"twin": 11074,
	"textbooks": 3367,
	"air": 3839,
	"synthetic": 1097
	}
	},
	"val": {
	"pages": 711,
	"books": 31,
	"chars": 1045871,
	"by_tier": {
	"gold_twin": 209,
	"gold_digital": 478,
	"synthetic": 24
	},
	"by_source": {
	"twin": 209,
	"textbooks": 36,
	"air": 442,
	"synthetic": 24
	}
	},
	"test": {
	"pages": 54,
	"books": 53,
	"chars": 86038,
	"by_tier": {
	"gold_bench": 54
	},
	"by_source": {
	"bench": 54
	}
	}
	},
	"dropped": {
	"train: synthetic page typesets a Vagartha (Sarvam) paragraph": 847,
	"train: tier llm_claude (llm_claude)": 420,
	"train: tier llm_codex (llm_codex)": 219,
	"train: tier silver_vision (air_vision)": 908,
	"train: tier silver_vision (textbooks_vision)": 2881,
	"val: synthetic page typesets a Vagartha (Sarvam) paragraph": 26,
	"val: tier silver_vision (air_vision)": 125
	}
	}