chandra-ocr-2
Datalab · datalab-to/chandra-ocr-2
Weights license: Publisher terms: OpenRAIL-M (modified, Datalab) · code and components may use other terms
Quality
No comparable score
We track no public, comparable benchmark that covers this model yet. No score is inferred.
Reported by the publisher
- ArXiv
- 86.9olmOCR Benchmark · Own benchmarks
- Old Scans Math
- 89.1olmOCR Benchmark · Own benchmarks
- Tables
- 92.1olmOCR Benchmark · Own benchmarks
- Old Scans
- 51.1olmOCR Benchmark · Own benchmarks
- Headers and Footers
- 91.4olmOCR Benchmark · Own benchmarks
- Multi column
- 82.1olmOCR Benchmark · Own benchmarks
- Long tiny text
- 93.7olmOCR Benchmark · Own benchmarks
- Base
- 99.9olmOCR Benchmark · Own benchmarks
Every reported result
- ar
- 68.4%Multilingual Benchmark (43 Languages)
- bn
- 72.8%Multilingual Benchmark (43 Languages)
- ca
- 85.1%Multilingual Benchmark (43 Languages)
- cs
- 85.3%Multilingual Benchmark (43 Languages)
- da
- 91.1%Multilingual Benchmark (43 Languages)
- de
- 94.8%Multilingual Benchmark (43 Languages)
- el
- 85.6%Multilingual Benchmark (43 Languages)
- es
- 89.3%Multilingual Benchmark (43 Languages)
- fa
- 75.1%Multilingual Benchmark (43 Languages)
- fi
- 83.4%Multilingual Benchmark (43 Languages)
- fr
- 93.7%Multilingual Benchmark (43 Languages)
- gu
- 70.8%Multilingual Benchmark (43 Languages)
- he
- 70.4%Multilingual Benchmark (43 Languages)
- hi
- 78.4%Multilingual Benchmark (43 Languages)
- hr
- 90.1%Multilingual Benchmark (43 Languages)
- hu
- 82.1%Multilingual Benchmark (43 Languages)
- id
- 91.6%Multilingual Benchmark (43 Languages)
- it
- 94.1%Multilingual Benchmark (43 Languages)
- ja
- 86.9%Multilingual Benchmark (43 Languages)
- jv
- 73.2%Multilingual Benchmark (43 Languages)
- kn
- 63.2%Multilingual Benchmark (43 Languages)
- ko
- 81.5%Multilingual Benchmark (43 Languages)
- la
- 73.8%Multilingual Benchmark (43 Languages)
- ml
- 64.3%Multilingual Benchmark (43 Languages)
- mr
- 75.0%Multilingual Benchmark (43 Languages)
- nl
- 88.6%Multilingual Benchmark (43 Languages)
- no
- 90.3%Multilingual Benchmark (43 Languages)
- pl
- 91.5%Multilingual Benchmark (43 Languages)
- pt
- 95.2%Multilingual Benchmark (43 Languages)
- ro
- 84.5%Multilingual Benchmark (43 Languages)
- ru
- 85.5%Multilingual Benchmark (43 Languages)
- sa
- 51.1%Multilingual Benchmark (43 Languages)
- sr
- 90.3%Multilingual Benchmark (43 Languages)
- sv
- 92.8%Multilingual Benchmark (43 Languages)
- ta
- 77.7%Multilingual Benchmark (43 Languages)
- te
- 58.6%Multilingual Benchmark (43 Languages)
- th
- 62.6%Multilingual Benchmark (43 Languages)
- tr
- 84.1%Multilingual Benchmark (43 Languages)
- uk
- 91.0%Multilingual Benchmark (43 Languages)
- ur
- 43.2%Multilingual Benchmark (43 Languages)
- vi
- 80.4%Multilingual Benchmark (43 Languages)
- zh
- 88.7%Multilingual Benchmark (43 Languages)
- Average
- 77.8%Multilingual Benchmark (43 Languages)
Values as printed on the card, with the publisher's own settings (card section: Benchmarks › olmOCR Benchmark; Examples › Multilingual Benchmark (43 Languages)); column “Chandra 2”. Not comparable across publishers and not used in any ranking. Model card
Runs on
Fast memory holds it 118
The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.
Show all 118
Within memory, outside the fast part 1
Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.
Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Not a measured fit or speed.
Datalab released chandra-ocr-2 on 2026-03: a vision-language model that answers about images, with 5.3B parameters on the Hugging Face count and the Publisher terms: OpenRAIL-M (modified, Datalab) license. Languages the publisher declares: 90+ languages.
License: commercial use unanswered (this license has not been reviewed for commercial use yet). At Q4 the weights load about 4 GB before each machine's reserve; image resolution, video frames and context add memory, and the vision encoder needs support in the runtime you pick.
Measured speed
No published measurement for this checkpoint
Tokens per second are shown only when a source publishes them with hardware, build and setup. They are never estimated.
How to run
Runtime compatibility and tokens per second have not been measured for this entry. Published metadata and a memory estimate do not prove a successful local run.
Library in source: transformers
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
from transformers import AutoModelForImageTextToText, AutoProcessor
from chandra.model.hf import generate_hf
from chandra.model.schema import BatchInputItem
from chandra.output import parse_markdown
from PIL import Image
import torch
model = AutoModelForImageTextToText.from_pretrained(
"datalab-to/chandra-ocr-2",
dtype=torch.bfloat16,
device_map="auto",
)
model.eval()
model.processor = AutoProcessor.from_pretrained("datalab-to/chandra-ocr-2")
model.processor.tokenizer.padding_side = "left"
batch = [
BatchInputItem(
image=Image.open("document.png"),
prompt_type="ocr_layout"
)
]
result = generate_hf(batch, model)[0]
markdown = parse_markdown(result.raw)
print(markdown)pip install chandra-ocr
# With vLLM (recommended, easy install)
chandra_vllm
chandra input.pdf ./output
# With HuggingFace (requires torch)
pip install chandra-ocr[hf]
chandra input.pdf ./output --method hfTechnical details
What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.
- Checkpoint / source
- datalab-to/chandra-ocr-2
- Task
- image-text-to-text
- Architecture
- Qwen3_5ForConditionalGeneration
- Library in source
- transformers
- Weight license
- Publisher terms: OpenRAIL-M (modified, Datalab)chandra-ocr-2 LICENSE: AI Pubs Open RAIL-M License (Modified) (Hugging Face) · read 2026-10-18source
- Commercial use
- This license has not been reviewed for commercial use yet
- Languages
- 90+ languages Publisher's statement · read 2026-10-18source
- Released
- Chandra README news (GitHub, Datalab) · read 2026-10-18source
Every stored field
- Revision checked
- af93b47dba1b47b6640c86ccf487ed2260ab9a09
- Architecture
- Qwen3_5ForConditionalGeneration
- Library
- transformers
- Hub task tag
- image-text-to-text
- Formats
- transformers, safetensors
- Access
- Open download
- Publisher kind
- Official
- Hub parameter count
- 5,295,564,288safetensors.total
- Repository created
- 2026-03-16creation is not the model's release
- Repository updated
- 2026-06-26a commit is not a release date
- Hub downloads (30 days) · likes
- 1,353,287 · 576this repository only · read 2026-10-25
- Catalog state
- Hub config or weights found; no inventory benchmark link
Sources
Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.
- Publisher model card
- License: chandra-ocr-2 LICENSE: AI Pubs Open RAIL-M License (Modified) (Hugging Face)
- Release date: Chandra README news (GitHub, Datalab)
Curated facts checked: 2026-10-18. The license and release date cite the publisher's sources; parameters, files and task come from the Hugging Face record, and memory is a labelled estimate from its parameter count. A value no source states stays empty; a repository commit is not a model release date.






























































