Tarun Menta retweetledi
We shipped Chandra OCR 1.5 to the Datalab API this week!
This is a big improvement over 1.1, especially on tables, lists, charts, chemistry, and math.
English
Tarun Menta
37 posts
Tarun Menta
@_tarunmenta
Founding Research Engineer @datalabto | Ex @Adobe MDSR | @IITHyderabad `23
Brooklyn, NY Katılım Ağustos 2023
225 Takip Edilen113 Takipçiler
Tarun Menta retweetledi
People evaluating OCR ask the same two questions:
1. Which models are actually good
2. How do they behave on my documents
Benchmarks help with the first, but they rarely show you the qualitative side of model behavior across real layouts, scripts, bad scans, and messy photos.
So we built two tools to solve this 🧵
English
Tarun Menta retweetledi
Launch Week - Day 4: Spreadsheet Parsing 🚀
We’ve added native spreadsheet support to the Datalab API.
Spreadsheets look structured until you hit real-world files:
- staggered / overlapping tables
- sparse regions with fake separators
- hidden columns, merged cells
- stray cells that break heuristics
- images of tables (why?)
Getting reliable structure out of grids is genuinely hard.
English
@AsfiShaheen @VikParuchuri @datalabto It's running by default, and live in the API as of a few days ago, so you should be good!
English
@VikParuchuri @datalabto @_tarunmenta Is this available by default when I call your API or do i need a setting for it? Have a long job running as we speak so wondering if I pause it. It’s a super useful feature.
English
Section headers can make or break the accuracy of your RAG system.
Today, we're shipping a feature that outputs perfect header levels, even across hundreds of pages.
English
Tarun Menta retweetledi
Super cool work from @zach_nussbaum training a custom speculative decoding model to speed up Chandra inference on our API 🚀
Datalab@datalabto
We shipped Chandra (our SOTA OCR model) but base latency wasn't good enough for production. So we trained an Eagle3 draft model: ✅3× lower p99 latency ✅40% higher throughput ✅zero accuracy loss Here's how we made Chandra OCR 3× faster with Eagle3 speculative decoding 🧵
English
Tarun Menta retweetledi
Chandra OCR scores 93.9 on the OlmOCR benchmark, if we correct for minor formatting differences.
Let's discuss what this says about OCR benchmarking 🧵
English
Tarun Menta retweetledi
We ran an experiment to test a simple idea: better OCR leads to better structured extraction.
Modern LLMs can follow schemas and parse complex documents, but only if they can actually read them. On real-world invoices with rotated scans, dense tables, and skewed text, that’s not always the case.
🧵 to see how Gemini 2.5 Flash and GPT 5-mini performed in our research:
English
@_avichawla @akshay_pachaar We provide structured extraction with citations backed by our OCR models! Read more about it it here - datalab.to/blog/structure…
English
@akshay_pachaar would be super useful to build citation-backed RAG!
English
Everyone is sleeping on this new OCR model!
Datalab's Chandra topped independent benchmarks and beat the previously best dots-ocr.
- Support for 40+ languages
- Handles text, tables, formulas seamlessly
I tested on Ramanujan's handwritten letter from 1913.
100% open-source.
English
Tarun Menta retweetledi
This OCR model was probably the best one with the least hype
Awesome release with both a serverless API and open models on @huggingface
The org only has 85 followers on the hub ?!
Datalab@datalabto
Last week we launched Chandra, the newest model in our OCR family 🚀 Despite a busy week for OCR releases, it topped independent benchmarks and received incredible community feedback.
English
Tarun Menta retweetledi
We're excited to collaborate with @datalabto to make high-throughput document intelligence accessible to all developers.
Instantly deploy and scale Datalab's best-in-class Marker pipeline on Modal GPUs 📑
English
Tarun Menta retweetledi
We just released Workflows (beta) — a way to chain document-processing steps like parsing, extraction, segmentation, and conditional logic into a single, reusable pipeline.
Read more here → datalab.to/blog/build-doc…
English
@martian_2090 @VikParuchuri We convert Chandra’s output to plain HTML, so this won’t be represented. However the bounding boxes are accurately extracted too (seen in the visualizations), so its possible with some heuristics plus basic CSS :)
English
@VikParuchuri Thanks @VikParuchuri for open sourcing this wonderful model. I can see that text extraction is very accurate but placement of texts especially in header is always left aligned after OCR even though it is center aligned or right aligned in the original image.
English
Many famous authors had messy handwriting. Let's see how well Chandra OCR does on some samples.
First up, Edgar Allen Poe. I really hope he got paid for these articles.
English
We've been training a lot of models @datalabto lately, and evaluating OCR quality is hard. Most benchmarks rely on brittle string-matching metrics
olmOCR-bench’s unit-test approach has been the most trustworthy and aligns well with our manual judging - we love this benchmark!
Jake Poznanski@jakepoznanski
It's officially the week of OCR! I thought to share some of the lessons learned since olmOCR v1: 1. You need a reliable way to measure your performance. v1 was done on vibes-only, but it was hard to be confident in any changes we wanted to make. olmOCR-bench was our answer.
English
Chandra OCR is now open source! 🔥
Thrilled for this launch - and there’s a lot more coming from @datalabto.
Vik Paruchuri@VikParuchuri
I'm excited to announce that Chandra OCR is open source! - Full layout information - Extracts and captions images and diagrams - Strong handwriting, form, table support - Works with transformers and vLLM
English
@protobluf @VikParuchuri @zach_nussbaum Combination of a lot of things, including models and heuristics
English
@_tarunmenta @VikParuchuri @zach_nussbaum oh, follow up question -- how do you know when to route? is that another model that makes the decision?
English
Introducing chandra OCR - now available on the Datalab API:
- Top scores in table, math benchmarks
- Handles messy handwriting
- Form support (incl checkboxes)
- 30+ language coverage
- Full layout infomation
- Open source (with HF + VLLM support) coming soon
English
@protobluf @VikParuchuri @zach_nussbaum The model in our blog post is a smaller model available through Marker/Surya. Chandra is a larger model that decodes the full page in a single shot. We route between both models for the best results on our API!
English
@VikParuchuri @_tarunmenta @zach_nussbaum is this the same model as described here? x.com/datalabto/stat…
Datalab@datalabto
Launch Day 3 of 6: Layout Model Updates 🚀 If your doc parser gets layout wrong, everything downstream breaks. ⚠️ Reading order scrambled → unusable output ⚠️ Blocks missed → footnotes, financial figures, legal text lost We just rolled out major layout upgrades at Datalab to fix this at the root.
English
Tarun Menta retweetledi
Gearing up for an exciting launch. Watch this space 👀
Vik Paruchuri@VikParuchuri
This is from a model I'm training - it's significantly better at reading handwriting than me, so I can't tell if it's correct or not...
English
We’ve got something major in the works at @datalabto. Stay tuned 🔥
Vik Paruchuri@VikParuchuri
This is from a model I'm training - it's significantly better at reading handwriting than me, so I can't tell if it's correct or not...
English
Poured a ton into this over the past few months - late nights, endless brainstorming, and lots of debugging marathons alongside @VikParuchuri and @zach_nussbaum. Couldn’t be prouder that it’s finally in the hands of users 🚀
Datalab@datalabto
Launch Day 3 of 6: Layout Model Updates 🚀 If your doc parser gets layout wrong, everything downstream breaks. ⚠️ Reading order scrambled → unusable output ⚠️ Blocks missed → footnotes, financial figures, legal text lost We just rolled out major layout upgrades at Datalab to fix this at the root.
English