datalab-to

A faster Tesseract alternative: OCR PDFs to markdown on CPU with Indra OCR Tesseract is still the default OCR engine in a lot of pipelines, mostly because it is free, local and needs no GPU. Indra OCR from Datalab keeps all three of those properties and adds layout, reading order and handwriting. This post shows how to install it, convert PDFs to markdown or JSON, and use it from Python. datalab-to/indra describes Indra OCR as a 1M-parameter OCR model that runs on CPU. It turns PDFs and images into markdown, text, or JSON with boxes, and reads handwriting, 80+ languages and pages at any angle. It runs on ONNX Runtime, with no PyTorch and no GPU. The numbers from the README - Accuracy: 66.7 on OmniParseBench (71.4 with --math), ahead of PaddleOCR's full PP-StructureV3 pipeline at 58.4. Its text score of 46.7 is ahead of PaddleOCR's PP-OCR text models (35.8) and Tesseract (19.3). - Speed: about 44 pages per second on 32 CPU cores, which the README puts at 15x Tesseract and 10x PaddleOCR's PP-OCR text models on the same machine type and pages, and about 210 ms per page on 4 cores. - Size: the default models are about 76 MB, 30 MB of networks and 46 MB of language models for unsure lines. These are the project's own benchmarks, so treat them as a reason to test on your documents, not as a guarantee. Step 1: install You need Python 3.10 or newer. The README lists testing on Linux (x86_64 and aarch64) and macOS on Apple silicon. pip install indra-ocr The first run downloads the models from Hugging Face and caches them. To use a local copy, set INDRA_OCR_MODEL_DIR or pass --model-dir, which is what you want on an air-gapped server. Step 2: convert PDFs from the command line indra-ocr paper.pdf # markdown to stdout indra-ocr *.pdf --out results/ # one .md per input, all cores indra-ocr scan.png --format json # lines, boxes, layout blocks, confidences indra-ocr paper.pdf --format text # plain text, one line per detected line indra-ocr paper.pdf --mode ocr --timings # ignore the text layer, OCR every page indra-ocr paper.pdf --math # write math as LaTeX The default --mode auto is the part that saves time on mixed archives: it reads the PDF text layer where it passes the checks and only OCRs what is missing. Use --mode ocr when a text layer is present but garbage, which happens a lot with old scans. Page headers and footers are left out of the markdown unless you pass --keep-furniture. The CLI keeps going past files it cannot read and exits with 1 if any failed, so it is safe to point at a big folder. Step 3: check results in the viewer indra-ocr-viewer --port 8765 Open localhost and drop in a PDF or image. Each page shows its lines, layout blocks and reading order drawn over the page, with the rendered markdown, the lines with their confidences and the JSON. Files stay on your machine. Install pip install "indra-ocr[tesseract]" plus a Tesseract binary and the viewer can compare the two side by side, which is the fastest way to decide whether to switch. Step 4: use it from Python import indra_ocr conv = indra_ocr.load(threads=4) # mode="auto" | "ocr" | "text" for i, page in conv.convert("paper.pdf"): print(page.source, page.timings["total_ms"]) print(page.markdown) # headings, lists, tables as HTML page.blocks # [{"label", "bbox", "text"}] in reading order page.lines # [Line(box, text, conf)] in reading order For batches, convert_many runs one single-threaded worker per core and returns results in input order. Keep the if __name__ == "__main__": guard on macOS and Windows: import indra_ocr if __name__ == "__main__": for path, result in indra_ocr.convert_many(["a.pdf", "b.pdf"], cores=None): if isinstance(result, indra_ocr.InputError): continue # unreadable or locked file for i, page in result: print(path, i, page.source) Every block carries a label (Text, Section-Header, Table, Caption, Footnote, Form and more) and a bounding box, so feeding clean chunks to an LLM or a search index is a filter, not a regex project. What people have built on it The builds collected on the Indra page are mostly about replacing Tesseract in existing tools: - An OCRmyPDF engine plugin that runs Indra OCR instead of Tesseract and ships in a Paperless-ngx image, so Paperless keeps its skip, redo and force modes and only the recognition engine changes. - A Paperless-ngx 3.x parser plugin that runs Indra on ONNX Runtime on the CPU and produces searchable PDF/A archives with a text layer, with no extra service. - A WebAssembly port with a TypeScript API on npm that does the whole thing in the browser with no server. - A C# and .NET port on NuGet that keeps the same command line, viewer and output, with no Python needed. If you run Paperless-ngx at home, those two plugins are the quickest real-world win. When to reach for something else Indra is built for speed on CPUs. For very long documents, Baidu's Unlimited OCR is an open OCR model aimed at parsing documents of any length within a finite context window, and it is worth comparing on that kind of input. The Indra README itself points to Datalab's larger Chandra model for the hardest handwriting, math, complex tables and degraded scans. Caveats - Handwriting works but is the weak spot: the README reports 25.2% on the OmniParseBench handwriting tests. - The code is Apache 2.0, but the model weights use a modified OpenRAIL-M license: free for research, personal use and startups under $20M funding or revenue, and not for use that competes with Datalab's API. Check that before you ship it in a product. - This walkthrough follows the README and has not been run end to end by us. Source repo: datalab-to/indra. Want the setup commands for 50 AI repos people actually ship with? Get the free AI Repos That Actually Ship pack: dedupedai.com

datalab-to/surya-ocr-2.5 · Hugging Face Surya is a 689M param OCR model with these features: - Accuracy - scores 85.9 on OmniParseBench, the highest of the open models under 1B parameters we have scored, and 83.7% on olmOCR-bench - Speed - throughput of 4.5 pages/s on one H100 at 192 DPI, 1.7x Surya 2 with MTP (see Throughput) - Multilingual - scores 74.9% on an internal benchmark set of 91 languages (more here) - Layout analysis (table, image, header, etc.) with reading order - Tables as HTML, with merged cells as colspan / rowspan (row and column geometry needs a Surya 2 checkpoint; see Limitations) Commercial usage The Surya code is licensed under Apache 2.0. The model weights use a modified AI Pubs Open Rail-M license (free for research, personal use, and startups under $5M funding/revenue). For broader commercial licensing of the model weights, visit our pricing page here. Limitations - This is specialized for document OCR. Performance on photos or natural scenes is not the goal. - Layout / OCR / table_rec all need a running inference backend (vllm or llama.cpp). Detection runs purely on torch and works without it. - Surya 2.5 reads tables as HTML (full-page OCR, block mode, surya_table), but not row and column geometry: surya_table --mode simple returns an error. Use a Surya 2 checkpoint (SURYA_MODEL_CHECKPOINT=datalab-to/surya-ocr-2) for geometry. Github : datalab-to/surya huggingface.co

You can find Surya here: Model - huggingface.co Code - datalab-to/surya Blog - datalab.to

They released a tesseract successor (datalab-to/indra) Haven't tried it yet but we all know how tesseract is and the benchmarks show a beautiful spike in quality and speed. It's a 1M parameter model! I thought we'd never see the day but sub B parameter models are popping up left and right now. datalab-to/indra

Which OCR Solution Works Best for Your Documents? Master Full-stack AI Engineering In today's newsletter: - This inference engine runs LLMs 4x faster than llama.cpp: - Which OCR solution works best for your documents? - The No. 1 Deep Researcher beats Claude/ChatGPT using a counterintuitive trick. TODAY'S ISSUE LLM INFERENCE This inference engine runs LLMs 4x faster than llama.cpp: We ran Qwen3.5 9B on a base M5 MacBook Pro with three local inference engines, one at a time. - llama.cpp reached 22.0 tokens/s - MLX reached 25.1 tokens/s - uzu reached 92.1 tokens/s The video shows all three runs side by side. The first two numbers don’t mean llama.cpp or MLX are badly built. Both sit right under a hard limit of local decoding. Every new token needs one full pass through the model, and that pass reads all of the model’s weights from memory. A base M5 reads memory at about 153 GB/s, and a 4-bit 9B model weighs about 5.2 GB. So at one token per pass, the ceiling is roughly 153 ÷ 5.2, or about 29 tokens/s, however fast the GPU computes. Ordinary decoding stops near 29 tokens/s, while uzu gets about 7.5 tokens out of every pass. uzu gets past that ceiling by producing several tokens per pass instead of one. ↳ A small drafter (DFlash) proposes candidates for the next 16 positions at once. ↳ A tiny adapter (Weaver, 56.7M params) picks among them left to right so the guesses form coherent text, and keeps a few alternative branches as a tree. ↳ The 9B model checks the whole tree in a single pass and keeps the longest path it agrees with. The main model still decides every token that ends up in the output. In our run, each pass of the 9B model produced 7.5 tokens on average. Ordinary decoding produces exactly 1. The weights and kernels are built around this path too. uzu ships its own 4-bit checkpoints, and on M5 chips it runs the verification step on the GPU’s int8 matrix path. uzu is an open-source inference engine by Mirai Labs, written in Rust with Metal kernels. It has Swift, Python, TypeScript and Rust bindings, so the same engine runs inside a Mac or iOS app. Find the GitHub repo here → (don’t forget to star it 🌟) OCR Which OCR solution works best for your documents? The OCR model with the best average score can still be the wrong pick for your documents. We ran DeepSeek-OCR, PaddleOCR-VL and Datalab Surya on the same 150 pages, from handwritten notes and rotated scans to dense tables and a filled tax form. The video shows every page, what each parser returned, and why each test passed or failed. Most OCR comparisons end with one number per parser. That number tells you who’s ahead on average. It doesn’t tell you where a parser breaks on documents like yours. And nobody parses an average document. An insurance team parses scanned claim forms, a research team parses two-column papers full of equations, and a finance team parses dense tables. The parser that leads the average can still fail on the one page type you care about. So instead of one score, we checked every page with simple yes/no tests: - Is this line present in the output? - Is this cell under the right headings? - Does A come before B? Each test passes or fails, and every failure comes with a reason, like “text missing” or “marked as table, not heading”. The scoring runs as open code. No LLM judges the output. Each check is a small yes/no question, and the score is the share of checks a parser passes. Here’s the overall scoreboard from our 150-page run: → Datalab: 88.2 → DeepSeek-OCR: 77.5 → PaddleOCR-VL: 73.4 Datalab led the overall score in our run, but DeepSeek-OCR led on tables. If we had stopped at the overall column, the answer would look obvious. DeepSeek-OCR was the best on tables at 87.2, ahead of Datalab at 78.9 and PaddleOCR-VL at 78.0. So a team whose documents are mostly dense tables would pick a different parser than the average suggests. On handwritten notes, the gap went the other way. One parser (Datalab) passed 5 of 6 tests, while the other two passed 0. On a handwritten page, Datalab passed 5 of 6 tests while PaddleOCR-VL misread most of the lines. The filled 1040 tax form showed the starkest failure. PaddleOCR-VL returned only the form’s title, so every test looking for a filled-in value failed with “text missing”. On the same page, Datalab passed 8 of 8 tests and DeepSeek-OCR passed 6 of 8. On the filled 1040 form, PaddleOCR-VL returned only the title and passed 0 of 8 tests. A parse that returns only a title is a different risk from one with a few wrong cells. A single average folds both into the same number. No single parser won on every type of page. The right pick depends on your documents, not on who has the best average. The tool we used is OmniParse Bench, and it’s 100% open-source. It has 16,288 tests across 2,937 pages in 92 languages, drawn from 2,343 source documents. It builds on what olmOCR-bench showed, that pass/fail unit tests are a simple proxy for parsing quality that anyone can check. Datalab’s view is that olmOCR-bench has saturated, and what now separates the top systems there is mostly output conventions and overfitting rather than reading. Every test belongs to one of three families: text, tables and layout. A family’s score is the share of its tests that pass, and the headline is the mean of the three. A page where the parser call fails counts as 0, so a parser can’t look better by quietly dropping the hard pages. The headline averages three family scores, and tags only decide which tests you look at. Each test is tagged by what makes its content hard, like handwriting, rotation, tiny text, degraded scans, forms, tables, math and multi-column layouts, along with its script and language. Tags never change whether a test passes. They only select tests, so you can score a parser on handwriting alone, or on French tables alone. Every score also comes with the number of tests and documents behind it, so a thin slice tells you it’s thin. Each test also records how its answer was established, such as the publisher’s own markup, a form the authors filled in themselves, or agreement across a panel of independent models. Tags turn one benchmark into many small ones, so you can score parsers only on pages that look like yours. One thing to note is that Datalab built this benchmark, and Surya is Datalab’s own model. That’s exactly why the open scorer matters. You don’t have to trust anyone’s leaderboard, since you can rerun every test yourself and read why each one passed or failed. So you can plug in any OCR solution and see exactly where it holds up and where it breaks. Before you commit to a parser, run it on the slices that match your documents and read the failures yourself. GitHub repo: datalab-to/omni_parse_bench (don’t forget to star 🌟) AI AGENTS The No. 1 deep researcher beats Claude and ChatGPT with a trick neither uses. We studied the open-source architecture behind it. A counterintuitive thing we found is that the orchestrator agent that runs the entire research strategy has no search access. It can’t query the web or open URLs. This looks wrong at first glance. Every other deep research system gives its coordinator far more capability. For instance: - OpenAI’s approach trains a single model for many consecutive tool calls. It searches, reads, reasons, and writes the report in a long sequential chain. ↳ The researchers behind the No. 1 system (Onyx; GitHub Repo) observed that this causes the model to spend cycles on low-value searches instead of maintaining a high-level research strategy. - Anthropic and Google use an orchestrator-researcher pattern similar to Onyx’s system. The key difference is how aggressively Onyx constrains the orchestrator. Most orchestrators have access to search and retrieval tools alongside dispatch capabilities. And the moment an orchestrator can search, it will. So instead of decomposing a query into focused research threads, it starts answering the question itself. It pulls a few results, skips proper task decomposition, and produces a surface-level report from whatever it found first. Stripping search from the orchestrator forces it to write self-contained and coherent task briefs for each research agent. The researchers also kept the architecture only two levels deep. When info passes through multiple agents, each one subtly distorts it through summarization/reinterpretation. Keeping it to two levels prevents this. These two constraints sit inside a larger three-phase pipeline (the visual below maps this): - Phase 1 decomposes the query into up to 6 research directions. No tool access prevents the model from prematurely answering. - Phase 2 dispatches 3 isolated research agents. Each runs up to 8 sub-cycles of search, read, and think, to produce an intermediate report with citations. The agents can also search internal enterprise docs (Confluence, Slack, 100+ connectors) with document-level permissions enforced, unlike proprietary solutions. - Phase 3 runs a deterministic step that renumbers and deduplicates to produce a report with a unified citation map. This pattern has been ranked No. 1 on DeepResearch Bench. The whole implementation is available on GitHub, and you can try it yourself. Here’s the Onyx Repo → (don’t forget to star ⭐ ) Onyx GitHub Repo 👉 Over to you: Which page type breaks your current OCR pipeline? THAT'S A WRAP NO-FLUFF RESOURCES TO... Succeed in AI Engineering roles All businesses care about impact. That’s it! - Can you reduce costs? - Drive revenue? - Can you scale ML models? - Predict trends before they happen? We have discussed several other topics (with implementations) in the past that align with such topics. Master full-stack AI engineering Here are some of them: - Learn MLOps from first principles to production in this course with 18 parts → - Learn everything about MCPs in this course with 9 parts → - Learn how to build Agentic systems in this course with 14 parts. - Learn how to build real-world RAG apps, evaluate, and scale them in this course. - Learn sophisticated graph architectures and how to train them on graph data in this course. - So many real-world NLP systems rely on pairwise context scoring. Learn scalable approaches here. - Learn how to run large models on small devices using Quantization techniques. - Learn how to generate prediction intervals or sets with strong statistical guarantees for increasing trust using Conformal Predictions. - Learn how to identify causal relationships and answer business questions using causal inference in this course. - Learn how to scale and implement ML model training in this practical guide. - Learn techniques to reliably test new models in production. - Learn how to build privacy-first ML systems using Federated Learning. - Learn 6 techniques with implementation to compress ML models. Master full-stack AI engineering All these resources will help you cultivate key skills that businesses and companies care about the most. Today’s email was brought to you by Avi Chawla and Akshay Pachaar. Looking for more? Unlock our premium DS/ML resources. © 2026 Daily Dose of Data Science

Chandra OCR：手寫表單、掃描檔直接變 Markdown，連打的勾都留著（Datalab 開源）#shorts Chandra 是 Datalab 開源的 OCR 模型：掃描檔、照片、PDF 丟進去，出來是分好標題、段落、表格的 Markdown（也能輸出 HTML、JSON）。開源一年就破 1.2 萬星。 3 個官方範例： 1. 手寫表單 → 日期、名字、信箱變成文字，打勾的選項照樣打勾 2. 整頁密密麻麻的數字表 → 出來還是一格一格的 Markdown 表格 3. 手寫微積分筆記 → 算式變成數學式，手畫的圖也留下來 為什麼大家在傳： ・官方公布的 olmOCR 測試分數，Chandra 2 拿 85.8，比 GPT-4o（69.9）、DeepSeek OCR（75.4）都高 ・官方多語測試，中文 88.7 分，Gemini 2.5 Flash 70.0 分 怎麼試： ・不想裝：Datalab 官網的 playground 直接試 ・自己電腦跑：pip install chandra-ocr[hf] → chandra input.pdf ./output --method hf 來源： GitHub：datalab-to/chandra Hugging Face 模型：huggingface.co 官方 playground：datalab.to 畫面來源： ・Datalab Chandra 官方範例圖（GitHub README Examples）：datalab-to/chandra ・olmOCR 排行圖（Chandra README Benchmarks）：datalab-to/chandra ・指令：Chandra README Quickstart datalab-to/chandra #Chandra #OCR #Markdown #AI工具

"Indra OCR is a 1M-parameter OCR model that runs on CPU. It turns PDFs and images into markdown, text, or JSON with boxes, and it reads handwriting, 80+ languages, and pages at any angle. ➡️ 1M param OCR model - runs on CPU, 15x faster than tesseract, close to 1B+ param models in accuracy" datalab-to/indra Community Overview: lifehubber.com Resources are shared for discovery and are not independently vetted—please do your own due diligence. New resources are added regularly — feel free to join the sub for updates. Full searchable archive of all resources posted so far on our community site, LifeHubber: lifehubber.com 300+ open-ish AI models, agents, tools, datasets, and related resources, with filtering and sorting.

"Indra OCR is a 1M-parameter OCR model that runs on CPU. It turns PDFs and images into markdown, text, or JSON with boxes, and it reads handwriting, 80+ languages, and pages at any angle. ➡️ 1M param OCR model - runs on CPU, 15x faster than tesseract, close to 1B+ param models in accuracy" datalab-to/indra Community Overview: lifehubber.com Resources are shared for discovery and are not independently vetted—please do your own due diligence. New resources are added regularly — feel free to join the sub for updates. Full searchable archive of all resources posted so far on our community site, LifeHubber: lifehubber.com 300+ open-ish AI models, agents, tools, datasets, and related resources, with filtering and sorting.

Open source do GitHub: #10 a #6 | Vol. 13 #shorts Os projetos #10 a #6 do Top 15 open source do GitHub, Vol. 13, 10 segundos para cada um. #10 House Control: 0xProbabillity/house-control #9 Talorys: rociiu/talorys #8 Indra OCR: datalab-to/indra #7 net-merge: rohit267/net-merge #6 opendesk: useopendesk/opendesk Vídeo completo com os 15: youtu.be Oferecimento: Hallen Tech, hallentech.com #shorts #opensource #github

Mais 15 projetos open source do GitHub que valem o seu tempo | Vol. 13 Mais 15 projetos open source do GitHub que valem o seu tempo, em contagem regressiva e com 10 segundos para cada um: um conversor de 255 formatos que funciona arrastando, uma TV sem anúncios, o Figma offline, janelas do Mac no iPad e muito mais. 📺 Décimo terceiro volume da série Open source do GitHub. Escolhidos pela análise da lista inteira (estrelas, frequência de commits, último commit, README, documentação, capturas e site), conferida em 23/10/2026. Qual desses você vai testar primeiro? Conta nos comentários! ⏱️ CAPÍTULOS E LINKS 0:00 Abertura 0:10 #15 HQTUI: painéis de terminal no estilo btop em 10 linguagens, sem dependências profullstack/hqtui 0:20 #14 Amagine3D: descreva o aparelho e as medidas e ganhe a caixa em 3D editável, em STEP, STL ou 3MF (6,3 mil estrelas) amagine-ai/Amagine3D 0:30 #13 slides-grab: o agente cria slides em HTML e você seleciona uma área para pedir o ajuste; exporta em PDF (1,2 mil estrelas) NomaDamas/slides-grab 0:40 #12 Snitch: monitor de rede em tempo real com firewall de saída: veja e bloqueie cada conexão aixisstudio/Snitch 0:50 #11 Glide: gravador de tela open source para macOS no estilo Screen Studio, com zoom automático nos cliques michaelshimeles/glide 1:00 #10 House Control: a sua casa em 3D na barra do Mac e no iPhone, controlando luzes e ar-condicionado 0xProbabillity/house-control 1:10 #9 Talorys: agente pessoal de IA que roda na sua conta da Cloudflare, com memória, tarefas e lembretes (1,4 mil estrelas) rociiu/talorys 1:20 #8 Indra OCR: modelo de OCR minúsculo que roda na CPU, em mais de 80 idiomas e até letra de mão datalab-to/indra 1:30 #7 net-merge: usa Wi-Fi, cabo e celular ao mesmo tempo, somando a velocidade das conexões (1,8 mil estrelas) rohit267/net-merge 1:40 #6 opendesk: central de atendimento open source com e-mail, chat do site e base de ajuda numa caixa só useopendesk/opendesk 1:50 #5 editonny: skill para Claude Code e Codex que edita vídeo por conversa: corta pausas, legenda e arruma o áudio soumatheusgomes/editonny 2:00 #4 Desigma: alternativa gratuita e offline ao Figma, feita em Rust, que importa arquivos do Figma deepfounder-ai/desigma 2:10 #3 Sill: qualquer janela do Mac no iPhone ou iPad, com toque, teclado e Apple Pencil, sem conta nem servidor (3,9 mil estrelas) Saffsanity/sill 2:20 #2 cleanyourtv: tira anúncios e apps de fábrica da TV pelo Wi-Fi, de forma reversível (24,8 mil estrelas) Gathin23/cleanyourtv 2:30 Oferecimento Hallen Tech 2:44 #1 Filee: conversor de arquivos offline: segure uma tecla, arraste e converta entre 255 formatos (10,3 mil estrelas) KnifeLemon/Filee 2:54 Encerramento 🧡 Oferecimento: Hallen Tech, liderança de tecnologia sob demanda, software, cloud e segurança para a sua empresa. Site: hallentech.com E-mail: contato@hallentech.com Instagram: instagram.com LinkedIn da Hallen Tech: linkedin.com LinkedIn do Eddwan: linkedin.com Se curtiu, deixa o like, segue o canal e ativa as notificações para a próxima lista. #opensource #github #programacao #ferramentas