Daniel van Strien

Uploaded a dataset of 98,877 historical newspaper pages (1700s–1940s) to the Hub, each with its original OCR text, word boxes and confidence scores. huggingface.co/datasets/big...

Newspapers from @europeana.bsky.social

huggingface.cobiglam/europeana_newspapers · Datasets at Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

The Europeana Newspapers dataset on @hf.co now has an `alto` config: the raw ALTO XML for all 5.9M pages. The coordinates for every word, line and block, per-word OCR confidence and font info that the flattened text dropped are back. huggingface.co/datasets/biglam/europeana_newspapers

16 Sept

Made some improvements to the OCR scripts onboarding in my uv-scripts collection. First run is now a small example: seven scanned NASA pages in, Markdown out, one HF jobs command. It also works on your own images or a folder of PDFs. You can pick from 28 OCR models. huggingface.co/datasets/uv-...

Daniel van Strien reposted

11 Sept

Need a historical illustration? Search 1.49 million images from British Library books and Britannica (1500s–1920s). Find similar images, download full-resolution images or transparent PNG cutouts, and explore the sources. huggingface.co/spaces/davan...

11 Sept

10 Sept

Used Astra + Jobs to see how well this model performs on the full olmOCR-bench Unsurprisingly, it doesn’t do brilliantly overall: 36.8% But for a ~16M-parameter recogniser, I think 74.4% on long/tiny text and 57.9% on multi-column pages are pretty interesting.

Daniel van Strien@danielvanstrien.bsky.social

A 15.9M-parameter specialist OCR model outperforms many VLMs hundreds of times larger. On a 2,165-page historical eval dataset, Kraken PP-OCRv6 ranks #4 on reading CER and #1 when long-s, ligatures and case are preserved. huggingface.co/spaces/fineb...

github.comocr-bench/experiments/kraken-olmocr-bench at main · davanstrien/ocr-benchPer-collection OCR leaderboards using VLM-as-judge - davanstrien/ocr-bench

10 Sept

This includes Kraken’s BLLA segmentation, with raw text output and no table/LaTeX reconstruction. Full results + scripts to reproduce the run: github.com/davanstrien/...

Daniel van Strien reposted

Adina Yakup

@adinayakup.bsky.social

10 Sept

DeepSeek v4.1 Flash is just another level 🤯 huggingface.co/deepseek-ai/... - Asymmetric Causal-Encoder-Decoder: 550B MoE, input 8B / output 16B - Native vision merged into one endpoint - KV cache crushed: ~1/4 the HBM vs last one, 437× smaller than their first model

huggingface.codeepseek-ai/DeepSeek-V4.1-Flash · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.coOCR on the Hub - a davanstrien CollectionCurated OCR models for documents, languages, handwriting and text in images. Browse four collections with short practical notes.

7 Sept

OCR for Japanese manga, Swedish handwriting or Arabic print? There’s a growing range of OCR models on @hf.co: VLMs, dedicated text recognisers and complete OCR pipelines. I’ve gathered 41 models into four collections, with short notes to help you choose: huggingface.co/collections/...

3 Sept

kraken.rekraken — kraken documentation

3 Sept

Fun fact: I had to adjust the plotting logic to get it to show properly because the model does so well! Kraken OCR library: kraken.re/main/index.h... Model: huggingface.co/small-models...

Daniel van Strien reposted

Tom Aarsen

@tomaarsen.com

26 Aug

📈 New blog post: Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers. As a practical example, I finetuned a ColBERT-style model for medical retrieval. 14.5 hours on one RTX 3090, and it beats every general-purpose retriever I could find. Thread 🧵

26 Aug

Every illustration in the Encyclopaedia Britannica dataset now has an instance mask: 411,385 cut-outs from 115,293 pages, 1768–1929, each linked to its full-resolution scan. Public domain, no image generation involved. Work in progress.

Daniel van Strien@danielvanstrien.bsky.social

Uploaded a dataset of 115,293 illustrated pages from the Encyclopaedia Britannica, 1st edition (1768) to 14th (1929) to the Hub huggingface.co/datasets/big...

25 Aug

Uploaded a dataset of 115,293 illustrated pages from the Encyclopaedia Britannica, 1st edition (1768) to 14th (1929) to the Hub huggingface.co/datasets/big...

20 Aug

Every illustration in the British Library's 19th-century book images dataset now has an instance mask: 1.02M clean cutouts of public-domain engravings, no image generation involved! One small segmentation model, $3.24 of compute. Model, masks and a search Space with cutout view all open.

huggingface.cobiglam/british-library-book-images · Datasets at Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

20 Aug

Dataset (crop_masks config): huggingface.co/datasets/big... Search + cutouts: huggingface.co/spaces/davan... Model: huggingface.co/small-models...

danielvanstrien.xyzDistilling Qwen3.8 with datatrove on Hugging Face Jobs – Daniel van Striendatatrove’s new Jobs backend plus a days-old 27B teacher: regenerating a 35,837-summary training dataset in one afternoon for $15.58, with a calibration-first workflow and length-controllable outputs.

18 Aug

Synthetic data at scale without owning a GPU: datatrove's new Jobs backend + Qwen3.8-27B → 35,837 length-controllable TL;DRs of Hugging Face cards, $0.43 per 1,000. Full guide: danielvanstrien.xyz/posts/2026/d...

FineBooks wants to help re-OCR the public domain to unlock a rich source of data for AI, for researchers, and for enjoyment. Step one: work out which modern OCR models are actually good enough.

So we built a leaderboard. 14 open models, scored on 2,165 pages of expert human transcription. The best read historical print at ~97.6% character accuracy, for under $2 per 1,000 pages.

huggingface.coFineBooks: are open OCR models good enough to unlock historical knowledge?A Blog post by FineBooks on Hugging Face

A new @hf.co and @eleutherai.bsky.social project, with @storytracer.com! Blog post: huggingface.co/blog/fineboo...

Daniel van Strien reposted

Sebastian Majstorovic

@storytracer.com

huggingface.coFineBooks: are open OCR models good enough to unlock historical knowledge?A Blog post by FineBooks on Hugging Face

Are open OCR models good enough to unlock historical knowledge? @danielvanstrien.bsky.social and I built a leaderboard using six expert-transcribed volumes from the @biodivlibrary.bsky.social sky.social. Meet FineBooks, from @hf.co and @eleutherai.bsky.social! huggingface.co/blog/fineboo...

7 Aug

Uploaded a dataset of 1,080,814 public domain images, mostly from 19th-century books, to the Hugging Face Hub: huggingface.co/datasets/big... You can also do semantic search against the images here: huggingface.co/spaces/davan...

3 Aug

Public data on which coding agents actually use the Hugging Face Hub, refreshed monthly. Claude Code now sends a majority of everything the Hub can attribute to a named agent. In May, it led on 4 days out of 30. In July, all 30. huggingface.co/datasets/hug...

3 Aug

Shares are relative, so a falling share doesn't mean falling usage; the attributed base is growing. 23% of agent traffic still comes from tools that haven't registered, down from 37% in June. Curious whether the open harnesses close the gap in August!

31 July

A film catalogue tells you what a film is about, not what happens inside it. So: 1,864 public-domain Prelinger films, described every ~60 seconds by an open 2B video model. 23,148 searchable moments. Search "typing on a computer keyboard", land on the second it happens.

huggingface.coPrelinger moments - a Hugging Face Space by davanstrienTimestamped search over VLM-captioned Prelinger films

31 July

The model describes what's on screen; it can't tell you who anyone is or why a film was made. That part stays human. ~$10 in compute on @hf.co Jobs Unreviewed output: good for finding, not citing. huggingface.co/spaces/davan...

30 July

New recipe: timestamped video captions on @hf.co Jobs. Point it at a bucket of videos → parquet dataset out: scene descriptions + second-precise <start – end> events. ~$0.05 per hour of footage on a single A10G (Marlin-2B on vLLM). huggingface.co/datasets/uv-...

Daniel van Strien reposted

14 July

Small open models are getting genuinely good at document parsing: OvisOCR2 (0.9B, Apache 2.0) is claiming SOTA on OmniDocBench v1.6. Day-1 recipe: OCR a whole HF image dataset — digitised newspapers, archives, zines — to markdown with one command. huggingface.co/datasets/uv-scripts/ocr

@jamesfeigenbaum.bsky.social

14 July

Daniel van Strien reposted

James Feigenbaum

I don't have it yet, but there's an econ history paper in these transcripts or transcripts like them (that we can now turn into textual data at the cost of pennies per hour)

Daniel van Strien@danielvanstrien.bsky.social

Open ASR models with speaker diarization are now fast and cheap: I diarized 174 hours of Apollo 11 mission audio (the real July 1969 NASA tapes) for $9.46 with a 0.9B open model. Search it, hear any moment on the original tape: huggingface.co/spaces/davan...

huggingface.codavanstrien/apollo-11-diarized · Datasets at Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

The transcript is a finding aid, not a replacement — every segment links back to the Internet Archive originals. And the pipeline flagged which reels are blank carrier hiss: useful collection metadata in its own right. Dataset (CC0): huggingface.co/datasets/dav...

huggingface.couv-scripts/transcription · Datasets at Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

It's unedited machine output over scratchy radio (it keeps hearing "Follow eleven, this is Houston"). The recipe runs on any audio collection in one command: huggingface.co/datasets/uv-...

I ran 10 newer OCR models on @ai2.bsky.social's olmOCR-bench "old scans" subset. The ranking flips depending on what you actually want.

View full thread

Two other things: a 1B model (LightOnOCR-2) has the best raw transcription in the field, and PaddleOCR-VL 1.6 sometimes hallucinates Chinese characters on English scans.

danielvanstrien.xyzBenchmarking OCR fairly is harder than it looks – Daniel van StrienI ran ten current OCR models on olmOCR-bench’s ‘old scans’ subset. The ranking flips depending on one thing: whether you want a model that keeps what’s on the page, or one that cleans it up. A note on...

Write-up + plus how I ran the whole thing on HF Jobs, no local GPU: danielvanstrien.xyz/posts/2026/o...

I ran 10 newer OCR models on @ai2.bsky.social's olmOCR-bench "old scans" subset. The ranking flips depending on what you actually want.

View full thread

The score rewards dropping boilerplate, i.e. letterheads, stamps, page numbers, so a model that reads the page more faithfully can rank lower.

IMO this is because a lot of VLM-based OCR models were made to provide tokens for training. It's less useful if you want faithful OCR of the whole page, like an archive where the letterhead is part of the record.

I ran 10 newer OCR models on @ai2.bsky.social's olmOCR-bench "old scans" subset. The ranking flips depending on what you actually want.

On the headline score, PaddleOCR-VL beats NuExtract3 (38.6 vs 37.8). But rank by how much of the page each model actually reads, and NuExtract3 is well ahead (41.6 vs 31.2). Same two models, opposite order.

2 July

Coding agents are real users of the @hf.co Hub! They're searching for models, building and pushing datasets, training models on Jobs, spinning up Spaces... Now there's public data: each agent's share of Hub traffic, updated monthly 👇

huggingface.cohuggingface/agent-usage · Datasets at Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

2 July

Explore it in the Dataset Viewer, no code needed: - the monthly leaderboard + how it shifts as new tools launch - request share vs user share (a few heavy pipelines vs many light users) - daily data: launch spikes, weekday vs weekend patterns... huggingface.co/datasets/hug...

2 July

The latest @commoncrawl.bsky.social crawl indexes way more than HTML pages. 20.9M PDFs. Plus calendars, BibTeX, markdown... The whole index now lives in a @hf.co Bucket, so I pulled this with one SQL query straight over it (new S3 API). 2.1B rows, nothing downloaded, $0 to read.

30 June

You can now use 100s of tools with @hf.co Buckets, thanks to the new S3 API! Usually just one or two lines to change. huggingface.co/docs/hub/sto...

29 June

Fairly benchmarking OCR models is hard! Ran a few newer OCR models on @ai2.bsky.social's olmOCR-bench "old scans" subset. Worth knowing: the score "punishes" models for extracting too much (letterheads, stamps, etc.)

29 June

A model that reads the page more faithfully can rank lower. Works well as a score if you want clean "reading-order" tokens for LLM training, but less useful as a metric if you care about is "faithful" OCR.

Daniel van Strien reposted