onprem unstructured data extraction with 4 lines of code
The traditional pipeline for unstructured data extraction typically follows these steps: 1. Image → OCR Model (e.g., Google Vision) → Layout Model (e.g. Surya) → LLM → Final Answer However, this can be streamlined using a Vision-Language Model (VLM): 2. Image → VLM → Final Answer Recently VLMs have improved a lot for OCR and document understanding tasks, specifically the Qwen-2.5-VL series. We can run the Qwen-2.5-VL-7B-AWQ model locally with just 16GB VRAM, and perform end-to-end information extraction (fields and table extraction) without any external models. Hallucination with VLMs One…
What it does
In the maker’s words, at launch
The traditional pipeline for unstructured data extraction typically follows these steps: 1. Image → OCR Model (e.g., Google Vision) → Layout Model (e.g. Surya) → LLM → Final Answer However, this can be streamlined using a Vision-Language Model (VLM): 2. Image → VLM → Final Answer Recently VLMs have improved a lot for OCR and document understanding tasks, specifically the Qwen-2.5-VL series. We can run the Qwen-2.5-VL-7B-AWQ model locally with just 16GB VRAM, and perform end-to-end information extraction (fields and table extraction) without any external models. Hallucination with VLMs One question I am asked often is hallucination with VLMs compared to OCR model. This is a valid point. But, even with correct OCR and Layout formatting, LLM can still hallucinate, and can give incorrect final answers. Layout models often struggle with complex documents (e.g., tables, complex sparse document). If the formatted text from the layout model is incorrect the LLM model will always produce incorrect extraction with high confidence. Check out our GitHub repo for implementation details: GitHub: https://github.com/NanoNets/docext Would love to hear suggestions for improvement!
Does the same job
all alternatives →- Q2Qwen-2.5-32B is now the best open source OCR model2025 · github.com · ▲211
Last week was big for open source LLMs. We got: - Qwen 2.5 VL (72b and 32b) - Gemma-3 (27b) - DeepSeek-v3-0324 And a couple weeks ago we got the new mistral-ocr model. We updated our OCR benchmark to include the new models. We evaluated 1,000 documents for JSON extraction accuracy. Major takeaways: - Qwen 2.5 VL (72b and 32b) are by far the most impressive. Both landed right around 75% accuracy (equivalent to GPT-4o’s performance). Qwen 72b was only 0.4% above 32b. Within the margin of error. - Both Qwen models passed mistral-ocr (72.2%), which is specifically trained for OCR. - Gemma-3…
- IJI just open sourced my document/website extractor for Vision-LLMs2024 · github.com · ▲37
Hi HackerNews, Lately, I have seen an explosion in posts offering paid APIs/services to get unstructured data into LLMs (i.e. langchain extract, ragflow, unstructured, unstract, just to name a few) and I have been largely disappointed by them, either because they fail to implement multimodal support, fail to give good context for "really tricky" PDFs / Word docs / Powerpoints, or are just plain difficult to use. In light of all these posts I figured I'd share my solution that has been working smoothly for me and my clients. I put it up on GitHub for free so you can check it…
- OPOCR pipeline for ML training (tables, diagrams, math, multilingual)2025 · github.com · ▲170
Hi HN, I’ve been working on an OCR pipeline specifically optimized for machine learning dataset preparation. It’s designed to process complex academic materials — including math formulas, tables, figures, and multilingual text — and output clean, structured formats like JSON and Markdown. Some features: • Multi-stage OCR combining DocLayout-YOLO, Google Vision, MathPix, and Gemini Pro Vision • Extracts and understands diagrams, tables, LaTeX-style math, and multilingual text (Japanese/Korean/English) • Highly tuned for ML training pipelines, including dataset generation and…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, April 2025
the whole month →- IB
Hi everyone, I built PyXL — a hardware processor that executes a custom assembly generated from Python programs, without using a traditional interpreter or virtual machine. It compiles Python -> CPython Bytecode -> Instruction set designed for direct hardware execution. I’m sharing an early benchmark: a GPIO test where PyXL achieves a 480ns round-trip toggle — compared to 14-25 micro seconds on a MicroPython Pyboard - even though PyXL runs at a lower clock (100MHz vs. 168MHz). The design is stack-based, fully pipelined, and preserves Python's dynamic typing without static type restrictions.…
Dev tools · 2025 · runpyxl.com
- UC
Life & fun · 2025 · filiph.github.io
- IB
https://the-pocket.github.io/Tutorial-Codebase-Knowledge/
AI · 2025 · github.com
