China open-sourced a peanut-sized OCR that parses entire 100-page PDFs in one shot..
It's called Unlimited-OCR. Only 3B params. Runs locally.
Every other OCR tool chops your doc into pages and loses the thread. this one reads the whole thing in a single pass.
→ One-shot "long-horizon" parsing (32K context window)
→ Multilingual, out of the box
→ 93% on the standard parsing benchmark (+6 over baseline)
→ <0.11 error rate past 40 pages
→ Runs 100% locally on your own hardware
→ Works with Transformers, vLLM, SGLang, Docker, Ollama, llama.cpp
Traditional cloud OCR (Textract, Google Vision, Azure Doc Intelligence) costs $1.50–$15 per 1,000 pages.
This runs on your machine. For free. Forever.
Baidu built it explicitly to push DeepSeek-OCR one step further. Already at 1.9M downloads on Hugging Face and most people have no idea it exists yet.
100% open source.
🗺️ Mapas Sincronizados en R para Análisis Comparativo
Si necesitas visualizar el cambio o la diferencia entre dos variables geográficas (tipo 'Antes' y 'Después'), te recomiendo crear gráficos dinámico con {mapview} y {leaflet.extras2} de R.
#Rstats#DataScience#Geospatial#GIS
Google engineer explained how to fine-tune a tiny LLM from 46% to 90% accuracy on your phone in 21 minutes - better than $1500 on-device AI bootcamps.
pick Gemma 270M -> generate synthetic task data -> fine-tune with LoRA -> quantize to int4 -> deploy to Pixel and hit 2000 tokens per second.
That loop is how a 270M model beats a 70B one on your task, running fully offline in your pocket.
Gemma 270M + synthetic data + LoRA + int4 quantization + on-device runtime - that's the stack.
Watch and save it, then fine-tune your own tiny agent tonight.
I built a Cloud Optimized Geotiff explorer: Take any local COG, it breaks down visually the size, shape & resolution of each overview & tile
I'm exploring Fable to make better technical visuals that I wish I had when learning all this
THE RULE THAT GIANT MODELS REQUIRE GIANT RAM IS OFFICIALLY DEAD
COLIBRI runs GLM-5.2, a 744B model, on a 25GB machine with no GPU.
Because the model only needs a fraction of its parameters at once, colibrì smartly holds the core in RAM and streams the rest from disk on the fly.
While disk speeds limit how fast it types, getting perfect responses from a massive model on a consumer rig is a stunning proof of concept!
★ 2.1k stars · Apache-2.0
100% free and open-source.
Repo in 🧵↓
Por primera vez, Colombia presenta el mapa más completo de emisiones de gases de efecto invernadero, con información histórica de 1990 a 2024 y un nivel de detalle que permitirá conocer las emisiones municipio por municipio y departamento por departamento.
Te invitamos a conocer estos resultados, el lanzamiento de SINGEI, la nueva plataforma para consultar esta información, y cómo estos datos fortalecen la toma de decisiones frente al cambio climático.
📅 Viernes 10 de julio
🕓 4:00 p. m.
📍 FIMA – Corferias, Bogotá | Sala Chiribiquete
🎟️ Entrada libre con cupo limitado
🔗 Inscríbete aquí: https://t.co/3o7rwdhjbp
#SomosClima cuando conocemos el territorio, entendemos nuestras emisiones y actuamos con información basada en evidencia. 🌱📊
#FIMA2026 #CambioClimático #SINGEI #Ideam #DelDatoALaAcción
mapgl 0.5.0 for R lets you sync up as many maps as your browser can handle.
Shown here: 16 distinct demographic maps for the Atlanta area; interact with one, the other 15 are kept in sync with a laser pointer tracking your position.
Now I'm not necessarily recommending a 16-panel map... but it's fun to see what's possible.
Alibaba just released ABot-Earth 0.5
A generative 3D model that builds seamless environments from satellite imagery in under 10 minutes per km².
It uses 3D Gaussian Splatting for real-time web maps and embodied AI like UAV navigation.