Building the first foundational model for agent personalization.
External link for Fastino
Fastino reposted this
One of my favorite features in GLiNER2 is how you can combine multiple tasks in a single model call 🐙. In this example, I'm parsing a hotel booking query and simultaneously extracting entities (hotel name, dates, budget) 📍 while performing text classification (urgency level, trip type) 🏷️ all in ONE forward pass, on CPU, in ~100ms ⚡ 💡 Try it: https://lnkd.in/eFzv3-Fj
Fastino reposted this
One of the core innovations in GLiNER2 is its ability to produce structured JSON directly ✨🎯 without relying on text generation. This unlocks fast ⚡, deterministic 🔒, and hallucination-free 🚫🌫️ information extraction. A key technical challenge addressed in this work is entity grouping 🔗🧩: ensuring that multiple attributes referring to the same underlying object instance are reliably associated. Achieving consistent parent–child aggregation 🌳👨👧👦 required architectural mechanisms that go beyond conventional span-based NER models. We also designed a schema language with predefined choice constraints 🧬, allowing fields like category to be restricted to values such as [hardware, phone, laptop, other]. This enables controlled and reliable structured extraction. Huge thanks to my colleagues at Fastino 🙌 for the brilliant collaboration and the countless research discussions that made this possible. More to come soon 🚀.
Amazing work from the NVIDIA AI team! NVIDIA fine-tuned Fastino's GLiNER to detect and classify a broad range of Personally Identifiable Information (PII) and Protected Health Information (PHI) in structured and unstructured text. NVIDIA’s GLiNER-PII reaches 92% recall on the Nemotron-PII benchmark and is already available on Hugging Face (link below). What they released includes: 🔹 Nemotron-PII Dataset — 100K fully synthetic records across 50+ industries, generated using NeMo Data Designer to ensure privacy-compliant training and evaluation. 🔹 GLiNER-PII Model — A fine-tuned version of Fastino’s GLiNER optimized for PII/PHI detection across emails, clinical notes, legal docs, and more. We’re thrilled to see GLiNER powering privacy-critical use cases at this scale. Millions of developers now fine-tune GLiNER for their own applications, and it’s awesome to see NVIDIA’s contribution! ➡️ We also just release GLiNER2 with open source and hosted versions. We’ll be releasing our internal fine-tuning tools soon to make it even easier for teams to build specialized GLiNER and GLiNER2 variants. Stay tuned! NVIDIA's GLiNER model 🔗 https://lnkd.in/gbPZDMFT GLiNER2 now available! 🔗 https://fastino.ai/gliner
Amazing work from the NVIDIA AI team! NVIDIA fine-tuned Fastino's GLiNER to detect and classify a broad range of Personally Identifiable Information (PII) and Protected Health Information (PHI) in structured and unstructured text. NVIDIA’s GLiNER-PII reaches 92% recall on the Nemotron-PII benchmark and is already available on Hugging Face. What they released today includes: 🔹 Nemotron-PII Dataset — 100K fully synthetic records across 50+ industries, generated using NeMo Data Designer to ensure privacy-compliant training and evaluation. 🔹 GLiNER-PII Model — A fine-tuned version of Fastino’s GLiNER optimized for PII/PHI detection across emails, clinical notes, legal docs, and more. We’re thrilled to see GLiNER powering privacy-critical use cases at this scale. Millions of developers now fine-tune GLiNER for their own applications, and it’s awesome to see NVIDIA’s contribution! ➡️ We’ll be releasing our internal fine-tuning tools soon to make it even easier for teams to build specialized GLiNER variants for their domains.
NVIDIA Nemotron-PII: Privacy-Safe Training Data + PII Detection 💡 Training or evaluating models on emails, clinical notes, or legal documents requires careful PII/PHI handling. We're releasing a fully synthetic dataset and open-source detection model to help. What's included 👇 • Nemotron-PII dataset: 100K synthetic records spanning 50+ industries, built with NeMo Data Designer • GLiNER-PII model: Fine-tuned for PII/PHI detection (92% recall, 64% F1) Available on Hugging Face: https://lnkd.in/d9bTnZTt
We are thrilled to announce the launch of GLiNER2, the next-generation open source model from Fastino built for unified entity extraction, classification and structured data parsing. After the success of the original GLiNER (200k+ monthly downloads currently on Hugging Face), we’ve taken a major leap forward: • One single model for NER, text classification and JSON-structure extraction in a single pass • CPU deployment with <150 ms latency • Open source (Apache-2.0) and Fastino hosted versions Special thanks to our lead researcher on the project Urchade Zaratiana after an awesome EMNLP 2025 presentation last week! 🔗 GitHub: https://lnkd.in/gEuMA-6g 🔗 Models: https://lnkd.in/g-bzbFkA 🔗 Docs: https://lnkd.in/gugS4_Z7 🔗 Hosted API: https://lnkd.in/gsA9WTf3 We invite the community to star, fork and build with GLiNER2 today!
Fastino reposted this
🎉 Excited to officially release #GLiNER2, presented today at at #EMNLP2025 🇨🇳 during the demo session! 💡 GLiNER2 extends beyond NER to support Text Classification 🏷️ and Structured Extraction 🧩 in a single unified model. Specifically, we demonstrate that it's possible to predict hierarchical (JSON-like) structures 🪆 using only a transformer encoder, without the need for text generation. Moreover, GLiNER2 can efficiently handle multiple classification and extraction tasks in a single forward pass within a multi-task setup. ✅ All-in-one model: NER, classification, and structured extraction 🔄 ✅ Runs anywhere: CPU-first, lightweight, and fast ⚙️ ✅ Powerful results: strong performance without compromise 💪 This work was done at Fastino with my amazing colleagues: Gil Pasternak, George Hurn-Maloney, and Ash Lewis 🚀 Get Started: 💻 GitHub: https://lnkd.in/eFzv3-Fj 🎮 Demo: https://lnkd.in/ez2v7XsX 🤗 Models: https://lnkd.in/enZ2dBvN 📄 Paper: https://lnkd.in/eAgBshHE #EMNLP2025 #NLP #InformationExtraction #GLiNER
We’re building some exciting new products at Fastino. But before releasing them, we want your input. We invite you to join a small group of professionals in our user testing cohort. ✅ Join 3 short 30-minute sessions over 6 weeks ✅ Get early access to Fastino’s latest products ✅ Receive a gift card for participating We’re especially looking for folks who are: - Using AI tools in their work - Curious about building or testing AI/LLM-powered tools - Currently working in a full-time role If that sounds like you, we’d love to hear from you. 👉 Apply in under 2 minutes here: https://lnkd.in/gWvVzuDJ We’re keeping the cohort small, so if you’re interested, please apply soon!
Thanks to PwC for hosting the Private Equity Digital Executive Forum event in NYC and assembling such an impressive lineup of speakers! Our founder George Hurn-Maloney from Fastino joined a State of AI panel alongside Scott Likens (PwC, Global Chief AI Engineer) and Shaown Nandi (Amazon Web Services (AWS), Director of Technology). We explored how foundation models are evolving to meet real enterprise needs - particularly for agentic tasks like function calling, summarization, and data structuring. Special thanks to Mahera (Walia) Mayer for moderating the thought-provoking session!
Huge thanks to the MLOps community for the shoutout! Come find us and other great teams at the AI Agent Builders Summit in San Francisco next week 👇 📍 May 28 | The Hibernia | 4:30 PM 🔗 https://lnkd.in/gt7d_-H5
Big ups to our friends at Fastino 💥 🚀 They build Task-specific Language Models (TLMs). Think of it this way: instead of using a giant general-purpose AI for every little thing, Fastino crafts smaller, highly focused AI models. 🚀 These are like precision instruments, expertly trained for particular jobs, whether it's quickly summarizing text, pulling out specific data, or even translating instructions into code. This means they often perform these specific tasks faster, more accurately, and usually with a more predictable price tag than their larger counterparts. Pretty cool, no? 🤘 Come meet them and bunch of other cool teams next Wednesday at the Hibernia : ) This one’s for builders who care about speed and scale. 🔗 https://lnkd.in/gt7d_-H5
What an incredible evening in New York! 🌇✨ A huge thanks to everyone who joined us and Insight Partners for our rooftop happy hour in NYC. We loved connecting with so many talented AI developers, builders, and investors. Special shoutout to Michael Spiro from Insight Partners for co-hosting with our founder, George Hurn-Maloney, and for sparking some great conversations about the future of AI and task-specific language models. Thank you for making this event memorable, see you next time big apple! If you missed us this time, fill out this form for first dibs on our next event: https://lnkd.in/e9ZsFC27
LinkedIn is better on the app
Don’t have the app? Get it in the Microsoft Store.
Open the app