Gino Brunner

💻 Why we still need small language models – even in the age of frontier AI Following a focused six-week sprint, our Project t0 team showed that small, open-weight language models can achieve near-frontier reasoning performance when applied to focused, domain-specific tasks.  By combining retrieval-augmented generation, reasoning trace fine-tuning and budget forcing at inference time, the team has built a 3B parameter model capable of tackling real-world health queries – and is small enough to run locally on a laptop. Federico Nanni, James Geddes, Ryan Chan and Tomas L. from the project team explore the details in this blog: https://bit.ly/4mfv4t7 To support others working on lean model deployment, we’re publishing: ✅ Two technical reports documenting our pipeline, data generation, fine-tuning on HPC infrastructure and evaluation process. ✅ Our code and trained models, open-sourced for reproducibility and reuse. ✅ The code for setting up a simple interface for multi-turn conversational interaction, allowing users to test the model in a realistic setting. This work is designed to support teams in compute-constrained or privacy-sensitive environments to build practical AI assistants that don’t require frontier-scale resources. 🔎 Learn more about our work and read the technical reports: https://bit.ly/4mfv4t7 #AI #ArtificialIntelligence

245

7 Kommentare

Command-line agents can get you really far in document search and analysis! We tested SemTools, our CLI toolkit for parsing and semantic search, with coding agents like Claude Code on 1000 Arxiv papers. The results show that combining Unix tools with semantic search capabilities creates surprisingly capable knowledge workers. 🔍 SemTools adds parse and search commands that let agents handle complex documents with fuzzy semantic keyword search 📊 Agents with semantic search provided more detailed, accurate answers across search, cross-reference, and temporal analysis tasks ⚡ CLI access proves incredibly powerful relative to effort - leveraging existing Unix tooling instead of building custom RAG infrastructure 🛠️ The combination of grep, find, and semantic search handles a wide variety of document tasks at high fidelity Learn about our SemTools experiment and see the full benchmark results: https://lnkd.in/eBq-vprb

57

2 Kommentare

gLSTM extends xLSTM to a graph neural network architecture: https://lnkd.in/da3at-Y9 "gLSTM mitigates sensitivity over-squashing and capacity over-squashing." "gLSTM achieves comfortably state of the art results on the Diameter and Eccentricity Graph Property Prediction tasks, and very strong performance on SSSP"

299

5 Kommentare

Our team has built the first Relational Foundation Model (KumoRFM) – a foundation model purpose-built for relational databases (https://kumorfm.ai). Just as LLMs transformed unstructured data like text and images, KumoRFM brings the power of foundation models to structured, relational, table-based data. 🔍 What does this mean? Instead of building and tuning separate machine learning models for every predictive task (fraud detection, churn prediction, recommendations, etc.), enterprises can now query their data warehouse directly—in natural language or with our Predictive Query Language (PQL)—and get predictions in seconds. 🧠 Under the hood, KumoRFM: - Represents relational data as a temporal heterogeneous graph. - Uses a Relational Graph Transformer for in-context learning. - Seamlessly adapts to unseen database schemas and diverse predictive tasks—no manual tuning required. ⚡ Key benefits: - Accurate predictions for any predictive task across industries (finance, e-commerce, healthcare). - Trust and transparency via explanations and quantitative evaluation. - Flexible deployment: Kumo SaaS or Data Warehouse Native (Snowflake, Databricks). I’m proud to have co-authored the KumoRFM paper, which details how we’re rethinking AI for enterprise data. This is just the beginning of a new era where relational data becomes truly predictive. 👉 Read more about the work and see the full article at Blocks & Files: https://lnkd.in/d4vTsUMT

304

13 Kommentare

Now on GitHub ⚙️ A new pull request adds a diffusion-based trajectory planner as part of a modular end-to-end approach in #Autoware, where machine learning supports both prediction and planning. This enables more natural driving behavior in complex scenarios 🚗💻 See details🔗 https://hubs.li/Q03G1MJP0 #EndtoEnd #AI #OpenSource #TIERIV

156

💭 What if you could train SOTA retrieval models without knowledge distillation? With PyLate, LightOn makes it possible: ⚡️ Direct large-scale contrastive training, no teacher needed  📚 Train on billions of passages with GradCache & distributed infra  🎯 Better generalization across datasets (BEIR, BRIGHT, …) 🌐 Seamless multi-node scaling 👉🏻 Faster, smarter, more robust retrieval training. Learn more : https://lnkd.in/euPXMaEv

47

🤔 Wondering if the new Astro IDE already recognizes the latest Airflow 3.1 features? What about comparing Astro IDE's grasp of Airflow, versus other popular LLMs and coding agents like ChatGPT with gpt-5, Cursor with gpt-5, and Cursor with claude-4.5-sonnet? Tamara Janina Fingerlin dives into which of these solutions can help you write the best human-in-the-loop Apache Airflow DAG in her new blog post: https://bit.ly/3KLMf89

60

1 Kommentar