Uphill Conf 2025
2025-04-24..2025-04-25
Bern, Switzerland
[Uphill Conf 2018, Uphill Conf 2019, Uphill Conf 2024, Uphill Conf 2025]
who it's for
machine learning engineers, software engineers building with LLMs, mobile and edge engineers, platform engineers, engineering managers, Java backend developers
topics
transformer attention mechanics; on-device LLM inference; policy gradients and RLHF; autonomous code-fixing agents; compact model families (SmolLM); vision-language models (SmolVLM); LoRA adapter fine-tuning; Java LLM integration with Quarkus; enterprise agent security and deployment; adapter hosting and swap latencies; WebGPU and WebAssembly acceleration; instruction tuning and function calling
| title | author | length |
|---|---|---|
| Keys, Queries, and Values: The Celestial Mechanics of Attention | Luis Serrano | 45m |
Claims that attention is what makes transformers understand context, demonstrates the actual scaled dot‑product formula softmax of QK^T/√dk and shows… | ||
| Let’s Build K.I.T.T. with JavaScript | Nico Martin | 33m |
A browser K.I.T.T. is actually built: on-device LLMs (Gemma family, small 1B variants) run via transforms.js and WebLM with WebGPU/WebAssembly acceleration… | ||
| Reinforcement Learning - What it is and How it Works | Joshua Starmer | 38m |
Walks through a concrete policy gradients example that starts with probability .5 then shifts to .6 then to .2 by updating a single bias, multiplies a… | ||
| Agent-Based LLMs for Data Analysis and Fixing Software Bugs | Nicole Königstein | 33m |
Demonstrates autonomous, multimodal agents that debug and patch real GitHub repositories via a tailored AGI ACI agent computer interface with commands like… | ||
| SmolLM: The Rise of Smol Models | Loubna Ben Allal | 18m |
Claims that compact models match much larger ones by training choices and data curation: SmolLM family has 100M, 300M, 1.7B variants, pretraining mixes curated… | ||
| Crafting Intelligent Applications in Java with Quarkus / LangChain4j | Dimitris Andreadis | 42m |
Demonstrates a reproducible Java stack: Quarkus DevMode plus LangChain4j and annotations that generate LLM-backed implementations from a Java interface, with… | ||
| Teaching Machines to See:Vision-Language Models Made Efficient with SmolVLM | Andrés Marafioti | 32m |
Claims a compact VLM family—SmolVLM—built with a simple vision encoder plus a linear embedding bridge and pixel shuffle token reduction, yielding models at 256… | ||
| Multi Adapter Hosting LoRA Adapter | Simon Peyer, Luca Perrozzi | 25m |
LoRA fine-tuning keeps the base model frozen and trains only two low‑rank matrices A and B so adapters add few parameters yet match full fine‑tuning on many… | ||
| Max Productivity, Minimal Compute, Zero Leaks: Pioneering Secure Enterprise Agents | Sandra Kublik | 32m |
Command A claims parity or better than GPT‑4o and Deepseek V3 on human evaluations while cutting compute: a ~110B design that fits on two A100s and uses… | ||
| Interview with Joshua Starmer | Joshua Starmer | 09m |
Claims that AI is just multiplication and addition, shows how equations become words so each step is tiny, and cites concrete examples switching from mouse… | ||
| Interview with Loubna Ben Allal | Loubna Ben Allal | 06m |
Describes building small LMs for phones that trade accuracy for much lower inference cost, outlines experiments on training and runtime efficiency, and… | ||
| interview of - the one and only - Réginald Bien-Aimé! | Réginald Bien-Aimé | 01m |
Transcript contains only conference praise and no technical details so pick this talk only if it actually contains the specifics you want: a crisp concrete… | ||
| Feedback from Our Guests and Speakers! | 01m | |
Attendee feedback praises the conference’s consistent high quality and many say they will return next year. Participants report concrete idea exchanges and… | ||
| What Our Visitors Say! | 01m | |
Attendees describe concrete, day-to-day takeaways from sessions that focus on agents and large language models, including how to tune and chain them. Several… | ||
| Interview with Chris Beyeler from SwissAI | Chris Beyeler | 01m |
An interview with Chris Beyeler of SwissAI describes a community shift from ideas to rapid prototyping, with intense hallway collaboration producing demo-ready… | ||
| Luis Serrano with ice cream, on top of Gurten with beautiful view of Berne! | Luis Serrano | 01m |
Title gives only scenery and small talk: the video delivers congratulations about the conference, praise for the gelato is absolutely amazing, a line about the… | ||
| nothing matches | ||