Martin Jaggi
mjaggi.bsky.social
Martin Jaggi
@mjaggi.bsky.social
Prof at EPFL
AI • Climbing
apertus also! (september release, same mission but multilingual)
October 5, 2025 at 10:13 PM
cool idea. let’s us know how it goes! btw maybe these can be useful github.com/swiss-ai/ape...
or, since today, also unsloth and llamacpp
GitHub - swiss-ai/apertus-finetuning-recipes
Contribute to swiss-ai/apertus-finetuning-recipes development by creating an account on GitHub.
github.com
October 3, 2025 at 5:53 PM
on the engineering track it renews yearly usually, but permanent is possible after some experience & paperwork. on the academic track see e.g. here www.epfl.ch/about/workin...
Faculty Positions in Computer & Communication Sciences – Learning Sciences
The School of Computer and Communication Sciences (IC) at EPFL invites applications for tenure-track faculty positions in learning sciences and educational technologies, with a focus on computational ...
www.epfl.ch
September 26, 2025 at 7:34 PM
Link to the first version of the Apertus open-data open-weights LLM - multilingual in >1000 languages, and compliant ethical AI huggingface.co/collections/...
Apertus LLM - a swiss-ai Collection
Democratizing Open and Compliant LLMs for Global Language Environments: 8B and 70B open-data open-weights models, multilingual in >1000 languages
huggingface.co
September 25, 2025 at 9:08 PM
Several open positions at EPFL Lausanne and ETH Zurich and, as part of the Swiss AI Initiative. We cover the entire stack of foundation model training. And we're open to international applicants of course (no H-1B required ;))
September 25, 2025 at 9:08 PM
funktioniert schon seit letzter woche im neusten LM Studio (mit MLX) huggingface.co/models?searc...

GGUF kommt auch bald die tage
September 18, 2025 at 10:07 PM
no. the commercial models like chatGPT and gemini still can do better swiss german than apertus.
September 13, 2025 at 4:14 PM
the 70b should work too if enough ram, you can just replace 8 by 70 in the swiss ai huggingface repo
September 6, 2025 at 9:50 PM
MLX provides already many quantizations. But yes still working on gguf github.com/ggml-org/lla...
September 6, 2025 at 10:34 AM
or this quantized version of the model, which is faster and is even easier as it requires no huggingface login:

pip install mlx-lm

mlx_lm.generate --model mlx-community/Apertus-8B-Instruct-2509-8bit --prompt "who are you?"
September 5, 2025 at 10:43 PM
volle transparenz der KI modelle inkl. ihrer trainingsdaten ist hier sehr wichtig, damit man solche biases indentifizieren und nachvollziehen kann (audit). dies ist bei llama & qwen modellen (die nur open weights sind) klar nicht möglich., bei apertus aber schon
September 5, 2025 at 9:11 PM
propaganda von russischen und auch anderen akteuren ist klar auf dem web vorhanden und kann trainingsdaten reinfliessen: www.newsguardtech.com/wp-content/u...
www.newsguardtech.com
September 5, 2025 at 9:11 PM
'wahrheitsgehalt' zu verifizieren ist nicht ganz einfach. grok war ja auch nicht grad so erfolgreich damit.

was wir machen ist dass russisch (und die anderen top 20 sprachen) nach den gleichen kriterien nach qualität der daten gefiltert werden. github.com/epfml/finewe...
September 5, 2025 at 9:11 PM