Research, engineering, and AI insights by Moncif El-mouden.
Training Apertus Nash on a single DGX Spark, the reward model lived in its own container and re-read its rubric on every request — because RL exploits whatever the reward forgets to punish, and that only shows up mid-run.
Memory-LoRA generates a repository-specific LoRA for Gemma-4-E2B in one forward pass. For weeks it looked like it worked. It was worse than no adapter at all, and the training loss never said so.
Légitus in four stages: continued pretraining on the LEGI corpus, 27k synthetic instructions built around citation and abstention, assistant-only SFT, and a weight merge that traded self-identification for clean stopping.
A walkthrough of my PhD project — scraping the French Competition Authority archive, ranking 18 LLMs on a manually annotated benchmark, and building a deep-research agent that turns the extracted data into econometric findings.