Why we started Oblio Labs
Catastrophic forgetting is well studied for full-parameter training. We want to know how much of that work survives the move to LoRA.
announcement catastrophic-forgetting lora
What models forget when you fine-tune them
We study catastrophic forgetting in large language models and build the tooling we need to measure it.
01
Catastrophic Forgetting
Which mitigations from the full-parameter literature still work on adapters.
02
Fine-Tuning
LoRA, QLoRA, and adapter merging for domain specialization.
03
Open Tooling
Benchmarks and code we publish as experiments run, so anyone can check the results.