# CRMA Continual learning

**URL:** <https://discuss.huggingface.co/t/crma-continual-learning/173752>\
**Category:** Spaces\
**Created:** [February 25, 2026, 3:52am UTC](https://discuss.huggingface.co/t/crma-continual-learning/173752 "2026-02-25T03:52:37Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Fourwheels2512](https://avatars.discourse-cdn.com/v4/letter/f/e9c0ed/32.png) [@Fourwheels2512](https://discuss.huggingface.co/u/Fourwheels2512)\
**Post date:** [February 25, 2026, 3:52am UTC](https://discuss.huggingface.co/t/crma-continual-learning/173752/1 "2026-02-25T03:52:38Z")

</div>

Been experimenting with CRMA — a stability layer for LoRA fine-tuning that handles sequential training without catastrophic forgetting. Tested on TinyLlama and Mistral-7B. Full results dropping soon.

---

<div class="post-metadata">

**Author:** ![Laimleon](https://avatars.discourse-cdn.com/v4/letter/l/838e76/32.png) [@Laimleon](https://discuss.huggingface.co/u/Laimleon)\
**Post date:** [February 25, 2026, 4:29am UTC](https://discuss.huggingface.co/t/crma-continual-learning/173752/2 "2026-02-25T04:29:40Z")

</div>

Thanks for this post.