Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: Looking for Mamba LLMs with pre-trained and fine-tuned weights available
1 point by rar00 17 days ago | hide | past | favorite
Hello HN, I'm conducting research on Mamba and I'm looking for another open-weights LLM based on this architecture with checkpoints for the pre-trained base as well as the instruction fine-tuned weights.

I found so far FalconMamba by TII UAE and Zamba2 by Zyphra but ideally need at least a third model given their architectural differences (Mamba1 vs Hybrid Mamba2). Models like Gu&Dao's original Mamba1&2, BlackMamba, Mamba-Codestral, Jamba only have a single checkpoint type.

Is anyone aware of another Mamba (or a SSM) with both types of checkpoints officially available?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: