# Update #3: Post training yandex/AliceAI-80B-A3B [instruct!] from scratch — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/8ab5df37968da26a

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-02 16:30 ingested 2026-10-03 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1wvyc3e/update_3_post_training_yandexaliceai80ba3b/)

## retained text



```
Last update for those who may be following: [https://www.reddit.com/r/LocalLLaMA/comments/1wv5h8x/update\_2\_post\_training\_yandexaliceai80ba3b/](https://www.reddit.com/r/LocalLLaMA/comments/1wv5h8x/update_2_post_training_yandexaliceai80ba3b/)

I screwed up guys 😂

Turns out my loss curve during my last run was legitimately unhealthy - as some of you, and myself, were concerned about. After evaluating my QLoRA, I found zero'd gradients in all but two layers. Turns out I had a NaN issue related to my custom v100 kernels that I didn't catch - so that run is cooked, I had to restart. I guess two layers training managed to emulate a loss curve I could at least derive a sensible explanation for until I actually got to evaluate.

Thankfully, checked to make sure gradients were applying again, and restarted the run. Once again, it's live streaming at [https://figure-bios-expect-cio.trycloudflare.com/](https://figure-bios-expect-cio.trycloudflare.com/)

Loss curve looks much healthier this time and is making me feel more confident that this is going to be okay. Stay tuned! I'm gonna release GGUFs and a llama.cpp patch when I have a working version.

[My first epoch loss curve from this run](https://preview.redd.it/rmstu418z2th1.png?width=1579&format=png&auto=webp&s=fddba685c7ff90076739d35e1af82e667095e79d)

[my first epoch loss curve on the failed first run \(note the differences in scale even if the pattern looks similar\) ](https://preview.redd.it/fbg5kc2bz2th1.png?width=487&format=png&auto=webp&s=5b9d47dd32a86b1693c79c175ed9c01b4d4c1274)
```



document fetched 2026-10-03 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1wvyc3e/update_3_post_training_yandexaliceai80ba3b
