|
---
|
|
language:
|
|
- en
|
|
license: cc-by-nc-4.0
|
|
---
|
|
|
|
![image](https://huggingface.co/Sao10K/MN-12B-Lyra-v2a1/resolve/main/image.webp)
|
|
**Mistral Nemo 12B Lyra v2 - Stage 1**
|
|
|
|
This is Stage 1 of Lyra v2. Only SFT and 2-Step Merge has been performed at this stage.
|
|
|
|
This is still raw, there may be issues.
|
|
|
|
Kinda wordy and verbose like Claude is, but that is fixable in stage 2, hopefully.
|
|
|
|
Stage 2 Plans - RL currently TBD
|
|
|
|
---
|
|
|
|
\- `[/INST]` may leak, let's say that is an unintended side-effect. Just simply add it to your custom stopping strings. I tried adding it as an EOS in model config but that ruined the model somehow, so I didn't.
|
|
|
|
\- min_p value of 0.1 is a sane value to use. Temp? 0.7-1.2, up to you.
|
|
|
|
\- Prompt Format: ChatML
|
|
|
|
Have a day. |