|
--- |
|
license: apache-2.0 |
|
base_model: alpindale/WizardLM-2-8x22B |
|
--- |
|
|
|
# This repo contains the copy of the original quantized to FP8. Original: [rAIfle/SorcererLM-8x22b-bf16](https://huggingface.co/rAIfle/SorcererLM-8x22b-bf16) |
|
|
|
# SorcererLM-8x22b-bf16 |
|
|
|
<img src="https://files.catbox.moe/1kohx8.png" width="400"/> |
|
|
|
<audio controls src="https://cdn-uploads.huggingface.co/production/uploads/6569a4ed2419be6072890cf8/L_uGojVkNUsK6QHvWgs9o.mpga"></audio> |
|
|
|
Oh boy, here we go. Low-rank (`r=16, alpha=32`) 16bit-LoRA on top of [WizardLM-2-8x22B](https://huggingface.co/alpindale/WizardLM-2-8x22B), trained on 2 epochs of (cleaned & deduped) c2-logs. As far as I can tell, this is an upgrade from `WizardLM-2-8x22B` for RP purposes. |
|
|
|
Alongside this ready-to-use release I'm also releasing [the LoRA itself](https://huggingface.co/rAIfle/SorcererLM-8x22b-epoch2-LoRA) as well as [the earlier `epoch1`-checkpoint of the LoRA](https://huggingface.co/rAIfle/SorcererLM-8x22b-epoch1-LoRA). |
|
|
|
## Why A LoRA? |
|
|
|
The choice was fully intentional. I briefly considered a FFT but for this particular use-case a LoRA seemed a better fit. `WizardLM-2-8x22B` is smart by itself but its used vocabulary leaves much to be desired when it comes to RP. By training a low-rank LoRA on top of it to teach it some of Claude's writing style, we remedy that. |
|
|
|
## Prompting |
|
|
|
- Use the templates in [Quant-Cartel/Recommended-Settings](https://huggingface.co/Quant-Cartel/Recommended-Settings) under the `SorcererLM`-folder. |
|
- Or Vicuna 1.1 and a sane context template. It's somewhat sensitive to samplers, I'd recommend Temperature 1, MinP 0.05 and a dash of DRY but YMMV. Shorter prompts seem to work better, too. |
|
|
|
## Quantized Versions |
|
|
|
- [iMat GGUFs](https://huggingface.co/Quant-Cartel/SorcererLM-8x22b-iMat-GGUF) |
|
- [longcal exl2s](https://huggingface.co/Quant-Cartel/SorcererLM-8x22b-exl2-longcal) |
|
|
|
## Acknowledgments |
|
|
|
The main shoutout I want to make is to my [Cartel](https://huggingface.co/Quant-Cartel) bros, [Envoid](https://huggingface.co/Envoid) and particularly [I^2](https://huggingface.co/InferenceIllusionist), for being amazing. I count this as a team effort, so they deserve kudos too if you like this. |
|
|
|
|
|
## Training |
|
|
|
Trained using [qlora-pipe](https://github.com/tdrussell/qlora-pipe). Configs included in the `train`-subfolder. |
|
|
|
## Safety |
|
|
|
... n/a |
|
|
|
|
