README.md

---
library_name: transformers
tags:
- not-for-all-audiences
- axolotl
- qlora
language:
- en
license: other
---

<div align="center">
  <b style="font-size: 36px;">MiS-Firefly-v0.2-22B</b>

  <img src="https://huggingface.co/invisietch/MiS-Firefly-v0.2-22B/resolve/main/header.png" style="width:60%">

<b>HF</b> :
<a href="https://huggingface.co/invisietch/MiS-Firefly-v0.2-22B">FP16</a>
&vert;
<b>GGUF</b> :
<a href="https://huggingface.co/mradermacher/MiS-Firefly-v0.2-22B-i1-GGUF">imatrix</a> &middot;
<a href="https://huggingface.co/mradermacher/MiS-Firefly-v0.2-22B-GGUF">static</a> &middot;
<a href="https://huggingface.co/invisietch/MiS-Firefly-v0.2-22B-Q6_K-GGUF">Q6_K</a> &middot;
<a href="https://huggingface.co/invisietch/MiS-Firefly-v0.2-22B-Q4_K_M-GGUF">Q4_K_M</a>
<br />
<i>Thanks to <a href="https://huggingface.co/SicariusSicariiStuff/">SicariusSicariiStuff</a> for the help with training &amp; 
<a href="https://huggingface.co/mradermacher/">mradermacher</a> for the imatrix &amp; static GGUFs.</i>
</div>

# Model Details

**This is a fix for the quantization issue in Firefly v0.1.**

Firefly is a Mistral Small 22B finetune designed for creative writing and roleplay. The model is largely uncensored and should support
context up to 32,768 tokens.

The model has been tested in various roleplay scenarios up to 16k context, as well as in a role as an assistant. It shows a broad
competency &amp; coherence across various scenarios.

Special thanks to <a href="https://huggingface.co/SicariusSicariiStuff">SicariusSicariiStuff</a> for bouncing ideas back &amp; forth on
training, and <a href="https://huggingface.co/SytanSD">SytanSD</a> for quants.

# Feedback

I appreciate all feedback on any of my models, you can use:

* [My Discord server](https://discord.gg/AJwZuu7Ncx) - requires Discord.
* [The Community tab](https://huggingface.co/invisietch/MiS-Firefly-v0.2-22B/discussions) - requires HF login.
* Discord DMs to **invisietch**.

Your feedback is how I improve these models for future versions.

# Disclaimer

This model is extensively uncensored. It can generate explicit, disturbing or offensive responses. Use responsibly. I am not responsible for 
your use of this model.

This model is a finetune of Mistral Small 22B (2409) and usage must follow the terms of Mistral's license. By downloading this model, you
agree not to use it for commercial purposes unless you have a valid Mistral commercial license. See [the base model card](https://huggingface.co/mistralai/Mistral-Small-Instruct-2409)
for more details.

# Prompting Format

I'd recommend Mistral v2v3 prompting format:

```
<s>[INST] User message here.[/INST] Bot response here</s>[INST] User message 2 here.
```

# Sampler Settings

I'm running the following sampler settings but this is an RC and they may not be optimal.

- **Temperature:** Dynamic 0.7-1.1
- **Min-P:** 0.07
- **Rep Pen:** 1.08
- **Rep Pen Range:** 1536
- **XTC:** 0.1/0.15

If you get completely incoherent responses, feel free to use these as a starting point.

# Training Strategy

I started with a finetune of Mistral Small 22B which had been trained on the Gutenberg dataset: [nbeerbower/Mistral-Small-Gutenberg-Doppel-22B](https://huggingface.co/nbeerbower/Mistral-Small-Gutenberg-Doppel-22B).

The first stage of my training was a single epoch at low LR over a 474 million token text completion dataset.

I followed this up with a coherence, decensorship & roleplay finetune over a 172 million token instruct dataset over two epochs.

I did a slerp merge of epoch 1 into epoch 2 at a light weight which resolved the name-spelling issues on quantized versions of Firefly v0.1.

Total training time was about 32hrs on 4x Nvidia A100 80GB.

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>