README.md

---
base_model:
- nothingiisreal/L3.1-8B-Celeste-V1.5
- Sao10K/Llama-3.1-8B-Stheno-v3.4
- Sao10K/L3.1-8B-Niitama-v1.1
- arcee-ai/Llama-3.1-SuperNova-Lite
- akjindal53244/Llama-3.1-Storm-8B
- arcee-ai/Llama-Spark
- grimjim/Llama-3-Instruct-abliteration-LoRA-8B
- crestf411/sunfall-peft
- v000000/L3.1-Celestial-Stone-2x8B
library_name: transformers
tags:
- merge
- llama
- mixtral
- dpo
datasets:
- jondurbin/gutenberg-dpo-v0.1
---

> [!WARNING]
> **Sampler:**<br>
> Likes a low temperature due to the MoE architecture. I use 0.3 personally.

# Llama-3.1-Celestial-Stone-2x8B-DPO (BF16)

* *DPO Trained, Mixture of Experts (14B).*

![image/png](https://cdn-uploads.huggingface.co/production/uploads/64f74b6e6389380c77562762/lyRa7z5maTqAaa43sxC2J.png)

* <b>2x Experts working together per token, Gutenberg novelwriting finetuning.</b>

------------------------------------------------------------------------------

*The first expert* is Instruct 405B distillation/RP vector merge <b>(Supernova-Lite, Niitama1.1, Storm)</b>

*The second expert* is ERP/Reddit data merge <b>(Celeste1.5, Stheno3.4, Storm)</b>

-------------------------------------------------------------------------------

*The base model* is <b>Sao10k/L3.1-Stheno-3.4</b> with the <b>Sunfall LoRa 0.6.1</b> to make it understand SillyTavern prompts and storywriting better.

-------------------------------------------------------------------------------

*Resultant merge finetuned* on [jondurbin/gutenberg-dpo-v0.1](https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1).

-------------------------------------------------------------------------------

*List of llama.cpp repos*

# Thanks mradermacher (GGUF):

* [GGUF static Q2-Q8](https://huggingface.co/mradermacher/L3.1-Celestial-Stone-2x8B-DPO-GGUF)
* [GGUF Imatrix Q2-Q6](https://huggingface.co/mradermacher/L3.1-Celestial-Stone-2x8B-DPO-i1-GGUF)

# Thanks QuantFactory (GGUF):

* [GGUF static Q2-Q8](https://huggingface.co/QuantFactory/L3.1-Celestial-Stone-2x8B-DPO-GGUF)

# Thanks Triangle104 (GGUF):

* [Q8_0](https://huggingface.co/Triangle104/L3.1-Celestial-Stone-2x8B-DPO-Q8_0-GGUF)
* [Q6_K](https://huggingface.co/Triangle104/L3.1-Celestial-Stone-2x8B-DPO-Q6_K-GGUF)
* [Q5_K_M](https://huggingface.co/Triangle104/L3.1-Celestial-Stone-2x8B-DPO-Q5_K_M-GGUF)
* [Q5_K_S](https://huggingface.co/Triangle104/L3.1-Celestial-Stone-2x8B-DPO-Q5_K_S-GGUF)
* [Q4_K_M](https://huggingface.co/Triangle104/L3.1-Celestial-Stone-2x8B-DPO-Q4_K_M-GGUF)
* [Q4_K_S](https://huggingface.co/Triangle104/L3.1-Celestial-Stone-2x8B-DPO-Q4_K_S-GGUF)

*Other*

* [GGUF Imatrix IQ4-Q8](https://huggingface.co/v000000/L3.1-Celestial-Stone-2x8B-DPO-GGUFs-IMATRIX)

---------------------------------------------------------------------------------

[L3.1-Celestial-Stone-2x8B](https://huggingface.co/v000000/L3.1-Celestial-Stone-2x8B) Finetuned on Nvidia A100. (See Base Model card for additional details.)

0.5 Epoch completed of dataset [jondurbin/gutenberg-dpo-v0.1](https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1) with learning_rate=8e-6

Result seems pretty good even with half epoch and low learning rate, the effect is smoother and less pronounced but its probably not *optimal*.

Outputs are more compliant and verbose, less sloppy and safety aligned.

------------------------------------------------------------------------------

# Prompt Template:
```bash
<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{output}<|eot_id|>

```

*Sometimes has false refusals but swiping and "uncensored" prompts work. I have no idea why this happens tbh, since none of the base models exhibit this behavior, it seems to be a random emergence, and extra abliteration has no impact? gating method has no impact.* 
*But it's still pretty good imo.*

*For Llama.cpp/LMStudio/etc Make sure "num_experts_used = 2"*