92 lines
4.3 KiB
Markdown
92 lines
4.3 KiB
Markdown
---
|
|
base_model: DreamFast/qwen3-4b-heretic
|
|
tags:
|
|
- Uncensored
|
|
- text-generation-inference
|
|
- transformers
|
|
- unsloth
|
|
- qwen3
|
|
- trl
|
|
- roleplay
|
|
- conversational
|
|
license: agpl-3.0
|
|
pipeline_tag: text-generation
|
|
language:
|
|
- en
|
|
datasets:
|
|
- N-Bot-Int/Iris-Uncensored-Reformat-R2
|
|
- N-Bot-Int/RP-Mixed-v1
|
|
library_name: transformers
|
|
---
|
|
# This Model is the Latest and "encouraged to use" version of the Adjusted model!
|
|
The old adjusted model is still available as always!
|
|
(This model is trained on 400+ dataset with 8096 tokens+ of DPO Datasets to further increase specialization of the model!)
|
|
|
|
# ElaNore3-4B - Is Now Released!
|
|
|
|
- IMAGE GENERATED USING CHATGPT!
|
|
|
|
# ElaNore3-4B, The Newest And BEST model WE HAVE MADE!
|
|
- Feast your Eyes on ElaNore3-4B, trained on Qwen3-4B(CREDIT TO DREAMFAST for the HERETIC Base)
|
|
- ElaNore3-4B, Is trained on Google Colab, with a goal of Making The **BEST** Smallest RP model
|
|
that can be Run on any hardware!
|
|
- ElaNore3 specializes in **Roleplaying** scenarios, with specialization on **ChatML** format!
|
|
|
|
*READ MORE FOR MORE INFO*
|
|
4 BILLIONS PARAMS MODEL
|
|
|
|
# ElaNore3-4B Model Procedure/Methodology:
|
|
- **ElaNore3-4B** is trained Using **DreamFast**'s Heretical Version of the Base Model(Qwen3-4B).
|
|
Dataset is prepared for ElaNore, with 6K Rows/Entry of Carefully Picked RP scenarios and Dataset From Iris-Uncensored-Reformat-R2,
|
|
Synthetically Made Dataset Entry(4k combined) from Hermes, and Human Roleplay Entries available here in Huggingface.
|
|
Forming The final Dataset Named: RP-MIXED-V2, which contains 60% Synthetic Dataset, 40% Human-Written Dataset all finetuned for RP in mind
|
|
|
|
- 4k synthetically made dataset contains the following:
|
|
- Single Roleplay
|
|
- MultiTurn Roleplay
|
|
- Narration Roleplay
|
|
|
|
- 2k Human Dataset contains the following:
|
|
- Human Written Roleplay
|
|
- Small Salvaged Dataset from Iris Uncensored Reformat R2
|
|
|
|
- **ElaNore3-4B** is Trained using Unsloth, SFT with 3 Epochs with final Training loss of 1.4 using the RP-MIXED-V2 dataset,
|
|
Trained on **GOOGLE COLAB FREE TIER** T4 GPU which took half a day to train(Lucky Me)
|
|
|
|
- **ElaNore3-4B** is Our Brand New Powerful Model, If you ever encountered any issue, Want to commission us, or have any suggestions, please email us directly through
|
|
[nexus.networkinteractives@gmail.com](mailto:nexus.networkinteractives@gmail.com)
|
|
we value any reports, suggestions to how we improve future Model,
|
|
Once again feel free to finetune the model to your likings, However please consider Adding this Page
|
|
for **CREDITS**
|
|
|
|
- Please handle the AI with Care and ethical considerations, when **FINETUNING** this AI model, due to its **UNCENSORED** Nature.
|
|
- We are not responsible for what this model generates. Use it responsibly and legally. You downloaded it, you own what you do with it.
|
|
|
|
|
|
- **ElaNore3-4B** is
|
|
- **Developed by:** N-Bot-Int
|
|
- **License:** gpl 3.0
|
|
|
|
**EQ Bench V2 score(LEGACY BUT V3 IS EXPENSIVE AS FUQ soooooooo)**
|
|
|
|
- **Slightly lowered due to EQ Bench not using ChatML which dipped the AI Model's performance slightly!**
|
|
- **METRIC SCORES ARE HIGHLY SUBJECTIVE, Feel free to use the model and judge them yourselves!**
|
|
|
|
- # Notice
|
|
- **For a Good Experience, Please use**
|
|
- (PLEASE CALIBRATE THE MODEL DEPENDING ON THE CHARACTER CARD YOU USE)
|
|
- USE A SYSTEM PROMPT IF YOU NEED ACTIONS WRAPPED IN "*", Hermes does not use it nor human Roleplay on the dataset,
|
|
hence the model obtained a bias to not use asterisk on actions, but use double-quotes on character's words
|
|
- USE **CHATML**, the AI MODEL IS FINETUNED TO USE CHATML more than any other format!
|
|
|
|
- # Detail card:
|
|
- Parameter
|
|
- 4 Billion Parameters
|
|
- (Please check your GPU Core, VRAM, CPU and RAM to see if you can comfortably run 4B models)
|
|
|
|
- Finetuning tool:
|
|
- Unsloth AI
|
|
- This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
|
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
|
- Fine-tuned Using:
|
|
- Google Colab Free Tier |