Update README.md
Browse files
README.md
CHANGED
|
@@ -1,22 +1,97 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
tags:
|
|
|
|
|
|
|
| 4 |
- text-generation-inference
|
| 5 |
- transformers
|
| 6 |
- unsloth
|
| 7 |
-
- mistral
|
| 8 |
- trl
|
| 9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
language:
|
| 11 |
- en
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
# Uploaded model
|
| 15 |
|
| 16 |
-
-
|
| 17 |
-
-
|
| 18 |
-
-
|
|
|
|
| 19 |
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
tags:
|
| 4 |
+
- mistral
|
| 5 |
+
- Uncensored
|
| 6 |
- text-generation-inference
|
| 7 |
- transformers
|
| 8 |
- unsloth
|
|
|
|
| 9 |
- trl
|
| 10 |
+
- roleplay
|
| 11 |
+
- conversational
|
| 12 |
+
- rp
|
| 13 |
+
datasets:
|
| 14 |
+
- N-Bot-Int/Iris-Uncensored-R1
|
| 15 |
+
- N-Bot-Int/Moshpit-Combined-R2-Uncensored
|
| 16 |
+
- N-Bot-Int/Mushed-Dataset-Uncensored
|
| 17 |
+
- N-Bot-Int/Muncher-R1-Uncensored
|
| 18 |
+
- unalignment/toxic-dpo-v0.2
|
| 19 |
language:
|
| 20 |
- en
|
| 21 |
+
base_model:
|
| 22 |
+
- unsloth/mistral-7b-instruct-v0.3-bnb-4bit
|
| 23 |
+
pipeline_tag: text-generation
|
| 24 |
+
library_name: peft
|
| 25 |
+
metrics:
|
| 26 |
+
- character
|
| 27 |
---
|
| 28 |
+
<a href="https://ibb.co/GvDjFcVp"><img src="https://raw.githubusercontent.com/Nexus-Network-Interactives/HuggingfacePage/refs/heads/main/MistThena7B.webp" alt="image" border="0"></a>
|
| 29 |
+
|
| 30 |
+
# MistThena7B - A.
|
| 31 |
+
- MistThena7B is our brand New AI boasting with An Even Bigger **7B** and Ditching **Llama3.2** for **Mistral** for **lightweight Finetuning**
|
| 32 |
+
And Fast Training and Output. MistThena7B is designed to Ditch its Outer-score and Prioritize Total Roleplaying, Trained with **5x More** Dataset
|
| 33 |
+
Compared to We used At **OpenElla3-Llama3.2B**, Making this New Model Even More Competitive **Against Hallucinations, and Even More Better
|
| 34 |
+
Textual Generations And Uncensored Output**
|
| 35 |
+
|
| 36 |
+
- MistThena7B Model **A** Does not suffer the same Prompting issue with **OpenElla3-Llama3.2B**, however please use ChatML style Prompting For Better
|
| 37 |
+
Experience, And Remember to be aware of bias with the training dataset used, The **AI** model is Under **Apache 2.0** however
|
| 38 |
+
**WE ARE NOT RESPONSIBLE TO YOUR USAGE, PROMPT AND TYPE IN YOUR OWN ACCORDING/WILL**
|
| 39 |
+
|
| 40 |
+
- MistThena7B Model **A** Outperforms OpenElla Family Model, However please keep in mind the Parameter Difference. It Outperforms Testing Benchmarks
|
| 41 |
+
In **Roleplaying and Engaging with RP or Generation of Prompts**, You are Free to release a Benchmark.
|
| 42 |
+
|
| 43 |
+
- MistThena7B contains more Fine-tuned Dataset so please Report any issues found through our email
|
| 44 |
+
<link src="mailto:nexus.networkinteractives@gmail.com">nexus.networkinteractives@gmail.com</link>,
|
| 45 |
+
about any overfitting, or improvements for the future Model **C**,
|
| 46 |
+
Once again feel free to Modify the LORA to your likings, However please consider Adding this Page
|
| 47 |
+
for credits and if you'll increase its **Dataset**, then please handle it with care and ethical considerations
|
| 48 |
+
|
| 49 |
+
- MistThena is
|
| 50 |
+
- **Developed by:** N-Bot-Int
|
| 51 |
+
- **License:** apache-2.0
|
| 52 |
+
- **Finetuned from model :** unsloth/mistral-7b-instruct-v0.3-bnb-4bit
|
| 53 |
+
- **Sequential Trained from Model:** N-Bot-Int/OpenElla3-Llama3.2A
|
| 54 |
+
- **Dataset Combined Using:** Mosher-R1(Propietary Software)
|
| 55 |
+
|
| 56 |
+
- Comparison Metric Score
|
| 57 |
+

|
| 58 |
+
- Metrics Made By **ItsMeDevRoland**
|
| 59 |
+
Which compares:
|
| 60 |
+
- **Deepseek R1 3B GGUF**
|
| 61 |
+
- **Dolphin 3B GGUF**
|
| 62 |
+
- **Hermes 3b Llama GGUFF**
|
| 63 |
+
- **OpenElla3-Llama3.2B GGUFF**
|
| 64 |
+
Which are All Ranked with the Same Prompt, Same Temperature, Same Hardware(Google Colab),
|
| 65 |
+
To Properly Showcase the differences and strength of the Models
|
| 66 |
+
|
| 67 |
+
- **THIS MODEL EXCELLS IN LONGER PROMPT AND STAYING IN CHARACTER BUT LAGS BEHIND DEEPSEEK-R1**
|
| 68 |
+
|
| 69 |
+
- # THERE ARE YET TO BE RELEASED METRIC SCORE FOR THIS MODEL, PLEASE REMAIN PATIENT WHILST **ItsMeDevRoland** Released an Updated Report
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
- # Notice
|
| 73 |
+
- **For a Good Experience, Please use**
|
| 74 |
+
- Low temperature 1.5, min_p = 0.1 and max_new_tokens = 128
|
| 75 |
|
|
|
|
| 76 |
|
| 77 |
+
- # Detail card:
|
| 78 |
+
- Parameter
|
| 79 |
+
- 7 Billion Parameters
|
| 80 |
+
- (Please visit your GPU Vendor if you can Run 7B models)
|
| 81 |
|
| 82 |
+
- Training
|
| 83 |
+
- 200 steps
|
| 84 |
+
- N-Bot-Int/Iris-Uncensored-R1
|
| 85 |
+
- 100
|
| 86 |
+
- N-Bot-Int/Iris-Uncensored-R1(Reinforcement Training)
|
| 87 |
+
- 100 steps
|
| 88 |
+
- M-Datasets
|
| 89 |
+
- 60 steps(DPO)
|
| 90 |
+
- Unalignment/Toxic-DPO
|
| 91 |
|
| 92 |
+
- Finetuning tool:
|
| 93 |
+
- Unsloth AI
|
| 94 |
+
- This mistral model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 95 |
+
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
| 96 |
+
- Fine-tuned Using:
|
| 97 |
+
- Google Colab
|