|
|
--- |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- mistral |
|
|
- Uncensored |
|
|
- text-generation-inference |
|
|
- transformers |
|
|
- unsloth |
|
|
- trl |
|
|
- roleplay |
|
|
- conversational |
|
|
- rp |
|
|
datasets: |
|
|
- N-Bot-Int/Iris-Uncensored-R2 |
|
|
- N-Bot-Int/Millie-R1_DPO |
|
|
- N-Bot-Int/Millia-R1_DPO |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- unsloth/mistral-7b-instruct-v0.3-bnb-4bit |
|
|
pipeline_tag: text-generation |
|
|
library_name: transformers |
|
|
metrics: |
|
|
- character |
|
|
--- |
|
|
 |
|
|
# Official Quants are Uploaded By Us |
|
|
- [MistThena7BV2-GGUF](https://huggingface.co/N-Bot-Int/MistThena7BV2-GGUF) |
|
|
|
|
|
# Support us on Ko-Fi! |
|
|
- [](https://ko-fi.com/J3J61D8NHV) |
|
|
|
|
|
# MistThena7B - V2. |
|
|
- Introducing our Mindboggling MistThena7B **V2**, This Version Offer an Upgraded RP experience, beyond Other AI model |
|
|
We've made, Outcompetting our 3B, 1B, MythoMax, Deepseek and Hermes for Roleplaying! |
|
|
|
|
|
- **MistThena7B-V2** Offer an expanded Roleplay capabilities, using our EmojiEmulsifyer Program to Train MistThena7B |
|
|
To Use Emojis, expanding the Roleplaying Immersiveness and Actionsets MistThena7B can do! |
|
|
|
|
|
- Activate MistThena's Expanded Actions, by mirroring it(ie using Emoji on your own prompts), to ensure MistThena's |
|
|
Use of Emoji or Actions! |
|
|
|
|
|
- MistThena7B-V2 is also trained on 160K Examples from our Latest Corpus **IRIS_UNCENSORED_R2**!, This shows that |
|
|
Iris_Uncensored_R2 shows huge potential to produce good outputs, and reveals MistThena7B's Good Roleplaying capabilities, |
|
|
Which were obtained through High and rigorous training, combined with Preventive measure to overfitting |
|
|
|
|
|
- MistThena7B contains more Fine-tuned Dataset so please Report any issues found through our email |
|
|
[nexus.networkinteractives@gmail.com](mailto:nexus.networkinteractives@gmail.com) |
|
|
about any overfitting, or improvements for the future Model **V3**, |
|
|
Once again feel free to Modify the LORA to your likings, However please consider Adding this Page |
|
|
for credits and if you'll increase its **Dataset**, then please handle it with care and ethical considerations |
|
|
|
|
|
- MistThena is |
|
|
- **Developed by:** N-Bot-Int |
|
|
- **License:** apache-2.0 |
|
|
- **Finetuned from model:** unsloth/mistral-7b-instruct-v0.3-bnb-4bit |
|
|
- **Sequential Trained from Model:** N-Bot-Int/OpenElla3-Llama3.2A |
|
|
- **Dataset Combined Using:** Mosher-R1(Propietary Software) |
|
|
|
|
|
- Comparison Metric Score |
|
|
 |
|
|
- Metrics Made By **ItsMeDevRoland** |
|
|
Which compares: |
|
|
- **MistThena7B-V1** |
|
|
- **MistThena7B-V2 : 60 STEP VERSION** |
|
|
Which are All Ranked with the Same Prompt, Same Temperature, Same Hardware(Google Colab), |
|
|
To Properly Showcase the differences and strength of the Models |
|
|
|
|
|
--- |
|
|
# 🌀 MistThema-7B V2: Slower Beats, Stronger Bonds — A Roleplay Revival |
|
|
> "She may not win the speed race, but when it comes to presence and performance — she owns the stage." |
|
|
 |
|
|
 |
|
|
--- |
|
|
|
|
|
# MistThema-7B V2 isn’t just an upgrade — she’s a reinvention. Built on V1’s storytelling roots, V2 shifts her focus inward: |
|
|
- **longer scenes, deeper characters, and dialogue that breathes.** |
|
|
 |
|
|
- 💬 **Roleplay Evaluation** |
|
|
- ✍️ **Length Score**: 0.34 → **1.00** (🚀) |
|
|
- 🧠 **Character Consistency**: 0.20 → **0.53** |
|
|
- 🌌 **Immersion**: 0.00 → **0.47** |
|
|
- 🎭 **Overall RP Score**: 0.17 → **0.67** |
|
|
> She no longer just responds — she *inhabits*. MistThema-7B V2 is the method actor of models, channeling roles with vivid coherence and creative depth. |
|
|
|
|
|
--- |
|
|
|
|
|
# ⚙️ The Cost of Craft: Time for Thought |
|
|
- 🕒 **Inference Time**: 114s → **179s** (↑) |
|
|
- ⚡ **Tokens/sec**: 1.51 → **1.28** (↓) |
|
|
> Yes, she’s slower — but that’s not a bug. That’s intention. Every word is more considered, every output more deliberate. |
|
|
|
|
|
--- |
|
|
|
|
|
# 📏 Traditional Metrics? A Trade-off |
|
|
- 📘 **BLEU Score**: 0.43 → **0.18** |
|
|
- 📕 **ROUGE-L**: 0.60 → **0.32** |
|
|
> While V1 outperforms on surface-level matching, V2 is optimized for *experiential fidelity*, not rigid overlap. |
|
|
|
|
|
--- |
|
|
|
|
|
# 🎯 Reimagined for Realness |
|
|
MistThema-7B V2 isn’t trying to mimic — she’s trying to *immerse*. Designed to tell stories, embody roles, and hold character in long-form exchanges. |
|
|
- 🧩 Tailored for: |
|
|
- Narrative-heavy use cases |
|
|
- Emotional continuity and consistency |
|
|
- Richer, longer interactions |
|
|
|
|
|
--- |
|
|
|
|
|
> “MistThema-7B V2 trades benchmarks for believability. Less about matching — more about meaning. Less polished — more *present*.” |
|
|
# MistThema-7B V2 is where slower feels *stronger*. |
|
|
--- |
|
|
|
|
|
- # Notice |
|
|
- **For a Good Experience, Please use** |
|
|
-  |
|
|
- {FOR KOBOLDCPP USERS, PLEASE USE COHERENT CREATIVITY LEGACY} |
|
|
- TL:DR I'm still looking for ways to improve a good settings, if you have any good settings, feel free to put them |
|
|
in the community! i'll credit ya out! |
|
|
|
|
|
|
|
|
|
|
|
- # Detail card: |
|
|
- Parameter |
|
|
- 7 Billion Parameters |
|
|
- (Please visit your GPU Vendor if you can Run 7B models) |
|
|
|
|
|
- Training |
|
|
- 250 Steps |
|
|
- N-Bot-Int/Iris_Uncensored_R2 |
|
|
- 60 Steps |
|
|
- N-Bot-Int/Millie_DPO |
|
|
|
|
|
- Finetuning tool: |
|
|
- Unsloth AI |
|
|
- This mistral 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 |
