Dark Scarlett 31B
Passion ignited. Boundaries dissolved.
chat_template_kwargs: {"enable_thinking": true, "reasoning_effort": "medium"} in SillyTavern.
Core Capabilities
Optimized for immersive adult narrative experiences
๐น What Is Dark Scarlett?
Inspired by the spirit of Melody1437, though trained on a distinct dataset, Dark Scarlett-31B refines the core architecture with specialized scenario prompts to enhance depth and responsiveness. This model focuses on delivering a cohesive narrative experience with a distinct perspective.
- ๐ง Core Heritage: Captures the spirit of Melody1437 with diversified scenario prompts
- ๐ญ Group Chat Support: Includes some training for dynamic group chat interactions
- ๐ฏ Perspective: Optimized for Male User โ Female AI interactions
- ๐ Narrative Flow: Improved context retention with longer conversation threads in select scenarios
๐งฌ Synthetic Life Engine
Dataset generated using our advanced Character Engine and Emotional Engine, creating genuine life and emotional resonance in every interaction.
๐ญ Character Engine
Ensures consistent personality traits, speech patterns, and behavioral logic across all contexts. Characters remain true to themselves throughout.
๐ Emotional Engine
Injects dynamic emotional states into responses, creating depth and realistic reactions that breathe life into every exchange.
โจ Quality Refinement
Automated detection and rewriting of repetitive phrases ensures fresh, high-quality dialogue in every turn.
๐ฌ Dialogue Integrity
Advanced quote normalization ensures balanced dialogue markers, preventing formatting errors and maintaining immersion.
๐ง Training Details & Parameters
Fine-tuned using LoRA (Low-Rank Adaptation) for efficient and targeted weight adjustment, preserving the base model's capabilities while imprinting new behavioral patterns.
๐ข Epochs
Full passes through the training dataset for thorough learning
๐ LoRA Rank
Optimized rank for balanced adaptation and efficiency
| Parameter | Value |
|---|---|
| Training Method | LoRA (Low-Rank Adaptation) |
| LoRA Rank (r) | 32 |
| Epochs | 1 |
| Trained Layers | Text layers only |
| Experimental Feature | Group Chat Support |
- ๐ฏ Method: LoRA fine-tuning for parameter-efficient adaptation, preserving base knowledge while imprinting new behaviors
- โก Efficiency: Only a fraction of parameters updated, keeping the model lean and responsive
- ๐ญ Result: Maintains coherence while exhibiting the desired personality traits and interaction style
๐ฎ Training Process
Model weights subjected to iterative refinement during data creation. Each conversation underwent multiple checks for stability and alignment.
- ๐ Multi-Turn Generation: Conversations built turn-by-turn for natural context flow
- ๐ก๏ธ Refusal Filtering: Automated systems detected and removed unwanted refusals
- ๐งน Slop Cleaning: Undesirable phrases identified and rewritten by dedicated assistant models
- ๐ช Group Chat Training: Includes some training data focused on multi-user group dynamics
๐ Dataset Overview
Model trained on a specialized adult-oriented roleplay dataset with diverse scenarios and emotional contexts, inspired by the spirit of Melody1437, though distinct in composition.
- ๐ Content Rating: Strictly 18+ (Adults Only)
- ๐ Dataset Size: Currently contains 8,130 specialized prompts.
- ๐ Conversation Depth: Portions include extended conversations for enhanced roleplay continuity.
- ๐ญ Focus: Mature themes, immersive roleplay, uncensored dialogue
- ๐ฅ Perspective: Male (User) โ Female (AI)
- ๐งน Formatting: Removed Em dashes, asterisks, and markdown formatting from the dataset for cleaner output.
๐ Version Notes
- โ ๏ธ Content Warning: This model can generate mature content. Use responsibly.
- ๐งช Test Build: This is an experimental release (v0.35) and not the final version.
- ๐ Data Limitations: Trained on a curated dataset (8,130 prompts); performance may vary.
- ๐งน Clean Output: Reasoning tags, Em dashes, asterisks, and markdown formatting stripped for cleaner roleplay
- ๐ Core Identity: Trained on the spirit of Melody1437, but utilizes a unique dataset
- ๐ช Experimental: Includes some training for group chat scenarios
โ๏ธ Configuration
๐๏ธ Sampler Settings
Recommended parameters for optimal output
๐ฆ GGUF Quantizations
Available formats for local inference
- Q4_K_MDecent Quality
- Q5_K_MRecommended
- Q6_KHigh Quality
- Q8_0Near Lossless
๐ Credits
-
GECFDO โ Dataset Generation
-
FrenzyBiscuit โ Fine-Tuning & Dataset Creation
๐ License & Usage
- ๐ Built upon the Apache 2.0 license
- ๐ก๏ธ You accept full responsibility for all outputs generated
- ๐ You confirm you are at least 18 years old
- ๐ Creators bear no responsibility for how the model is used
- ๐ก For personal use only (non-profit/non-commercial) to the extent legally allowed