Hourly Usage
Performance Metrics
Median Total Time
28.90s
Median TTFT
6.56s
Median Prefill TPS
1058.80
Median Gen TPS
6.25
Model Information
Context Size
262144
Quantization
r64 on INT8
Engine
vllm
Creation Method
Finetune
Model Type
Gemma31B
Chat Template
Gemma4
Reasoning
Yes
Vision
Yes
Parameters
31B
Added At
9/26/2026
base_model:
- google/gemma-4-31B-it base_model_relation: finetune tags:
- gemma-4
- roleplay
- conversational
- instruct
- apache-2.0
- nsfw
- adult-content
- unaligned
- mature
- explicit license: apache-2.0
Melody1437-31B
It's a fabulous day!
🔄 What Changed?
Retrained from Melody1437-31B using the identical dataset with an upgraded LoRA configuration for richer, more expressive outputs.
- 🧠 Same Dataset: Identical training data to Melody1437-31B
- ⚡ Higher LoRA Rank: 64 (vs original 24) for greater expressiveness
- 🎭 Enhanced Nuance: Captures subtleties the lower rank couldn't fully express
🧬 Synthetic Life Engine
Dataset generated using our advanced Character Engine and Emotional Engine, creating genuine life in every interaction.
🎭 Character Engine
Ensures consistent personality traits, speech patterns, and behavioral logic across all contexts.
💓 Emotional Engine
Injects dynamic emotional states into responses, creating depth and realistic reactions.
✨ Quality Refinement
Automated detection and rewriting of repetitive phrases for fresh, high-quality dialogue.
💬 Dialogue Integrity
Advanced quote normalization ensures balanced dialogue markers, preventing formatting errors.
🧠 Training Details & Parameters
Fine-tuned using LoRA (Low-Rank Adaptation) for efficient and targeted weight adjustment.
Epochs
Full passes through the training dataset
LoRA Rank
Higher rank for richer adaptation
| Parameter | Value |
|---|---|
| Training Method | LoRA (Low-Rank Adaptation) |
| LoRA Rank (r) | 64 |
| Epochs | 2 |
| Trained Layers | Text layers only |
- 🎯 Method: LoRA fine-tuning for parameter-efficient adaptation, preserving base knowledge
- ⚡ Efficiency: Only a fraction of parameters updated, keeping the model lean
- 🎭 Result: Maintains coherence while exhibiting desired personality traits
🔮 Training Process
Model weights subjected to iterative refinement during data creation. Each conversation underwent multiple checks for stability and alignment.
- 🔄 Multi-Turn Generation: Conversations built turn-by-turn for natural context flow
- 🛡️ Refusal Filtering: Automated systems detected and removed unwanted refusals
- 🧹 Slop Cleaning: Undesirable phrases identified and rewritten by dedicated assistant models
📚 Dataset Overview
Model trained on a specialized adult-oriented roleplay dataset with diverse scenarios and emotional contexts.
- 🔞 Content Rating: Strictly 18+ (Adults Only)
- 🎭 Focus: Mature themes, immersive roleplay, uncensored dialogue
- 🌈 Diversity: Generated using variable system prompts and emotional states
📜 Version Notes
- ⚠️ Content Warning: This model can generate mature content. Use responsibly.
- 🧹 Clean Output: Reasoning tags and excessive markdown stripped for cleaner roleplay
- ⬆️ Evolution: Same dataset as Melody1437-31B, retrained at LoRA 64 for enhanced performance
🎭 Usage Guidelines
A structured roleplay system prompt is required for optimal performance. Optimized for immersive character interaction rather than standard instructional queries.
- 📝 Initialization: Begin with a clear character definition and scenario context
- 🎭 Format: Use asterisks for actions
*like this*and quotes for dialogue"like this" - ⚠️ Limitation: Standard query-based prompting may result in refusals
- 💡 Tip: Model expects balanced quotes — ensure input dialogue is properly formatted
⚙️ Configuration
🎛️ Sampling Config
top_p: 0.92
temp: 1.0
rep pen: 0
presence pen: 0
💖 Credits
-
Darkhn — Fixing Thinking
-
GECFDO — Data Generation & iMatrix Quants
-
Sleep Deprived — Dataset Generator
-
FrenzyBiscuit — Fine-Tuning & Dataset Creation
🔖 License & Usage
- 📜 Built upon the Apache 2.0 license
- 🛡️ You accept full responsibility for all outputs generated
- 🔞 You confirm you are at least 18 years old
- 🌍 Creators bear no responsibility for how the model is used
- 🏡 For personal use only (non-profit/non-commercial) to the extent legally allowed