Minimaid L2

🚀 MiniMaid-L2

MiniMaid-L2 is a finetuned model based on MiniMaid-L1. It uses larger and higher-quality datasets to enhance role - playing capabilities. Knowledge distillation from the popular role - playing model NoroMaid - 7B - DPO is also applied to improve its performance in coherent and good role - playing. It outperforms its predecessor by leveraging clever knowledge distillation and finetuning on top of MiniMaid-L1, sacrificing a bit of non - noticeable token - generation speed but offering a near - perfect and competitive model against 3b alternatives.

🚀 Quick Start

Notice

⚠️ Important Note

For a good experience, please use low temperature 1.5, min_p = 0.1 and max_new_tokens = 128.

✨ Features

• Enhanced Role - playing Capabilities: MiniMaid-L2 is fine - tuned with large and high - quality datasets and knowledge distillation from NoroMaid - 7B - DPO, providing better role - playing performance.
• Outperforming Predecessor: It outcompetes MiniMaid-L1 by using knowledge distillation and finetuning, offering better overall performance while sacrificing a bit of token - generation speed.
• Competitive Against 3B Alternatives: It can compete with 3B models in major role - playing metrics such as character consistency, immersion, and length score.
• Efficient and Smart: With fast inference time, high tokens/sec, and strong n - gram overlap (BLEU/ROUGE - L), it shows that distilled models can outperform larger ones.
• Built for Specific Use - Cases: It is suitable for high - fidelity RP generation, lower - latency systems, and custom, character - driven storytelling.

📦 Installation

No installation steps are provided in the original document.

💻 Usage Examples

No code examples are provided in the original document.

📚 Documentation

MiniMaid-L1 Base - Model Card Procedure

• MiniMaid-L1 achieved good performance through DPO and combined heavy finetuning. To prevent overfitting, high LR decays and randomization techniques were used. However, due to the difficulty of training on Google Colab, the model might underperform, underfit on specific tasks, or overfit on certain knowledge.
• If you find any issues, please email nexus.networkinteractives@gmail.com regarding overfitting or improvements for the future Model V3. You can modify the LORA as you like, but please add this page for credits. If you increase its Dataset, handle it with care and ethical considerations.

MiniMaid-L2 Evaluation

• Roleplay Evaluation (v1):
  • Character Consistency: 0.84
  • Immersion: 0.47
  • Overall RP Score: 0.76
  • Length Score: 1.00
  • L2 scored +0.25 higher overall than L1 and outperformed top - tier 3B models in every major RP metric.
• Efficiency Metrics:
  • Inference Time: 54.2s (still 3x faster than Hermes)
  • Tokens/sec: 6.88 (near - instant on consumer GPUs)
  • BLEU/ROUGE - L: Stronger n - gram overlap than any 3B rival

Detail Card

Property	Details
Parameter	1 Billion Parameters (Please visit your GPU Vendor if you can run 1B models)
Finetuning tool	Unsloth AI. This llama model was trained 2x faster with Unsloth and Huggingface's TRL library. Fine - tuned using Google Colab.

MiniMaid-L1 Official Metric Score

• Metrics are made by ItsMeDevRoland, comparing MiniMaid-L1 GGUFF and MiniMaid-L2 GGUFF with the same prompt, same temperature, and same hardware (Google Colab) to show the differences and strengths of the models.
• Visit below to see details!

🔧 Technical Details

No specific technical details (more than 50 words) are provided in the original document.

📄 License

The model is licensed under the Apache - 2.0 license.