N-Bot-Int/ElaNore3-4B_ADJUSTED_merged
N-Bot-Int/ElaNore3-4B_ADJUSTED_merged is a 4 billion parameter causal language model, fine-tuned from Qwen3-4B by N-Bot-Int. This model is specifically optimized for roleplaying scenarios, demonstrating enhanced context handling and responsiveness compared to its base version. It specializes in generating multi-turn and narrative roleplay content, particularly when utilizing the ChatML format.
Loading preview...
ElaNore3-4B_ADJUSTED_merged: Roleplay Optimized Model
ElaNore3-4B_ADJUSTED_merged is a 4 billion parameter model developed by N-Bot-Int, fine-tuned from DreamFast's Heretical Version of Qwen3-4B. This model aims to be a highly effective, small-scale roleplaying (RP) model capable of running on various hardware configurations. It is an "ADJUSTED" version, indicating improved responsiveness and context handling over the base ElaNore3-4B.
Key Capabilities & Training:
- Roleplaying Specialization: The model excels in diverse roleplaying scenarios, including single-turn, multi-turn, and narrative roleplay.
- ChatML Optimization: It is specifically fine-tuned to perform optimally with the ChatML format, which is recommended for best results.
- Dataset: Trained on "RP-MIXED-V2," a dataset comprising 60% synthetically generated RP scenarios (from sources like Hermes) and 40% human-written RP entries, including salvaged data from Iris-Uncensored-Reformat-R2.
- Training Methodology: Fine-tuned using Unsloth and Huggingface's TRL library over 3 epochs, achieving a final training loss of 1.4.
Considerations for Use:
- Hardware Compatibility: Designed to be runnable on various hardware, making it accessible for users with different GPU, VRAM, CPU, and RAM specifications.
- Uncensored Nature: Users are advised to handle the model with care and ethical considerations due to its uncensored nature, especially when fine-tuning.
- System Prompts: For specific action formatting (e.g., actions wrapped in asterisks), using a system prompt is recommended as the model tends to use double-quotes for character words by default.