NeverSleep/Noromaid-13B-0.4-DPO is a 13 billion parameter language model developed by IkariDev and Undi, fine-tuned for enhanced human-like behavior and roleplay scenarios. It utilizes a full finetune approach, incorporating unique private datasets alongside established DPO training data. This model is particularly optimized for generating engaging and fresh conversational outputs, making it suitable for applications requiring nuanced character interaction.
Loading preview...
Noromaid-13B-0.4-DPO: An Overview
Noromaid-13B-0.4-DPO is a 13 billion parameter language model developed by IkariDev and Undi, featuring a 4096-token context length. This model has undergone a full finetune process, distinguishing itself through its specialized training data and DPO (Direct Preference Optimization) approach.
Key Capabilities & Training
- Enhanced Human-like Behavior: The model was trained on the
no_robotsdataset to improve its ability to generate human-like responses and enhance output quality. - Fresh Roleplay Data: It incorporates new, private Aesir RP datasets, specifically designed to avoid common 'LimaRP spam' and introduce novel conversational patterns.
- DPO Training: Utilizes DPO datasets such as
Intel/orca_dpo_pairsandNobodyExistsOnTheInternet/ToxicDPOqafor refined response generation. - NsChatml Prompt Format: Designed to work with the NsChatml prompt format, facilitating structured interactions.
Good For
- Roleplay and Conversational Agents: Its specialized training on unique roleplay datasets makes it well-suited for generating engaging and fresh character interactions.
- Applications Requiring Nuanced Dialogue: The focus on human-like behavior and DPO training suggests strong performance in generating natural and contextually appropriate dialogue.
- Exploration of New Datasets: Developers interested in models trained on less common, private datasets for unique output characteristics.