RP-SAINEMO is a 12 billion parameter language model developed by Moraliane, created by merging pre-trained models using the della_linear method. It is specifically designed to combine strong Russian language capabilities with roleplay (RP) elements, leveraging a base of IlyaGusev_saiga_nemo_12b. The model aims to provide a balanced performance for Russian-language roleplaying scenarios, with a context length of 32768 tokens.
Loading preview...
Model Overview
RP-SAINEMO is a 12 billion parameter language model developed by Moraliane, built upon the NeMo framework and created through a merge of existing models. The primary goal of this model is to integrate robust Russian language processing with roleplay (RP) capabilities, making it suitable for interactive narrative generation in Russian.
Merge Details
This model was constructed using the della_linear merge method via mergekit. The merge prioritized Russian context, with IlyaGusev_saiga_nemo_12b serving as the base model and receiving a higher weight (0.8) during the merge process. MarinaraSpaghetti_NemoMix-Unleashed-12B was also included with a lower weight (0.2) to retain roleplay elements while ensuring strong Russian language performance.
Key Characteristics
- Parameter Count: 12 billion parameters.
- Context Length: Supports a context window of 32768 tokens.
- Language Focus: Optimized for Russian language understanding and generation.
- Roleplay Integration: Designed to incorporate roleplay elements effectively.
Recommended Usage
Optimal performance is achieved with specific generation settings, including a temperature of 1.2, TopA of 0.1, and TopP of 1, as suggested by the developer. GGUF quantized versions are also available for broader compatibility.