DS-Archive/CalliopeDS-v2-L2-13B
CalliopeDS-v2-L2-13B is a 13 billion parameter Llama 2-based model developed by DS-Archive, created through a SLERP merge of several PEFT-adapted models. This model is specifically designed to excel at creative writing and roleplay, generating descriptive and verbose responses while maintaining strong general intelligence and instruction-following capabilities. It is optimized for nuanced conversational interactions and story generation, making it suitable for applications requiring detailed narrative output.
Loading preview...
Overview
CalliopeDS-v2-L2-13B is a 13 billion parameter language model built upon the Llama 2 architecture. Developed by DS-Archive, it is a product of merging multiple PEFT-adapted models, including PygmalionAI/pygmalion-2-13b, NousResearch/Nous-Hermes-Llama2-13b, Doctor-Shotgun/llama-2-supercot-lora, lemonilia/LimaRP-Llama2-13B-v3-EXPERIMENT, and Undi95/Storytelling-v2-13B-lora, using Charles Goddard's mergekit.
Key Capabilities
- Creative Writing & Roleplay: Specifically engineered to produce descriptive, verbose, and contextually rich responses for roleplaying and creative narrative generation.
- Instruction Following: Maintains general intelligence and instruction-following capabilities despite its specialized focus.
- Response Length Control: Incorporates a unique feature from LimaRP v3, allowing users to specify desired response lengths (e.g.,
tiny,medium,unlimited) directly within the prompt for granular control over output verbosity.
Usage & Prompt Formats
This model supports various prompt formats due to its merged nature. Recommended formats include the Alpaca instruction format used by LIMARP v3 and the Pygmalion/Metharme format. It has also been tested with system prompts for character-driven roleplay.
Limitations
- Bias: May exhibit biases similar to those found in niche online roleplaying communities, in addition to base model biases.
- Factual Accuracy: Not intended for providing factual information or advice.