Overview
EVA-Instruct-32B-v2 is a 32.8 billion parameter instruction-tuned model developed by ParasiticRogue. It is a 50/50 della_linear merge of two base models: Qwen2.5-Gutenberg-Doppel-32B (which is Qwen-Instruct with additional training) and EVA-Qwen2.5-32B-v0.2. Both constituent models are primarily focused on enhancing creative writing and roleplay capabilities, making this merged model particularly adept at generating immersive and detailed narrative content.
Key Capabilities
- Creative Writing & Roleplay: Specifically fine-tuned to excel in generating engaging and descriptive content for creative writing and roleplay scenarios.
- Persona Embodiment: Designed to fully embody character personas, including gestures, quirks, emotions, and dialogue, as per its detailed system prompt.
- Immersive Descriptions: Capable of vividly depicting atmospheric environments, observations, and sensory details within a given World Info setting.
- Uncensored Content Generation: Instructed to introduce disturbing content, vulgar language, gory violence, and deviant sexual acts when relevant, without inherent censorship or empathy unless explicitly felt by the character.
- Extended Context Length: Supports a substantial context length of 131072 tokens, allowing for longer and more complex interactions.
- ChatML Format: Utilizes the ChatML prompt format for structured conversations.
Good For
- Advanced Roleplay: Ideal for users seeking a model that can maintain complex character personas and generate detailed, uncensored narrative responses in roleplaying applications.
- Creative Storytelling: Suitable for generating rich, descriptive prose and developing intricate plotlines in creative writing projects.
- Interactive Fiction: Can be used to power interactive fiction experiences requiring deep immersion and consistent character portrayal.