Guilherme34/Firefly-V3.2

Firefly-V3.2: An Enhanced Roleplay-First Model

Firefly-V3.2 is a significantly refined language model developed by Guilherme34, building upon the foundation of Firefly-V3. It is designed to deliver superior performance in roleplay scenarios, offering improved coherency and even mathematical abilities. The model's development involved a novel method inspired by the Sakana AI Text to Lora paper, alongside extensive additional training and fine-tuning.

Key Capabilities & Features

• Roleplay-First Design: Optimized for immersive character interaction, emotional pacing, scene continuity, and expressive dialogue.
• Enhanced Coherency: Demonstrates improved ability to maintain logical flow and context throughout conversations.
• "Brain Upgrade": Features a "literal brain upgrade" over its predecessor, Firefly-V3, with better performance in roleplay, coherency, and math.
• Advanced Training: Built from a merge foundation (including Guilherme34/Firefly-V2.5 and SicariusSicariiStuff/Impish_LLAMA_3B) and further refined through additional training and a Text-to-Lora method for increased creativity and awareness.
• Character Presence: Aims to make characters feel more distinct, expressive, and consistent across extended chats.
• Creative Writing: Strong capabilities for prose, atmosphere, dialogue, character descriptions, and interactive fiction.

Good For

• Roleplay chats: Character-driven conversations and interactive scenes.
• Character cards: Creating bots with personality, style, and scene memory.
• Creative writing: Generating narration, dialogue, atmosphere, and prose.
• Long scenes: Extended interactions requiring pacing and continuity.