OddTheGreat/Lastman_12B_V.2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 3, 2025Architecture:Transformer0.0K Featherless Exclusive Cold

OddTheGreat/Lastman_12B_V.2 is a 12 billion parameter merged language model designed for versatile everyday use. Optimized for creative writing, casual chat, and roleplay, it offers good instruction following and context sensitivity within its size class. This model aims to provide a balanced performance for users with limited VRAM, including acceptable quality for Russian language tasks at higher quantization levels.

Loading preview...

OddTheGreat/Lastman_12B_V.2: A Merged Model for Creative Applications

This model, developed by OddTheGreat, is a 12 billion parameter merge of pre-trained language models. Its primary design goal was to create a robust, all-around model suitable for creative writing, casual conversations, and roleplay scenarios, particularly for users operating with limited VRAM.

Key Capabilities

  • Creative Writing & Roleplay: Engineered to be sufficiently creative and effective in generating engaging narratives and participating in roleplay, including the ability to reply on behalf of a user if described in a character card.
  • Instruction Following: Demonstrates good adherence to instructions, making it reliable for various prompt-based tasks.
  • Context Sensitivity: Offers acceptable context awareness for a 12B Mistral-based model, contributing to coherent and relevant responses.
  • Multilingual Support: While primarily English-focused, it has been tested with Russian (RU) and performs well at Q5 quantization levels and higher, providing non-dry and intelligent responses.

Good For

  • Users seeking a balanced model for daily creative tasks and interactive chat.
  • Applications requiring good instruction following and narrative generation.
  • Environments with limited VRAM where a 12B model needs to perform efficiently.
  • Casual chat and roleplay where context sensitivity and creative output are valued.