OddTheGreat/lastman_12B_V.1

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 2, 2025Architecture:Transformer0.0K Featherless Exclusive Cold

OddTheGreat/lastman_12B_V.1 is a 12 billion parameter merged language model, built to enhance roleplay capabilities and general performance. With a 32768 token context length, it excels at attentive character card adherence and creative assistance. This model is optimized for roleplay, creative writing, and general conversational tasks, demonstrating stability even at extended context lengths.

Loading preview...

OddTheGreat/lastman_12B_V.1 Overview

OddTheGreat/lastman_12B_V.1 is a 12 billion parameter language model resulting from a merge of pre-trained models. Its primary development goal was to significantly improve roleplay capabilities and overall performance, building upon the foundation of Nomad_12b. The model has been rigorously tested across various applications, including roleplay, general work assistance, and creative writing.

Key Capabilities

  • Enhanced Roleplay: Demonstrates strong attentiveness to character cards and general context, making it highly effective for immersive roleplay scenarios.
  • Creative Assistance: Capable of generating creative content and following complex instructions.
  • Stability: Maintains quality even at extended context lengths (up to 16k tokens tested) and exhibits robustness to suboptimal settings.
  • Uncensored: Like other Mistral-based models, it is uncensored and suitable for ERP (Erotic Roleplay).
  • Multilingual Support: Offers good performance for Russian as an assistant or chat model, with English being preferred for roleplay.

Use Cases

  • Roleplay: Ideal for detailed and context-aware roleplaying, especially in English.
  • Creative Writing: Useful as a creative assistant for generating ideas and content.
  • General Assistance: Can serve as a capable assistant for various tasks, including conversational AI.

This model was tested using the Mistral v7 template at a temperature of 1.04, confirming its stability and performance across different applications.