SteelStorage/L3.3-MS-Evayale-70B

TEXT GENERATIONConcurrent Unit Cost:4Model Size:70BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Dec 11, 2024Architecture:Transformer0.0K Featherless Exclusive Cold

SteelStorage/L3.3-MS-Evayale-70B is a 70 billion parameter Llama 3.3-based model created by SteelSkull, designed for enhanced storytelling and detailed prose. This model merges the robust narrative capabilities of EVA-LLaMA-3.33-70B-v0.0 with the descriptive strengths of L3.3-70B-Euryale-v2.3. It is specifically optimized to maintain the positive attributes of both base models, making it suitable for applications requiring rich, immersive textual generation.

Loading preview...

L3.3-MS-Evayale-70B: Merged Storytelling and Prose Model

L3.3-MS-Evayale-70B is a 70 billion parameter model developed by SteelSkull, built upon the Llama 3.3 architecture. Its name signifies its foundation (Llama 3.3), its "Model Stock" merge method, and the combination of "EVA-LLaMA-3.33-70B-v0.0" and "L3.3-70B-Euryale-v2.3" models.

Key Capabilities & Design Philosophy

  • Enhanced Storytelling: The primary goal of Evayale-70B is to combine the strong storytelling abilities of the EVA-LLaMA model with the detailed prose and scene descriptions from the EURYALE model.
  • Balanced Strengths: It aims to maintain the positive aspects of both constituent models, offering a balanced output that is both narratively compelling and rich in descriptive detail.
  • Llama 3.3 Base: Built on unsloth/Llama-3.3-70B-Instruct, ensuring a robust and capable foundation.
  • Bfloat16 Precision: Utilizes bfloat16 for training and inference, balancing performance and memory efficiency.

Recommended Usage

  • Creative Writing: Ideal for generating immersive stories, detailed narratives, and descriptive content where both plot and prose quality are crucial.
  • Roleplay Scenarios: Its merged strengths make it well-suited for complex roleplaying environments requiring rich character interactions and environmental descriptions.

Users are encouraged to explore community-contributed prompt templates like "LLam@ception" and "DWK-SP0.02" for optimal performance.