allura-org/Llama-3.3-70B-Joyous

TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kPublished:Dec 26, 2025License:llama3.3Architecture:Transformer0.0K Cold

Llama-3.3-70B-Joyous is a 70 billion parameter language model developed by allura-org, fine-tuned from the Llama 3.3 architecture. Optimized primarily for roleplay tasks, it also demonstrates unexpectedly strong performance in general assistant use cases, surpassing its base model in subjective evaluations. With a context length of 32768 tokens, this model offers robust capabilities for both creative and practical applications.

Loading preview...

Joyous 70B: A Llama 3.3 Finetune

Joyous 70B is a specialized finetune of the Llama 3.3 70B model by allura-org, designed with a primary focus on roleplay tasks. Despite its intended specialization, the model has shown surprisingly strong performance in general assistant use cases, outperforming its base model in subjective evaluations.

Key Capabilities

  • Enhanced Roleplay: Specifically tuned for engaging and dynamic roleplay scenarios.
  • Strong Assistant Performance: Demonstrates unexpected proficiency in general conversational and task-oriented assistant functions.
  • Llama 3.3 Architecture: Leverages the robust foundation of the Llama 3.3 70B model.

Good for

  • Creative Roleplay Applications: Ideal for scenarios requiring character interaction and narrative generation.
  • General Conversational AI: Suitable for building helpful and harmless assistant agents.
  • Experimentation with Llama 3.3 Derivatives: Offers a unique finetune for developers exploring specialized Llama 3.3 variants.