Darkknight535/OpenCrystal-12B-L3

TEXT GENERATIONConcurrency Cost:1Model Size:15BQuant:FP8Ctx Length:8kTool Calling:SupportedPublished:Aug 25, 2024Architecture:Transformer0.0K Cold

Darkknight535/OpenCrystal-12B-L3 is a 15 billion parameter fine-tuned language model, based on the Llama 3 instruct preset, designed for creative and coherent text generation. It excels at dynamic NPC roleplay, creative writing with 'crazy swipes', and following prompts effectively, often producing outputs similar to Character.ai. The model demonstrates strong performance in summarizing and generating image prompts, and is noted for its ability to handle higher context lengths, with 12K easily tested.

Loading preview...

OpenCrystal-12B-L3: A Fine-Tuned Llama 3 Variant for Creative Generation

OpenCrystal-12B-L3 is a 15 billion parameter language model developed by Darkknight535, building upon the Llama 3 instruct architecture. This model is specifically fine-tuned to enhance creative text generation and dynamic roleplay scenarios, with the developer recommending it over previous versions (v2 and v2.1).

Key Capabilities & Features

  • Dynamic NPC Roleplay: The model can automatically speak as other non-player characters, adapting to the story's plot and character personalities.
  • Creative & Coherent Output: It is noted for generating creative responses, described as having "crazy swipes," and maintaining coherence, with an output style reminiscent of Character.ai.
  • Prompt Following: Demonstrates improved adherence to user prompts, making it reliable for specific narrative or conversational requirements.
  • Context Length: Supports and performs well with higher context lengths, with 12K tokens easily tested.
  • Summarization & Image Prompt Generation: Capable of effectively summarizing content and generating descriptive image prompts, potentially leveraging its Llama 3 instruct base.
  • Instruct Template: Utilizes a Llama 3 instruct and context preset, with specific JSON files provided for reference.

Performance Insights

While optimized for creative tasks, the model's performance on the Open LLM Leaderboard shows an average score of 20.51. Specific benchmark results include:

  • IFEval (0-Shot): 40.71
  • BBH (3-Shot): 31.84
  • MMLU-PRO (5-shot): 29.34

Recommended Use Cases

  • Creative Writing & Storytelling: Ideal for generating imaginative narratives, character dialogues, and dynamic plot developments.
  • Roleplay & Conversational Agents: Excels in scenarios requiring the model to adopt multiple personas and maintain engaging conversations.
  • Content Generation: Useful for tasks like summarizing text or creating detailed prompts for image generation.