Undi95/CreativityEngine

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Sep 5, 2023License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Warm

Undi95/CreativityEngine is a 13 billion parameter language model, a merge of jondurbin/airoboros-lmoe-13b-2.1 (creative adapter) and elinas/chronos-13b-v2, designed for creative applications. This model demonstrates a balanced performance across various benchmarks, including strong results in HellaSwag and Winogrande, making it suitable for tasks requiring nuanced language understanding and generation. Its architecture is optimized for creative text generation and complex reasoning, leveraging its merged base models.

Loading preview...

Overview

Undi95/CreativityEngine is a 13 billion parameter language model created by Undi95, resulting from a merge of two distinct models: jondurbin/airoboros-lmoe-13b-2.1 (specifically its creative adapter) and elinas/chronos-13b-v2, with a 0.38 weight applied to the latter. This combination aims to leverage the strengths of both base models, particularly for creative text generation.

Key Capabilities & Performance

Evaluated on the Open LLM Leaderboard, CreativityEngine demonstrates a balanced performance profile:

  • Avg. Score: 52.07
  • HellaSwag (10-shot): 82.42, indicating strong common-sense reasoning.
  • Winogrande (5-shot): 74.19, showing proficiency in resolving pronoun ambiguity.
  • ARC (25-shot): 59.3
  • MMLU (5-shot): 53.55
  • TruthfulQA (0-shot): 52.46
  • GSM8K (5-shot): 9.55, suggesting limitations in complex mathematical problem-solving.
  • DROP (3-shot): 32.98

Good For

  • Creative Text Generation: The inclusion of a "creative adapter" from airoboros-lmoe-13b-2.1 suggests an optimization for imaginative and diverse text outputs.
  • General Language Understanding: Its solid performance on HellaSwag and Winogrande indicates good capabilities for tasks requiring contextual understanding and common sense.
  • Applications requiring a 13B parameter model: Offers a balance of performance and computational efficiency for various NLP tasks.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p