tuantran1632001/Psyfighter2-Orca2-13B-ties

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Jan 11, 2024License:otherArchitecture:Transformer0.0K Cold

The tuantran1632001/Psyfighter2-Orca2-13B-ties is a 13 billion parameter language model created by tuantran1632001, formed by merging KoboldAI/LLaMA2-13B-Psyfighter2 and microsoft/Orca-2-13b using the TIES merging method. This model aims to replicate the performance of larger models like jebcarter/psyonic-cetacean-20B at a smaller scale. It achieves an average score of 63.48 on the Open LLM Leaderboard, indicating strong general performance across various benchmarks.

Loading preview...

Model Overview

Psyfighter2-Orca2-13B-ties is a 13 billion parameter language model developed by tuantran1632001. It was created by merging two distinct models: KoboldAI/LLaMA2-13B-Psyfighter2 and microsoft/Orca-2-13b, utilizing the TIES merging method via mergekit.

Key Characteristics

  • Merged Architecture: Combines the strengths of a LLaMA2-based model with Microsoft's Orca-2, aiming for a balanced performance profile.
  • Performance: Achieves a competitive average score of 63.48 on the Open LLM Leaderboard, outperforming its base models and even the 20B parameter target model, jebcarter/psyonic-cetacean-20B.
  • Benchmark Results:
    • AI2 Reasoning Challenge (25-Shot): 62.46
    • HellaSwag (10-Shot): 81.74
    • MMLU (5-Shot): 60.31
    • TruthfulQA (0-shot): 55.40
    • Winogrande (5-shot): 77.27
    • GSM8k (5-shot): 43.67

Use Cases

This model is suitable for general-purpose language tasks where a 13B parameter model is desired, especially for users seeking a model with strong reasoning and common-sense capabilities, as indicated by its benchmark scores. Its performance suggests it can be a viable option for applications requiring a balance of size and capability.