brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties-ExtremeDensity
brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties-ExtremeDensity is a 34 billion parameter language model created by brucethemoose, built using a DARE TIES merge of several Yi-34B-200K variants and other 34B models. This model is specifically designed as a test of a very high-density DARE TIES merge for benchmarking on the Open LLM Leaderboard. It achieves an average score of 71.57 on the leaderboard, with notable performance on HellaSwag (85.69) and MMLU (77.35). Its primary purpose is experimental evaluation of merge techniques rather than general application.
Loading preview...
Model Overview
brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties-ExtremeDensity is a 34 billion parameter language model developed by brucethemoose. This model is an experimental DARE TIES merge, specifically configured for very high density, and was created primarily for benchmarking purposes on the Open LLM Leaderboard.
Key Characteristics
- Architecture: A DARE TIES merge of multiple 34B models, including variants of Yi-34B-200K, Tess-34B, Airoboros-3.1-Yi-34B-200K, Nous-Capybara-34B, PlatYi-34B-200K-Q, Dolphin-2.2-Yi-34B-200K, and Una-Xaberius-34B-v1beta.
- Parameter Count: 34 billion parameters.
- Context Length: Supports a context length of 32768 tokens.
- Merge Method: Utilizes the
dare_tiesmerge method with specific density and weight parameters for each constituent model.
Performance on Open LLM Leaderboard
This model has been evaluated on the Hugging Face Open LLM Leaderboard, achieving an overall average score of 71.57. Key benchmark results include:
- HellaSwag (10-Shot): 85.69
- MMLU (5-Shot): 77.35
- AI2 Reasoning Challenge (25-Shot): 66.89
- Winogrande (5-Shot): 82.00
- GSM8k (5-Shot): 59.82
- TruthfulQA (0-shot): 57.63
Intended Use
This model is explicitly described as a test of a very high-density DARE TIES merge for benchmarking. The creator suggests that users should likely consider an alternative model, specifically brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties-HighDensity, for general use cases, indicating this model's primary role is experimental evaluation rather than practical application.