Gille/StrangeMerges_53-7B-model_stock
Gille/StrangeMerges_53-7B-model_stock is a 7 billion parameter language model created by Gille, formed by merging five distinct models using the model_stock method. This merge combines models like Gille/StrangeMerges_52-7B-dare_ties and Kukedlc/NeuralMaths-Experiment-7b, resulting in a model with an 8192 token context length. It achieves an average score of 76.07 on the Open LLM Leaderboard, demonstrating balanced performance across various reasoning and language understanding tasks, including 72.78 on ARC and 64.97 on MMLU.
Loading preview...
StrangeMerges_53-7B-model_stock Overview
StrangeMerges_53-7B-model_stock is a 7 billion parameter language model developed by Gille. It is a product of merging five different base models, including Gille/StrangeMerges_52-7B-dare_ties, rwitz/experiment26-truthy-iter-0, Gille/StrangeMerges_32-7B-slerp, AurelPx/Percival_01-7b-slerp, and Kukedlc/NeuralMaths-Experiment-7b, utilizing the model_stock merge method via LazyMergekit.
Key Capabilities & Performance
This merged model demonstrates solid performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard:
- Average Score: 76.07
- AI2 Reasoning Challenge (25-Shot): 72.78
- HellaSwag (10-Shot): 88.46
- MMLU (5-Shot): 64.97
- TruthfulQA (0-shot): 73.86
- Winogrande (5-shot): 83.66
- GSM8k (5-shot): 72.71
With an 8192 token context length, it is suitable for tasks requiring moderate context understanding and generation. The diverse origins of its constituent models suggest a broad range of general-purpose language understanding and generation capabilities.
Good for
- General-purpose text generation: Its balanced benchmark scores indicate suitability for various conversational and creative tasks.
- Reasoning tasks: Performance on ARC and GSM8k suggests competence in logical and mathematical reasoning.
- Experimentation with merged models: Developers interested in exploring the outcomes of complex model merges may find this a valuable base.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.