nlpguy/AlloyIngotNeoX: A Merged 7B Language Model
nlpguy/AlloyIngotNeoX is a 7 billion parameter language model developed by nlpguy, constructed through a strategic merge of two pre-trained models: bardsai/jaskier-7b-dpo-v4.3 and Gille/StrangeMerges_20-7B-slerp. This model leverages the SLERP (Spherical Linear Interpolation) merge method, a technique known for combining the strengths of its constituent models while maintaining coherence.
Key Capabilities & Performance
This model exhibits robust performance across a suite of benchmarks, as evaluated on the Open LLM Leaderboard. Its average score of 76.21 indicates strong general capabilities. Specific benchmark results include:
- AI2 Reasoning Challenge (25-Shot): 74.32
- HellaSwag (10-Shot): 89.07
- MMLU (5-Shot): 64.97
- TruthfulQA (0-shot): 74.57
- Winogrande (5-shot): 84.53
- GSM8k (5-shot): 69.83
These scores suggest proficiency in areas such as common sense reasoning, factual recall, and mathematical problem-solving.
Good For
- General-purpose text generation: Capable of handling a wide array of natural language tasks.
- Reasoning and logical inference: Demonstrated by strong performance on ARC and Winogrande.
- Question Answering: Supported by its TruthfulQA and MMLU scores.
- Exploratory research: As a merged model, it offers a unique blend of characteristics from its base models, making it interesting for further fine-tuning or analysis.