Overview

eren23/dpo-binarized-NeuralTrix-7B is a 7 billion parameter language model developed by eren23. It leverages the CultriX/NeuralTrix-7B-dpo architecture and has been fine-tuned using Direct Preference Optimization (DPO). The training utilized the argilla/OpenHermes2.5-dpo-binarized-alpha dataset, which is derived from teknium/OpenHermes-2.5 and processed with argilla-io/distilabel.

Key Capabilities & Performance

This model demonstrates strong performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard. Its average score is 76.17, indicating solid general-purpose capabilities. Specific benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 72.35
• HellaSwag (10-Shot): 88.89
• MMLU (5-Shot): 64.09
• TruthfulQA (0-shot): 79.07
• Winogrande (5-shot): 84.61
• GSM8k (5-shot): 68.01

Good For

• General text generation and understanding tasks.
• Applications requiring a model with a balanced performance across reasoning, common sense, and language comprehension.
• Use cases where a 7B parameter model offers a good balance between performance and computational efficiency.