kevin009/llamaRAGdrama: Factually Reliable RAG and Q&A

kevin009/llamaRAGdrama is a 7 billion parameter model specifically fine-tuned for Question Answering (Q&A) and Retrieval Augmented Generation (RAG). Its core objective is to synthesize text content that remains factual and reliable, even when dealing with dramatic or creative scenarios.

Key Capabilities

• Factual Reliability: Trained on a diverse dataset including dramatic texts and factual databases, ensuring generated content adheres to real-world facts.
• Q&A and RAG Optimization: Designed for applications requiring accurate information retrieval and synthesis.
• Balanced Training: Data preprocessing distinguished between creative and fact-dependent elements to achieve a balanced output.

Performance Highlights

Evaluated on the Open LLM Leaderboard, kevin009/llamaRAGdrama achieved an average score of 74.65. Notable benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 72.01
• HellaSwag (10-Shot): 88.83
• MMLU (5-Shot): 64.50
• TruthfulQA (0-shot): 70.24
• Winogrande (5-shot): 86.66
• GSM8k (5-shot): 65.66

Intended Use Cases

This model is ideal for:

• Retrieval Augmented Generation (RAG) systems: Where factual accuracy is paramount.
• Question Answering (Q&A) applications: Providing reliable answers from given contexts.

Limitations

While designed for truthfulness, the model may still exhibit biases and inaccuracies. Users should be aware that despite its factual grounding, it may not always be considered entirely "safe" in all content generation scenarios.