pandafish-dt-7b: A Merged 7B Language Model

pandafish-dt-7b is a 7 billion parameter language model developed by ichigoberry, created through a 'dare_ties' merge of Experiment26-7B and MergeCeption-7B-v3 using the LazyMergekit framework. This merging technique aims to combine the strengths of its constituent models to achieve enhanced performance.

Key Capabilities & Performance

This model has been evaluated on the Nous Benchmark suite, demonstrating competitive results among 7B models. Notably, pandafish-dt-7b achieves:

• 77.19 on GPT4All, indicating strong general knowledge and reasoning.
• 78.41 on TruthfulQA, suggesting good factual accuracy and resistance to hallucination.
• A respectable average score of 62.65 across the benchmark suite, placing it among top-performing merged models like AlphaMonarch-7B and Monarch-7B.

Usage and Availability

pandafish-dt-7b is readily available for use, with a Huggingface Space playground for easy interaction. For deployment, various quantized versions are provided:

• GGUF: Available from ichigoberry and mradermacher (with IQ).
• MLX: 4-bit and 8-bit versions are available from the MLX community.

Good for

• General-purpose text generation: Excels in producing coherent and contextually relevant text.
• Question Answering: Strong performance in factual recall and answering diverse queries.
• Reasoning tasks: Its competitive benchmark scores suggest good capabilities in general reasoning scenarios.