GenZ 13B v2

About Model

GenZ 13B v2 is a newly developed instruction fine-tuned model that capably manages complex instructions, providing a substantial 4K input length capacity. This model has been carefully fine-tuned using the LLaMA 2 pre-trained model. A notable feature of GenZ 13B v2 is its capability to understand detailed instructions and queries, making it suitable for tasks that require detailed and thorough responses.

HuggingFace



Github

Results

GenZ-13B v2 is an instruction fine-tuned LLM, and it can manage complex instructions, providing a substantial 4K input length capacity. A notable feature of the model is its capability to understand detailed instructions and queries, making it suitable for tasks that require detailed and thorough responses.

Benchmark

The performance of the Genz 13B v2 model is evaluated across several measures. On the MT Bench and Vicuna Bench, the model achieved a notable scores of 6.79 and 87.2 respectively. When tested on the MMLU, it produced a respectable score of 53.68. In HumanEval, the model scored 21.95, Hellaswag score of 77.48 and BBH score of 38.1

Performance

Genz 13B v2 has a total of 13 billion parameters. Even with its comparatively smaller size than the hefty GPT 3.5, Genz 13B has demonstrated impressive competitiveness. In terms of performance, the model has reached 87% of what GPT 3.5 can accomplish, which is quite an achievement, given the latter's size and compute requirements.