Abstract: GenZ 13B v2 is an Instruction Fine-tuned LLM open-sourced by Accubits Technologies and Bud Ecosystem Inc. It can be run even on your laptop. The model is primarily instruct-tuned for better reasoning, roleplay, and writing capabilities and achieved SOTA in its category for the MT Bench benchmark with 87% accuracy compared to ChatGPT and on par with the 70B model of the LLaMA2 chat model. With GenZ, we are on a mission to build an open-source foundational model with the knowledge and reasoning capabilities of GPT4, which is privacy first and can be hosted even on a mobile device.
Accubits Technologies and Bud Ecosystem open-sourced its third Large Language Model (LLM), GenZ 13B v2, an Instruction Fine-tuned LLM offering license for commercial use. It consists of three models. The first is a 32-bit model, and the second is a 4-bit quantized model that can be run even on low-spec hardware to do inference. The third model doesn’t require any GPU and can run on CPUs. This is a step toward our efforts to democratize access to LLMs and foundation models for everyone, irrespective of their access to high-performance computing systems.
GenZ-13B v2 models are primarily instruct-tuned for better reasoning, roleplay, and writing capabilities. Our initial evaluation shows that the model achieved SOTA in its category for the MT Bench benchmark with 87% accuracy compared to ChatGPT and on par with the 70B model of the LLaMA2 chat model, which is a 5X bigger model that requires 40X more GPU memory. The model also gives better accuracy than Stanford instruction trained Alpaca models, Vicuna, etc.
Large language models (LLMs) are at the forefront of what could be a fundamental shift in human evolution. They are not merely tools to optimize our present capabilities but represent a whole new trajectory for progress, akin to the onset of the industrial revolution or the advent of the digital age. Democratization of skills is one of the most exciting prospects ushered in by LLMs. Today, the threshold to becoming an artist, story writer, or programmer is significantly lower than it once was. LLMs and foundation models enable the non-experts to express themselves creatively, write compelling narratives, or code software solutions, thus, fostering an egalitarian culture of talent and skill.
However, the accessibility of LLMs remains a significant challenge. Currently, only large corporations and research institutions can bear the high infrastructural costs associated with LLMs. The expense of running these models can be prohibitive, thus limiting their accessibility and potential for widespread innovative applications.
With GenZ, we are on a mission to build an open-source foundational model with the knowledge and reasoning capabilities of GPT4, which is privacy first and can be hosted even on a mobile device. We believe everyone should have access to this revolutionary technology. Each developer, entrepreneur, and business should have the means to experiment with these models, to build creative, efficient, and ground-breaking solutions. The power of LLMs should not be exclusive but should be leveraged for the collective advancement of society. After all, technological progress reaches its full potential when it can be harnessed by all, not just by a privileged few. GenZ 13B v2 is taking us one step closer to this mission.
| Model Name | MT Bench | Vicuna Bench | MMLU | Human Eval | Hellaswag | BBH |
|---|---|---|---|---|---|---|
| Genz 13B v1 | 6.12 | 86.1 | 53.62 | 17.68 | 77.38 | 37.76 |
| Genz 13B v2 | 6.79 | 87.2 | 53.68 | 21.95 | 77.48 | 38.1 |
GenZ-13B v2 has a 4K input length capacity and 13 billion parameters. The model’s capabilities have been tested across various benchmarks. On the MT Bench and Vicuna Bench, the model achieved notable scores of 6.79 and 87.2, respectively. In the MMLU benchmark, the model produced a respectable score of 53.68. In HumanEval, Hellaswag, and BBH, the model scored 21.95, 77.48, and 38.1, respectively.
We observed that the model has a slight drop in handling complex code. We acknowledge this challenge and are devoted to rectifying it in the next iteration. Quantized models show lesser reasoning power and QnA capabilities. We are further evaluating the model and will publish the results as they become available.
By streamlining the models for local hardware, we aim to democratize access to this powerful technology, opening up new possibilities for innovation and exploration for all. Accubits Technologies invites all interested parties to explore the potential applications of GenZ-13B in their operations. The model’s unique ability to understand and respond to detailed instructions will revolutionize sectors requiring complex and nuanced interactions, leading to more accurate and efficient results.
Download Models
| Genz-13b-v2-ggml | HuggingFace | GitHub |
| Genz-13b-v2-4bit | HuggingFace | GitHub |
| Genz-13b-v2 | HuggingFace | GitHub |
| Genz-70B | HuggingFace | GitHub |
Drop us a word if you want to know more about the model or need help adopting LLMs in your enterprise.
Intended Use
When we created GenZ 13B, we had a clear vision of how it could be used to push the boundaries of what’s possible with large language models. We also understand the importance of using such models responsibly. Here’s a brief overview of the intended and out-of-scope uses for GenZ 13B.
Direct Use:
GenZ 13B is designed to be a powerful tool for research on large language models. It’s also an excellent foundation for further specialization and fine-tuning for specific use cases, such as:
Out-of-Scope Use
While GenZ 13B is versatile, there are certain uses that are out of scope:
Remember, GenZ 13B, like any large language model, is trained on a large-scale corpora representative of the web, and therefore, may carry the stereotypes and biases commonly encountered online.
Recommendations
We recommend users of GenZ 13B to consider fine-tuning it for the specific set of tasks of interest. Appropriate precautions and guardrails should be taken for any production use. Using GenZ 13B responsibly is key to unlocking its full potential while maintaining a safe and respectful environment.
