What is alpindale/gemma-7b?
alpindale/gemma-7b is a base version of Google's Gemma family of lightweight, open large language models. Derived from the same research and technology as the Gemini models, Gemma-7b is a text-to-text, decoder-only model available with open weights. It is designed to be accessible and deployable in environments with limited resources, such as laptops or desktops, fostering innovation across various applications.
Key Capabilities & Features
- Text Generation: Excels at diverse text generation tasks, including question answering, summarization, and reasoning.
- Open Weights: Provides full access to model weights, enabling extensive customization and fine-tuning.
- Resource-Efficient Deployment: Its 7 billion parameter size makes it suitable for deployment on consumer-grade hardware.
- Training Data: Trained on a diverse dataset of 6 trillion tokens, including web documents, code, and mathematical texts, to enhance its versatility.
- Responsible AI Focus: Developed with rigorous CSAM and sensitive data filtering, and evaluated against various ethics and safety benchmarks.
Why Choose Gemma-7b?
- Accessibility: Offers state-of-the-art AI capabilities in a format that is easier to deploy and experiment with compared to larger models.
- Performance: Benchmarks indicate strong performance across a range of tasks, including MMLU (64.3), HellaSwag (81.2), and HumanEval (32.3).
- Fine-tuning Support: Provides examples and scripts for supervised fine-tuning (SFT) using techniques like QLoRA and FSDP.
- Hardware Optimization: Optimized for efficient training and inference on Google's Tensor Processing Units (TPUs) and supports various precision levels (float16, bfloat16, 8-bit, 4-bit quantization) for GPU usage.