--- language: - en pipeline_tag: text-generation tags: - facebook - meta - llama - llama-3 - llava license: other license_name: llama3 license_link: LICENSE --- ## Disclaimer These models are research experiments and may generate incorrect or harmful content. Outputs from these models should not be taken as factual or representative of the views of myself or the model's creator or any other individual. The creator(s) of these models and I are not responsible for any harm or damage caused by the models outputs. I did not train these models or have any say in their creation, I merely converted these models from the sources available below. To report issues or concerns, please contact the model maker via the links provided in this README. ## Conversions I have used llama.cpp to convert and quantize each of the models available in this repository. Currently, I have quantized: - `meta` Llama 3 [8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B). Q4_K_M and Q5_K_M. - `meta` Llama 3 [8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct). Q4_K_M and Q5_K_M. - `xtuner` Llava Llama 3 [Llava-Llama-3-8B-v1_1](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1). Q4_K_M and Q5_K_M. **Important information related to each model can be found in the links above** **Model Architecture** Llama 3 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
Training Data | Params | Context length | GQA | Token count | Knowledge cutoff | |
Llama 3 | A new mix of publicly available online data. | 8B | 8k | Yes | 15T+ | March, 2023 |
70B | 8k | Yes | December, 2023 |