Some GGUF v3 quantizations of the model sachith-surge/open-llama-v2-lamini-orca-evol-qlora-checkpoint-safetensors

open-llama-v2-lamini-orca-evol-qlora-checkpoint-safetensors

This is a instruction tuned model based on Open-LLaMA-3b-v2. It is trained on a large corpus of text from various domains and can generate realistic and coherent texts on various topics. The model is created by Team Indigo and is licensed under apache-2.0. The model has 3.43B parameters and uses F32 tensor type. The model is too large to load onto the free Inference API, but you can try it on Inference Endpoints instead. The model is named open-llama-v2-lamini-orca-evol-qlora-checkpoint-safetensors and is part of the Safetensors llama project. The model is intended for research and educational purposes only and should not be used for any harmful or malicious purposes.

Model description

The model is based on Open-LLaMA-3b-v2, a large-scale language model that can generate natural language texts given a prompt. The model is fine-tuned using the alpaca training prompt, a method that allows the model to learn from multiple sources of information without forgetting previous knowledge. The alpaca training prompt consists of a prefix, a query, and a suffix that guide the model to generate relevant and diverse texts.

The model is trained on custom datasets that are created using three different schemes: LaMini scheme, Orca scheme, and evol-instruct scheme. These schemes are designed to enhance the quality and diversity of the generated texts by providing different types of information and instructions to the model.

• The LaMini scheme uses a large and diverse corpus of text from various domains, such as news, books, blogs, social media, etc. The scheme also uses a small set of keywords to provide topical information to the model.
• The Orca scheme uses a smaller and more focused corpus of text from specific domains, such as science, technology, art, etc. The scheme also uses a longer set of keywords to provide more detailed information to the model.
• The evol-instruct scheme uses an evolutionary algorithm to generate and select the best instructions for the model. The scheme also uses a feedback mechanism to reward or penalize the model based on its performance.

Limitations and bias

The model is trained on a large corpus of text from various sources, which may contain biases or inaccuracies. The model may also generate texts that are offensive, harmful, or misleading. The model should not be used for any critical or sensitive applications that require high accuracy or ethical standards.

The model is also limited by its size and complexity, which may affect its speed and performance. The model may not be able to handle long or complex prompts or queries, or generate long or coherent texts. The model may also produce repetitive or nonsensical texts if it encounters unfamiliar or ambiguous inputs.

The model is still a work in progress and may have bugs or errors. The model is constantly being improved and updated based on feedback and evaluation. If you encounter any issues or have any suggestions for improvement.