Local LLMs

This is a fine-tuned model based on the LLama 13b base model created by Meta. It was fine-tuned with a broad set of instruction and creative writing prompts to enable it to perform a vast number of tasks. Its capabilities are comparable to those of text-davinci-003, the base model of GPT 3.5. However, it cannot match the quality of interaction that GPT 3.5 is capable of.

This language model is state-of-the-art and was fine-tuned on over 300,000 instructions. It is based on the LLama 13b base model by meta. The result is an enhanced Llama 13b model that performs similarly to GPT-3.5-turbo across a variety of tasks. It is an excellent model to use if you require high performance but want to process your data locally.

This model is based on the gpt-j-6b model, but has been fine-tuned with a broad set of instruction and creative writing prompts, enabling it to perform a wide range of tasks. Although its capabilities are less pronounced than the models based on LLama, it provides a faster response time.

Falcon-7B is a strong base model, outperforming comparable open-source models (e.g., MPT-7B, StableLM, RedPajama etc.), thanks to being trained on 1,500B tokens of RefinedWeb enhanced with curated corpora.

A highly advanced 2.7B Causal Language Model designed specifically for code completion tasks. Trained on a diverse dataset of 20 programming languages, including Java, JavaScript, Python, and more, it has a comprehensive understanding of code syntax and semantics. With its state-of-the-art techniques such as Flash Attention, AliBi positional embeddings, and LionW optimizer, replit-code-v1-3b offers fast and accurate code suggestions, making it a powerful tool for developers seeking intelligent code completion assistance.

Flan-T5 Large is a language model that has been fine-tuned on a large corpus of text data using the instruction-based fine-tuning approach. It is designed to perform well on a variety of natural language processing tasks, including zero-shot and few-shot learning. The main difference between Flan-T5 Large and other models is the fine-tuning process, where Flan-T5 Large is trained using instructions as prompts, which helps improve its performance and generalization to unseen tasks. Flan-T5 Large is particularly good at complex reasoning tasks and can generate creative and contextually appropriate responses.

The Flan-T5 XL model is a language model that has been fine-tuned on a large corpus of text data using instruction-based training. It is designed to excel in zero-shot and few-shot learning tasks, such as reasoning and question answering. Compared to other models, Flan-T5 XL has a larger number of parameters, specifically 3 billion, which allows it to capture more complex patterns and generate more accurate responses. It also outperforms other models in terms of performance on various benchmarks, including achieving state-of-the-art results on tasks like Massive Multi-task Language Understanding (MMLU). Overall, Flan-T5 XL offers improved performance and versatility compared to other models.

The Flan-T5 XXL model is a large language model that has been fine-tuned on a wide range of tasks and instructions. It is designed to excel in zero-shot and few-shot learning scenarios, where it can generate accurate and contextually appropriate responses without specific training on the task at hand. Compared to other Flan-T5 models, the XXL variant has the highest number of parameters (11 billion), allowing it to capture more complex patterns and generate more detailed and nuanced responses. It has been shown to outperform other models, including PaLM, on various benchmark tasks, demonstrating its superior performance and versatility.

Falcon 7B model optimized for CPU inference. Considered by many to be the best open-source 7B model. Fast response time and good overall quality.

Orca is a model based on the LLama 13b model and fine-tuned with creative methods. It is believed to outperform other LLama models at reasoning and logic tasks. Although the model is generally slow, it delivers excellent results in the areas for which it was fine-tuned. Based on its performance, Orca would be a suitable choice for tasks that require strong reasoning and logic capabilities. However, users should be aware that the model's slow speed could impact its usability for certain applications.

Wizard LM is a fast model, finetuned by Microsoft and Peking University, based on the smaller LLama 7b model.

Manticore 13B (previously known as Wizard Mega) is a language model developed by OpenAccess AI Collective. It is based on the Llama 13B model and has been fine-tuned on various datasets to improve its performance and capabilities. Manticore 13B provides detailed responses to a wide range of prompts, including role-play, conceptual physics, logical fallacies, and more. It has been optimized for both concise and detailed responses, depending on the specific dataset.

LLM ChatGPT 3.5 is an advanced conversational AI model developed by OpenAI. It excels at understanding and generating natural language, providing contextually relevant responses. Trained on a vast amount of text data, it has wide-ranging knowledge and can provide information on various topics. However, it lacks real-time information beyond its September 2021 training cut-off. While it comprehends many aspects of conversation, it may struggle with nuanced queries and lack critical thinking. Users should verify information from reliable sources and be aware of its limitations.

Similar to ChatGPT3.5-Turbo 4k, this model can process four times more data in a single query. This makes it highly effective in tasks that require a significant amount of input data or for generating long-form outputs, such as blog posts or other extended content formats.

This model is an optimization of the Luminous-extended, fine-tuned to better follow user instructions across a diverse set of tasks. It excels in fields such as information extraction, language simplification, and other tasks needing explicit instruction. Its major advantage lies in its superior zero-shot performance, which minimizes token usage in inputs, reducing costs while improving efficiency. This blend of affordability and adaptability makes it an excellent choice for various applications where precise control and prompt responses are essential.

This model represents the pinnacle of the Luminous family, combining the immense capacity of Luminous-supreme with the fine-tuned instruction-following capability seen in our control models. Excelling at complex and creative tasks, it brings an unmatched level of comprehension and versatility to your fingertips. Its ability to perform highly in zero-shot scenarios makes it an optimal choice for demanding applications where precision, nuanced understanding, and complex problem-solving are required. Despite being the largest model, its efficient use of input tokens ensures cost-effectiveness.