Sciphi
Models by this creator
⚙️
Triplex
219
Triplex is a state-of-the-art large language model (LLM) developed by SciPhi.AI for the task of knowledge graph construction. It is a finetuned version of the Phi3-3.8B model that excels at extracting triplets - simple statements consisting of a subject, predicate, and object - from text or other data sources. Compared to GPT-4, Triplex is able to construct knowledge graphs at a 98% cost reduction while maintaining strong performance. Unlike more expensive knowledge graph approaches like Microsoft's Graph RAG, Triplex enables local graph building at a fraction of the cost using SciPhi's R2R system. It outperforms GPT-4 on benchmark tasks related to knowledge graph construction, making it a compelling option for projects that require building knowledge graphs from unstructured data. Model inputs and outputs Inputs Unstructured text data**: Triplex takes in raw text as input and extracts knowledge graph triplets from it. Entity types and predicates**: The model also takes in a list of entity types and predicates that it should focus on when extracting triplets. Outputs Knowledge graph triplets**: The main output of Triplex is a set of extracted triplets representing relationships between entities in the input text. Capabilities Triplex excels at the task of knowledge graph construction, outperforming GPT-4 while costing 1/60th as much. It is able to rapidly extract high-quality triplets from text, enabling users to build knowledge graphs at a fraction of the typical cost. This makes it a powerful tool for applications that require structured knowledge extracted from unstructured data sources. What can I use it for? Triplex is well-suited for any project that requires building knowledge graphs from text data. This could include applications in areas like: Business intelligence**: Extracting insights and relationships from corporate documents, reports, and other internal data sources. Scientific research**: Mapping out connections between concepts, entities, and findings in academic papers and other technical literature. Public sector**: Aggregating and structuring information from government reports, legislation, and other public documents. The cost-effectiveness of Triplex makes it an appealing option for organizations that need to build knowledge graphs but have limited budgets or computational resources. Things to try One interesting aspect of Triplex is its ability to focus on specific entity types and predicates when extracting knowledge graph triplets. This allows users to tailor the model's output to their particular needs and use cases. For example, you could experiment with different sets of entity types and predicates to see how the extracted triplets vary, and then select the configuration that is most relevant for your project. Another thing to try is using Triplex in conjunction with SciPhi's R2R system for local knowledge graph building. By leveraging R2R, you can quickly and efficiently construct knowledge graphs from text data without the need for expensive cloud-based infrastructure.
Updated 8/23/2024
📶
Sensei-7B-V1
84
The Sensei-7B-V1 is a Large Language Model (LLM) fine-tuned from the mistral-ft-optimized-1218 model, which is based on the Mistral-7B model. Sensei-7B-V1 was fine-tuned with a fully synthetic dataset to specialize in performing retrieval-augmented generation (RAG) over detailed web search results. This model aims to generate accurate and well-cited summaries from a range of search results, providing more precise answers to user queries. Similar models include the Mistral-7B-Instruct-v0.1, merlinite-7b, Mistral-7B-Instruct-v0.2, and Mixtral-8x7B-Instruct-v0.1. These models share similarities in their base architecture and fine-tuning approaches, though they may differ in specific capabilities and performance characteristics. Model inputs and outputs Inputs Single search query**: The model is designed to take a single search query as input and use it to generate a response. Outputs Retrieval-augmented generation**: The model returns an answer that is generated using the context of the search results as background information. JSON format**: The model's output is structured in a JSON format that includes a summary of the search results and a list of related queries. Capabilities The Sensei-7B-V1 model specializes in using search to generate accurate and well-cited summaries. It can leverage detailed web search results to provide more precise answers to user queries, drawing upon the contextual information to produce informative responses. What can I use it for? The Sensei-7B-V1 model can be useful for applications that require generating detailed, fact-based responses to user questions or information requests. This could include chatbots, virtual assistants, or knowledge-based systems that need to provide accurate and well-supported information to users. Things to try One interesting aspect of the Sensei-7B-V1 model is its ability to utilize search results as context for generating responses. You could experiment with providing the model with different types of search queries, from factual questions to more open-ended information requests, and observe how it leverages the search context to formulate its answers. Additionally, you could explore the model's performance on tasks that require synthesizing information from multiple sources, such as summarizing a set of web pages on a given topic or answering follow-up questions that build upon the initial search results.
Updated 5/28/2024
🌀
SciPhi-Self-RAG-Mistral-7B-32k
83
SciPhi-Self-RAG-Mistral-7B-32k is a Large Language Model (LLM) fine-tuned from the Mistral-7B-v0.1 model. It underwent further fine-tuning on the recently released self-rag dataset, as well as other RAG-related instruct datasets, in an effort to improve its conversational abilities. The model benchmarks well, but requires additional tuning to become an excellent conversationalist. Similar models include the Sensei-7B-V1, which is also fine-tuned from the Mistral-7B base model but specializes in retrieval-augmented generation (RAG) over detailed web search results. Model inputs and outputs Inputs Conversation**: The model accepts a list of messages in the format {"role": "system|user|assistant", "content": "message text"}, where the "system" message provides additional instructions for the assistant. Outputs Response**: The model generates a response text, which can be a continuation of the conversation. Capabilities The SciPhi-Self-RAG-Mistral-7B-32k model is capable of engaging in open-ended conversations and leveraging search results to provide more accurate and well-cited responses to user queries. It has been fine-tuned to specialize in using search, such as AgentSearch, to generate summaries from a range of search results. What can I use it for? You can use the SciPhi-Self-RAG-Mistral-7B-32k model for a variety of natural language processing tasks, such as open-ended conversation, question-answering, and retrieval-augmented generation. The model could be particularly useful for applications that require accurate and well-cited responses, such as customer service chatbots, virtual assistants, or knowledge management systems. Things to try One interesting thing to try with the SciPhi-Self-RAG-Mistral-7B-32k model is to provide it with specific search queries and observe how it leverages the search results to generate responses. You can also experiment with different prompting techniques, such as providing the model with additional context or instructions, to see how it affects the quality and coherence of the generated responses.
Updated 5/28/2024
👀
SciPhi-Mistral-7B-32k
68
The SciPhi-Mistral-7B-32k is a Large Language Model (LLM) fine-tuned from the Mistral-7B-v0.1 model. This model underwent a fine-tuning process over four epochs using more than 1 billion tokens, which include regular instruction tuning data and synthetic textbooks. The objective of this work was to increase the model's scientific reasoning and educational abilities. Similar models include the SciPhi-Self-RAG-Mistral-7B-32k, which was further fine-tuned on the self-rag dataset, and the Sensei-7B-V1 which specializes in retrieval-augmented generation (RAG) over detailed web search results. Model inputs and outputs The SciPhi-Mistral-7B-32k is a text-to-text model that can take in a variety of prompts and generate relevant responses. For best results, it is recommended to follow the Alpaca prompting guidelines. Inputs Prompts**: Natural language instructions or questions that the model should respond to. Outputs Text responses**: The model will generate relevant text responses based on the input prompt. Capabilities The SciPhi-Mistral-7B-32k model has been trained to excel at scientific reasoning and educational tasks. It can provide informative and well-cited responses to questions on a wide range of scientific topics. The model also demonstrates strong language understanding and generation capabilities, allowing it to engage in natural conversations. What can I use it for? The SciPhi-Mistral-7B-32k model can be utilized in a variety of applications that require scientific knowledge or educational capabilities. This could include: Developing interactive educational tools or virtual assistants Generating summaries or explanations of complex scientific concepts Answering questions and providing information on scientific topics Assisting with research and literature review tasks Things to try One interesting aspect of the SciPhi-Mistral-7B-32k model is its ability to provide well-cited responses. By following the Alpaca prompting guidelines, you can prompt the model to generate responses that incorporate relevant information from the provided context. This can be useful for tasks that require factual accuracy and transparency, such as research assistance or explainable AI applications. Another interesting feature is the model's potential for conversational abilities. By framing prompts as natural language dialogues, you can explore the model's ability to engage in coherent and contextual exchanges, potentially uncovering new use cases or areas for further development.
Updated 5/27/2024