NEWSLETTER
In artificial intelligence and automatic learning, high quality data sets play a crucial role in the development of precise and reliable models. However, the collection of extensive and verified data, particularly in specialized domains such as mathematics, coding and science, increases a challenge. Traditional data collection methods often cannot produce data sets that effectively train complex reasoning tasks. This gap highlights the need for new approaches to the creation and verification of the data set.
Prime Intellect has entered Synthetic-1, an open source data set designed to provide traces of math, coding and science. Built with the Deepseek-R1 support, this data set consists of 1.4 million tasks and structured verifiers. The objective of Synthetic-1 is to improve reasoning models by providing well organized and reliable data, addressing the deficiencies of existing resources.
Synthetic-1 includes a range of types of tasks, each designed to guarantee quality and relevance:
- 777,000 mathematical problems with symbolic verifiers: These problems, obtained from the Numinamath data set, focus on the questions at the competence of high school. A LLM -based filtering process eliminates non -verifiable problems, such as those that require evidence, and reformulate multiple choice questions in direct response formats.
- 144,000 coding problems with unit tests: Extracted from data sets such as applications, codects, codeforces and taco, these problems come with unit tests to verify the solutions. The data set initially contained python problems, which then expanded to include JavaScript, Rust and C ++, increasing the variety and depth of the challenges.
- 313,000 questions about Stem open with LLM Evaluation: Using the Stackexchange data set, this subset covers a broad spectrum of technical and scientific topics. The selection process prioritizes the questions that require reasoning instead of recovering simple information. A Judge of LLM obtains responses based on his alignment with the community responses of the best votes.
- 70,000 real -world software engineering tasks: These tasks, extracted from Github's commitments in the Commitpack data set, involve modifying code files based on confirmation instructions. A Judge of LLM evaluates the solutions by comparing them with the real post-compromise states.
- 61,000 Code output prediction tasks: Focus on predicting the output of code transformations into the chains, this subset defies models with increasingly complex chain handling tasks. These problems are designed to be particularly difficult for modern ai models.
The structured nature of Synthetic-1 makes it a valuable resource for training models in structured reasoning. By including programmatically verifiable problems, such as coding tasks with unit tests, the data set guarantees clear correction criteria. In addition, open reasoning questions verified by LLM judges provide challenges that drive the limits of the current abilities of ai. The collaborative framework of the data set also allows continuous improvement and expansion, promoting a shared effort to refine the training resources of ai.
Synthetic-1 represents a step forward in the creation of high quality data sets for ai models based on reasoning. When addressing the gaps in existing data sets, it provides a structured basis to improve the reasoning of machines in mathematics, coding and science. The project also encourages continuous contributions, so it is an evolving resource for researchers and developers who work to advance the abilities of ai in the structured resolution of problems.
Verify he <a target="_blank" href="https://www.primeintellect.ai/blog/synthetic-1″ target=”_blank” rel=”noreferrer noopener”>Details and Data set in the hugged face. All credit for this investigation goes to the researchers of this project. Besides, don't forget to follow us <a target="_blank" href="https://x.com/intent/follow?screen_name=marktechpost” target=”_blank” rel=”noreferrer noopener”>twitter and join our Telegram channel and LINKEDIN GRsplash. Do not forget to join our 75K+ ml of submen.
Recommended open source ai platform: “Intellagent is a multiple open source agent frame to evaluate the complex conversational system” (promoted)
Asif Razzaq is the CEO of Marktechpost Media Inc .. as a visionary entrepreneur and engineer, Asif undertakes to take advantage of the potential of artificial intelligence for the social good. Its most recent effort is the launch of an artificial intelligence media platform, Marktechpost, which stands out for its deep coverage of automatic learning and deep learning news that is technically solid and easily understandable by a broad audience. The platform has more than 2 million monthly views, illustrating its popularity among the public.
