IFM Institute in Abu Dhabi opens code and data for six AI models — Asharq Al-Awsat
The IFM research institute in Abu Dhabi released six K2 Horizon artificial intelligence models on September 3 and published the data and methods used to create them. As Asharq Al-Awsat reports, the release also includes model weights, software code and intermediate training checkpoints.
Reproducing the training process
IFM founder Eric Xing told Reuters that the published set of materials allows researchers to trace the models’ development process and reproduce the results. According to him, the aim of the release is to set a benchmark for what a fully open release of AI models can look like.
Xing also said that openness and competitive model performance do not contradict each other. According to him, the release is intended to demonstrate this to policymakers, regulators and organizations advocating for the public interest.
More current news is available on the UA.News Telegram channel Telegram.
Difference from open-weight models
IFM’s approach differs from the practice of developers who publish only model weights for download but do not fully disclose the training data and methods. The report notes that OpenAI and Anthropic do not make their models available for download and do not disclose the data and technologies used to create them.
The K2 Horizon lineup includes models for various tasks: from a system designed for smartwatches and other resource-constrained devices to a model with 375 billion parameters for enterprise use. The launch is part of the United Arab Emirates’ course to develop the country as a global artificial intelligence hub.
