Codex is Open Sourcing AI models (huggingface.co)

OpenAI has announced that it is open-sourcing Codex, its AI coding agent, to further enhance capabilities for the AI/ML community. Integrating Codex with the Hugging Face Skills repository allows it to fine-tune models using reinforcement learning, review live training metrics, evaluate model checkpoints, and auto-generate reports. The introduction of HF-skills enables Codex to manage complex machine learning experiments autonomously, streamlining workflows and facilitating hands-off model training and evaluation processes. This development is significant as it empowers engineers to delegate model training tasks effectively, leading to more efficient experimentation. Codex can automatically validate datasets, select appropriate GPU resources, and maintain detailed training reports, making it a powerful tool for ML practitioners. The extension supports production-level training methods and model sizes ranging from 0.5B to 7B parameters, indicating robust capabilities for both exploratory and applied machine learning projects. Users will need a Hugging Face Pro account to access certain features, further aligning the infrastructure for collaborative and scalable AI research.