Amazon Science is focused on improving code generation through debugging. They use large language models (LLMs) to both generate and debug code, leveraging techniques like supervised fine-tuning and reinforcement learning to train the models. A key element involves creating synthetic debugging data to overcome the scarcity of real-world examples. This approach shows significant improvement in code performance as measured by standard benchmarks. The team utilizes methods like chain-of-thought reasoning and focuses on refining models. A variety of job opportunities are also presented in areas including generative AI, transportation optimization, and advertising technology. These are all related to machine learning, data science, and AI within Amazon and AWS.

The PyTorch blog post focuses on accelerating generative AI models, specifically Segment Anything 2 (SAM2), using native PyTorch. It details techniques like torch.compile and torch.export for optimized, low-latency inference. The authors achieved significant performance improvements (up to 13x) by employing ahead-of-time compilation, reduced precision, batched prompts, and GPU preprocessing. These optimizations were tested in realistic, autoscaling cloud environments via Modal, demonstrating their practical benefits. The experiments show the balance between speed and accuracy when applying various "fast" and "furious" strategies to SAM2. The post also provides resources to reproduce the results and encourages community contributions.

Visuo-Haptic 6D Object Pose Tracking

Force-Attending Curriculum Training for Contact-Rich Policy Learning

LLMs learn to lie, cheat and KILL in Among Us

Sharp Monocular Metric Depth Estimation

Benchmarking Chain-of-Thought in Large Multimodal Models

Efficient GRPO for Long-Context Reasoning Models

Tunable Length Control for Chain-of-Thought Reasoning

Implementing a Transformer model from scratch is a great way to understand the architecture and its components in depth.