Toolformer: Language Models Can Teach Themselves to Use Tools
Presents Toolformer, a model trained to decide which APIs to call, when to call them, what arguments to pass, and how to best incorporate the results into future token prediction.
Toolformer achieves substantially improved zero-shot performance across a variety of downstream tasks, often competitive with much larger models, without sacrificing its core language modeling abilities. This can improve workflow automation and processing in businesses through the use of tool integration.
Q-Diffusion: Quantizing Diffusion Models
Can directly quantize full-precision diffusion models into 8-bit or 4-bit models while maintaining comparable performance in a training-free manner.
This can speed up the image synthesis process in businesses that use diffusion models. It can also improve text-guided image generation in industries such as advertising and design.
Re-ViLM: Retrieval-Augmented Visual Language Model for Zero and Few-Shot Image Captioning
Significantly improved perf for image-to-text generation, esp. for zero-shot and few-shot generation in OOD with 4x less parameters compared with baseline methods.
This can improve image-to-text generation in industries that require large amounts of visual data, such as e-commerce and social media. It can also make the process more efficient and cost-effective.