SuperAPI AI Directory

AI developer platforms

3,662 results · Page 58 of 102

BiWM

AI developer platforms

BiWM is the first open-source minimalist training framework for bidirectional autoregressive video world models. Through two training stages, it transforms a pretrained bidirectional video diffusion model into an action-controllable autoregressive world model: stage one performs camera-control fine-tuning with 81 discrete camera actions, and stage two performs few-step DMD distillation to achieve chunk-by-chunk autoregressive generation. Unlike causal attention pipelines, BiWM preserves full bidirectional attention within the current chunk and its history, and supports t2v (text-to-video), i2v (image-to-video), and v2v (video-to-video) conditional generation in the same model. The framework supports backbone models such as Wan2.1 / Wan2.2 and HunyuanVideo-1.5 / LTX-2.3, and is released under the Apache License 2.0.

Read more

Book-of-MLM

AI developer platforms

Supporting teaching resources for "Multimodal Large Language Models: A New Paradigm of Artificial Intelligence Technology," open-sourced and maintained by HCP-Lab at Sun Yat-sen University on GitHub. The book systematically introduces the key technologies, foundation models, and typical applications of multimodal large models. The book has 5 chapters in total, covering large model architectures, core multimodal technologies, representative multimodal large models, visual question answering/AIGC/embodied intelligence applications, and frontier discussions toward artificial general intelligence (causal reasoning, world models, embodied intelligence, and multi-agent systems, etc.). The resource repository provides undergraduate teaching courseware from the School of Computer Science and Engineering at Sun Yat-sen University, with 12 lectures in total, and has been used as teaching PPTs by more than ten universities including Tongji University, Chongqing University, Northwestern Polytechnical University, and Shandong University. This book has been selected for the Sun Yat-sen University undergraduate textbook series. It is suitable as a reference for senior undergraduates, graduate students, and IT practitioners at universities, and can accommodate teaching arrangements of 32 to 56 class hours.

Read more