SuperAPI AI Directory
Partner with us中文AI newsFavorites

All AI tools

35,738 results · Page 228 of 993

DreamBox Learning

AI learning

DreamBox Learning is an online math and reading learning platform for K-8 students, offering adaptive, personalized interactive instruction. It uses intelligent adaptive learning technology to assess student actions in real time and dynamically adjust learning paths, combining step-by-step lessons with continuous formative assessment to help learners at different levels progress at their own pace. The platform also serves families and educators and can be used for intervention, remediation, or enrichment learning.

Read more

DreamLite

AI images

DreamLite is a lightweight on-device unified diffusion model (0.39B) released by ByteVisionLab, supporting both text-to-image generation and text-guided image editing within a single network architecture. It is based on a pruned mobile U-Net backbone, unifies multimodal conditions through in-context spatial concatenation in latent space, and achieves 4-step ultra-fast inference via progressive step distillation, generating or editing 1024×1024 images in about 3 seconds on an iPhone 17 Pro, running entirely on-device with no cloud required. The project has been merged into the Hugging Face Diffusers library, providing a CLI, Gradio demo, LoRA fine-tuning, and iOS on-device deployment references; model weights require access approval and are restricted to non-commercial research use.

Read more

DRT-o1

AI office

DRT-o1 is a Deep Reasoning Translation project that explores introducing long chain-of-thought reasoning into neural machine translation (MT). The project synthesizes translation samples with long thought processes through a multi-agent framework and has trained models such as DRT-7B, DRT-8B, and DRT-14B. Related papers include DRT (ACL 2025 Findings), DeepTrans (TACL 2025), and ExTrans (arXiv 2025). The project has released the MetaphorTrans synthetic data (22,264 entries), as well as model checkpoints such as DeepTrans-7B, ExTrans-7B, and mExTrans-7B. During inference, the DRT models first output the thought content and then provide the final translation, using specific system prompts and user prompts. DeepTrans improves free translation ability through reinforcement learning, using DeepSeek-v3 as the reward model; ExTrans adopts exemplar-enhanced reinforcement learning, and mExTrans extends to 11 languages. Evaluation is based on two literary books, among which the content of "The Essential O. Henry Collection" has been released, while "Orbital" requires an application to obtain due to copyright protection. The project uses the cc-by-nc-sa-4.0 license.

Read more