Search Results for "qa"

...By combining text and structured image representations, it aims to facilitate tasks where both descriptive and structural understanding are important, such as detailed image QA, interactive image editing via prompt layers, and image-conditioned generation with structural control. The layered approach supports training signals that help the model learn how visual elements relate to each other and to textual context, rather than simply learning global image embeddings.

...It is built to operate end-to-end: planning a research strategy, gathering sources, reasoning over conflicting claims, and writing a grounded response. The repository includes evaluation results on multi-step QA and research benchmarks, illustrating how web-time context boosts accuracy. Because the system is modular, you can swap the search component, reader, or policy to fit private deployments or different data domains. It’s aimed at developers who want a transparent, hackable research agent they can run locally or wire into existing workflows.

...Trained on a large bilingual dataset — including 30 million high-quality Chinese image-text pairs from CogView and 300 million English pairs — VisualGLM-6B is designed for image understanding, description, and question answering. Fine-tuning on long visual QA datasets further aligns the model’s responses with human preferences. The repository provides inference APIs, command-line demos, web demos, and efficient fine-tuning options like LoRA, QLoRA, and P-tuning. It also supports quantization down to INT4, enabling local deployment on consumer GPUs with as little as 6.3 GB VRAM.

Vidi is a family of large multimodal models developed for deep video understanding and editing tasks, integrating vision, audio, and language to allow sophisticated querying and manipulation of video content. It’s designed to process long-form, real-world videos and answer complex queries such as “when in this clip does X happen?” or “where in the frame is object Y during that moment?” — offering temporal retrieval, spatio-temporal grounding (i.e. locating objects over time + space), and...