Jinming Wuhttps://kim1ng.github.io/Recent content on Jinming WuHugoen-USkiming1206@gmail.com (Jinming Wu)kiming1206@gmail.com (Jinming Wu)Wed, 25 Jun 2025 00:00:00 +0000MMSearch-R1: Incentivizing LMMs to Searchhttps://kim1ng.github.io/projects/mmsearch_r1/Wed, 25 Jun 2025 00:00:00 +0000kiming1206@gmail.com (Jinming Wu)https://kim1ng.github.io/projects/mmsearch_r1/MMSearch-R1 is an e2e RL framework that enables LMMs to perform on-demand, multi-turn search with real-world multimodal search tools. MMSearch-R1-7B model outperforms same-size traditional RAG baselines and cuts search calls by over 30%.LLaVA-Video: Video Instruction Tuning With Synthetic Datahttps://kim1ng.github.io/projects/llava_video/Thu, 03 Oct 2024 00:00:00 +0000kiming1206@gmail.com (Jinming Wu)https://kim1ng.github.io/projects/llava_video/LLaVA-Video series models are Video-LLMs fully trained on our high-quality synthetic dataset, LLaVA-Video-178K, which comprises 178K video captions and 1.15M video QAs. Our models demonstrate strong performance across 10+ video understanding tasks.MDR: Model-Specific Demonstration Retrieval at Inference Time for In-Context Learninghttps://kim1ng.github.io/projects/mdr/Tue, 04 Jun 2024 00:00:00 +0000kiming1206@gmail.com (Jinming Wu)https://kim1ng.github.io/projects/mdr/MDR proposed a simple and effective metric to measure the preference of different LLMs for demonstrations and leveraged this metric to improve demonstration retrieval frameworks at inference stage.