Wei Ping

Building LLMs, with a focus on post-training and reasoning: • Led and made equal contributions to building Nemotron-Cascade 1 & 2, a family of general-purpose LLMs trained with cascade RL and multi-domain on-policy distillation. Nemotron-Cascade-2-30B-A3B is the second open-weight LLM, after DeepSeekV3.2-Speciale-671B-A37B, to achieve Gold Medal-level performance in the 2025 International Mathematical Olympiad (IMO), the International Olympiad in Informatics (IOI), and the ICPC World Finals, demonstrating remarkably high intelligence density with 20× fewer parameters. • Led AceReason effort. The 1.0 and 1.1 versions are the best-in-class 7B and 14B open LLMs for math and code reasoning at the time of release.

I led and contributed to LLM efforts in multimodality, embedding models, RAG, and long-context modeling: • Multimodal LLMs: I led and equally contributed to NVIDIA Vision Language Model (NVLM), a family of frontier multimodal LLMs. I also contributed to small/mid-sized audio/vision language models, including Audio Flamingo and VILA. • Embedding models: I led the development of NV-Embed-v1 and v2, the No. 1 ranking embedding models on the MTEB leaderboard, along with the multimodal extension, MM-Embed. • RAG and long-context: I led the development of ChatQA, ChatQA 1.5, ChatQA 2, and RankRAG which set new benchmarks for RAG and long-context generations at their respective releases. I also drove several exploratory studies, including Retrieval Meets Long-Context LLMs.

Developed and trained many generative models and LLMs: • DiffWave, BigVGAN. • In the realm of LLMs, my work focused on enhancing factuality and mitigating toxicity.

Led a team of talented researchers and engineers working on text-to-speech. • Awarded "Star Employee of the Quarter" in Q3 2018 at Baidu. • Received double promotion in one quarter. Selected (Co)-First-Author Work: • ClariNet (ICLR'19) • WaveFlow (ICML'20) • Non-autoregressive neural TTS (ICML'20)

Trained many deep neural networks at Baidu SVAIL (founded & led by Andrew Ng). Selected (Co)-First-Author Work: • Deep Voice 3 (ICLR'18), • Neural Voice Cloning with A Few Samples (NeurIPS'18) • Deep Voice 2 (NeurIPS'17)

• Conducted research in machine learning. • Developed large-scale video retrieval systems.

Research area: Machine Learning