Tracked topic
… Two practical questions: Is nvfp4 RTF and vram on a 24gb consumer card. fp8 sits ~30gb, bf16 ~60, so nvfp4 is the only realistic path on a 4090. is megatron-bridge the inference route you'd point at, or NeMo-RL? …
… Nemotron 3 Ultra is our most capable model yet, employing multiple key technologies - LatentMoE , Multi Token Prediction MTP , NVFP4 pre-training, multi-environment RLVR , MOPD, and reasoning budget control . …
… Driven by these designs, SANA-WMdemonstrates remarkable efficiency across data, training compute, and inference hardware: it uses only sim213K public video clips with metric-scale pose supervision , completes training in 15 days on 64 H100s, and generates each 60s clip on a single GPU; its distille… …
… The following papers were recommended by the Semantic Scholar API Wan-Streamer v0.1: End-to-end Real-time Interactive Foundation Models 2026 StreamChar: Long-Horizon Streaming Character Audio-Video Generation with Decoupled Orchestration 2026 StreamingEffect: Real-Time Human-Centric Video Effect Ge… …
… As far as I can tell, NVIDIA’s current stack is pretty much all-in on NVFP4, which is also reflected in the recent Transformer Engine updates. …
… The following papers were recommended by the Semantic Scholar API StreamingEffect: Real-Time Human-Centric Video Effect Generation 2026 LongLive-2.0: An NVFP4 Parallel Infrastructure for Long Video Generation 2026 SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transform… …
… The following papers were recommended by the Semantic Scholar API DEMON: Diffusion Engine for Musical Orchestrated Noise 2026 A 35B Hybrid-Attention Mixture-of-Experts Model on a 6GB 2011 GPU: Hand-Written 4-bit CUDA Inference for Fermi 2026 FLASH-MAXSIM: IO-Aware Fused Kernels for Late-Interaction… …