Tracked topic
… Looking ahead, the NVIDIA Rubin platform delivers large leaps in NVFP4 capability for training and inference, offering 35 petaFLOPS of NVFP4 training compute, and 50 petaFLOPs of NVFP4 Transformer Engine inference compute. …
… Im Zentrum steht NVFP4, also ein 4-Bit-Gleitkommaformat, das in diesem System für Training, Inferenz und KV-Cache eine zentrale Rolle spielt. Der aktuelle GitHub-Stand nennt für den 25. Mai 2026 eine Optimierung des NVFP4-Inferenzpfads. …
… Selective BF16 layers are essential for NVFP4: Ablation studies show that fully NVFP4 models diverge. Stable training requires keeping some layers in BF16, particularly near the end of the network, to mitigate NVFP4 quantization error. …
… To enable it, set the quantization flag in MaxText to switch into the NVFP4 path. Two modes are exposed: quantization=te nvfp4 : NVFP4 with Random Hadamard Transform. Recommended when the convergence under te nvfp4 no rht is not satisfactory. quantization=te nvfp4 no rht : NVFP4 without RHT. …
Safetensors, llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved: https://huggingface.co/llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved GGUFs, llmfan46/Qwen3.5-35B-A3B-uncensored-here…
Build is done. 16 DGX Sparks on the fabric, all hitting line rate. Setup was time consuming but honestly smoother than I expected. Each Spark runs Nvidia’s flavor of Ubuntu out of the box with mostly everything pre insta…
Added a 16x Spark Cluster to my homelab over the last few days. Curious if this is the largest Spark cluster anyone has built. About 2 years ago I had renovated my basement and built a personal lab/datacenter into my off…
Added a 16x Spark Cluster to my homelab over the last few days. Curious if this is the largest Spark cluster anyone has built. About 2 years ago I had renovated my basement and built a personal lab/datacenter into my off…
… The second part enables native NVFP4 support in ComfyUI Flux.1, Flux.2, and Quen Image. …
… At CES in January, NVIDIA announced several models released with NVFP4 and FP8 support. And now more NVFP4 and FP8 models are available — LTX-2.3, with NVFP4 support coming soon, FLUX.2 Klein 4B , and FLUX.2 Klein 9B — directly in ComfyUI. …
… BF16 and NVFP4 The model is available today on Hugging Face with BF16 checkpoints, and an NVFP4 quantized checkpoint for DiffusionGemma is also available using NVIDIA Model Optimizer . …
… Performance Unlocked: NVFP4 Accelerates LLM Training Key to the outstanding results this round was performing calculations using NVFP4 precision — a first in the history of MLPerf Training. …
DeepSeek V4 is out , bringing major optimizations, including up to 1.6T model sizes, and NVIDIA is ready with Day-0 support on Blackwell GPUs using NVFP4. …
… The chip itself is designed purely for AI-intensive workloads, offering 50 FLOPs of NVFP4 Inference, 35 PFLOPs of NVFP4 Training performance, a 5x and 3.5x increase over Blackwell, respectively. …