NVIDIA Nemotron 3.5 Super VL: Open Weights, Data and Training Recipes
Releasing data and recipes alongside weights lets practitioners reproduce or fine-tune the full training pipeline rather than just the checkpoint, which is rare even among open model releases. Builders running multimodal workloads can compare it against other open vision-language models and adapt it for their own inference setups.
NVIDIA detailed how it built Nemotron 3.5 Super VL, covering its post-training approach, multi-token prediction and inference optimizations. The model is released as open and customizable, with weights, training data and recipes included. It is available now on Hugging Face under the nvidia account.
How we built Nemotron 3.5 Super VL, including its post-training approach, multi-token prediction, and inference optimizations: nvda.ws/3Wqw7i6 As always, NVIDIA Nemotron 3.5 Super VL is open and customizable. This includes weights, data and recipes. Available now on @huggingface → huggingface.co/nvidia/NVIDIA-… 💬 0 🔄 3 ❤️ 17 👀 1604 📊 3 ⚡
