Open Models Almanac
SourcesNVIDIA

publisher · 19 sources

NVIDIA

19 sources read from this publisher, newest first, with the methods each one describes.

nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 · Hugging Face

first party release · model card · 2026-10-23 · 24 methods

nvidia/nemotron-3.5-super-vl-120b-a12b

first party release · vendor docs · 2026-10-17 · 29 methods

NVIDIA Nemotron 3.5 Super and Lightning

first party release · vendor docs · 2026-10-14 · 30 methods

nemotron-3.5-lightning-30b-a3b Model by NVIDIA | NVIDIA NIM

first party release · vendor docs · 2026-09-02 · 21 methods

nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 · Hugging Face

first party release · model card · 2026-08-14 · 11 methods

nvidia/nemotron-3-ultra-550b-a55b

first party release · vendor docs · 2026-07-23 · 77 methods

nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · Hugging Face

first party release · model card · 2026-06-24 · 22 methods

NVIDIA Isaac GR00T N1.7: Open Reasoning VLA Model for Humanoid Robots

first party release · official blog · 2026-04-17 · 6 methods

NVIDIA Nemotron 3: Efficient and Open Intelligence

first party release · technical report · 2025-12-24 · 19 methods

Reinforcement learningGroup Relative Policy OptimizationFlow matchingKnowledge insulationAutoregressive discrete-action predictionSeparate flow-matching action expertCausally grounded, decision-linked reasoning tracesChain of Causation (CoC) labeling frameworkDisagreement-based curation of RL post-training dataReasoning-action consistency rewardRule-based scenario clip selection and balancingSupervised fine-tuning on Chain of Causation dataChain-of-Causation reasoningClosed-set decision labels aligned with ego trajectoriesConditional flow matching objective for the action expertContinuous trajectory decoding with unicycle dynamics and flow matchingDecision-grounded keyframe selectionDiscrete action-token representationEnd-to-end autonomous drivingEuler integration for flow-matching trajectory denoisingFlex multi-camera vision tokenizationFlow-matching diffusion action expert for trajectory decodingGaussian conditional optimal-transport path for flow matchingHuman-verified and LLM-based evaluationHybrid human and automatic labelingHybrid human-in-the-loop and automated labelingHybrid reasoning-trace labeling with automated labeling and human reviewIndependent annotator quality checksJoint cross-entropy training of reasoning and trajectory tokensJoint discrete trajectory-token and continuous flow-matching action-expert representationKeyframe selection for causal reasoning annotationKeyframe-range annotation for proactive scenariosKL regularization to a reference policyLarge reasoning model as evaluatorMulti-stage training strategyPhysical-AI-focused pretrainingPre-action keyframe placement using a short temporal bufferReasoning reward from LRM feedbackRL-based post-training framework with complementary rewardsRule-based matching of inferred intent against trajectory-derived meta-actions

NVIDIA Nemotron AI Models

first party release · vendor docs · 2025-09-15 · 12 methods

GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

research paper · technical report · 2025-03-18 · 44 methods

Low-Rank AdaptationFlow matchingAction chunkingHeterogeneous-data co-trainingCross-attentionEmbodiment-specific state and action projectorsHeterogeneous training mixtureInverse-dynamics-model pseudo-action labelingMultimodal LLM instruction-adherence judging1:1 co-training of real-world and neural trajectoriesAuxiliary object-detection lossAxis-angle representation for end-effector rotation actionsCo-training on synthetic and real-world dataCo-training with generated neural trajectoriesDiffusion Transformer IDM using SigLIP-2 vision embeddingsDual-system compositional architectureExplicit modality configuration for state and action vectorsExplicit rotational-data representation specificationFiltering and re-captioning generated neural trajectoriesFine-tuning video generation models on robot teleoperation dataFlow-matching training objective for the IDMFreeze the language component of the vision-language backboneHeterogeneous data-mixture batch samplingHeterogeneous training data pyramidInstruction-adherence filtering of generated videosInverse-kinematics retargetingLatent-action representation learned from videoMiddle-layer VLM representations for policy conditioningMin-max normalization of robot states and actionsMultiple annotation types in a single datasetNeural trajectory generation using video generative modelsObject-centric subtask decomposition for simulation demonstrationsObject-position alignment for adapting demonstration segmentsOWL-v2 target-object bounding-box annotationRecaptioning videos that fail instruction-adherence reviewRepresent end-effector rotations in 6D for state normalizationSemantically partitioned state and action vectors with field metadataSimulation trajectory generation through demonstration transformation and replaySingle-embodiment post-training fine-tuningSquared-error loss for normalized object-center coordinates

nvidia/GR00T-N1.7-3B · Hugging Face

first party release · model card · 2025-03-18 · 20 methods

nvidia/Alpamayo-1.5-10B · Hugging Face

first party release · model card · 2010-05-01 · 9 methods

NVIDIA Nemotron 3 Ultra Technical Report

technical report · 121 methods

NVIDIA/Isaac-GR00T

first party release · code repo · 14 methods