NVIDIA releases 2B automotive reasoning VLA model Alpamayo 2 Edge
Understand this story
NVIDIA releases Alpamayo 2 Edge, a 2B-parameter-class reasoning vision-language-action (VLA) model for autonomous driving that can plan trajectories and provide explanations in the vehicle, targeting DRIVE AGX Thor hardware. NVIDIA previously said that large models provide the foundation for such capabilities, but to truly bring models into vehicles, it is still necessary to shrink model size, retain effective reasoning ability, and validate performance on target hardware. On usage conditions, the model weights use the OpenMDW-1.1 license and the code uses Apache 2.0; real-time operation on Thor requires the recipes from the Alpamayo Developer Hub; on workstation GPUs, at least 16 GB of VRAM is recommended; other GPU architectures have not yet been validated.
AI overview based on source reports · Updated 23 min ago
Developments
2 developments
- Nov 23 02:37 · 1 reportsNVIDIA releases automotive VLA model Alpamayo 2 EdgeMarkTechPost:NVIDIA Releases Alpamayo 2 Edge: A 2B Reasoning VLA for DRIVE Thor
- Nov 11 00:00 · 1 reportsNVIDIA describes running AV reasoning models within in-vehicle compute and latency constraintsNVIDIA Technical Blog:Running AV Reasoning Models Within Onboard Compute and Latency Limits
Coverage timeline
Follow the coverage to explore different perspectives.
- MarkTechPostNVIDIA Releases Alpamayo 2 Edge: A 2B Reasoning VLA for DRIVE Thor
NVIDIA released Alpamayo 2 Edge, a 2B-class reasoning vision-language-action model that turns surround-camera video into 6.4-second trajectory plans and runs on DRIVE AGX Thor with an FP8 plus TensorRT Edge-LLM recipe at 160 to 190 ms pixel-to-trajectory latency. It uses adaptive reasoning, emitting <NOMINAL> tokens instead of full Chain-of-Causation traces during routine driving, and scored 15.15 km per at-fault incident in AlpaSim versus 0.25 km for Qwen-Drive-1.0 4B, while its LingoQA score of 75.2 trails Alpamayo 2 Super (79.2). Weights use the OpenMDW-1.1 license and code uses Apache 2.0, with BF16 PyTorch validated only on an H100 80GB.
- NVIDIA Technical BlogRunning AV Reasoning Models Within Onboard Compute and Latency Limits
AV developers need reasoning models that fit onboard compute and latency constraints; larger models provide a foundation, but deploying them in a vehicle requires reducing model size, preserving useful reasoning, and validating performance on target hardware. The source mentions NVIDIA Alpamayo 2 Edge in this context, but the supplied material is truncated before detailing it.
Activity for this story
Not enough continuous observations to show a trend yet.