TensorFlow Adaptation
Adaptation to Atlas AI Processors
Supports native TensorFlow Python APIs and extended Python APIs.
Core Framework Functions
Supports TensorFlow graph execution, optimization, and more.
Custom Operator Development
Allows adding custom operators within the TensorFlow framework.
Distributed Training
Supports parallel training of distributed data, including single-server multi-device and multi-server multi-device scenarios.
Mixed Precision
Supports the use of FP16 and FP32 together to improve performance while retaining accuracy.
Model Inference
Outputs standard TensorFlow PB models and converts them into offline inference models using ATC.
Development Scenarios
Development Resources
TensorFlow Model Library
Mainstream TensorFlow models pre-trained on Atlas AI Processors to allow you to deploy and train models quickly