inference/research
verifiedhigh · 0.90updated 2026-10-04

Neuron Graph Compiler (neuronx-cc)

spec-sheet

Fields

id
sources/alt-neuron-compiler
Publisher
Amazon Web Services
Kind
spec-sheet
Published
2026-08-17
Accessed
2026-10-04
notes
neuronx-cc (XLA-based, NeuronCores v2) and neuron-cc (TVM-based, v1). Auto-casts FP32 to BF16. Optimization levels 1-3. NEFF output. Mixed precision: FP32/TF32/FP16/BF16/FP8. Model-specific optimizations (Transformer, U-Net). Logical NeuronCore sharding.

Notes

neuronx-cc (XLA-based, NeuronCores v2) and neuron-cc (TVM-based, v1). Auto-casts FP32 to BF16. Optimization levels 1-3. NEFF output. Mixed precision: FP32/TF32/FP16/BF16/FP8. Model-specific optimizations (Transformer, U-Net). Logical NeuronCore sharding.

Cross-references

This record references no other records.