thunder.plugins

DDP([bucket_size_in_mb, broadcast_from, ...])

Plugin for enabling Distributed Data Parallel (DDP) training in Thunder.

FSDP([device, broadcast_from, ...])

Plugin for enabling Fully Sharded Data Parallel (FSDP) training in Thunder.

QuantizeInt4()

Plugin for 4-bit integer quantization using BitsAndBytes.

FP8()

Plugin for enabling FP8 precision via NVIDIA Transformer Engine, enabling higher throughput of matrix operations in FP8.

ReduceOverhead()

Plugin to enable CUDA Graphs and reduce CPU overhead.