Compiling a Transformer using torch.compile and TensorRT

This interactive script is intended as a sample of the Torch-TensorRT workflow with torch.compile on a transformer-based model.

Imports and Model Definition

import torch
import torch_tensorrt
from transformers import BertModel

# Initialize model with float precision and sample inputs
model = BertModel.from_pretrained("bert-base-uncased").eval().to("cuda")
inputs = [
    torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"),
    torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"),
]

Optional Input Arguments to torch_tensorrt.compile

# Enabled precision for TensorRT optimization
enabled_precisions = {torch.float}

# Whether to print verbose logs
debug = True

# Workspace size for TensorRT
workspace_size = 20 << 30

# Maximum number of TRT Engines
# (Lower value allows more graph segmentation)
min_block_size = 7

# Operations to Run in Torch, regardless of converter support
torch_executed_ops = {}

Compilation with torch.compile

# Define backend compilation keyword arguments
compilation_kwargs = {
    "enabled_precisions": enabled_precisions,
    "debug": debug,
    "workspace_size": workspace_size,
    "min_block_size": min_block_size,
    "torch_executed_ops": torch_executed_ops,
}

# Build and compile the model with torch.compile, using Torch-TensorRT backend
optimized_model = torch.compile(
    model,
    backend="torch_tensorrt",
    options=compilation_kwargs,
)
optimized_model(*inputs)

Equivalently, we could have run the above via the convenience frontend, as so: torch_tensorrt.compile(model, ir=”torch_compile”, inputs=inputs, **compilation_kwargs)

Inference

# Does not cause recompilation (same batch size as input)
new_inputs = [
    torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"),
    torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"),
]
new_outputs = optimized_model(*new_inputs)

# Does cause recompilation (new batch size)
new_inputs = [
    torch.randint(0, 2, (4, 14), dtype=torch.int32).to("cuda"),
    torch.randint(0, 2, (4, 14), dtype=torch.int32).to("cuda"),
]
new_outputs = optimized_model(*new_inputs)

Cleanup

# Finally, we use Torch utilities to clean up the workspace
torch._dynamo.reset()

Cuda Driver Error Note

Occasionally, upon exiting the Python runtime after Dynamo compilation with torch_tensorrt, one may encounter a Cuda Driver Error. This issue is related to https://github.com/NVIDIA/TensorRT/issues/2052 and can be resolved by wrapping the compilation/inference in a function and using a scoped call, as in:

if __name__ == '__main__':
    compile_engine_and_infer()