aot_precompile() — pytorch Function Reference
Architecture documentation for the aot_precompile() function in common.py from the pytorch codebase.
Entity Profile
Dependency Diagram
graph TD 5130216c_70e0_8051_b662_1aed31c12627["aot_precompile()"] 04a3a4a6_8db3_854d_a893_02c9542bf9dd["speedup_experiment()"] 04a3a4a6_8db3_854d_a893_02c9542bf9dd -->|calls| 5130216c_70e0_8051_b662_1aed31c12627 1eff8423_1f23_d138_6815_07d8dc29a749["_normalize_bench_inputs()"] 5130216c_70e0_8051_b662_1aed31c12627 -->|calls| 1eff8423_1f23_d138_6815_07d8dc29a749 style 5130216c_70e0_8051_b662_1aed31c12627 fill:#6366f1,stroke:#818cf8,color:#fff
Relationship Graph
Source Code
benchmarks/dynamo/common.py lines 1517–1545
def aot_precompile(model, example_inputs):
example_args, example_kwargs = _normalize_bench_inputs(example_inputs)
with tempfile.NamedTemporaryFile(suffix=".pt", delete=False) as f:
save_path = f.name
with fresh_cache(), torch._dynamo.config.patch("enable_aot_compile", True):
compiled_fn = torch.compile(
model,
fullgraph=True,
options={"guard_filter_fn": lambda guards: [False for _ in guards]},
).forward.aot_compile((example_args, example_kwargs))
compiled_fn.save_compiled_function(save_path)
torch._dynamo.reset()
with open(save_path, "rb") as f:
load_start_time = time.perf_counter()
loaded_fn = torch.compiler.load_compiled_function(f)
load_end_time = time.perf_counter()
print(
f"AOT Precompile loading time: {load_end_time - load_start_time} seconds"
)
def opt_aot_precompile(_, example_inputs, collect_outputs=False):
example_args, example_kwargs = _normalize_bench_inputs(example_inputs)
return loaded_fn(model, *example_args, **example_kwargs)
return opt_aot_precompile
Domain
Subdomains
Called By
Source
Frequently Asked Questions
What does aot_precompile() do?
aot_precompile() is a function in the pytorch codebase.
What does aot_precompile() call?
aot_precompile() calls 1 function(s): _normalize_bench_inputs.
What calls aot_precompile()?
aot_precompile() is called by 1 function(s): speedup_experiment.
Analyze Your Own Codebase
Get architecture documentation, dependency graphs, and domain analysis for your codebase in minutes.
Try Supermodel Free