torch.compile

torch._dynamo PyTorch’s compiler frontend which can output subgraphs when programs aren’t traceable. FX (Functional Transformations) FX is a toolkit for developers to use to transform nn.Module instances. FX consists of three main components: a symbolic tracer, an intermediate representation , and Python code generation . torch._inductor The default compiler in torch.compile, builds loops schedules and codegens triton or C++ kernels.

Topic	Replies	Views	Activity
About the torch.compile category torch.compile	0	1193	January 9, 2023
Torch compile: optimizer.step Generates Excessive Warning Messages torch.compile	10	278	May 8, 2025
Where the grades from backward graph will be copied to param.grade in cuda graphs torch.compile	0	15	May 7, 2025
Recompilations leading to timeout torch.compile	0	11	May 1, 2025
torch._dynamo.exc.Unsupported: call_method NNModuleVariable() _sa_block [TensorVariable(), LazyVariableTracker(), LazyVariableTracker()] {} torch._dynamo	1	445	April 29, 2025
How to use triton kernels with pytorch? torch.compile	1	29	April 29, 2025
What is ComputedBuffer? torch._inductor	1	204	April 24, 2025
Compiled matmul is slower than vanilla matmul torch.compile	1	25	April 24, 2025
FP8 `torch.empty` doesn't work under `inductor` of pytorch 2.4.1 torch.compile	2	17	April 24, 2025
Device "meta" and device "cuda:0" error torch.compile	0	22	April 23, 2025
Resetting cache in benchmark torch._dynamo	4	1448	April 23, 2025
Torch.compile cpp backend not fusing OuterLoopFusedSchedulerNode properly torch._inductor	2	13	April 22, 2025
Higher Peak Memory on First Run with torch.compile and Inductor Backend torch.compile	0	17	April 21, 2025
How to prevent dispatching `torch._C._nn.upsample_bilinear2d` to vectorize implementation `torch.ops.aten.upsample_bilinear2d.vec` torch.compile	0	11	April 21, 2025
Why adaptive_avg_pool1d became adaptive_avg_pool2d in make_fx? FX (Functional Transformations)	1	8	April 21, 2025
C++/Cuda / aot.compile and cuda graph torch.compile	1	476	April 18, 2025
How does torch dynamo set custom eval frame callback? torch._dynamo	1	18	April 16, 2025
TorchRuntimeError: Failed running call_module L__self___position_embedding_table((FakeTensor(..., device='mps', ...), *{}): Unhandled FakeTensor Device Propagation for aten.index_select.default, found two different devices mps:0, mps torch._inductor	0	12	April 14, 2025
Torch.compile emits double backward error even when the corresponding function is disabled torch.compile	5	50	April 14, 2025
Is the use of CUDA Graph Python API discouraged? torch.compile	2	36	April 13, 2025
Inductor codegen debugging torch._inductor	0	21	April 12, 2025
When will re-autotune mm kernel torch.compile	0	31	April 9, 2025
Compiling variable-size cross entropy torch.compile	0	17	April 7, 2025
How does pytorch triton work with dynamic shapes? torch._inductor	0	36	March 31, 2025
Torch.compile() causes RuntimeError: element 0 of tensors does not require torch.compile	1	18	March 31, 2025
Error: accessing tensor output of CUDAGraphs that has been overwritten by a subsequent run torch.compile	0	116	March 30, 2025
Need help debugging torch.compile issue torch._inductor	0	47	March 26, 2025
Dynamic slicing torch.compile	5	84	March 24, 2025
Best way to detect and fix graphs breaks torch.compile	5	132	March 24, 2025
Is it possible for torch SDPA to be slower than manual attention? torch.compile	5	62	March 24, 2025