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	1350	January 9, 2023
SubgraphMatcher can't be initialized with MOE pattern FX (Functional Transformations)	0	27	May 7, 2026
Understanding torch.compile() torch.compile	2	64	April 20, 2026
Variable sized outputs in custom operator? torch.compile	1	26	April 15, 2026
JITed nn.TransformerDecoderLayer runs significantly slower than in eager mode torch.compile	0	27	March 20, 2026
Inductor codegen debugging torch._inductor	1	190	March 18, 2026
Mark_dynamic does not work when the dimension is 1 torch.compile	1	73	March 18, 2026
Flex_attention: BlockMask conversion? torch.compile	1	72	February 25, 2026
Nested torch.compile-d function calls with different options / CUDA graph options torch.compile	0	34	February 16, 2026
Flex_attention and torch.compile torch.compile	1	85	February 12, 2026
ONNX Community Survey 2026 torch.compile	0	43	February 11, 2026
"Offset increment outside graph capture encountered unexpectedly" is a specific PyTorch Windows issue that's known to occur with CUDA Graphs. This is a bug in PyTorch Windows builds FX (Functional Transformations)	0	111	February 10, 2026
torch.compile(fullgraph=True) fails on nn.utils.rnn.pad_sequence with padding_side, while fullgraph=False succeeds torch.compile	0	25	February 5, 2026
Error: accessing tensor output of CUDAGraphs that has been overwritten by a subsequent run torch.compile	1	807	February 5, 2026
Investigating toch custom ops using torch.compile torch.compile	2	77	February 1, 2026
Optimizing a TTS Model torch.compile	0	55	January 30, 2026
Torch.compile on train_step incl both fwd and bwd torch.compile	1	48	January 21, 2026
AOTInductor on Windows torch.compile	0	103	January 19, 2026
Torch.compile Diagnostic Dashboard torch.compile	1	56	January 18, 2026
Compiling a method other than forward torch.compile	3	379	January 14, 2026
Torch.compile - getting `nans` with the latest version of torch torch.compile	3	237	January 6, 2026
Torch.compile results in loss nan while it converges without torch.compile on mac cpu and mps torch.compile	0	31	January 6, 2026
How to verify that default_decompositions successfully reduce operators to the Core ATen IR set? FX (Functional Transformations)	0	25	December 30, 2025
NaNs with torch.compile (inductor, mode="reduce-overhead"), OOM with backend="cudagraphs" torch._inductor	0	75	December 23, 2025
How to save torch.compile so we don't need to re-compile torch.compile	2	460	December 23, 2025
How to load state_dict if saved as torch.save(model.state_dict(), fs) torch.compile	12	199	December 21, 2025
Issue when using aot_module with torch.autograd torch.compile	0	31	December 19, 2025
Let HOF (torch.cond, torch.while_loop) skip compilation of input functions if needed torch._dynamo	0	28	December 17, 2025
Autotune cache key for matrix multiplication torch._inductor	0	29	December 11, 2025
Torch compile cache sharing across different GPUs torch.compile	3	230	December 10, 2025