pytorch/torchdynamo
A Python-level JIT compiler designed to make unmodified PyTorch programs faster.
PythonBSD-3-Clause
Issues
- 1
- 3
LoweringException: IndexError: list index out of range
#2054 opened by soumickmj - 2
dead links in README and TROUBLESHOOTING.
#2051 opened by kevint324 - 0
Add support for tagging parameters in export
#2050 opened by voznesenskym - 1
- 1
[Inductor] [CPU] Accuracy failure in huggingface model XLNetLMHeadModel on WW03.2
#2036 opened by ESI-SYD - 1
- 1
- 4
- 1
- 1
- 5
[Inductor] [CPU] Crash failure in timm models (mkldnn._convolution_pointwise.binary)
#1975 opened by ESI-SYD - 2
- 2
[Inductor] [CPU] Crash failure in timm_model dla102
#1945 opened by ESI-SYD - 4
Minifier fails for wav2vec
#1959 opened by wconstab - 3
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
#1962 opened by yanboliang - 2
PT2.0 max-autotune mode fails for resnet34
#2022 opened by Puneet2000 - 0
Dynamo "You can suppress this exception and fall back to eager" garbles KeyError
#1953 opened by ezyang - 20
- 1
torch._inductor.lowering: [WARNING] using triton random, expect difference from eager
#2042 opened by vgoklani - 3
[dynamo] CUDA graphs not working on recent PyTorch nighties, crash during warmup because of FakeTensors
#1987 opened by pommedeterresautee - 2
3% Memory Regression when Compiling the Optimizer
#2033 opened by eellison - 3
How to use the new autotune intoruced in https://github.com/pytorch/torchdynamo/pull/1338
#2023 opened by ayoub-louati - 0
Minifier doesn't work with torch.compile
#1964 opened by williamwen42 - 3
wrap_fx_proxy_cls doesn't cover torch.cuda.Stream
#2021 opened by kehuanfeng - 6
MMCLS Resnet-50 failed training, got 'Triton Error [CUDA]: invalid device context'
#2026 opened by C1rN09 - 2
Fix support **kwargs for torch._dynamo.export
#1997 opened by thiagocrepaldi - 0
TMPDIR ignored
#2004 opened by timlacroix - 1
crash when enabled triton matmul
#2015 opened by stephen-youn - 3
Unable to reporduce Hugging Face model example shown for PyTorch 2.0
#2019 opened by aidan-jackson-data - 1
[CPP] Support Meta tensor for conv/linear fused ops
#1982 opened by jgong5 - 2
how to get the path of the generated triton codes (triton code and its generated IRs)?
#2007 opened by stephen-youn - 6
compiled model run in v100 GPU is slower
#2008 opened by stephen-youn - 3
Fake Tensor Fails for codegen model
#2012 opened by PoodleWang - 1
- 0
- 3
Delete dynamic_propagation config
#1949 opened by ezyang - 3
HuggingFace models on CI are quite flaky
#1985 opened by desertfire - 0
- 0
Correctness regression: dlrm
#1984 opened by desertfire - 5
Direct calls to logging are suppressed inside benchmarks/dynamo (and possibly other situations)
#1981 opened by ezyang - 1
Dynamo does a lot of unnecessary input cloning for real_value_cache, even if cache is not used
#1950 opened by ezyang - 0
Preserve dynamo/inductor config during minifier
#1965 opened by williamwen42 - 2
- 1
- 2
Troubleshooting guide is hard to find
#1971 opened by wconstab - 4
- 2
HF Models Triggering AssertError in Functorch
#1937 opened by williamwen42 - 0
Inference mode dupe specializes without guards
#1939 opened by ezyang - 3
Inductor segfault when compiling Adam
#1942 opened by mlazos