Lightning-AI lightning-thunder issues

Lightning-AI / lightning-thunder

Make PyTorch models up to 40% faster! Thunder is a source to source compiler for PyTorch. It enables using different hardware executors at once; across one or thousands of GPUs.

Apache License 2.0

1.06k stars 59 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

"nvFuser" illegal memory access with falcon-7b model

#659 tfogal opened 1 minute ago
0
LayerNorm support

#658 tfogal opened 1 hour ago
0
Desired inplace idioms (add yours here)

#657 t-vi opened 4 hours ago
0
CUDAGraphs as executor/transform/fusion pass

#656 nikitaved opened 6 hours ago
0
When comparing Thunder Torch Executor to Torch Eager, the ResNet18 gradients are not close for FP32.

#655 kiya00 opened 7 hours ago
0
try to fix up pyproject for twine

#654 t-vi opened 12 hours ago
0
[in-place][functionalization] Correct `.from_bsym_swap_proxies` on `new_bsym`, not `bsym`

#653 crcrpar closed 8 hours ago
0
functionalization fails to replace in-place operands with new tensors

#652 crcrpar closed 8 hours ago
0
Bumps cudnn FE to v1.5.2

#651 vedaanta opened 16 hours ago
3
[DUMMY][NO MERGE] commit to build new contianers

#650 vedaanta closed 16 hours ago
0
skip quant test using bnb if bitsandbytes is not available

#649 t-vi closed 22 hours ago
0
ThunderModule - fix load_original_state_dict to work with modules with buffers

#648 kshitij12345 closed 1 day ago
0
ThunderModule - `state_dict` followed by `load_original_state_dict` returns `state_dict` of the original module.

#647 kshitij12345 opened 1 day ago
1
Updating Module's buffer leads to crash in thunder.jit

#646 kshitij12345 opened 1 day ago
0
Test for resnet18 accuracy

#645 kiya00 closed 6 hours ago
1
Add `is_complex` op

#644 riccardofelluga closed 1 day ago
0
Indexing with slice and list

#643 t-vi opened 1 day ago
1
Check number of proxy args before indexing

#642 crcrpar closed 1 day ago
0
Turn `tf32` off for resnet18 test cases

#641 crcrpar closed 2 days ago
0
Transforms as EarlyTransforms

#640 t-vi opened 2 days ago
1
reenable cudnn sdpa

#639 t-vi closed 2 days ago
0
Detailed `__repr__`

#638 k223kim opened 2 days ago
0
pass results through epilogue if present

#637 t-vi opened 3 days ago
1
always return epilogue inputs from prologue

#636 t-vi closed 4 days ago
0
Make CudaGraph wrapping a transform (but call it explicitly)

#635 t-vi opened 5 days ago
1
Investigate numerical error of fp32 resnet18 of `train=True`

#634 crcrpar closed 2 days ago
0
Fix in-place to views condition

#633 crcrpar closed 5 days ago
1
jit: support dataclass as an output for jitted function

#632 kshitij12345 closed 1 day ago
3
Remove robieta from CODEOWNERS

#631 robieta closed 5 days ago
0
Disallow in-place to view tensors

#630 crcrpar closed 5 days ago
0
normalize test fix

#629 k223kim closed 5 days ago
0
Make epilogue bits more uniform and useful

#628 t-vi opened 6 days ago
0
Fix issue template

#627 t-vi closed 6 days ago
0
recursive error fix

#626 k223kim closed 2 days ago
3
cuDNN executor: No valid engine configs for MUL_Reduction_MUL_Matmul_MUL_ADD_SUB_EXP_Reshape_Matmul_Matmul_MUL_SUB_MUL_Reduction_MUL_Reshape_Matmul_Reshape_Matmul_

#625 t-vi closed 6 days ago
2
[inplace][distributed] support `torch.distributed` common ops with `async_op=False`

#624 crcrpar opened 1 week ago
0
support returning dataclasses with tensors

#623 t-vi closed 1 day ago
0
Handles non-contiguous input strides

#622 vedaanta opened 1 week ago
2
use `torch.get_default_dtype` and `torch.get_default_device` for factory method in `thunder/torch/__init__.py`

#621 jjsjann123 opened 1 week ago
2
Update error messages.

#620 tfogal closed 1 week ago
0
`saved_tensors_list` might contain `None`s

#619 nikitaved opened 1 week ago
0
[feature request] Numpy 2 compatibility

#618 nikitaved opened 1 week ago
1
fix numpy version for tests

#617 nikitaved closed 1 week ago
0
test: no Nones in saved_tensors_list

#616 nikitaved opened 1 week ago
1
Match torch signature for `torch.nn.functional.silu`

#615 riccardofelluga closed 1 week ago
0
[Feature request] Optional debugging option to get trace with information on tensor strides along with tensor shapes

#614 parthmannan opened 1 week ago
3
Allowing static constraint in torch/__init__.py

#613 jjsjann123 closed 2 days ago
2
[pre-commit.ci] pre-commit suggestions

#612 pre-commit-ci[bot] closed 1 week ago
1
`numpy` version requirement fix

#611 nikitaved closed 1 week ago
1
numpy 2.0.0 dtypes changes

#610 crcrpar closed 1 week ago
2