-
Notifications
You must be signed in to change notification settings - Fork 80
Issues: Lightning-AI/lightning-thunder
Label tracking meta-issue (edit me to get automatically CC'ed...
#72
opened Mar 25, 2024 by
carmocca
Open
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Author
Label
Projects
Milestones
Assignee
Sort
Issues list
Make trace introspection available as methods or attribute accesses on functions returned by Thunder
enhancement
New feature or request
#1418
opened Nov 8, 2024 by
IvanYashchuk
dce
inside torch.autograd.Function
could fail as its traces miss unpacking bsyms
#1413
opened Nov 8, 2024 by
crcrpar
Extend catching empty autocast regions to catch other empty contexts
dynamo+thunder
for things that could be applicable to the dynamo+thunder frontend
#1412
opened Nov 8, 2024 by
IvanYashchuk
HF Qwen 2 with Thunder returns a slightly different loss function output
huggingface
For supporting HF models
#1407
opened Nov 7, 2024 by
IvanYashchuk
AssertionError from process_recorded_modifications
huggingface
For supporting HF models
jit
#1405
opened Nov 7, 2024 by
IvanYashchuk
Handle HF dynamic cache
huggingface
For supporting HF models
interpreter
program-coverage
Requests for model and program coverage
#1396
opened Nov 4, 2024 by
t-vi
ThunderFX: support graph-by-graph benchmarking for PyTorch native checkpointing
benchmarking
#1381
opened Oct 31, 2024 by
kiya00
Loss function backward output kept in memory longer than expected
bug
Something isn't working
memory use
#1379
opened Oct 31, 2024 by
riccardofelluga
Thunder saves too many tensors for backward for a Transformer's residual connection pattern
bug
Something isn't working
memory use
rematerialization
#1368
opened Oct 30, 2024 by
IvanYashchuk
ThunderFX is slower than torch.compile for vicuna-33b-v1.3, Platypus-30B and falcon-40b with FSDP & zero3.
mixology
Issues that the mixology team has surfaced
#1366
opened Oct 30, 2024 by
mpatel31415
Thunder and ThunderFX are slower than torch.compile for FP8 and falcon-7b and other models
#1365
opened Oct 30, 2024 by
mpatel31415
Thunder is slower than torch.compile for FP8 and Phi-3-mini-4k-instruct (by 18.6% for DDP and 7.9% for FSDP)
blocks NeMo
mixology
Issues that the mixology team has surfaced
#1364
opened Oct 30, 2024 by
mpatel31415
Add OpInfo for torch.Tensor.__setitem__ (to test prims.copy_with_setitem)
testing
#1357
opened Oct 28, 2024 by
kshitij12345
materialize frames / tracebacks less often
enhancement
New feature or request
interpreter
#1354
opened Oct 28, 2024 by
t-vi
test_transformer_engine_executor.py is not runnable in the PyTorch build without distributed support
ci / tests
#1350
opened Oct 28, 2024 by
IvanYashchuk
Empty autocast regions are preserved by Dynamo and given to compilers to optimize
dynamo+thunder
for things that could be applicable to the dynamo+thunder frontend
#1342
opened Oct 23, 2024 by
IvanYashchuk
Investigate the difference in speedup between Llama3-8b and Mistral-7B-v0.1
enhancement
New feature or request
performance
#1341
opened Oct 22, 2024 by
mpatel31415
Incorrect prologue for a while loop over a list of tensors
jit
#1335
opened Oct 21, 2024 by
IvanYashchuk
profiling tutorial
documentation
Improvements or additions to documentation
#1323
opened Oct 17, 2024 by
t-vi
Document Python+PyTorch version compatibility matrix in README
enhancement
New feature or request
#1310
opened Oct 15, 2024 by
mjmikulski
UX: adding an executor requires grabbing the default list
enhancement
New feature or request
#1309
opened Oct 15, 2024 by
mjmikulski
Previous Next
ProTip!
Exclude everything labeled
bug
with -label:bug.