Skip to content

Issues: Lightning-AI/lightning-thunder

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.

Sign up for GitHub

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

295 Open 349 Closed
295 Open 349 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Issues list

thunderfx produces results with incorrect requires_grad
#1733 opened Feb 1, 2025 by jjsjann123
Grad Transform generates inconsistent saved_for_backward between forward and backward trace.
#1732 opened Feb 1, 2025 by jjsjann123
dividing a float16 tensor by a python float is inaccurate with nvfuser nvfuser
#1724 opened Jan 30, 2025 by beverlylytle
1
div with nvfuser returns incorrect dtype nvfuser
#1723 opened Jan 30, 2025 by beverlylytle
1
@jjsjann123
4
cudnn SDPA : cudnn sdpa is not used for bigcode/starcoder2-7b cudnn
#1722 opened Jan 30, 2025 by kshitij12345
2
Add HF models PEFT benchmarks benchmarking nemo Issues needed to support NVIDIA NeMo models.
#1716 opened Jan 29, 2025 by riccardofelluga
4 tasks done
1
@riccardofelluga
Run LitGPT benchmarking with a custom Attention implementation priority. benchmarking
#1714 opened Jan 29, 2025 by wprazuch
1
@riccardofelluga
1
Input upcast is missing in Thunder's implementation of torch.nn.functional.rms_norm operators
#1713 opened Jan 29, 2025 by IvanYashchuk
3
symbolic cache policy can't handle string inputs properly. symbolic values
#1710 opened Jan 28, 2025 by jjsjann123
@jjsjann123
[Reporting Tool] Modular report classes for saving non-Thunder-specific repro scripts reporting thunderfx for things that could be applicable to the dynamo+thunder frontend
#1700 opened Jan 27, 2025 by kiya00
1
@kiya00
Implement max_norm argument for torch.nn.functional.embedding enhancement New feature or request in-place operators
#1699 opened Jan 27, 2025 by IvanYashchuk
check memory location of things tagged STATIC_MEMORY_LOCATION by default cudagraphs enhancement New feature or request
#1686 opened Jan 24, 2025 by t-vi
Transforming traces should always precede a domination check enhancement New feature or request
#1684 opened Jan 22, 2025 by ali-alshaar7
@ali-alshaar7
1
Investigate bf16 rms norm numerics numerical accuracy operators thunderfx for things that could be applicable to the dynamo+thunder frontend
#1678 opened Jan 22, 2025 by t-vi
@riccardofelluga
2
Connect prims.copy_with_setitem to nvFuser's Executor enhancement New feature or request nvfuser thunderfx for things that could be applicable to the dynamo+thunder frontend
#1676 opened Jan 22, 2025 by kevinstephano
@kevinstephano
1
Make ThunderModule.no_sync behave as PyTorch's Distributed one
#1670 opened Jan 21, 2025 by t-vi
Cannot turn off rematerialization for an individual lang op with hand specified grad transform autograd randomness rematerialization
#1658 opened Jan 17, 2025 by kevinstephano
@IvanYashchuk @mruberry
4
thunder.jit has a relatively high CPU overhead when processing small graphs with small inputs. performance
#1657 opened Jan 17, 2025 by kiya00
4
nvFuser using more memory than inductor for HF CausalLMLoss memory use nvfuser
#1654 opened Jan 17, 2025 by riccardofelluga
1
3
backward creates inconsistent proxies between args and unpacking them autograd tracing architecture
#1633 opened Jan 10, 2025 by t-vi
@t-vi
avoid joint trace in rematerialize forward backward rematerialization
#1618 opened Jan 8, 2025 by t-vi
1
make traces own proxies and bsyms enhancement New feature or request tracing architecture
#1606 opened Jan 6, 2025 by t-vi
2
nvFuser has a faster RMSNorm fusion definition than thunder's RMSNorm decomposition operators performance
#1582 opened Dec 23, 2024 by mruberry
3
Get dynamic shapes to work with Phi-3-mini-128k-instruct enhancement New feature or request nemo Issues needed to support NVIDIA NeMo models.
#1579 opened Dec 20, 2024 by tfogal
@jjsjann123
7
Consider adding is_leaf attribute to TensorProxies enhancement New feature or request
#1577 opened Dec 20, 2024 by beverlylytle
ProTip! Follow long discussions with comments:>50.