[pull] main from pytorch:main#1648
Merged
Merged
Conversation
fixes: #136452 docstrings for `linalg.norm` and `linalg.matrix_norm` contain a confusing contradiction between the `dtype` kwarg and the return for complex dtypes. This is clarified by simply stating that the input is cast. The output dtype is already implied by the definitions of the operations. All three of `norm`, `vector_norm`, and `matrix_norm` are now consistent. They no longer state that the return type will be guaranteed to be the same. * `torch.linalg.norm` flows to either `torch.linalg.vector_norm` or `torch.linalg.matrix_norm` depending on the conditions documented. * All of `norm`, `vector_norm` and `matrix_norm` indicate that the return value will be real if input is complex. * Both of `vector_norm` and `matrix_norm` indicate that for complex input the norm of `A.abs()` or `x.abs()` is computed instead. `vector_norm` did not contain a contradiction but the description for `dtype` was also far less clear. IMO saying less with respect to dtype is more clear here. Pull Request resolved: #190381 Approved by: https://github.com/lezcano
Disables TestNVUniversalGemmEpilogueFusion.test_workspace_runtime_integration, failing in CI and tracked/disabled via #188674. Test Plan: test-only; the test is skipped. Authored with Claude. Pull Request resolved: #190101 Approved by: https://github.com/janeyx99
Disables test_p2p_ipc_expandable_segments (P2PIpcTest), which is failing in CI (one of several rotating failures on the B200 distributed job). Tracked and disabled via #189878; this makes the skip explicit in-tree. Test Plan: test-only change; the test is skipped. Authored with Claude. Pull Request resolved: #190096 Approved by: https://github.com/janeyx99
## Why When one `aot_autograd` backend compiles a second, distinct graph, `AotAutograd.__call__` re-enters with `bw_compiler` already wrapped. The `_is_wrapped_bw_compiler` branch then ran `bw_compiler.compiler_fn = bw_compiler`, grafting a self-referential attribute (and needing a `type: ignore`) instead of leaving the wrapper untouched. The branch only exists to be idempotent. ## How - Leave `bw_compiler` as-is when it is already wrapped - Add `test_aot_autograd_reentrant_bw_compiler`, which fails before this change Pull Request resolved: #189325 Approved by: https://github.com/bobrenjc93, https://github.com/jansel
…#189525) ## Related Issue Fixes #188151 ## Why `_bound_stride` registers expected stride keys as `stride * size`, but fake tensors compute contiguous strides as `stride * max(size, 1)`. Int-input-derived symbols under `dynamic=True` have ranges including 0/1, so the Max does not simplify, the lookup misses, and an internal RuntimeError escapes. Regression from #147130 (2.12+). ## How - Also register the `max(size, 1)` key form (via `torch.sym_max`, no guards) mapped to the same canonical merged stride expr Pull Request resolved: #189525 Approved by: https://github.com/aorenste
…cpu_float32` (#190318) `PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=16 python functorch/test_ops.py TestOperatorsCPU.test_vmap_autograd_grad_nn_functional_conv2d_cpu_float32` fails with > Greatest absolute difference: 1.1444091796875e-05 at index (0, 4, 0, 0, 2) (up to 1e-05 allowed) > Greatest relative difference: 2.064850013994146e-05 at index (0, 4, 0, 0, 2) (up to 1.3e-06 allowed) So use the same tolerance used for CUDA unconditionally. Fixes #151113 Pull Request resolved: #190318 Approved by: https://github.com/janeyx99
Advances the Triton 3.8 pin from [`43422b04`](triton-lang/triton@43422b0) to [`4774139f`](triton-lang/triton@4774139), the current `release/3.8.x` tip. The new pin includes: - triton-lang/triton#10900: combines TMEM loads followed by row reductions on sm103+. - triton-lang/triton#10787: release-branch revert of triton-lang/triton#7067. - triton-lang/triton#10779: disables LLVM SLP for sm90 bf16 reductions, fixing the H100 bf16 Swin accuracy regression reported in #188492. - triton-lang/triton#10935: uses `torch.cuda.is_available()` when torch is already imported, avoiding forked-child CUDA state poisoning reported in #189287. Pull Request resolved: #190349 Approved by: https://github.com/atalman
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )