Topic · Edition
Changes since langchain-openai==1.5.1 release(openai): 1.5.2 (#39719) fix(openai): preserve reasoning item boundaries (#39278) release(openai): 1.5.2a1 (#39709) feat(openai): extract gateway metadata from response…
Descriptions: Root cause: Commit 08c29108 ("Remove the obsolete folder yaml/") in torch-xpu-ops migrated XPU dispatch registrations from torch-xpu-ops/yaml/native/native_functions.yaml to PyTorch's…
Fix Inductor wrapper codegen for user-defined Triton kernels so literal Python float launch arguments use Triton's normal fp32 signature behavior instead of being forced through Inductor's fp64 policy for Python floats.…
Issue Fixes #193705 Summary remove_redundant_views() (torch/_inductor/fx_passes/joint_graph.py) erases any zero-user node it's tracking, including graph placeholders — shrinking the compiled function's arity while the…
Motivation As title Additional Context CI passed. Pull Request resolved: #192495 Approved by: https://github.com/gujinghui, https://github.com/albanD
[Testcase Refactoring] Make test_cuda_event_eq device-agnostic via in…
[Testcase Refactoring] Make test_cuda_event_eq device-agnostic via in…
[Inductor] Use schedule-first bounded regions for combo kernel memory…
[pt2] Fix barrier deps to exclude nodes defined after sync point (#19…
[pt2] Fix barrier deps to exclude nodes defined after sync point (#19…
Provide script to swap between upstream Triton and FBTriton for torchTLX dev works. python tools/torchtlx/dev.py doctor python tools/torchtlx/dev.py switch fbtriton --from-source /data/users/daohang/fbtriton python…
Provide script to swap between upstream Triton and FBTriton for torchTLX dev works. python tools/torchtlx/dev.py doctor python tools/torchtlx/dev.py switch fbtriton --from-source /data/users/daohang/fbtriton python…
[c10][cuda] Per-stream expandable-segment reserve downsizing (#192269…
[c10][cuda] Per-stream expandable-segment reserve downsizing (#192269…
[qnnpack] Add 1..4x8 dynamic-quant aarch64 q8gemm microkernel for sma…
[qnnpack] Add 1..4x8 dynamic-quant aarch64 q8gemm microkernel for sma…
[TEST][CUDA] Bump reflection_pad_2d decomp test to use bfloat16 thr…
Initial release release(openai): 1.5.2a1 (#39709) feat(openai): extract gateway metadata from response headers when available (#39706) chore(openai): update snapshots (#39657) fix(openai): support o-series models in…
Summary: Mirror TORCH_COMPILE_DYNAMIC_SOURCES, we add TORCH_COMPILE_STATIC_SOURCES (or torch.compiler.config.static_sources) to pass a list of source names that should be marked as static. This is useful when automatic…
Teach SIMD scheduling and codegen to fuse reduction epilogues that consume interleaved pairs from a parent tile. This enables standalone NVFP4 packing while rejecting ambiguous, mutated, and non-leaf consumers.…
This reverts commit e734072. Reverted #193729 on behalf of https://github.com/jeanschmidt due to it have problems with CCCL 2.4.0 and older (comment)
Cases of test_small_mm_pointwise are failing flakily in test/inductor/test_mmdecomp.py on GB200 because the lowering pathway runs autotuning to determine whether to pad or not- this generally favors the pointwise…
Cases of test_small_mm_pointwise are failing flakily in test/inductor/test_mmdecomp.py on GB200 because the lowering pathway runs autotuning to determine whether to pad or not- this generally favors the pointwise…
[ATen][CPU] Speed up sort by packing (value, index) into a contiguous…
[dynamo] Split FrozensetVariable off SetVariable to match CPython (#1…
[dynamo] Split FrozensetVariable off SetVariable to match CPython (#1…
