Daily edition · AI Tools
TILens turns technical updates into a focused daily brief: official releases, trusted reporting, and practitioner analysis, deduplicated and organized by topic.
TORCH_HIP_VERSION is defined by cmake as HIP major*100+minor (currently 7xx), so < 305 in ATenNVRTC.h and >= 301 in nccl.h always resolve to the modern branch; hardcode it. The surviving…
Changes since langchain-core==1.5.5 chore(core): release 1.5.6 (#39704) feat(core): incorporate gateway metadata to traces (#39703)
[inductor] Fix missing int_array declarations in AOTI C++ wrapper (#1…
[pytorch][PR] Migrate fastAtomicAdd to headeronly (#193176) (#193604)…
[AI Codemod][PerfAICT-General] Add contiguous fast-path to scalar exe…
[XPU] Add bf16 tolerance override for LinearCrossEntropyLoss test_sav…
[inductor] Test cvt_e8m0_rceil software fallback correctness on non-S…
[Testcase Refactoring] Refactor test_python_dispatcher.py for test ca…
[inductor] batch_linear_lhs: stop walk at contiguous, guard non-str t…
[Testcase Refactoring] Add proper hw_classification attribute to test…
[Testcase Refactoring] Add proper hw_classification attribute to test…
mean/var/std (and var_mean/std_mean), nanmean and gradient reject integer input with a plain RuntimeError, but are well-defined for integers (NumPy promotes to float) and merely unimplemented, so raise…
This PR is auto-generated nightly by this action. Update the pinned vllm hash. Pull Request resolved: #193748 Approved by: https://github.com/pytorchbot
This PR is auto-generated nightly by this action. Update the pinned vllm hash. Pull Request resolved: #193748 Approved by: https://github.com/pytorchbot
install pytest-timeout for xpu test config
Replace repeated matrix rank and contraction-dimension checks across matrix multiplication implementations with the shared check_mm_shapes helper, producing consistent operation-prefixed diagnostics. Authored with…
Replace repeated matrix rank and contraction-dimension checks across matrix multiplication implementations with the shared check_mm_shapes helper, producing consistent operation-prefixed diagnostics. Authored with…
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
Related Part of #170487 Why Setting tensor.requires_grad = True inside torch.compile always broke the graph, even on tensors created in the compiled region The equivalent .requires_grad_(True) call already worked, so…
Related Part of #170487 Why Setting tensor.requires_grad = True inside torch.compile always broke the graph, even on tensors created in the compiled region The equivalent .requires_grad_(True) call already worked, so…
torch.compiler.export_python wraps torch.compiler.precompile in a decorator keyed by a readable Python artifact on disk. The first call captures and lowers the decorated function, publishes the self-contained source,…
update vllm commit hash
