GitHub

238 articles in the TruthFoundry index. Each links out to the original.

• PyTorch raises NotImplementedError for unsupported linalg dtypes · PyTorch developers raise NotImplementedError for linalg ops rejecting integer, bool, and half precision inputs.
• PyTorch adopts eager cuBLAS workspace allocation to fix CUDA errors · PyTorch developers resolved a CUDA graph error by switching to eager cuBLAS workspace allocation despite minor performance overhead.
• Pytorch Replaces Global Device Mapping with DeviceOpOverrides · Pytorch removes global device mapping in favor of authoritative DeviceOpOverrides for C++ wrappers.
• PyTorch fixes autocast object identity guard bug · PyTorch developers resolve a bug where replacing autocast objects caused unnecessary recompiles due to identity guards.
• PyTorch PR Ensures CuTeDSL Tests Run on B200 Hardware · PyTorch developers fix CI configuration to run CuTeDSL tests on the B200 job.
• PyTorch PR Fixes CuTeDSL Test Execution for Blackwell GPUs · PyTorch developers fix CI configuration to run CuTeDSL tests on Blackwell hardware.
• PyTorch rejects import-time rebuilds to prevent native AOT corruption · PyTorch developers block import-time rebuilds to prevent interference with native AOT compilation.
• PyTorch Adds Logic to Validate Native AOT Build Conditions · PyTorch introduces a verification function to determine if Native AOT builds should proceed based on runtime and environment constraints.
• PyTorch Generates Native AOT Kernel Launchers and Gating Functions · PyTorch introduces a generator to create C++ launchers and conditional gating for native AOT kernels.
• PyTorch developer jananisriram tags ciflow/torchtitan update · PyTorch contributor jananisriram tagged a torchtitan update commit on GitHub.
• PyTorch developer Janani Srinam tags update commit · PyTorch contributor Janani Srinam tagged a repository update commit on GitHub.
• PyTorch developer jananisriram tags ciflow/torchtitan/195955 update · PyTorch developer jananisriram tags a new update for the ciflow/torchtitan/195955 repository.
• PyTorch releases update tag ciflow/trunk/185466 · PyTorch repository user jansel tagged a new update commit ciflow/trunk/185466.
• PyTorch fixes TensorList storage sanity-check codegen bug · PyTorch developers resolved a bug where TensorList storage was incorrectly constructed with double the intended capacity.
• PyTorch Dynamo Optimizes Bytecode Constant Lookup via Identity Mapping · PyTorch's Dynamo engine improves performance by indexing bytecode constants using object identity to avoid quadratic lookup times.
• PyTorch PR #195168 migrates DispatchKeySetVariable methods to tp_methods declarative handlers · PyTorch pull request #195168 migrates DispatchKeySetVariable methods 'has' and 'highestPriorityTypeId' to declarative tp_methods handlers.
• PyTorch Merges MPS Graph Optimization for Log Softmax · PyTorch merges a PR optimizing log_softmax memory usage and inference speed on Apple MPS.
• PyTorch Adds OOM Support to XPU Caching Allocator · PyTorch trunk introduces use_oom support for XPU caching allocator to align with CUDA distributed usage.
• PyTorch Updates torch-xpu-ops Commit to Fix Bugs and Build Warnings · PyTorch updates torch-xpu-ops commit to address safe-softmax issues, int4 matmul fixes, and build warnings.
• PyTorch updates torch-xpu-ops commit to fix XPU bugs · PyTorch resolves a PR to update torch-xpu-ops with bug fixes for XPU operations.
• PyTorch Enables Nested Reductions by Default in OSS · PyTorch enables nested reductions by default in its OSS build following internal stabilization.
• PyTorch developer tags XPU update commit · PyTorch developer guangyey tags an XPU update commit on September 3.
• PyTorch developer tags XPU update commit · Guangyey tags a PyTorch XPU update commit on GitHub.
• PyTorch Adds Host TMA Support for C++ Wrappers in AOTI · PyTorch introduces host-side TensorDescriptor support for AOTI C++ wrappers to fix view preservation issues.
• PyTorch AOTI Enhances Host TMA Support in C++ Wrappers · PyTorch AOTI improves host TMA support in C++ wrappers by decomposing descriptors into base tensors and metadata
• PyTorch updates ROCm preview version to August 2026 build · PyTorch resolves ROCm 10.1 HIPRTC test failures and updates package versions in a new preview release.
• PyTorch PR Updates ROCm Preview to 10.1.0a20260821, Skips Failing HIPRTC Tests · PyTorch PR #193181 updates the ROCm preview package to version 10.1.0a20260821 and skips HIPRTC tests on ROCm 10.1+.
• PyTorch developer tags update commit to repository · PyTorch contributor bobrenjc93 tagged a commit with an update release note.
• PyTorch developer tags repository update commit · Bobrenjc93 tagged a PyTorch repository update commit on GitHub.
• PyTorch developer tags repository update commit · Bobrenjc93 tagged a PyTorch repository update commit on GitHub.