PyTorch — full archive
117 daily briefings.
- TENSORDICT 0.14 LANDS WITH BREAKING API DEFAULTS, TORCHTITAN GAINS ASYNC-TP BACKEND
- DYNAMO TIGHTENS PYTHON SEMANTICS AS PYTORCH SHIPS TEST REFACTOR WAVE
- PYTORCH FUSES MATMUL REDUCTION INTO TRITON, FIXES HYDRA 1.4 COMPATIBILITY
- EXECUTORCH BACKENDS PATCH METADATA BUG, PYTORCH FIXES TEST BLOAT
- HELION RUNTIME REORGANIZED, DYNAMO FIXES NESTED GRAPH BREAKS, EXECUTORCH LANDS MOE QUANTIZATION PIPELINE
- COLLECTIVE OPS STANDARDIZATION LANDS ACROSS C10D BACKENDS
- EXECUTORCH ROUTES REASONING SEPARATELY, PYTORCH MODERNIZES COMPILER STACK
- DYNAMO FIXES GRAPH-BREAKING SDPA PARAMS, PAT SVD LEAK BREAKS DOWNSTREAM TESTS
- CRITICAL OIDC INJECTION IN DOCS PREVIEW WORKFLOW PATCHED
- PYTORCH AUTOGRAD GETS 7% FASTER, AOTI FIXES SILENT FAILURES
- DYNAMO REVERTS BREAKING CHANGE, EXECUTORCH CLEANS UP DEPRECATED TYPES
- PYTORCH SHIPS BUILD FIX WHILE HELION TUNES H100 KERNELS TO DEFAULT
- HELION BACKEND REFACTOR UNLOCKS MULTI-DEVICE COMPILATION; DYNAMO FIXES FORWARD-AD TANGENT BUG
- HELION SHIPS B200 AUTOTUNER, PYTORCH PLUGS CUDA DUPLICATION, TORCHRL ADDS PULL-BASED MONITORING
- EXECUTORCH SLASHES CONV LATENCY, PYTORCH CORE FIXES FLEXATTENTION SPECIALIZATION BUG
- INDUCTOR FIXES NESTED REDUCTION BLOCK POINTER BUG, TORCHRL UNIFIES CHECKPOINTING
- PYTORCH FIXES CRITICAL MPS MATH BUG WHILE TAMING CI GHOSTS
- PYTORCH FIXES H100 FLASH ATTENTION CI BREAKAGE, HELION CONSOLIDATES PALLAS LAUNCHERS
- DYNAMO FIXES ITERATOR CORRUPTION AND FAKE INT ARITHMETIC
- TORCHRL CHECKPOINTS NOW MEMORY-MAPPED BY DEFAULT, FIXING WINDOWS FILE-LOCK CORRUPTION
- PYTORCH 2.13 SHIPS WITH FLEXATTENTION ON APPLE SILICON, NCCL2 BACKEND LANDS IN-TREE
- DYNAMO BACKENDS UNBROKEN, PYTORCH COMPLETES OSDC MIGRATION
- PYTORCH FLUSHES CI FLAKES, TIGHTENS MSVC GATES, TORCHTITAN CRASHES ON NAN
- TORCH.COMPILE GETS CUDA SYNC FIX, DYNAMO TIGHTENS TYPE CHECKING
- CUDA BACKWARD KERNEL BUG FIXED AS EXECUTORCH EXPANDS DYNAMIC SHAPE SUPPORT
- PYTORCH RELEASE WHEELS SHEDDING 80MB OF BLOAT, QUANTIZED EXPORTS FIXED
- INDUCTOR CAT_LINEAR FUSION CUTS MATERIALIZATION, ROCM GAINS CDNA5 SUPPORT
- PYTORCH FIXES CRYPTIC LINALG ERROR, ACCELERATES EXECUTORCH BACKENDS
- DISTRIBUTED BACKEND REORG UNBLOCKS CUSTOM BACKENDS, EXECUTORCH LANDS ARM TOSA OPS
- EXECUTORCH FUSES TORCHAO 4-BIT EMBEDDINGS; CUDA TUNING LANDS
- EXECUTORCH FLASHDECODING LANDS, INDUCTOR FIXES BACKWARD CRASH, TORCHTITAN RL PIPELINE GETS WEIGHT SYNC OVERLAP
- EXECUTORCH CLOSES CROSS-COMPILE GAP, RL SUITE SHIPS MAJOR REPLAY BUFFER OVERHAUL
- PYTORCH SHIPS METAL ACCELERATORS AND SPARSE DTYPE SUPPORT WHILE TORCHTITAN ADVANCES EXPERT PARALLELISM
- DYNAMO CLOSES GENERATOR AND RANGE GAPS, HELION SHIPS REDUCTION HEURISTICS
- PYTORCH FIXES CRITICAL FAKETEN OR EDGE CASE AS EXECUTORCH ADDS ARM TOSA SUPPORT
- INDUCTOR DOUBLES DOWN ON PROFILING; C10D BACKENDS GET UNIFIED API
- HALIDE FUSION BUG CRUSH, ROCM DEADLOCK FIXES ACROSS THE STACK
- GLOO SECURITY FIX SHIPS, CUDA 13.0.3 LANDS, FBGEMM MIGRATION ADVANCES
- TORCHTITAN SHIPS FLEXATTENTION INDUCTOR BOOST, GRAPH TRAINER UNLOCKS CPU OFFLOADING
- DYNAMO GAINS ARITHMETIC OPERATORS AS FSDP2 CUTS BACKWARD COMPUTE STALLS
- PALLAS JAGGED KERNELS NOW PIPELINE, FSDP2 UNBLOCKS BACKWARD COMPUTE
- PYTORCH ROLLS BACK NAMED TENSOR REMOVAL, FIXES CHROMIUM PROFILER LEAK
- PYTORCH KILLS CUDA 11 SUPPORT, SHIPS METAL RANDPERM OVERHAUL
- EXECUTORCH ADDS TURBOQUANT TO GEMMA4, PYTORCH PURGES FLAKY DYNAMO TESTS
- PYTORCH SHIPS TRITON CACHE HOT-LOAD AND PROFILER RESILIENCE FIXES
- MPS CONVOLUTION FAST PATH LANDS, DYNAMO GAINS RICH COMPARISON OPS
- SHAPESSPEC UNLOCKS VARIADIC ARGS ACROSS DYNAMO AND QUANTIZED TENSORS GET DEPRECATION WARNING
- PYTORCH FIXES PRECISION DRIFT IN INDUCTOR, EXECUTORCH BACKS OUT XNNPACK SERIALIZATION
- TORCH.COMPILE GAINS PYTHON COMPARISON OPERATORS, TESTS GO DEVICE-AGNOSTIC ACROSS THREE REPOS
- PYTORCH TIGHTENS BUILD STANDARDS WITH CLANG-TIDY COVERAGE AND C++20 MODERNIZATION
- PYTORCH SHIPS L2 CROSS-REPO CI RELAY, PATCHES PYTHON 3.15 WHEEL BUILDS
- DYNAMO STANDARDIZES CORE API, PROFILER BYPASSES C++ SERIALIZATION
- FSDP2 CUDA GRAPHS STREAM EXPLOSION FIXED, EXECUTORCH FUSION PIPELINE PATCHED
- TORCHTITAN SHIPS TOKEN-IN-TOKEN-OUT GENERATOR FOR RL, FIXES CRITICAL CI BREAKS ACROSS PLATFORMS
- INDUCTOR STACK FIXES CRITICAL NANOUTPUTS AND DTYPE BUGS WHILE CI INFRA PIVOTS TO IPV6
- FLEX_ATTENTION TIGHTENS, VMAP GAINS VECTORIZATION, MACOS CLEARS WARNINGS
- ROCM DISTRIBUTED JOBS GRADUATE FROM SANDBOX, HELION SHIPS CUDAGRAPH OPTIMIZATION
Never miss a PyTorch briefing.
Get PyTorch's week in review every Monday - and security advisories the day they land. Free, unsubscribe in one click.
One click. Every Monday: what mattered across the projects you've been reading (or our AI picks) — security news same-day. You refine anytime. Choose specific journals · Read a sample issue →