v26.06.00

What's Changed

🚨 Breaking Changes

• Replace legacy SCC with the new SCC implementation and SCC performance improvements by @seunghwak in #5469
• Migrate RMM usage to CCCL MR design by @bdice in #5483
• cuGraph common refactor by @bdice in #5502

🐛 Bug Fixes

• Add missing CCCL headers for cuda:: symbols by @bdice in #5466
• Fix build errors in debug build by @seunghwak in #5474
• Fix k-truss test failure on Blackwell by @jnke2016 in #5495
• Typo in selecting hungarian implementation by @ChuckHastings in #5506
• Re-enable test_benchmarking_datasets test by @rlratzel in #5504

🚀 New Features

• Strongly Connected Components Part 1/2 by @seunghwak in #5442
• Strongly Connected Components 2 (Find trivial SCCs in components & Intersect reachable sets) by @ngokulakrish in #5468
• Move cugraph C++ primitives from src/prims to include/cugraph/prims by @seunghwak in #5477
• Add topological sort algorithm by @ngokulakrish in #5492

🛠️ Improvements

• Update C++ test build to use C++ 20 by @seunghwak in #5459
• Cut binary size in similarity coefficients. by @seunghwak in #5455
• Improve BFS performance for large-scale multi-node NVLink systems by @seunghwak in #5322
• Forward-merge release/26.04 into main by @jameslamb in #5470
• Forward-merge release/26.04 into main by @jameslamb in #5472
• depend on libnvjitlink-dev at build time unconditionally by @jameslamb in #5479
• Main merge release/26.04 by @ChuckHastings in #5478
• update pip devcontainers' base image tags by @trxcllnt in #5481
• Remove a couple of variant branches to shrink libcugraph_mg about 6% by @ChuckHastings in #5482
• wheels CI: update to torch 2.10 for 'oldest' configuration by @jameslamb in #5485
• Temporarily skip the k-truss tests by @jnke2016 in #5487
• Update to clang 20.1.8 by @bdice in #5488
• Revert "wheels CI: update to torch 2.10 for 'oldest' configuration (#5485) by @jameslamb in #5493
• Reduce binary size part2 by @ChuckHastings in #5486
• Addressing Issue 5397 5384 by @BradReesWork in #5490
• Deduplicate compute_number_of_edges CUB/Thrust kernels via explicit instantiation by @bdice in #5489
• Use token.rapids.nvidia.com when issuing S3 bucket creds in devcontainers by @trxcllnt in #5505
• Remove NO_CUDART_DEP property by @KyleFromNVIDIA in #5508
• Fix symbol export by @vyasr in #5510
• fix(ci): resolve all zizmor findings and add zizmor pre-commit checks by @gforsyth in #5509
• Link libcugraph dynamically against libcuvs by @bdice in #5499
• fix(ci): add explicit actions: write permission for telemetry-summarize
  by @gforsyth in #5513
• Build and test with CUDA 13.2.0 by @bdice in #5511
• split testing libraries from run libraries in dependencies.yaml by @jayavenkatesh19 in #5447
• add support for multigraph to disjoint sampling by @ChuckHastings in #5520
• skip CuPy 14.1.0 by @jameslamb in #5531

New Contributors

• @ngokulakrish made their first contribution in #5468

Full Changelog: v26.06.00a...v26.06.00

v26.04.00

What's Changed

🚨 Breaking Changes

• cuGraph now is two shared libraries due to linker requirements by @robertmaynard in #5452

🐛 Bug Fixes

• We updated RAPIDS to use C++ 20, but missed updating cugraph etl by @ChuckHastings in #5408
• Add our own is_discard_iterator as thrust removed it in 3.2 by @robertmaynard in #5416
• Fix memory allocation failure in SG betweenness centrality for small-diameter graphs by @seunghwak in #5432
• Fix null optional dereference in temporal_partition_vertices by @ChuckHastings in #5446

🚀 New Features

• Add RAPIDS Doctor Smoke Test/Smoke Check for cuGraph by @alexbarghi-nv in #5451

🛠️ Improvements

• Drop Python 3.10 support by @gforsyth in #5412
• Use verify-hardcoded-version pre-commit hook by @KyleFromNVIDIA in #5411
• tighten wheel size limits, expand CI-skipping logic, other small build changes by @jameslamb in #5413
• remove pip.conf migration code in CI scripts, update CI-skipping rules by @jameslamb in #5415
• Replace .values_host with .to_numpy() by @bdice in #5420
• refactor: build wheels and conda packages using Python limited API by @gforsyth in #5419
• Use GHA id-token for sccache-dist auth token by @trxcllnt in #5425
• Temporarily disable RTX PRO 6000 PR CI by @bdice in #5422
• refactor C++ examples for new RAPIDS branching strategy by @jameslamb in #5423
• check-nightly-ci: update to new version by @jameslamb in #5433
• Temporarily disables tests on RTX Pro 6000 for nightly runs by @rlratzel in #5434
• remove references to 'nvidia' and 'pytorch' conda channels by @jameslamb in #5426
• check-nightly-ci: remove testing config by @jameslamb in #5436
• Update raft headers by @aamijar in #5429
• Drop thrust::constant_iterator in favor of cuda::constant_iterator by @miscco in #5430
• Drop usage of thrust function objects in favor of cuda::std ones by @miscco in #5431
• Fix ucx link errors in pip devcontainers by @trxcllnt in #5435
• Drop thrust::transform_iterator in favor of cuda::transform_iterator by @miscco in #5428
• refactor(limited api): add explicit wheel.py-api to pyproject.toml
  by @gforsyth in #5440
• Add make_initialized_edge_(src|dst_)property utility functions. by @seunghwak in #5441
• Update Cython lower bound pin to 3.2.2 by @vyasr in #5449
• Remove pytest upper bound pin by @vyasr in #5450
• Add support for Python 3.14 by @gforsyth in #5439
• wheels: use cuda-toolkit runtime wheels by @jameslamb in #5444
• Avoid deprecated jit_unspill in dask tests by @TomAugspurger in #5456
• add no_pytorch matrix option in dependencies.yaml by @trxcllnt in #5454
• fix verify-hardcoded-versions issues by @jameslamb in #5462
• ensure 'torch' CUDA wheels are installed in CI, remove unused dependencies by @jameslamb in #5453
• Re-enables tests on RTX Pro 6000 by @rlratzel in #5445
• enforce a floor on libnvjitlink, build wheels with CUDA 13.0.x, test wheels against mix of CTK versions by @jameslamb in #5457
• C++ Disjoint sampling implementation by @ChuckHastings in #5414

New Contributors

• @aamijar made their first contribution in #5429

Full Changelog: v26.04.00a...v26.04.00

v26.02.00

What's Changed

🚨 Breaking Changes

• Default to static linking of libcudart by @bdice in #5366
• Require C++20 by @bdice in #5310

🐛 Bug Fixes

• Main merge release/25.12 (unblock forward merge) by @alexbarghi-nv in #5354
• Include <thrust/distance.h> where it is used by @bdice in #5357
• Bug fix in per_v_transform_reduce(_if)_incoming|outgoing_e by @seunghwak in #5360
• Specify argument types inside proclaim_return_type by @vyasr in #5379
• Fix bug in multi-GPU sampling by @ChuckHastings in #5372
• Update examples to incorporate the latest changes to the API by @ChuckHastings in #5389

🚀 New Features

🛠️ Improvements

• Addressing FIXMEs by @BradReesWork in #5330
• Use strict priority in CI conda tests by @bdice in #5352
• Use strict priority in CI conda tests by @bdice in #5359
• CUDA 13: use releases, not nightlies, for PyTorch wheels by @jameslamb in #5356
• Remove alpha specs from non-RAPIDS dependencies by @bdice in #5362
• Remove deprecated raft::random::uniformInt usage by @jnke2016 in #5365
• Compatibility fixes for CCCL 3.2 by @bdice in #5363
• Add devcontainer fallback for C++ test location by @bdice in #5369
• Removes experimental PropertyGraph (SG and MG) and all references by @rlratzel in #5374
• Proclaim return types and other fixes needed for CCCL 3.2 by @bdice in #5375
• Empty commit to trigger a build by @bdice in #5377
• Use SPDX license identifiers in pyproject.toml, bump build dependency floors by @jameslamb in #5380
• Add CUDA 13.1 support by @bdice in #5373
• Remove cumlprims from clang-format by @bdice in #5382
• Cleanup empty by @BradReesWork in #5385
• Empty commit to trigger a build by @jameslamb in #5391
• Fix negative sampling with biases by @ChuckHastings in #5386
• Improve Error Messages from MissingUCXX by @alexbarghi-nv in #5378
• build and test against CUDA 13.1.0 by @jameslamb in #5383
• Use main shared-workflows branch by @jameslamb in #5396
• Replace deprecated thrust facilities in favor of libcu++ ones by @miscco in #5394
• Remove incorrect documentation main-to-release replacements by @AyodeAwe in #5400
• Update RMM memory resource APIs to ref-based equivalents by @bdice in #5392
• Fix MG Leiden UB in std::move, re-enable MG Python tests by @rlratzel in #5398
• wheel builds: react to changes in pip's handling of build constraints by @mmccarty in #5399
• Replace deprecated thrust tuple facilities with cuda::std equivalents by @bdice in #5404
• fix(build): build package on merge to release/* branch by @gforsyth in #5409

Full Changelog: v26.02.00a...v26.02.00

v25.12.02

What's Changed

🛠️ Improvements

• Unset SCCACHE_NO_CACHE by @trxcllnt in #5371

Full Changelog: v25.12.01...v25.12.02

v25.12.01

What's Changed

🛠️ Improvements

• Don't use the build cluster for releases due to runner disk constraints by @trxcllnt in #5370

Full Changelog: v25.12.00...v25.12.01

v25.12.00

What's Changed

🚨 Breaking Changes

• Add support for other temporal comparisons by @ChuckHastings in #5283
• Dropping cuGraph Service by @BradReesWork in #5325
• Require CUDA 12.2+ by @jakirkham in #5327
• Remove Old Sampling API by @alexbarghi-nv in #5335
• Time filtering hop 0 by @ChuckHastings in #5328
• Deprecate Distributed Sampling API by @alexbarghi-nv in #5336

🐛 Bug Fixes

• Skip failing dataset test for now by @ChuckHastings in #5301
• Fix force_atlas2 on empty graphs by @eriknw in #5303
• Fix G.nodes() returning internal vertex IDs by @eriknw in #5304
• Use pinned_host_memory_resource for device-accessible memory in renumbering.cu by @bdice in #5317
• Fix warning in unreachable device code by @bdice in #5316
• Pin Cython pre-3.2.0 and PyTest pre-9 by @jakirkham in #5340
• refactored update-version.sh to handle new branching strategy by @rockhowse in #5341

📖 Documentation

• Use Docker CUDA major-version tags by @bdice in #5291

🚀 New Features

• [FEA] Update pylibcugraph to support different temporal comparisons by @alexbarghi-nv in #5345

🛠️ Improvements

• Update RAPIDS_BRANCH, codify changes in update-version.sh by @KyleFromNVIDIA in #5278
• Update for cudf.DataFrame.from_pandas deprecation by @TomAugspurger in #5286
• Update to rapids-logger 0.2 by @bdice in #5298
• force_atlas2: allow vertex mass to be given by the user by @eriknw in #5293
• Use pinned_host_memory_resource instead of pinned_memory_resource. by @bdice in #5312
• Enable sccache-dist connection pool by @trxcllnt in #5314
• Use main in RAPIDS_BRANCH by @bdice in #5318
• Fix ForceAtlas2 speed factor in prevent_overlapping mode by @MathisHammel in #5295
• Use main shared-workflows branch by @bdice in #5319
• Update the code to be compatible with the new cuco stream-ordered allocator by @PointKernel in #5311
• Use SPDX for all copyright headers by @KyleFromNVIDIA in #5320
• Remove shims for CCCL < 3.1 compatibility by @bdice in #5331
• Migrate cugraph to CCCL memory resource interface by @bdice in #5337
• Use ruff-check, ruff-format instead of flake8, black by @KyleFromNVIDIA in #5334
• Fixes networkx version checks for skipping certain tests. by @rlratzel in #5333
• rename edge_time_t to time_stamp_t by @ChuckHastings in #5339
• Enable merge barriers by @KyleFromNVIDIA in #5344
• Update RMM includes from <rmm/mr/device/*> to <rmm/mr/*> by @bdice in #5343
• New sampling flag for disjoint sampling by @ChuckHastings in #5342
• Update API to leverage cuVS Spectral clustering by @jnke2016 in #5326
• Use sccache-dist build cluster for conda and wheel builds by @trxcllnt in #5332
• Requires NX 3.5 for edge BC tests by @rlratzel in #5355

New Contributors

• @rockhowse made their first contribution in #5341

Full Changelog: v25.12.00a...v25.12.00

v25.10.01

🚨 Breaking Changes

• Fix cugraph-service Failure and Re-enable CI Tests (#5268) @nv-rliu
• ForceAtlas2: continue #5213 for prevent_overlapping and mobility (#5260) @eriknw
• Update create graph to use arithmetic types (#5248) @ChuckHastings
• Remove subgraph extraction (#5247) @BradReesWork
• Update shuffle_ext_vertices, shuffle_int_vertices, shuffle_local_edge_srcs to take std::vector<arithemtic_devcie_uvector_t> (#5225) @seunghwak
• cleanup and dropping NX (#5194) @BradReesWork
• Re-enable spectral clustering (#5151) @jnke2016

🐛 Bug Fixes

• [CRITICAL BUG] Fix sample result destructor (hotfix) (#5306) @alexbarghi-nv
• bug fix - incorrect if statement causing GNN failure (#5297) @ChuckHastings
• Fix cugraph-service Failure and Re-enable CI Tests (#5268) @nv-rliu
• [BUG] Fix Sampling Output Parameters for Temporal Sampling API (#5267) @alexbarghi-nv
• Skip failing cugraph-service tests for 25.10 CI (#5265) @nv-rliu
• check that optional is set before resizing (#5245) @ChuckHastings
• mg_util supports running on tegra like machines (#5241) @robertmaynard
• Fix incorrect filter in ForceAtlas2 attraction kernel (#5229) @MathisHammel

📖 Documentation

• removing change blocks and some diagram improvements (#5231) @acostadon

🛠️ Improvements

• Fix missed deps in update-version.sh (#5287) @AyodeAwe
• Empty commit to trigger a build (#5274) @msarahan
• Clarify supported type combination (#5273) @jnke2016
• deprecate cugraph-service (#5269) @BradReesWork
• Update balanced_edge_test (#5262) @ChuckHastings
• ForceAtlas2: continue #5213 for prevent_overlapping and mobility (#5260) @eriknw
• Configure repo for automatic release notes generation (#5258) @AyodeAwe
• update dependencies: numba (>=0.60.0) (#5253) @jameslamb
• Implement two level near far method in SSSP and initial Graph 500 SSSP driver code (#5252) @seunghwak
• Adds Leiden benchmark (#5251) @rlratzel
• Temporal Sampling PLC API (#5250) @jnke2016
• Use branch-25.10 again (#5249) @jameslamb
• Update create graph to use arithmetic types (#5248) @ChuckHastings
• Remove subgraph extraction (#5247) @BradReesWork
• fix pytorch requirement for CUDA 13 builds (#5244) @jameslamb
• use dask-cuda[cu12, cu13] extras for wheel dependencies (#5243) @jameslamb
• Update rapids-dependency-file-generator (#5240) @KyleFromNVIDIA
• Move from UCX-Py to UCXX (#5238) @pentschev
• remove cugraph-pyg and wholegraph configuration (#5237) @jameslamb
• Build and test with CUDA 13.0.0 (#5236) @jameslamb
• Refactor Graph 500 BFS code and replace thrust::tuple with cuda::std::tuple (#5235) @seunghwak
• Temporal Sampling C API (#5232) @ChuckHastings
• Raise wheel size threshold for libcugraph (#5230) @ChuckHastings
• Migrate from thrust::tuple to cuda::std::tuple (#5226) @ChuckHastings
• Update shuffle_ext_vertices, shuffle_int_vertices, shuffle_local_edge_srcs to take std::vector<arithemtic_devcie_uvector_t> (#5225) @seunghwak
• Use build cluster in devcontainers (#5223) @trxcllnt
• Use rapids_cuda_enable_fatbin_compression (#5222) @robertmaynard
• Update rapids_config to handle user defined branch name (#5218) @robertmaynard
• performance notebook improvements (#5217) @acostadon
• Update rapids-build-backend to 0.4.1 (#5211) @KyleFromNVIDIA
• Concurrent multisource backwardpass (#5206) @HowardHuang1
• ci(labeler): update labeler action to @v5 (#5205) @gforsyth
• Allows Dataset method get_edgelist() to be called on a CPU-only system (#5196) @rlratzel
• cugraph RAPIDS dependency now use the new rapids-cmake branch info (#5195) @robertmaynard
• cleanup and dropping NX (#5194) @BradReesWork
• Allow latest OS in devcontainers (#5188) @bdice
• Update build infra to support new branching strategy (#5172) @robertmaynard
• Temporal Graph Construction C API (#5161) @ChuckHastings
• Re-enable spectral clustering (#5151) @jnke2016
• Improve betweenness centrality performance (#5146) @HowardHuang1
• Use GCC 14 in conda builds. (#5124) @vyasr

v25.10.00

🚨 Breaking Changes

• Fix cugraph-service Failure and Re-enable CI Tests (#5268) @nv-rliu
• ForceAtlas2: continue #5213 for prevent_overlapping and mobility (#5260) @eriknw
• Update create graph to use arithmetic types (#5248) @ChuckHastings
• Remove subgraph extraction (#5247) @BradReesWork
• Update shuffle_ext_vertices, shuffle_int_vertices, shuffle_local_edge_srcs to take std::vector<arithemtic_devcie_uvector_t> (#5225) @seunghwak
• cleanup and dropping NX (#5194) @BradReesWork
• Re-enable spectral clustering (#5151) @jnke2016

🐛 Bug Fixes

• bug fix - incorrect if statement causing GNN failure (#5297) @ChuckHastings
• Fix cugraph-service Failure and Re-enable CI Tests (#5268) @nv-rliu
• [BUG] Fix Sampling Output Parameters for Temporal Sampling API (#5267) @alexbarghi-nv
• Skip failing cugraph-service tests for 25.10 CI (#5265) @nv-rliu
• check that optional is set before resizing (#5245) @ChuckHastings
• mg_util supports running on tegra like machines (#5241) @robertmaynard
• Fix incorrect filter in ForceAtlas2 attraction kernel (#5229) @MathisHammel

📖 Documentation

• removing change blocks and some diagram improvements (#5231) @acostadon

🛠️ Improvements

• Fix missed deps in update-version.sh (#5287) @AyodeAwe
• Empty commit to trigger a build (#5274) @msarahan
• Clarify supported type combination (#5273) @jnke2016
• deprecate cugraph-service (#5269) @BradReesWork
• Update balanced_edge_test (#5262) @ChuckHastings
• ForceAtlas2: continue #5213 for prevent_overlapping and mobility (#5260) @eriknw
• Configure repo for automatic release notes generation (#5258) @AyodeAwe
• update dependencies: numba (>=0.60.0) (#5253) @jameslamb
• Implement two level near far method in SSSP and initial Graph 500 SSSP driver code (#5252) @seunghwak
• Adds Leiden benchmark (#5251) @rlratzel
• Temporal Sampling PLC API (#5250) @jnke2016
• Use branch-25.10 again (#5249) @jameslamb
• Update create graph to use arithmetic types (#5248) @ChuckHastings
• Remove subgraph extraction (#5247) @BradReesWork
• fix pytorch requirement for CUDA 13 builds (#5244) @jameslamb
• use dask-cuda[cu12, cu13] extras for wheel dependencies (#5243) @jameslamb
• Update rapids-dependency-file-generator (#5240) @KyleFromNVIDIA
• Move from UCX-Py to UCXX (#5238) @pentschev
• remove cugraph-pyg and wholegraph configuration (#5237) @jameslamb
• Build and test with CUDA 13.0.0 (#5236) @jameslamb
• Refactor Graph 500 BFS code and replace thrust::tuple with cuda::std::tuple (#5235) @seunghwak
• Temporal Sampling C API (#5232) @ChuckHastings
• Raise wheel size threshold for libcugraph (#5230) @ChuckHastings
• Migrate from thrust::tuple to cuda::std::tuple (#5226) @ChuckHastings
• Update shuffle_ext_vertices, shuffle_int_vertices, shuffle_local_edge_srcs to take std::vector<arithemtic_devcie_uvector_t> (#5225) @seunghwak
• Use build cluster in devcontainers (#5223) @trxcllnt
• Use rapids_cuda_enable_fatbin_compression (#5222) @robertmaynard
• Update rapids_config to handle user defined branch name (#5218) @robertmaynard
• performance notebook improvements (#5217) @acostadon
• Update rapids-build-backend to 0.4.1 (#5211) @KyleFromNVIDIA
• Concurrent multisource backwardpass (#5206) @HowardHuang1
• ci(labeler): update labeler action to @v5 (#5205) @gforsyth
• Allows Dataset method get_edgelist() to be called on a CPU-only system (#5196) @rlratzel
• cugraph RAPIDS dependency now use the new rapids-cmake branch info (#5195) @robertmaynard
• cleanup and dropping NX (#5194) @BradReesWork
• Allow latest OS in devcontainers (#5188) @bdice
• Update build infra to support new branching strategy (#5172) @robertmaynard
• Temporal Graph Construction C API (#5161) @ChuckHastings
• Re-enable spectral clustering (#5151) @jnke2016
• Improve betweenness centrality performance (#5146) @HowardHuang1
• Use GCC 14 in conda builds. (#5124) @vyasr

[NIGHTLY] v25.12.00

🔗 Links

• Development Branch
• Compare with main branch

🚨 Breaking Changes

• Add support for other temporal comparisons (#5283) @ChuckHastings

🐛 Bug Fixes

• Use pinned_host_memory_resource for device-accessible memory in renumbering.cu (#5317) @bdice
• Fix warning in unreachable device code (#5316) @bdice
• Fix G.nodes() returning internal vertex IDs (#5304) @eriknw
• Fix force_atlas2 on empty graphs (#5303) @eriknw
• Skip failing dataset test for now (#5301) @ChuckHastings

📖 Documentation

• Use Docker CUDA major-version tags (#5291) @bdice

🛠️ Improvements

• Enable sccache-dist connection pool (#5314) @trxcllnt
• Use pinned_host_memory_resource instead of pinned_memory_resource. (#5312) @bdice
• Update to rapids-logger 0.2 (#5298) @bdice
• force_atlas2: allow vertex mass to be given by the user (#5293) @eriknw
• Update for cudf.DataFrame.from_pandas deprecation (#5286) @TomAugspurger
• Add support for other temporal comparisons (#5283) @ChuckHastings
• Update RAPIDS_BRANCH, codify changes in update-version.sh (#5278) @KyleFromNVIDIA

v25.08.00

🚨 Breaking Changes

• Move gnn.comms to PLC (#5166) @nv-rliu
• Update decompress_to_edgelist to take an optional large buffer type as an input parameter. (#5165) @seunghwak
• Optionally create a graph object in a large memory buffer (#5148) @seunghwak
• Compile Optimizations - batch 3 (#5119) @ChuckHastings
• Optionally store the generated edge list in pinned memory buffer (#5116) @seunghwak
• Resolve inconsistencies in algorithms that do not support self-loops & multi-edges. (#5109) @seunghwak
• Use pytest-xdist on Python Tests (#5102) @nv-rliu
• Remove CUDA 11 from dependencies.yaml (#5101) @KyleFromNVIDIA
• stop uploading packages to downloads.rapids.ai (#5096) @jameslamb

🐛 Bug Fixes

• [HOTFIX] Skips PG benchmarks for cyber datasets (#5209) @rlratzel
• Fix sampling bug introduced by compile time optimizations (#5199) @ChuckHastings
• [BUG] Fix Broken gnn.comms Import (#5183) @alexbarghi-nv
• Bug fix in assert() (#5142) @seunghwak
• Fix build error with gcc 14. (#5127) @seunghwak
• Fix Resultset Tests for Compat with xdist (#5125) @nv-rliu
• Allow Graph.from_numpy_array to accept nodes that are types not supported by cupy (#5123) @mroeschke
• Adding GH_TOKEN pass-through to summarize job (#5118) @msarahan
• [BUG] Fix Comms Shutdown Function (#5115) @alexbarghi-nv
• Add support for isolated vertices and fix bug in the link prediction algorithms (#5110) @jnke2016
• Update BC computation to address normalization edge conditions (#5105) @ChuckHastings
• Fix pylibcudf dependency (#5085) @vyasr

📖 Documentation

• remove references to DGL (#5197) @jameslamb
• redirecting GNN content to the docs pages and fixed libcugraph link (#5168) @acostadon
• [DOC] Remove DGL Reference from README (#5167) @alexbarghi-nv
• removed cuda 11 reference (#5163) @acostadon
• add docs on CI workflow inputs (#5139) @jameslamb

🚀 New Features

• Optionally create a graph object in a large memory buffer (#5148) @seunghwak
• Optionally store the generated edge list in pinned memory buffer (#5116) @seunghwak
• Implement transform_gather_e (#5106) @seunghwak
• Add edge_multi_index_property_t (#5104) @seunghwak
• added new benchmark nb to run mg or sg with more algorithms (#4933) @acostadon

🛠️ Improvements

• Drop old CUDA pre-10 logic (#5187) @jakirkham
• Disable codecov comments (#5185) @bdice
• Updates UCX test scripts and Dockerfile for CUDA 12.x (#5174) @rlratzel
• Update update_edge_src|dst_property behavior (with vertex list) (#5173) @seunghwak
• Move gnn.comms to PLC (#5166) @nv-rliu
• Update decompress_to_edgelist to take an optional large buffer type as an input parameter. (#5165) @seunghwak
• fix(docker): use versioned -latest tag for all rapidsai images (#5164) @gforsyth
• Remove cuda11 reference from CMake file (#5162) @ChuckHastings
• Relax PyTorch pinning to >=2.3. (#5159) @bdice
• [IMP] Allow Metadata Specification in Distributed Sampler (#5152) @alexbarghi-nv
• Add test for Cosine Similarity algorithm (#5150) @jnke2016
• Fix calls to Client.scheduler_info (#5147) @TomAugspurger
• Update MST test (#5143) @jnke2016
• Use CUDA 12.9 in Conda, Devcontainers, Spark, GHA, etc. (#5137) @jakirkham
• Correct C API test for uniform neighborhood sampling (#5136) @ChuckHastings
• Re-enable cugraph service test (#5134) @jnke2016
• First set of updates - Centrality and community (#5133) @BradReesWork
• Compile Optimizations - batch 4 (#5132) @ChuckHastings
• Remove nvidia and dask channels (#5129) @vyasr
• Temporarily Disable Cugraph Service test (#5128) @jnke2016
• refactor(rattler): remove cuda11 options and general cleanup (#5126) @gforsyth
• Remove the deprecated random walks API (#5122) @jnke2016
• Remove deprecation warning when symmetrizing the edgelist (#5121) @jnke2016
• Compile Optimizations - batch 3 (#5119) @ChuckHastings
• Compile optimizations - batch 2 (#5117) @ChuckHastings
• Refactor cugraph.datasets tests to be xdist compatible (#5114) @nv-rliu
• Compilation optimizations - batch 1 (#5112) @ChuckHastings
• Resolve inconsistencies in algorithms that do not support self-loops & multi-edges. (#5109) @seunghwak
• Use pytest-xdist on Python Tests (#5102) @nv-rliu
• Remove CUDA 11 from dependencies.yaml (#5101) @KyleFromNVIDIA
• Remove CUDA 11 devcontainers and update CI scripts (#5100) @bdice
• Require Cython>=3.1.2 (#5097) @vyasr
• stop uploading packages to downloads.rapids.ai (#5096) @jameslamb
• Branch 25.08 merge 25.06 (#5090) @rlratzel
• Forward-merge branch-25.06 into branch-25.08 (#5079) @gforsyth
• Forward-merge branch-25.06 into branch-25.08 (#5062) @gforsyth
• Temporal sampling implementation (#4994) @ChuckHastings