| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| uccl-0.1.1.tar.gz | 2026-05-10 | 4.8 MB | |
| uccl-0.1.1-cp312-abi3-manylinux_2_35_x86_64.whl | 2026-05-10 | 12.9 MB | |
| uccl-0.1.1+cu13.efa-cp312-abi3-manylinux_2_35_x86_64.whl | 2026-05-10 | 42.4 MB | |
| uccl-0.1.1+rocm-cp312-abi3-manylinux_2_35_x86_64.whl | 2026-05-10 | 15.7 MB | |
| uccl-0.1.1+cu13-cp312-abi3-manylinux_2_35_x86_64.whl | 2026-05-10 | 12.6 MB | |
| uccl-0.1.1+cu12.efa-cp312-abi3-manylinux_2_35_x86_64.whl | 2026-05-10 | 56.1 MB | |
| README.md | 2026-05-10 | 71.3 kB | |
| UCCL v0.1.1 source code.tar.gz | 2026-05-10 | 4.8 MB | |
| UCCL v0.1.1 source code.zip | 2026-05-10 | 5.4 MB | |
| Totals: 9 Items | 154.8 MB | 0 | |
What's Changed
New Features
- Introduce UCCL_PARAM by @zhongjiechen in https://github.com/uccl-project/uccl/pull/76
- P2P all gather collective by @viralbhadeshiya in https://github.com/uccl-project/uccl/pull/580
Other Changes
- renaming libnccl-net.so to libnccl-net-uccl.so for rdma_cuda, so that… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/7
- [EFA] Enable per-GPU proxy and shared-GPU proxy by @MaoZiming in https://github.com/uccl-project/uccl/pull/5
- Update README with citation by @MaoZiming in https://github.com/uccl-project/uccl/pull/6
- Update README.md Typo by @MaoZiming in https://github.com/uccl-project/uccl/pull/8
- fixing readme to be high-level and extensbile; add instruction for ru… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/9
- Update README.md Typo by @MaoZiming in https://github.com/uccl-project/uccl/pull/11
- Adding test script for using CPU proxy to initiate GPUDirect Communication by @MaoZiming in https://github.com/uccl-project/uccl/pull/15
- Add Uccl slides and more performance figures by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/21
- Added clang format by @MaoZiming in https://github.com/uccl-project/uccl/pull/23
- format cuda files by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/24
- Fig and slides by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/27
- Test and evaluate on 6 HGX nodes over 400G RoCE across two racks by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/20
- Added push workflow for formatting. by @MaoZiming in https://github.com/uccl-project/uccl/pull/29
- Unify cuda and hip by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/34
- P2P Transfer based on GPUDirect and CPU Proxy by @MaoZiming in https://github.com/uccl-project/uccl/pull/33
- Merge gpu_driven directory into p2p by @MaoZiming in https://github.com/uccl-project/uccl/pull/35
- fixing afxdp build by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/43
- Refactor
node_ipsinto a folder. by @MaoZiming in https://github.com/uccl-project/uccl/pull/40 - Create the uccl_home on other nodes if doesn't exist by @MaoZiming in https://github.com/uccl-project/uccl/pull/39
- fixing format and readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/45
- update fig font by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/46
- P2P add in CPU Proxy notifies GPU kernel command completion by @MaoZiming in https://github.com/uccl-project/uccl/pull/36
- Utils Refactoring by @MaoZiming in https://github.com/uccl-project/uccl/pull/42
- Adding explanation to the sudden drop in allreduce on 6 hgx by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/48
- Fixing font by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/49
- Refine roadmap by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/50
- Adding write with immediate for remote proxy nvlink forwarding by @MaoZiming in https://github.com/uccl-project/uccl/pull/47
- Minor format.sh fix by @MaoZiming in https://github.com/uccl-project/uccl/pull/52
- Reorder perf figs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/53
- fixing misc build by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/56
- Tpu by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/61
- specify more files for building in afxdp by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/62
- Optimize connection setup and refactor QP usgae for rdma. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/65
- Enabling multi-process for IB cluster by @MaoZiming in https://github.com/uccl-project/uccl/pull/60
- Remote CPU Proxy NVLink-based forwarding by @MaoZiming in https://github.com/uccl-project/uccl/pull/51
- Fix connection setup bug. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/67
- supporting lazy create in transport_test, make rdma_test work on ib by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/68
- Simple KV transfer engine exposed via pybind11 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/63
- Make UCCL easy to install by @praveingk in https://github.com/uccl-project/uccl/pull/59
- Adding benchmark scripts for kvcache transfer and fix cuda issue in pybind by @MaoZiming in https://github.com/uccl-project/uccl/pull/70
- Remove endif in transport.cc by @MaoZiming in https://github.com/uccl-project/uccl/pull/78
- Fix bug by @zhongjiechen in https://github.com/uccl-project/uccl/pull/79
- Fix kvtrans because of the RDMAEndpoint API change. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/80
- Make UCCL work on AMD + Broadcom NICs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/64
- revert hostfile by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/83
- Environment Variables in UCCL. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/87
- Readme for P2P by @MaoZiming in https://github.com/uccl-project/uccl/pull/85
- Optimize remote scatter kernel + GPU-driven Synchronous execution by @MaoZiming in https://github.com/uccl-project/uccl/pull/71
- Optimize UCCL performance on AMD+Broadcom by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/84
- using https instead of git to allow git clone without ssh keys by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/91
- Update readme. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/94
- Add PyTorch examples for AMD GPUs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/97
- Refactoring Readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/98
- Persistent Kernel for NVLink forwarding + Batching multiple tokens in one rdma write. by @MaoZiming in https://github.com/uccl-project/uccl/pull/88
- auto detect the best NIC based on GPU rank by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/100
- Rename KVTrans to general P2P by @MaoZiming in https://github.com/uccl-project/uccl/pull/102
- Rename p2p and kvtrans folders by @MaoZiming in https://github.com/uccl-project/uccl/pull/103
- kvtrans chunking by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/81
- UCCL_IB_HCA, UCCL_SOCKET_IFNAME by @zhongjiechen in https://github.com/uccl-project/uccl/pull/105
- Refactor rdma code by @zhongjiechen in https://github.com/uccl-project/uccl/pull/106
- Update rdma readme. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/107
- P2P rendezvous and register membership via Redis by @MaoZiming in https://github.com/uccl-project/uccl/pull/101
- fixing build bugs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/111
- Fix arm installation (for GH-200) by @MaoZiming in https://github.com/uccl-project/uccl/pull/112
- Run UCCL on GH200 by @MaoZiming in https://github.com/uccl-project/uccl/pull/113
- initial pip upload by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/104
- Badge to Readme by @MaoZiming in https://github.com/uccl-project/uccl/pull/126
- removing Macro of BROADCOM_NIC by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/128
gpu_drivenon GH200 by @MaoZiming in https://github.com/uccl-project/uccl/pull/122- Add pip for EFA and restructure readme based on pip by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/129
- Add pip for GH by @MaoZiming in https://github.com/uccl-project/uccl/pull/127
- Better logging by @zhongjiechen in https://github.com/uccl-project/uccl/pull/130
- Increase the resource log level to 1. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/133
- add nccl-sg to pip by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/136
- fixing readme typo by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/141
- Refactoring
gpu-drivenby @MaoZiming in https://github.com/uccl-project/uccl/pull/142 - auto-detect ib and ctrl nic similar to nccl by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/143
- Fix bugs in transport by @zhongjiechen in https://github.com/uccl-project/uccl/pull/145
- Making P2P pip-installable by @MaoZiming in https://github.com/uccl-project/uccl/pull/134
- Making python binding react to ctrl+c and add async send/recv by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/144
- readme for multi-tensor p2p transfer by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/148
- Fix CPU check. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/151
- small fix on incast exp setup by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/152
- Add comments for retransmission logic in rdma, as it is complex. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/156
- Fix benchmark script with installation by @praveingk in https://github.com/uccl-project/uccl/pull/149
- add async transfer for p2p by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/158
- switch to ephemeral port for efa by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/165
- Fair NIC allocation by @praveingk in https://github.com/uccl-project/uccl/pull/160
- New logo by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/168
- Fixed a retransmission BUG. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/169
- p2p flush endpoint metadata by @MaoZiming in https://github.com/uccl-project/uccl/pull/164
- fixing incast readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/173
- minor fix of ARCH_SUFFIX in docker_build_and_install by @MaoZiming in https://github.com/uccl-project/uccl/pull/171
- avoiding specifying any listen port for nccl plugin by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/174
- Fix docker build on amd by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/175
- Update docker_build_and_install.sh from post2 to post3 by @MaoZiming in https://github.com/uccl-project/uccl/pull/178
- Removing the need of specifying ROCE_GID_IDX or IB_GID_IDX by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/179
- Scale RID and FID by @zhongjiechen in https://github.com/uccl-project/uccl/pull/183
- P2p add an API for direct connection with metadata. by @MaoZiming in https://github.com/uccl-project/uccl/pull/177
- Optimizing nixl benchmark by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/180
- Add benchmark for nixl with mooncake backend by @praveingk in https://github.com/uccl-project/uccl/pull/170
- Uccl p2p dual by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/185
- Adding regv API for P2P by @MaoZiming in https://github.com/uccl-project/uccl/pull/184
- fixing efa examples by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/192
- small fix on typo in efa doc by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/193
- borrow nccl style of reg buffer for relaxed ordering using ibv_reg_mr… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/200
- make p2p work on amd by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/195
- fixing relaxed order capacity detection by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/201
- GPU p2p on gh200 now works using ibv_reg_mr_iova2 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/202
- Adding read API to uccl by @MaoZiming in https://github.com/uccl-project/uccl/pull/191
- Add error warning under uccl_regmr by @MaoZiming in https://github.com/uccl-project/uccl/pull/190
- using torchrun for benchmark_uccl by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/204
- Set RC mode env variable for P2P read by @MaoZiming in https://github.com/uccl-project/uccl/pull/206
- Add Python formatter with Black by @MaoZiming in https://github.com/uccl-project/uccl/pull/209
- refine docker build for p2p by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/207
- Add Nvlink copy for p2p by @MaoZiming in https://github.com/uccl-project/uccl/pull/208
- simplify recv logic by assuming recv size is know OOB by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/212
- small fix on ep init using arg-provided num_cpus by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/214
- specifying num kvblocks for p2p benchmarks by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/216
- benchmarking read, migrating read to use torchrun by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/218
- P2p read by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/219
- all p2p tests pass on amd by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/221
- Add mooncake build instructions by @praveingk in https://github.com/uccl-project/uccl/pull/189
- fixing run_nccl_test.sh and run_ib.sh by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/224
- Refactor gpu_driven proxy class by @MaoZiming in https://github.com/uccl-project/uccl/pull/211
- Fix gpu_driven tests by @MaoZiming in https://github.com/uccl-project/uccl/pull/226
- Fix util.h minor compiler warning by @MaoZiming in https://github.com/uccl-project/uccl/pull/227
- simplify rdma installation guide by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/232
- [GPU-driven] Remove thread local and rename variables by @MaoZiming in https://github.com/uccl-project/uccl/pull/231
- GPU driven clean code and fix segfault by @MaoZiming in https://github.com/uccl-project/uccl/pull/233
- update supporter, add davis artsy lab, remove personal email but use … by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/234
- Add dual proxy to gpu-driven by @MaoZiming in https://github.com/uccl-project/uccl/pull/220
- Removing LAZY_CREATE_ENGINE for p2p and rdma, refactoring test_listen_fd (now p2p_listen_fd) by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/230
- Initial attempt to translate gpu_driven to amd. Two parts left: __builti… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/236
- rerun amd evaluation by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/240
- update to latest ucx by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/241
- Automatically detecting EFA devices and ENA devices by @steven4354 in https://github.com/uccl-project/uccl/pull/167
- Adding Python binding for GPU-driven by @MaoZiming in https://github.com/uccl-project/uccl/pull/235
- add cluster monitor by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/242
- Nit on p2p interfaces by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/239
- Add an initial copy-only ccl for sendrecv with pytorch.dist like APIs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/223
- transparently handle rcmode by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/244
- Adding gpu-driven to install script by @MaoZiming in https://github.com/uccl-project/uccl/pull/248
- Optimizing uccl collective performance by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/246
- basic DeepEP codebase by @HermesCui in https://github.com/uccl-project/uccl/pull/251
- Gpu driven support dispatch and combine by @MaoZiming in https://github.com/uccl-project/uccl/pull/253
- Minor: move
benchmark_remote.pyto bench folder by @MaoZiming in https://github.com/uccl-project/uccl/pull/254 - Add UCCL proxy initialization before dispatch/combine buffer by @MaoZiming in https://github.com/uccl-project/uccl/pull/255
- Refactor Python class bench, proxy, peer copy by @MaoZiming in https://github.com/uccl-project/uccl/pull/258
- quick fix on shared_pool by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/257
- Running nixl on amd+broadcom by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/252
- Add
buffer.pyand bring torch to uccl_ep.cc by @MaoZiming in https://github.com/uccl-project/uccl/pull/259 - removing efa, misc, rdma usage of dedicate hostfile by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/262
- Adding
internode_ll.cuto UCCL by @MaoZiming in https://github.com/uccl-project/uccl/pull/261 - Mooncake by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/265
- fixing dev detection bugs when specifying incomplete ib_hca by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/266
- Gpu driven uccl ibgda placeholders by @MaoZiming in https://github.com/uccl-project/uccl/pull/263
- Minor fix on gpu_driven compile by @MaoZiming in https://github.com/uccl-project/uccl/pull/268
- add dual directional test for nixl by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/269
- Make UCCL run Primus. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/247
- [gpu_driven] Fix LowLatencyBuffer and workspace_ by @MaoZiming in https://github.com/uccl-project/uccl/pull/270
- Minor gpu_driven remove redundant comments by @MaoZiming in https://github.com/uccl-project/uccl/pull/271
- Remove nvshmemi_get_p2p_ptr and nvshmemi_ibgda_device_state_t dependencies. by @MaoZiming in https://github.com/uccl-project/uccl/pull/272
- Passing rb to internode_ll.cu by @MaoZiming in https://github.com/uccl-project/uccl/pull/273
- Implement nvshmemi_ibgda_put_nbi_warp by @MaoZiming in https://github.com/uccl-project/uccl/pull/275
- Add normalization for GPU ID by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/277
- Add p5en by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/281
- Fix gpu nic mapping by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/282
- P2p collective by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/274
- allowing building p2p on aws by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/284
- tune some parameters for congestion-free clusters by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/283
- uccl ep enable gpu-driven dual proxy by @MaoZiming in https://github.com/uccl-project/uccl/pull/278
- Add P2P write semantics by @zhongjiechen in https://github.com/uccl-project/uccl/pull/291
- Add vector version of write/send/advertise. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/292
- Update readme for p2p. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/294
- Minor fix on torch run for gpu_driven benchmark by @MaoZiming in https://github.com/uccl-project/uccl/pull/297
- [Minor] fix cmd=0 check in GPU-driven by @MaoZiming in https://github.com/uccl-project/uccl/pull/298
- Automatically detecting ib_hca in test_internode_simple by @MaoZiming in https://github.com/uccl-project/uccl/pull/299
- [gpu_driven] Minor Fix benchmark_remote by @MaoZiming in https://github.com/uccl-project/uccl/pull/300
- Add APIs for NIXL's UCCL backend by @praveingk in https://github.com/uccl-project/uccl/pull/186
- Fix gpu_driven readme by @MaoZiming in https://github.com/uccl-project/uccl/pull/304
- add atomic test on gh200 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/306
- Clean REMOTE_PERSISTENT_KERNEL_FLAG by @MaoZiming in https://github.com/uccl-project/uccl/pull/305
- PR template by @MaoZiming in https://github.com/uccl-project/uccl/pull/308
- uccl_deepep address passing & atomic add by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/289
- [GPU-driven] Clean merge compile issues by @MaoZiming in https://github.com/uccl-project/uccl/pull/312
- Eliminate Nvshmem get p2p by @HermesCui in https://github.com/uccl-project/uccl/pull/287
- Multiple CPU proxies rendezvous and manage memory regions by @MaoZiming in https://github.com/uccl-project/uccl/pull/296
- Add
is_atomicinTransferCmdby @MaoZiming in https://github.com/uccl-project/uccl/pull/313 - Add compilation verification step to PR template by @MaoZiming in https://github.com/uccl-project/uccl/pull/317
- GPU-Driven for EFA by @zhongjiechen in https://github.com/uccl-project/uccl/pull/285
- Dynamically detecting EFA from makefile by @MaoZiming in https://github.com/uccl-project/uccl/pull/318
- test_internode_simple.py detect whether IB or efa by @MaoZiming in https://github.com/uccl-project/uccl/pull/319
- Gpu driven fix post recv and atomic dst_rank by @MaoZiming in https://github.com/uccl-project/uccl/pull/316
- Add Efa atomic and make DeepEP test work on EFA by @MaoZiming in https://github.com/uccl-project/uccl/pull/320
- UCCL build integration test by @MaoZiming in https://github.com/uccl-project/uccl/pull/321
- Build integration test by @MaoZiming in https://github.com/uccl-project/uccl/pull/322
- Integration test
rsyncandbuild_and_installall. by @MaoZiming in https://github.com/uccl-project/uccl/pull/323 - GPU driven offset fix by @MaoZiming in https://github.com/uccl-project/uccl/pull/324
- Update workflow to remove pull request trigger by @MaoZiming in https://github.com/uccl-project/uccl/pull/326
- Fix GPU-NIC mapping initialization by @jinqinn in https://github.com/uccl-project/uccl/pull/325
- Rename
gpu-drivenfolder toepby @MaoZiming in https://github.com/uccl-project/uccl/pull/327 - Combine
uccl_epwith pybind proxy by @MaoZiming in https://github.com/uccl-project/uccl/pull/328 - Fix ep build pybind, add helper
initialize_ucclanddetroy_uccl. by @MaoZiming in https://github.com/uccl-project/uccl/pull/330 - fixing rdma finding bug by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/331
- Updated the README to reflect the new folder name and added details about running DeepEP-style dispatch tests. by @MaoZiming in https://github.com/uccl-project/uccl/pull/332
- Add IPC mem handle cache to P2P; fix UDS connection error by @derekwin in https://github.com/uccl-project/uccl/pull/302
- Add
test_low_latency.pyto ep by @MaoZiming in https://github.com/uccl-project/uccl/pull/329 - Update README to avoid mismatching python version issue by @zhangnju in https://github.com/uccl-project/uccl/pull/340
- removing vscode config to be generic to all editors by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/343
- removing redis from p2p, restructure function order and name in engin… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/344
- add target for build by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/345
- Adding write/read_ipc and its async versions by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/349
- add efa_nv_peermem by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/353
- Add
requirements.txttoepby @MaoZiming in https://github.com/uccl-project/uccl/pull/350 - [EP] Fix EP bugs and run
test_low_latency.pyby @MaoZiming in https://github.com/uccl-project/uccl/pull/352 - adding initial transfer apis by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/354
- Run benchmarks in
test_low_latency.pyby @MaoZiming in https://github.com/uccl-project/uccl/pull/358 - EP debug gpu zombies by @MaoZiming in https://github.com/uccl-project/uccl/pull/363
- Move PTX instructions to
ep_utilsby @MaoZiming in https://github.com/uccl-project/uccl/pull/365 - Debugging transfer API bugs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/362
- clean up: merge jrings by @monopodium in https://github.com/uccl-project/uccl/pull/361
- Support NIXL_READ in uccl by @praveingk in https://github.com/uccl-project/uccl/pull/357
- EP remove
time.waitfor combine by @MaoZiming in https://github.com/uccl-project/uccl/pull/366 - Add target 'therock' for ROCm through TheRock project. by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/364
- removing redis dep by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/368
- Minor: ep rename
sm_idtowarp_idby @MaoZiming in https://github.com/uccl-project/uccl/pull/369 - Move
internodeto uccl ep by @MaoZiming in https://github.com/uccl-project/uccl/pull/334 - Making ep work on gh200 by @MaoZiming in https://github.com/uccl-project/uccl/pull/367
- EP: fix uint8 numeric problem. by @MaoZiming in https://github.com/uccl-project/uccl/pull/371
- Enrich readme with collectives, p2p, and ep by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/372
- Initialize
intranodein uccl ep by @MaoZiming in https://github.com/uccl-project/uccl/pull/333 - Add eccl by @derekwin in https://github.com/uccl-project/uccl/pull/375
- EP Fix bench kineto by @MaoZiming in https://github.com/uccl-project/uccl/pull/379
- EP fix newline compile warning by @MaoZiming in https://github.com/uccl-project/uccl/pull/381
- Add Primus as a submodule and provide doc for running it. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/280
- quick fix on AMD build by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/382
- Cleanup for nixl plugin by @praveingk in https://github.com/uccl-project/uccl/pull/383
- Debug why p5en two GPU internode ll only has 3GB/s bandwidth by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/374
- Making intra-node via IPC work for low-latency by @MaoZiming in https://github.com/uccl-project/uccl/pull/380
- Minor tweaks to therock build by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/388
- quick fix on amd readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/389
- fix circular import on amd by @derekwin in https://github.com/uccl-project/uccl/pull/390
- support for building with different cuda/rocm versions by @derekwin in https://github.com/uccl-project/uccl/pull/391
- moving rdma, afxdp, efa into collective folders by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/393
- fix p2p test bug by @derekwin in https://github.com/uccl-project/uccl/pull/395
- EP: track counter in cpu proxy to prevent reordering by @MaoZiming in https://github.com/uccl-project/uccl/pull/387
- Intranode test ll by @HermesCui in https://github.com/uccl-project/uccl/pull/384
- Use multiple NICs for EP by @zhongjiechen in https://github.com/uccl-project/uccl/pull/378
- EP: rename block_idx to thread_idx by @MaoZiming in https://github.com/uccl-project/uccl/pull/397
- EP clean immediate and postponed atomics logics by @MaoZiming in https://github.com/uccl-project/uccl/pull/398
- Multi-Writer-Multi-Reader communication by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/373
- rename MakefileHip to Makefile.rocm, Hip.therock to .therock by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/400
- feat: add async sendv and recvv function by @FirwoodLin in https://github.com/uccl-project/uccl/pull/351
- Support st_na and ld_nc by @HermesCui in https://github.com/uccl-project/uccl/pull/402
- Warp-to-ring_buffer mapping optimization by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/401
- EP: sender imposes barrier by @MaoZiming in https://github.com/uccl-project/uccl/pull/399
- EP: minor, remove guard and lock for ep by @MaoZiming in https://github.com/uccl-project/uccl/pull/404
- make benchmark_remote run by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/405
- EP: Enhance README with API examples and benchmarks by @MaoZiming in https://github.com/uccl-project/uccl/pull/407
- Add notif support for vllm 's nixl_connector by @praveingk in https://github.com/uccl-project/uccl/pull/409
- EP: Fix GH200 & automatically detecting down NIC by @MaoZiming in https://github.com/uccl-project/uccl/pull/408
- Fixing failing CI test on uccl build by @MaoZiming in https://github.com/uccl-project/uccl/pull/410
- try to run nccl on tcpx by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/403
- Adding new EP CI tests. by @MaoZiming in https://github.com/uccl-project/uccl/pull/411
- EP: fix test intranode by @MaoZiming in https://github.com/uccl-project/uccl/pull/412
- Running original deepep by @MaoZiming in https://github.com/uccl-project/uccl/pull/413
- fixing cite in readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/416
- compare performance of NCCL and UCCL copy-only collective && fix GIL problem when building connection by @monopodium in https://github.com/uccl-project/uccl/pull/386
- feat: Add TCPX-based GPU-to-GPU data transfer implementation by @DanielDanyang in https://github.com/uccl-project/uccl/pull/385
- fixing benchmark_uccl_read/write error 12 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/419
- Stabilize number in DeepEP dispatch and combine phase by @viralbhadeshiya in https://github.com/uccl-project/uccl/pull/418
- try to merge redundant code in transport.cc by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/420
- Add TCPX P2P Performance Test and Fix Unpack Kernel by @DanielDanyang in https://github.com/uccl-project/uccl/pull/421
- Ep normal by @MaoZiming in https://github.com/uccl-project/uccl/pull/415
- Fix benchmark_remote test by @MaoZiming in https://github.com/uccl-project/uccl/pull/423
- add pin numa to avoid pin cpu that might suffer from core contention by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/424
- Add Batched transfer metadata exchange for NIXL plugin by @praveingk in https://github.com/uccl-project/uccl/pull/428
- CPU init RDMA all-to-all test by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/422
- Add TCPX P2P Multi-NIC Performance Test by @DanielDanyang in https://github.com/uccl-project/uccl/pull/427
- Running ep benchmarks by @MaoZiming in https://github.com/uccl-project/uccl/pull/430
- Changes to support nixl test by @praveingk in https://github.com/uccl-project/uccl/pull/431
- RDMA rail-optimized (no cross-rail traffic) all-to-all results by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/435
- Fix mix comm by @monopodium in https://github.com/uccl-project/uccl/pull/436
- TCPX Unpack Kernel Optimization by @DanielDanyang in https://github.com/uccl-project/uccl/pull/437
- fix: small bug of last commit by @monopodium in https://github.com/uccl-project/uccl/pull/439
- update latest slides by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/442
- fixing readme format by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/443
- Update "therock" target to build in a PyPA "manylinux" container by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/438
- Disable push and pull_request triggers for low-latency test by @MaoZiming in https://github.com/uccl-project/uccl/pull/447
- Tuning combine performance by @MaoZiming in https://github.com/uccl-project/uccl/pull/433
- Testing on larger cx7 testbed by @MaoZiming in https://github.com/uccl-project/uccl/pull/441
- deepep on cx7 by @MaoZiming in https://github.com/uccl-project/uccl/pull/444
- TCPX P2P Library: Modular Architecture Refactor by @DanielDanyang in https://github.com/uccl-project/uccl/pull/448
- Remove unhelpful combine tuning by @MaoZiming in https://github.com/uccl-project/uccl/pull/451
- removing p4d by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/452
- adding mscclpp version of FIFO by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/454
- benchmark_remote with FIFO vs RingBuffer by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/455
- fixing miniconda too old glic by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/456
- [EP] Improve combine - receiver buffer atomics, improve
notify_gpu_completionby @MaoZiming in https://github.com/uccl-project/uccl/pull/453 - renaming bench_fifo by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/458
- Fixed benchmark remote by @MaoZiming in https://github.com/uccl-project/uccl/pull/459
- Reduce kMaxInflight from 32 to 8 by @MaoZiming in https://github.com/uccl-project/uccl/pull/461
- feat: simplify TransferCmd to 128 bits by @MaoZiming in https://github.com/uccl-project/uccl/pull/460
- Replace Uccl p2p intra comm with dist nccl backend by @monopodium in https://github.com/uccl-project/uccl/pull/446
- feat: Extend mscclpp fifo for EP by @MaoZiming in https://github.com/uccl-project/uccl/pull/465
- [Typo] Fix typo async by @MaoZiming in https://github.com/uccl-project/uccl/pull/467
- feat: EP port to amd by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/457
- fixing CUDA_ARCH bugs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/469
- [EP]: UCCL_IB_HCA, removing libnuma, better query sysfs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/470
- fixing errno 110 with auto-detected right gid_idex by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/472
- using UCCL_IB_HCA and more by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/473
- add rocm7 install to readme.md by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/474
- [EP] minor fix on EP benchmarks by @MaoZiming in https://github.com/uccl-project/uccl/pull/475
- [EP] Minor fix on readme by @MaoZiming in https://github.com/uccl-project/uccl/pull/476
- [EP] Fine tune readme, adding tldr by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/477
- simplest memory region register by @monopodium in https://github.com/uccl-project/uccl/pull/462
- fixing install deps by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/478
- fixing rocm build by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/479
- [EP] Update README with environmental variable setup by @MaoZiming in https://github.com/uccl-project/uccl/pull/480
- [EP] Typo fix in readme by @MaoZiming in https://github.com/uccl-project/uccl/pull/481
- [EP] Integrated nvshmem and torch all 2all as our baseline test by @HermesCui in https://github.com/uccl-project/uccl/pull/425
- Ep readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/482
- add links to homereadme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/484
- Change UCCL backend name in nixl benchmark by @praveingk in https://github.com/uccl-project/uccl/pull/483
- Tcpx plugin by @derekwin in https://github.com/uccl-project/uccl/pull/463
- Remove gpu_idx from uccl p2p intitialization by @praveingk in https://github.com/uccl-project/uccl/pull/489
- [EP] push initialization, destroy, rdma_buffer to buffer by @MaoZiming in https://github.com/uccl-project/uccl/pull/491
- [EP] remove
MAKE_NORMAL_MODEmacro by @MaoZiming in https://github.com/uccl-project/uccl/pull/490 - [EP] Minor: fix stall. by @MaoZiming in https://github.com/uccl-project/uccl/pull/497
- [EP] Fix compile issue on AMD by @MaoZiming in https://github.com/uccl-project/uccl/pull/500
- [EP] fix wr_id_to_wr_ids on non-EFA by @MaoZiming in https://github.com/uccl-project/uccl/pull/501
- [CI] Remove stale low-latency CI test by @MaoZiming in https://github.com/uccl-project/uccl/pull/502
- [EP] subset barrier by @MaoZiming in https://github.com/uccl-project/uccl/pull/498
- [EP] fix intranode.py by @MaoZiming in https://github.com/uccl-project/uccl/pull/504
- Fix missing install requirement on intervaltree. by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/508
- [P2P] Infer device (gpu/net) after memory registration by @praveingk in https://github.com/uccl-project/uccl/pull/509
- add async version of read write v by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/513
- Switch TheRock PyPI URL from nightlies to prerelease channel. by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/514
- [EP] port internode kernel to amd. by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/506
- [EP] Remove debug print for RDMASenderCoordinator by @MaoZiming in https://github.com/uccl-project/uccl/pull/515
- Fix wheel name parsing and python-packaged ROCm suffix. by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/517
- [EP] README with dispatch/combine latency by @MaoZiming in https://github.com/uccl-project/uccl/pull/516
- [Collective] update rccl building instructions for rocm7 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/520
- fix destory mr by @derekwin in https://github.com/uccl-project/uccl/pull/524
- [EP] Trying to optimizing performance by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/505
- [EP] Adding some utils to EP by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/527
- [P2P] fixing alignment issues in p2p collective by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/528
- P2p intranode by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/529
- [EP] Porting to p6-b200 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/531
- [EP] Fix kMaxOutstandingRecvs in common.hpp by @MaoZiming in https://github.com/uccl-project/uccl/pull/533
- [EP] Cleanup unused files and variables by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/535
- [EP] moving tcp listen port negotiation into initialize_uccl by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/537
- [EP] fixing tcp rendezous bugs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/539
- removing rkey!=0 check by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/540
- [ALL] changing DCHECK to CHECK for necessary operations inside DCHECK() by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/545
- [P2P] Full TCPX Plugin for NIXL by @DanielDanyang in https://github.com/uccl-project/uccl/pull/521
- [EP] fixing ep building by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/546
- [EP] fixing connection bugs by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/547
- [EP] fixing normal runs on cx7 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/549
- [EP] fixing cpu alltoall for p6 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/550
- [EP] tuning combine performance, barely reaching 20GB/s but require t… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/552
- [EP] Refine building in setup.py by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/553
- feat: add DPDK packet I/O implementation by @Nelson-Cheung in https://github.com/uccl-project/uccl/pull/450
- Rename to uccl p2p, and install in pip by @praveingk in https://github.com/uccl-project/uccl/pull/543
- [doc] fixing readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/555
- [EP] Support sglang by @MaoZiming in https://github.com/uccl-project/uccl/pull/496
- Add detection of stale builder image. by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/556
- [EP] Fixing ep docker b200 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/558
- [P2P] Multi-channel TCPX engine by @DanielDanyang in https://github.com/uccl-project/uccl/pull/559
- [EP] Fix missing env handling in utils module by @MaoZiming in https://github.com/uccl-project/uccl/pull/561
- Bench nvshmem3.4.5 vs nvshmem 3.5.0 by @HermesCui in https://github.com/uccl-project/uccl/pull/536
- [P2P] NCCL version of the NIXL plugin by @DanielDanyang in https://github.com/uccl-project/uccl/pull/560
- [P2P]Changes to support NIXL CI/CD and benchmarks by @praveingk in https://github.com/uccl-project/uccl/pull/563
- EP: optimize low-latency kernel on amd-gpu. by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/565
- [P2P] NCCL_TCPX multi-GPUs bug fix by @derekwin in https://github.com/uccl-project/uccl/pull/566
- [P2P]Support notifs of large objects by @praveingk in https://github.com/uccl-project/uccl/pull/567
- EP: fix get_rdma_buffer on cuda platform. by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/574
- Fix missing semicolon by @pbelevich in https://github.com/uccl-project/uccl/pull/576
- [EP] debugging amd normal by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/548
- EP: auto-detect rocm arch by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/579
- [EP] Nebius init script by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/572
- Support rcmode for write operations by @praveingk in https://github.com/uccl-project/uccl/pull/577
- [P2P] add intervaltree to requirements by @monopodium in https://github.com/uccl-project/uccl/pull/526
- Fifobench tests by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/584
- [EP] Fixing sender barrier by @MaoZiming in https://github.com/uccl-project/uccl/pull/583
- [EP] Debug EP=32 DeepEP on Sglang by @MaoZiming in https://github.com/uccl-project/uccl/pull/562
- [eccl] update by @derekwin in https://github.com/uccl-project/uccl/pull/585
- [EP] load SourceMeta with ld_nc_global by @MaoZiming in https://github.com/uccl-project/uccl/pull/586
- WIP: first version efa p2p by @monopodium in https://github.com/uccl-project/uccl/pull/571
- [P2P] fixing readme and default to NCCL/RCCL for intranode comm (as the current copy engine is very slow on AMD) by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/587
- [P2P] introduce P2Pop and batch_isend_irecv, fixing isend and irecv s… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/588
- move misc, eccl to experimental by @derekwin in https://github.com/uccl-project/uccl/pull/591
- EP: support internode combine on amd gpus. by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/582
- [EP] Revise README for UCCL-EP communication by @MaoZiming in https://github.com/uccl-project/uccl/pull/594
- [EP] Add AMD results to EP README by @MaoZiming in https://github.com/uccl-project/uccl/pull/595
- [EP] Fix typo in README title by @MaoZiming in https://github.com/uccl-project/uccl/pull/596
- Fix consistency check by @praveingk in https://github.com/uccl-project/uccl/pull/598
- [DOC] Revise Road Map for EP by @MaoZiming in https://github.com/uccl-project/uccl/pull/599
- [eccl] Add gdrcopy-based cpu->gpu fifo by @derekwin in https://github.com/uccl-project/uccl/pull/602
- [P2P] Fix EFA build for NIXL CI by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/603
- [DOC] add more performance figure by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/604
- Update citations in README.md by @MaoZiming in https://github.com/uccl-project/uccl/pull/606
- [Build] Fix Makefile header dependency tracking by @yurekami in https://github.com/uccl-project/uccl/pull/608
- [Examples] Add NVIDIA CUDA examples for PyTorch DDP training by @yurekami in https://github.com/uccl-project/uccl/pull/607
- [FIX] Gracefully handle missing InfiniBand directory by @yurekami in https://github.com/uccl-project/uccl/pull/611
- [P2P] Adding p2p-native RDMA support for both EFA and non-EFA (IB) by @zhongjiechen in https://github.com/uccl-project/uccl/pull/597
- EP: Enable CUDA Graph for internode dispatch by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/620
- [eccl] add IpcEndpoint & UdsExchanger by @derekwin in https://github.com/uccl-project/uccl/pull/622
- [p2p]add dietgpu to third party by @monopodium in https://github.com/uccl-project/uccl/pull/623
- [EP] Support vLLM; high-throughput and low-latency by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/493
- [EP] Minor, move SGLang scripts under bench folder by @MaoZiming in https://github.com/uccl-project/uccl/pull/625
- [EP] fix small error of ep build by @monopodium in https://github.com/uccl-project/uccl/pull/626
- [p2p] deitgpu amd verion by @monopodium in https://github.com/uccl-project/uccl/pull/627
- [P2P] nits on p2p bench parameters, cleanup comments by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/629
- [EP] Revise README for vLLM deployment and benchmarks by @MaoZiming in https://github.com/uccl-project/uccl/pull/630
- [P2P] Fix bug in P2P read/write. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/633
- [eccl] Refactor Task by @derekwin in https://github.com/uccl-project/uccl/pull/631
- [P2P] Support async transfer for readv and writev. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/634
- Change dest qpn from uint16_t to uint32_t by @manojgop in https://github.com/uccl-project/uccl/pull/635
- [eccl] fifo perf benchmark by @derekwin in https://github.com/uccl-project/uccl/pull/639
- [EP] adding env variables for uccl ep by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/642
- [P2P] clean tcpx by @derekwin in https://github.com/uccl-project/uccl/pull/636
- [EP] debug AMD mem consistency issues by @MaoZiming in https://github.com/uccl-project/uccl/pull/600
- [eccl] add multi fifos & benchmark by @derekwin in https://github.com/uccl-project/uccl/pull/643
- [UKernel] rename eccl to ukernel by @derekwin in https://github.com/uccl-project/uccl/pull/646
- [UK] Reorganize by @derekwin in https://github.com/uccl-project/uccl/pull/652
- [P2P] Fix cleanup crash when [readv/sendv]_async by @praveingk in https://github.com/uccl-project/uccl/pull/649
- Fix warnings about jring by @zhongjiechen in https://github.com/uccl-project/uccl/pull/654
- [P2P] Remove collective backend from P2P by @zhongjiechen in https://github.com/uccl-project/uccl/pull/624
- [P2P] Fix GID index by @zhongjiechen in https://github.com/uccl-project/uccl/pull/656
- [P2P] cleanup p2p prints, defines, etc by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/657
- [P2P] Add input check. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/658
- [P2P] Miss some input checks. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/659
- [P2P] Fix bugs in Endpoint API. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/660
- EP: fix combine hang when num_sms=24 or 20 by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/661
- [EP] Low latency pressure test and fix bug on EFA by @MaoZiming in https://github.com/uccl-project/uccl/pull/648
- [EP] add print for calc_diff by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/663
- EP: add Primus-Megatron launch script primus_cli by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/640
- [P2P] Fix sl and tc and add input check. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/666
- EP: fix ep compile error in new released rocm7.2 image by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/668
- [EP] refine env config by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/664
- [P2P] run_ep.sh ll should not do stress test by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/674
- bug: fix rdma internode test buffer overflow by @crazyguitar in https://github.com/uccl-project/uccl/pull/670
- [EP] Low-latency
bench_kinetoin rare condition fails to create some samples by @MaoZiming in https://github.com/uccl-project/uccl/pull/669 - [EP] p5 using all 4 nics by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/676
- Add Triton-based MoE pack/unpack kernels and corresponding tests by @liad-hub in https://github.com/uccl-project/uccl/pull/647
- [EP] disable aggressive atomic on amd by default, as it fails stress … by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/680
- [UK] cpu-side arch by @derekwin in https://github.com/uccl-project/uccl/pull/678
- EP: fix cached_notify when use num_sms>32 on amdgpu. by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/682
- [EP] Limit max inflight bytes per GPU and do IBV_SEND_SIGNALED for ea… by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/681
- [UK] clean Request; add notifier to communicator by @derekwin in https://github.com/uccl-project/uccl/pull/683
- [P2P] Add NIXL support with Native RDMA by @praveingk in https://github.com/uccl-project/uccl/pull/677
- [P2P] Fix the issue of numa_node_ not being initialized. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/687
- [UK] Add big Moe operator for the application by @DanielDanyang in https://github.com/uccl-project/uccl/pull/688
- [EFA] Fix segmentation fault error in EMULATE_RC_ZC by @manojgop in https://github.com/uccl-project/uccl/pull/686
- [P2P] Fix QP resource release. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/690
- [UK] notify fifo by @derekwin in https://github.com/uccl-project/uccl/pull/691
- [P2P] Don't use ibv_wr_xxx API for AMD NIC. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/693
- [rdma] Make CQ Ex flags configurable for NIC compatibility. by @manojgop in https://github.com/uccl-project/uccl/pull/689
- [UK] add sm-to-sm fifo by @derekwin in https://github.com/uccl-project/uccl/pull/696
- [UK] add transport p2p to scheduler by @derekwin in https://github.com/uccl-project/uccl/pull/698
- TK GEMM kernel integration draft by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/699
- collective(rdma): Configurable CQ IGNORE_OVERRUN via kEnableCQIgnoreO… by @manojgop in https://github.com/uccl-project/uccl/pull/701
- TK GEMM integration optimization by @CalebZ9909 in https://github.com/uccl-project/uccl/pull/700
- [P2P] flow control and fix high latency issue by @praveingk in https://github.com/uccl-project/uccl/pull/703
- collective(efa): Add MANAGED macro for NICs incompatible with cudaMal… by @manojgop in https://github.com/uccl-project/uccl/pull/702
- [P2P] TCP support based on NCCL by @DanielDanyang in https://github.com/uccl-project/uccl/pull/638
- [P2P] Don't use inline for RDMA read. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/708
- [P2P] Detect gid index by @praveingk in https://github.com/uccl-project/uccl/pull/711
- Build nccl_endpoint only on USE_TCP by @praveingk in https://github.com/uccl-project/uccl/pull/712
- [EP]: fix fp8 error of internode_ll on amd gfx950 arch. by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/710
- Support both shared memory and HBM for IOV access based on kMaxIovs. by @manojgop in https://github.com/uccl-project/uccl/pull/705
- collective/rdma: Add Intel RDMA NIC support with conditional CQ confi… by @manojgop in https://github.com/uccl-project/uccl/pull/713
- [UK] fix TK submodule by @derekwin in https://github.com/uccl-project/uccl/pull/717
- [P2P] Replaced UDS with Shm-JRing by @derekwin in https://github.com/uccl-project/uccl/pull/718
- collective(rdma and efa): Add Intel RDMA NIC support with conditional… by @manojgop in https://github.com/uccl-project/uccl/pull/714
- [P2P] API for Integrating with Ray by @zhongjiechen in https://github.com/uccl-project/uccl/pull/662
- [P2P] Fix compiler warnings. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/719
- collective: Modify build script to retag wheel platform for Rocky Lin… by @manojgop in https://github.com/uccl-project/uccl/pull/720
- [EP] bug fixed with worse performance by @HermesCui in https://github.com/uccl-project/uccl/pull/723
- Add podman support as alternative to docker for builds by @manojgop in https://github.com/uccl-project/uccl/pull/721
- [P2P] fix EFA device selection on p5 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/724
- [P2P] remove unused streams_ by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/725
- [EP] reorg EP bench folder structure by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/726
- [EP] fixing bench alltoall rail by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/727
- [P2p] make it work on gb10 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/729
- [EP] Add Intel RDMA NIC support for UCCL EP by @manojgop in https://github.com/uccl-project/uccl/pull/731
- Enable TCP for NIXL's UCCL backend by @praveingk in https://github.com/uccl-project/uccl/pull/730
- [P2P]Add uccl shared lib install by @praveingk in https://github.com/uccl-project/uccl/pull/733
- [EP] minor fix on cuda 13 for cudaMemPrefetchAsync signature change by @MaoZiming in https://github.com/uccl-project/uccl/pull/742
- [EP] fixing vllm run by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/740
- [EP] Support LL on GB10 by @MaoZiming in https://github.com/uccl-project/uccl/pull/741
- [EP] Add benchmarking option for FP8 dispatch. by @laochonlam in https://github.com/uccl-project/uccl/pull/743
- [EP] Add pplx-style low-latency benchmark entrypoint by @laochonlam in https://github.com/uccl-project/uccl/pull/744
- [EP] refine vllm scripts and readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/746
- [EP] Update vLLM results on README.md by @MaoZiming in https://github.com/uccl-project/uccl/pull/747
- CPU support for IPC operations by @liranschour in https://github.com/uccl-project/uccl/pull/736
- Improved onsided intra-node operations by @liranschour in https://github.com/uccl-project/uccl/pull/750
- [P2P] code clean by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/751
- [EP] Share RDMA context/PD/MR across proxy threads on same NIC by @manojgop in https://github.com/uccl-project/uccl/pull/737
- [Collective] test on l4 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/754
- Add stop_accept API by @praveingk in https://github.com/uccl-project/uccl/pull/755
- [EP] Improve LL performance by per-expert token batching by @laochonlam in https://github.com/uccl-project/uccl/pull/745
- [EP] fix dmabuf perf regression by @manojgop in https://github.com/uccl-project/uccl/pull/756
- [EP] added ci tests for test_low_latency.py by @ShawnWeiChew in https://github.com/uccl-project/uccl/pull/758
- Fix bug in EFA benchmarks BW calculation by @yehuday in https://github.com/uccl-project/uccl/pull/759
- [EP] test on actual synced folder and fixed duplicate artifact name error by @ShawnWeiChew in https://github.com/uccl-project/uccl/pull/762
- update to random instead of incast by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/763
- [EP] AMD IONIC: Fix context freed with active resources by @MaoZiming in https://github.com/uccl-project/uccl/pull/695
- [EP] Fix ibv_create_cq_ex for Intel irdma NIC by @manojgop in https://github.com/uccl-project/uccl/pull/765
- [P2P]fix problem caused by sock_fd check by @monopodium in https://github.com/uccl-project/uccl/pull/767
- [Build] a variety of refinement to make build sys easy to understand by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/768
- [Build] fixing some small issues by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/769
- [P2P] hiding intra and inter behind Ray-inspired APis by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/770
- [P2P] fix problem of torch.distributed hanging by @monopodium in https://github.com/uccl-project/uccl/pull/771
- [P2P] switch to cp38-abi3 to be compatible with any python version by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/772
- [EP]: chunked DMA-BUF MR registration for IOMMU mapping limits. by @manojgop in https://github.com/uccl-project/uccl/pull/761
- [Build] support format.sh on macOS by @laochonlam in https://github.com/uccl-project/uccl/pull/774
- [EP] fix H100 runtime bring-up failures by @laochonlam in https://github.com/uccl-project/uccl/pull/773
- [EP] Fix off13 overflow and PER_EXPERT_BATCHING receiver barrier stride by @whn09 in https://github.com/uccl-project/uccl/pull/766
- [EP] remove torch dependency from uccl_ep.cc by @MaoZiming in https://github.com/uccl-project/uccl/pull/781
- [EP] make sudo optional in install_deps script by @MaoZiming in https://github.com/uccl-project/uccl/pull/782
- [collective/EFA] Fix get actual devices by @manojgop in https://github.com/uccl-project/uccl/pull/777
- Set DSCP bits only in TOS for irdma by @andrzej-k in https://github.com/uccl-project/uccl/pull/780
- [Build] fix auditwheel repair failure on hosts with older glibc by @manojgop in https://github.com/uccl-project/uccl/pull/776
- [EP]: extend TransferCmd WRITE offset range from 4 GiB to 64 GiB by @manojgop in https://github.com/uccl-project/uccl/pull/785
- [EP] remove aten depedency fromm uccl_ep by @MaoZiming in https://github.com/uccl-project/uccl/pull/784
- [p2p] Add compression for send/recv by @monopodium in https://github.com/uccl-project/uccl/pull/778
- [EP] Fix brooken GB10 CI path by @MaoZiming in https://github.com/uccl-project/uccl/pull/788
- [EP] Fix low-latency mode offset by @MaoZiming in https://github.com/uccl-project/uccl/pull/791
- [util] add internal logging system by @ShawnWeiChew in https://github.com/uccl-project/uccl/pull/783
- Revert "[util] add internal logging system" by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/792
- [util] add internal logging system (sec try) by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/794
- [util] WARNING->WARN for simplicity by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/796
- [EP] GB10 docker build with cuda13 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/797
- [P2P] fix dietgpu and amd ep by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/798
- [EP] Switch from Pybind11 to nanobind by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/790
- [EP] add per-expert batching results by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/800
- [EP] fix HIP __syncwarp compatibility on ROCm by @laochonlam in https://github.com/uccl-project/uccl/pull/801
- [EP] Fix RoCE GID auto-detection for Broadcom bnxt_re NICs by @laochonlam in https://github.com/uccl-project/uccl/pull/803
- [P2P] Enable nixlbench CI for P2P by @praveingk in https://github.com/uccl-project/uccl/pull/806
- [Util] update logging behaviors by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/809
- [UK] integrate collecive/rdma transport layer by @derekwin in https://github.com/uccl-project/uccl/pull/808
- EQDS enabling by @andrzej-k in https://github.com/uccl-project/uccl/pull/807
- [P2P] Add APIs for Ray integration by @zhongjiechen in https://github.com/uccl-project/uccl/pull/786
- [P2P] Improve small message performance for async WRITE/READ by @zhongjiechen in https://github.com/uccl-project/uccl/pull/812
- [p2p] Add Intel irdma support with DMA-BUF GPUDirect RDMA by @manojgop in https://github.com/uccl-project/uccl/pull/802
- [EP] fixing EP build CI on L40s by overwriting TORCH_CUDA_ARCH_LIST=9.0 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/813
- Add Release workflows by @praveingk in https://github.com/uccl-project/uccl/pull/760
- [EP] Remove copy_manager from EP by @ShawnWeiChew in https://github.com/uccl-project/uccl/pull/810
- fp8 support by @monopodium in https://github.com/uccl-project/uccl/pull/815
- Fix release by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/814
- [All] Fix release2 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/816
- Test Releases/v0.0.1.post6 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/817
- fix(build): use container glibc for auditwheel repair symbol validation by @manojgop in https://github.com/uccl-project/uccl/pull/818
- [P2P] enable multiple NICs for RDMA + refactoring code by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/819
- [P2P] use green context to limit SM usage for compression by @monopodium in https://github.com/uccl-project/uccl/pull/824
- [P2P] Deprecate num_cpus in Endpoint constructor. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/825
- Support Local transfers for NIXL by @praveingk in https://github.com/uccl-project/uccl/pull/820
- [EP] Fix FP8 E4M3 dtype mismatch on AMD MI300X/MI325X (gfx942) by @laochonlam in https://github.com/uccl-project/uccl/pull/804
- [build] Apptainer support by @sa1g in https://github.com/uccl-project/uccl/pull/822
- add amd ci test by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/829
- EP: update amd benchmark performance on MI355X with Pollara by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/831
- [Mint] import inital code from mscclpp by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/830
- [Mint] port mscclpp and build successfully by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/832
- [Mint] Dedup files by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/833
- [P2P] Share RDMA context/PD/MR across context slots on same NIC. by @manojgop in https://github.com/uccl-project/uccl/pull/826
- [P2P] fixing nixlbench by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/834
- [CI] disable AMD CI temporarily per amd requests by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/835
- Add vLLM EP launch script and SGLang scripts for Intel irdma NICs by @manojgop in https://github.com/uccl-project/uccl/pull/823
- [CI] reenable amd ci by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/841
- [P2P] fixing p2p wheel name and folder in readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/844
- fix wrong numa nodes in p2p by @miss-x-incognita in https://github.com/uccl-project/uccl/pull/845
- [P2P] allowing specifying multiple devices by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/846
- [EP] Allocate contiguous tensor buffer from uccl_ep.cc by @MaoZiming in https://github.com/uccl-project/uccl/pull/840
- [P2P] Remove UCCL_RCMODE from p2p. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/849
- [P2P] Memory region cache by @zhongjiechen in https://github.com/uccl-project/uccl/pull/827
- [P2P] Fix RoCE GID auto-detection to prefer IPv4-mapped over link-local by @manojgop in https://github.com/uccl-project/uccl/pull/847
- [CI] fixing l4 ip by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/850
- [Doc] Update roadmap by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/852
- [EP] Remove redundant check to allow for empty topk_idx tensors by @MaoZiming in https://github.com/uccl-project/uccl/pull/838
- [EP]: Add NUMA-aware NIC selection tie-breaker by @manojgop in https://github.com/uccl-project/uccl/pull/848
- [EP]: Add vLLM PD disaggregation scripts for Intel RDMA NICs by @manojgop in https://github.com/uccl-project/uccl/pull/853
- [All] adding easy boostrap script and cleanup submodules by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/854
- [Lite] Renaming from mint by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/856
- [EP] disable aggressive atomic in both setup.py and makefile by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/858
- [Lite] Better host memory staging for inter/intra P2P by @zhongjiechen in https://github.com/uccl-project/uccl/pull/855
- [Lite] Fix staging buffer bugs. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/860
- [Lite] Topology-aware NIC selection. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/861
- [P2P] NUMA aware host staging buffer allocation. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/862
- [Lite] Clean scripts. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/863
- [EP] Runtime aggressive atomic control and AMD memory ordering fixes by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/859
- [Build] rocm to roc7, rocm6 to roc6 by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/866
- [EP] enforce roc7 for ep amd build by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/867
- [EP] fixing amd per-expert batching by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/865
- [UK] Refactor ukernel by @derekwin in https://github.com/uccl-project/uccl/pull/869
- [UK] Split transport memory into MR/IPC/SHM managers and clean up bounce/util integration by @derekwin in https://github.com/uccl-project/uccl/pull/870
- [UK] simplify Communicator & IPCadapter by @derekwin in https://github.com/uccl-project/uccl/pull/871
- [UK] uccl performance by @derekwin in https://github.com/uccl-project/uccl/pull/872
- [UK] Support IPC and Device Bench on AMD by @DanielDanyang in https://github.com/uccl-project/uccl/pull/873
- [CI] fixing AMD CI by using recent version of rocm by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/880
- [EP] fix internode_dispatch crashes with null pointer assertion by @MaoZiming in https://github.com/uccl-project/uccl/pull/882
- [CI] AMD using torch rocm7.1 which contains libmagma.so by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/883
- [EP] Update latest vLLM LL results and gracefully handle num_tokens >= 256 by @MaoZiming in https://github.com/uccl-project/uccl/pull/884
- [P2P] Merge TCPX onto the NCCL endpoint and fix the NCCL/TCPX runtime path by @DanielDanyang in https://github.com/uccl-project/uccl/pull/881
- [P2P] Fix: Add NIC availability check to prevent IB assertion failures by @derekwin in https://github.com/uccl-project/uccl/pull/877
- [DOC]Fix typo in README by @yultang in https://github.com/uccl-project/uccl/pull/885
- [CI] Make AMD CI more robust by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/886
- Fix p2p build regression for therock target. by @amd-ivaganev in https://github.com/uccl-project/uccl/pull/887
- [P2P] fix: Use BDF as identity instead of local_gpu_idx_, which is not friendly to CUDA_VISIBLE_DEVICE. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/864
- [P2P] Fix compatibility issues with Nixl. by @zhongjiechen in https://github.com/uccl-project/uccl/pull/889
- [CI] Futher strengthen AMD CI by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/890
- [Doc] update readme by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/891
- [UK] Imporve ipc performance by @DanielDanyang in https://github.com/uccl-project/uccl/pull/876
- [CI] fixing amd CI by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/896
- [P2P] release bug fix by @derekwin in https://github.com/uccl-project/uccl/pull/897
- [EP]: fix async event chaining and auto-detect intranode topology by @zhenhuang12 in https://github.com/uccl-project/uccl/pull/892
- [EP] Handle zero-token intranode dispatch/combine paths by @Matt-Harvill in https://github.com/uccl-project/uccl/pull/898
- [EP] fixing test_internode.py by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/900
- Fix/bdf connect API and bug fixes by @praveingk in https://github.com/uccl-project/uccl/pull/899
- [UK] oob refactor by @derekwin in https://github.com/uccl-project/uccl/pull/902
- [EP] Allow runtime override of CPU recv timeout via UCCL_EP_CPU_TIMEOUT_SECS by @KevinZhao in https://github.com/uccl-project/uccl/pull/904
- [UK] Unify buffer_id semantics across Transport/CCL/Python and harden OOB/IPC runtime paths by @derekwin in https://github.com/uccl-project/uccl/pull/907
- [P2P] Multi-transport Build and Dispatcher by @praveingk in https://github.com/uccl-project/uccl/pull/879
- Fix for multi bdf by @praveingk in https://github.com/uccl-project/uccl/pull/908
- [P2P] Update readme post multi-backend support by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/909
- [EP] supress nb leak warning by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/910
- [P2P] Enable nccl internode CI by @praveingk in https://github.com/uccl-project/uccl/pull/911
- [EP] chore: isolate warning suppression to uccl_ep only by @ShawnWeiChew in https://github.com/uccl-project/uccl/pull/912
- [EP] switch efa from per-peer qp to per-peer ah by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/916
- [EP] Add documentation for MegatronLM with UCCL-EP guide by @MaoZiming in https://github.com/uccl-project/uccl/pull/914
- [Doc] utran osdi26 AE by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/917
- [Doc ]fixing utran OSDI26 AE testbed by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/918
- [P2P] Added congestion control support (Timely, Swift) by @andrzej-k in https://github.com/uccl-project/uccl/pull/837
- [Doc] update paper title and add uccl-tran to uccl-collective by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/920
- [EP] Fix internode dispatch deadlock and memory ordering issues by @yuhuiaws in https://github.com/uccl-project/uccl/pull/903
- [UK] unidirection adapter by @derekwin in https://github.com/uccl-project/uccl/pull/927
- [EP] Allow high-throughput + low-latency Buffers to coexist in one process by @MaoZiming in https://github.com/uccl-project/uccl/pull/928
- [EP] Accept DeepEP-compatible overlap kwargs in low_latency_combine (stub) by @KevinZhao in https://github.com/uccl-project/uccl/pull/919
- [CI] all codeowner by @YangZhou1997 in https://github.com/uccl-project/uccl/pull/929
New Contributors
- @zhongjiechen made their first contribution in https://github.com/uccl-project/uccl/pull/65
- @steven4354 made their first contribution in https://github.com/uccl-project/uccl/pull/167
- @HermesCui made their first contribution in https://github.com/uccl-project/uccl/pull/251
- @jinqinn made their first contribution in https://github.com/uccl-project/uccl/pull/325
- @derekwin made their first contribution in https://github.com/uccl-project/uccl/pull/302
- @zhangnju made their first contribution in https://github.com/uccl-project/uccl/pull/340
- @monopodium made their first contribution in https://github.com/uccl-project/uccl/pull/361
- @amd-ivaganev made their first contribution in https://github.com/uccl-project/uccl/pull/364
- @FirwoodLin made their first contribution in https://github.com/uccl-project/uccl/pull/351
- @DanielDanyang made their first contribution in https://github.com/uccl-project/uccl/pull/385
- @viralbhadeshiya made their first contribution in https://github.com/uccl-project/uccl/pull/418
- @zhenhuang12 made their first contribution in https://github.com/uccl-project/uccl/pull/457
- @Nelson-Cheung made their first contribution in https://github.com/uccl-project/uccl/pull/450
- @pbelevich made their first contribution in https://github.com/uccl-project/uccl/pull/576
- @yurekami made their first contribution in https://github.com/uccl-project/uccl/pull/608
- @manojgop made their first contribution in https://github.com/uccl-project/uccl/pull/635
- @crazyguitar made their first contribution in https://github.com/uccl-project/uccl/pull/670
- @liad-hub made their first contribution in https://github.com/uccl-project/uccl/pull/647
- @laochonlam made their first contribution in https://github.com/uccl-project/uccl/pull/743
- @liranschour made their first contribution in https://github.com/uccl-project/uccl/pull/736
- @ShawnWeiChew made their first contribution in https://github.com/uccl-project/uccl/pull/758
- @yehuday made their first contribution in https://github.com/uccl-project/uccl/pull/759
- @whn09 made their first contribution in https://github.com/uccl-project/uccl/pull/766
- @andrzej-k made their first contribution in https://github.com/uccl-project/uccl/pull/780
- @sa1g made their first contribution in https://github.com/uccl-project/uccl/pull/822
- @miss-x-incognita made their first contribution in https://github.com/uccl-project/uccl/pull/845
- @yultang made their first contribution in https://github.com/uccl-project/uccl/pull/885
- @Matt-Harvill made their first contribution in https://github.com/uccl-project/uccl/pull/898
- @KevinZhao made their first contribution in https://github.com/uccl-project/uccl/pull/904
- @yuhuiaws made their first contribution in https://github.com/uccl-project/uccl/pull/903
Full Changelog: https://github.com/uccl-project/uccl/commits/v0.1.1
