EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

support 1 shot allreduce in 1-node and 2-node using mscclpp (#6277)

Browse Source
This commit is contained in:
zyksir
2025-06-05 13:11:24 +08:00
committed by GitHub
parent 4474eaf552
commit 8e3797be1c
20 changed files with 2177 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
docs/backend/server_arguments.md
View File

@@ -201,6 +201,7 @@ Please consult the documentation below and [server_args.py](https://github.com/s
| `disable_cuda_graph_padding` | Disable CUDA Graph when padding is needed; otherwise, still use CUDA Graph. | `False` |
| `disable_outlines_disk_cache` | Disable disk cache for outlines grammar backend. | `False` |
| `disable_custom_all_reduce` | Disable usage of custom all-reduce kernel. | `False` |
| `enable_mscclpp` | Enable usage of mscclpp kernel for small message all-reduce. | `False` |
| `disable_overlap_schedule` | Disable the [Overhead-Scheduler](https://lmsys.org/blog/2024-12-04-sglang-v0-4/#zero-overhead-batch-scheduler). | `False` |
| `enable_nan_detection` | Enable warning if the logits contain `NaN`. | `False` |
| `enable_p2p_check` | Turns off the default of always allowing P2P checks when accessing GPU. | `False` |