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

[5/N] MoE Refactor: Update MoE parallelism arguments (#8658)

Browse Source
This commit is contained in:
Cheng Wan
2025-08-01 01:20:03 -07:00
committed by GitHub
parent c8d3a402c1
commit 6c88f6c8d9
38 changed files with 342 additions and 299 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
docs/backend/pd_disaggregation.md
View File

@@ -39,13 +39,13 @@ $ python -m sglang.srt.disaggregation.mini_lb --prefill http://127.0.0.1:30000 -
```bash
# prefill 0
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode normal --mem-fraction-static 0.8
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8
# prefill 1
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode normal --mem-fraction-static 0.8
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8
# decode 0
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode low_latency --mem-fraction-static 0.8 --max-running-requests 128
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8 --max-running-requests 128
# decode 1
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode low_latency --mem-fraction-static 0.8 --max-running-requests 128
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 --disaggregation-ib-device ${device_name} --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8 --max-running-requests 128
```
### Advanced Configuration
@@ -103,13 +103,13 @@ $ python -m sglang.srt.disaggregation.mini_lb --prefill http://127.0.0.1:30000 -
```bash
# prefill 0
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode normal --mem-fraction-static 0.8
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8
# prefill 1
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode normal --mem-fraction-static 0.8
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode prefill --host ${local_ip} --port 30000 --trust-remote-code --dist-init-addr ${prefill_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8
# decode 0
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode low_latency --mem-fraction-static 0.8 --max-running-requests 128
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 0 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8 --max-running-requests 128
# decode 1
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --enable-deepep-moe --deepep-mode low_latency --mem-fraction-static 0.8 --max-running-requests 128
$ python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3-0324 ---disaggregation-transfer-backend nixl --disaggregation-mode decode --host ${local_ip} --port 30001 --trust-remote-code --dist-init-addr ${decode_master_ip}:5000 --nnodes 2 --node-rank 1 --tp-size 16 --dp-size 8 --enable-dp-attention --moe-a2a-backend deepep --mem-fraction-static 0.8 --max-running-requests 128
```
## ASCEND

3
docs/backend/server_arguments.md
View File

@@ -212,8 +212,7 @@ Please consult the documentation below and [server_args.py](https://github.com/s
| Arguments | Description | Defaults |
|-----------|-------------|----------|
| `--ep-size` | The expert parallelism size. | 1 |
| `--enable-ep-moe` | Enabling expert parallelism for moe. The ep size is equal to the tp size. | False |
| `--enable-deepep-moe` | Enabling DeepEP MoE implementation for EP MoE. | False |
| `--moe-a2a-backend` | Select the backend for all-to-all communication for expert parallelism. | None |
| `--enable-flashinfer-cutlass-moe` | Enabling Flashinfer Cutlass MoE implementation for high throughput. | False |
| `--enable-flashinfer-trtllm-moe` | Enabling Flashinfer Trtllm MoE implementation for low latency. | False |
| `--deepep-mode` | Select the mode when enable DeepEP MoE, could be `normal`, `low_latency` or `auto`. Default is `auto`, which means `low_latency` for decode batch and `normal` for prefill batch. | auto |

10
docs/references/disaggregation/lws-examples/d.yaml
View File

@@ -28,9 +28,8 @@ spec:
- --enable-dp-lm-head
- --dp-size
- "16"
- --enable-deepep-moe
- --deepep-mode
- low_latency
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- decode
- --mem-fraction-static
@@ -166,9 +165,8 @@ spec:
- --enable-dp-lm-head
- --dp-size
- "16"
- --enable-deepep-moe
- --deepep-mode
- low_latency
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- decode
- --mem-fraction-static

10
docs/references/disaggregation/lws-examples/p.yaml
View File

@@ -38,9 +38,8 @@ spec:
- --dp-size
- "16"
- --disable-radix-cache
- --enable-deepep-moe
- --deepep-mode
- normal
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- prefill
- --mem-fraction-static
@@ -184,9 +183,8 @@ spec:
- --dp-size
- "16"
- --disable-radix-cache
- --enable-deepep-moe
- --deepep-mode
- normal
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- prefill
- --mem-fraction-static

20
docs/references/disaggregation/lws_pd_deploy.md
View File

@@ -64,9 +64,8 @@ spec:
- --dp-size
- "16"
- --disable-radix-cache
- --enable-deepep-moe
- --deepep-mode
- normal
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- prefill
- --mem-fraction-static
@@ -212,9 +211,8 @@ spec:
- --dp-size
- "16"
- --disable-radix-cache
- --enable-deepep-moe
- --deepep-mode
- normal
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- prefill
- --mem-fraction-static
@@ -373,9 +371,8 @@ spec:
- --enable-dp-lm-head
- --dp-size
- "16"
- --enable-deepep-moe
- --deepep-mode
- low_latency
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- decode
- --mem-fraction-static
@@ -508,9 +505,8 @@ spec:
#- --enable-two-batch-overlap
- --dp-size
- "16"
- --enable-deepep-moe
- --deepep-mode
- low_latency
- --moe-a2a-backend
- deepep
- --disaggregation-mode
- decode
- --mem-fraction-static