EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Doc] EPLB update the documentation (#8795)

Browse Source 
### What this PR does / why we need it?
1. Update documentation：In the current version, we recommend using the
following: policy of swift balancer(2).

### Does this PR introduce _any_ user-facing change?
  --additional-config '{ "eplb_config": {
    "dynamic_eplb": true,
    "expert_heat_collection_interval": 600,
    "algorithm_execution_interval": 50,
    "eplb_policy_type": 2,
    "num_redundant_experts": {ep_size}
    }}'

### How was this patch tested?
Test in DSV3.1 EP32

Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
This commit is contained in:
LI SHENGYONG
2026-04-29 17:15:29 +08:00
committed by GitHub
parent bc5ca2c856
commit 0cb4bca1ff
1 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
docs/source/user_guide/feature_guide/eplb_swift_balancer.md
View File

@@ -26,7 +26,7 @@ W8A8-Dynamic
### Dynamic EPLB
We need to add environment variable `export DYNAMIC_EPLB="true"` to enable vLLM EPLB. Enable dynamic balancing with auto-tuned parameters. Adjust expert_heat_collection_interval and algorithm_execution_interval based on workload patterns.
We need to add environment variable `export DYNAMIC_EPLB="true"` to enable vLLM EPLB. Enable dynamic balancing with auto-tuned parameters. Adjust expert_heat_collection_interval and algorithm_execution_interval based on workload patterns. In the current version, we recommend using the following: policy of swift balancer(2).
```shell
vllm serve Qwen/Qwen3-235B-A22 \
@@ -34,8 +34,10 @@ vllm serve Qwen/Qwen3-235B-A22 \
--enable-expert-parallel \
--additional-config '{ "eplb_config": {
"dynamic_eplb": true,
"expert_heat_collection_interval": 400,
"algorithm_execution_interval": 30
"expert_heat_collection_interval": 600,
"algorithm_execution_interval": 50,
"eplb_policy_type": 2,
"num_redundant_experts": {ep_size},
}}'
```