biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix][EPLB] Prevent user-provided EPLB config from being overwritten with defaults (#29911)

Browse Source 
Signed-off-by: Sage Moore <sage@neuralmagic.com>
This commit is contained in:
Sage Moore
2025-12-02 14:20:22 -08:00
committed by GitHub
parent 6fc5841db1
commit e6f114ac25
2 changed files with 9 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/distributed/test_eplb_spec_decode.py
View File

@@ -22,7 +22,14 @@ def get_model_args(
"num_speculative_tokens": 1,
"max_model_len": model_max_len,
}
eplb_config = {
"num_redundant_experts": tp_size,
"window_size": 128,
"step_interval": 1024,
"log_balancedness": False,
}
if use_async:
eplb_config["use_async"] = True
model_args = {
"pretrained": model_name,
"dtype": "auto",
@@ -31,15 +38,10 @@ def get_model_args(
"gpu_memory_utilization": 0.7,
"speculative_config": speculative_config,
"enable_expert_parallel": True,
"num_redundant_experts": tp_size,
"eplb_window_size": 128,
"eplb_step_interval": 1024,
"eplb_log_balancedness": False,
"eplb_config": eplb_config,
"enable_eplb": True,
"max_model_len": model_max_len,
}
if use_async:
model_args["eplb_config"] = {"use_async": True}
return model_args