[Distributed][MoonEP] BF16 integration of MoonEP balanced EP backend (#52101)
Signed-off-by: kevinli <306243285+kaijunli-infr@users.noreply.github.com> Co-authored-by: kevinli <306243285+kaijunli-infr@users.noreply.github.com>
This commit is contained in:
@@ -0,0 +1,5 @@
|
||||
model_name: "Qwen/Qwen3-30B-A3B"
|
||||
accuracy_threshold: 0.88
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
server_args: "--enforce-eager --max-model-len 8192 --data-parallel-size 2 --enable-expert-parallel --all2all-backend=moonep"
|
||||
@@ -10,3 +10,4 @@ Qwen3-30B-A3B-Fp8-CT-Block-deepgemm-deepep-ht.yaml
|
||||
Qwen3-30B-A3B-Fp8-CT-Block-deepgemm-deepep-ll.yaml
|
||||
Qwen3-30B-A3B-Fp8-CT-Block-deepgemm.yaml
|
||||
Qwen3-30B-A3B-BF16-triton.yaml
|
||||
Qwen3-30B-A3B-BF16-moonep.yaml
|
||||
|
||||
@@ -128,6 +128,12 @@ def test_gsm8k_correctness(config_filename):
|
||||
"Skipping Qwen3.5-35B-A3B-MXFP4-AITER-TP2 on non-GFX950 platforms. "
|
||||
"The quantization scheme is not supported on non-GFX950 platforms."
|
||||
)
|
||||
if "--all2all-backend=moonep" in eval_config.get("server_args", ""):
|
||||
from vllm.utils.import_utils import has_moonep
|
||||
|
||||
if not has_moonep():
|
||||
pytest.skip("Skipping MoonEP config: the moonep package is not installed.")
|
||||
|
||||
# Parse server arguments from config (use shlex to handle quoted strings)
|
||||
server_args_str = eval_config.get("server_args", "")
|
||||
server_args = shlex.split(server_args_str) if server_args_str else []
|
||||
|
||||
Reference in New Issue
Block a user