kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[None][fix] chore: fixing the math on asymmetric tp+pp tests (#7098)

Browse Source 
Signed-off-by: raayandhar <rdhar@nvidia.com>
This commit is contained in:
Raayan Dhar 2025-09-07 11:27:46 -07:00 committed by GitHub
parent 045d2cf761
commit 8f3121ac81
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tests/integration/defs/accuracy/test_disaggregated_serving.py
View File

@ -602,9 +602,9 @@ class TestLlama3_1_8BInstruct(LlmapiAccuracyTestHarness):
@parametrize_with_ids("gen_tp", [1, 2])
@pytest.mark.parametrize("testset", ["GSM8K", "MMLU"])
def test_ctx_pp_gen_tp_asymmetric(self, ctx_pp, gen_tp, testset):
if ctx_pp * gen_tp * 2 > get_device_count():
if ctx_pp + gen_tp > get_device_count():
pytest.skip(
f"Not enough devices for ctx_pp={ctx_pp}*gen_tp={gen_tp} test")
f"Not enough devices for ctx_pp={ctx_pp}+gen_tp={gen_tp} test")
return run_parallel_test(self.MODEL_NAME, self.MODEL_PATH, ctx_pp, 1, 1,
gen_tp, 1, 1, [get_accuracy_task(testset)])