-
Notifications
You must be signed in to change notification settings - Fork 24
Description
The DLRM Ubench benchmark will trigger assertion when execute './run_all.sh':
Command in run_all.sh
./run_dlrm_ubench_train_embeddingbag.sh -l results -c "[(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]" # Config not real
Output:
=== Launching FB5 ===
Benchmark: dlrm
Implementation: ubench
Mode: train
Config: embeddingbag_[(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]float
Saving FB5 Logger File: results/dlrm_ubench_train_embeddingbag[(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]_float.logRunning Command:
++ python dlrm/ubench/dlrm_ubench_train_driver.py --steps=100 --device=cpu '--fb5logger=results/dlrm_ubench_train_embeddingbag_[(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]_float.log' emb '--dataset=[(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]'
Measuring the performance of emb on device = cpu
Steps = 100 warmups = 10
with emb dataset [(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]
Traceback (most recent call last):
File "dlrm/ubench/dlrm_ubench_train_driver.py", line 83, in
assert(len(run_dataset) == 1)
AssertionError
=== Completed Run ===
If use -c "[(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2),(2,2,2,2)]", the length of run_dataset should be 5, not 1.
Could you take a look?