Errors encounter while running the replay bench #213
Description
Describe the Bug
Just trying to run a replay of collected traces based on these instructions when I get the following error:
Traceback (most recent call last):
File "/home/amir/param/train/comms/pt/commsTraceReplay.py", line 1606, in
Traceback (most recent call last):
File "/home/amir/param/train/comms/pt/commsTraceReplay.py", line 1606, in
main()
File "/home/amir/param/train/comms/pt/commsTraceReplay.py", line 1597, in main
main()
File "/home/amir/param/train/comms/pt/commsTraceReplay.py", line 1597, in main
commsParams = commsParamsHolderBase(args)
File "/home/amir/param/.venv/lib/python3.10/site-packages/param_bench/train/comms/pt/comms_utils.py", line 827, in init
commsParams = commsParamsHolderBase(args)
File "/home/amir/param/.venv/lib/python3.10/site-packages/param_bench/train/comms/pt/comms_utils.py", line 827, in init
self.use_device_time = args.use_device_time
AttributeError: 'Namespace' object has no attribute 'use_device_time'
self.use_device_time = args.use_device_time
AttributeError: 'Namespace' object has no attribute 'use_device_time'
Steps to Reproduce
I followed these steps:
- Cloned the repo
- Created a virtual env:
uv venv - Installed the dependencies:
uv pip install -r requirements.txt - Installed the et_replay module under the
train/comms/ptdirectory:uv pip install . - Ran the job:
mpirun -n 2 --host 192.168.3.11:2 python3 commsTraceReplay.py --trace-path logs/torch/test-20250908 --trace-type et --master-ip 192.168.3.11
Discussion
Following the class inheritance chains, I've found that the code adds arguments through the readArgs methods of the CommsTraceReplayBench and ParamCommsReplayBench, and none of them includes the --use-device-time switch. Is that possibly a stale argument?