Missing benchmark file and empty parameter list errors #1
Description
Environment
- Python: 3.10 (
conda env: pfeife310) - Torch: 2.3+
- GPUs: 8× NVIDIA H200
- Repo:
pfeife(latestmainas of Oct 22, 2025) - TorchBench path:
/m-coriander/coriander/yfzhang3/benchmark
Missing benchmark_partial.py in coverage script
Running:
bash coverage_test.shproduces the error:
python: can't open file '/m-coriander/coriander/yfzhang3/pfeife/benchmark_partial.py': [Errno 2] No such file or directory
The script calls:
python mp_launch.py --world_size 2 benchmark_partial.py ...
However, the repository only contains:
benchmark_local.py
benchmark_mp.py
The error is that benchmark_partial.py is missing from the current branch or the coverage script references an outdated filename.
Model loader mismatch → empty parameter list
After manually switching to use an existing benchmark file, running:
python benchmark_local.py \
--device_cnt=2 \
--batch_size=16 \
--batch_cnt=4 \
--repeat=2 \
--model timm_vision_transformer
leads to the following error:
ValueError: optimizer got an empty parameter list
Traceback excerpt:
File "graph_bench.py", line 76, in measure_optimizer
optims = [optim_cls(p, lr=0.0) for p in params_list]
ValueError: optimizer got an empty parameter list
This occurs during:
measure_optimizer(gm, graph, devices, option)
which is called from:
gen_comp_graph() → gen_partition_graph() → forward()