In a sweep, complexa design <inf_*> runs every stage against the same config (cli_runner.py:648). generate writes to root_path (baked into inf_*), but evaluate reads sample_storage_path — which is absent from inf_* — so it rebuilds a different path (evaluate.py:752,760) that was never written → FileNotFoundError. (nb03 works because its config bakes neither path, so both stages rebuild the same one.)
Severity: High · Status: workaround verified; upstream open
Steps to reproduce
complexa generate <sweep> → produces configs/inference_configs/inf_*.yaml.
complexa design configs/inference_configs/inf_0_*.yaml.
- generate succeeds; evaluate →
FileNotFoundError on the sample path.
Fix
Pass both paths explicitly: ++sample_storage_path=./inference/<stem> ++output_dir=./evaluation_results/<stem>, or run complexa analysis configs/eval_configs/eval_*.yaml. Upstream: bake sample_storage_path and output_dir into inf_* (generate_inference_configs.py:327); a sample_storage_path-only fix is insufficient (output_dir is None still trips the guard).
Related
run_name is appended to the inference/ dir (generate.py:62) but not evaluation_results/ (evaluate.py:756), which compounds the confusion when locating results.
Environment
NVIDIA A100 80GB PCIe · driver 565.57.01 · CUDA 12.7 · repo branch dev @ 916eaae · UV runtime.
In a sweep,
complexa design <inf_*>runs every stage against the same config (cli_runner.py:648).generatewrites toroot_path(baked intoinf_*), butevaluatereadssample_storage_path— which is absent frominf_*— so it rebuilds a different path (evaluate.py:752,760) that was never written →FileNotFoundError. (nb03 works because its config bakes neither path, so both stages rebuild the same one.)Severity: High · Status: workaround verified; upstream open
Steps to reproduce
complexa generate <sweep>→ producesconfigs/inference_configs/inf_*.yaml.complexa design configs/inference_configs/inf_0_*.yaml.FileNotFoundErroron the sample path.Fix
Pass both paths explicitly:
++sample_storage_path=./inference/<stem> ++output_dir=./evaluation_results/<stem>, or runcomplexa analysis configs/eval_configs/eval_*.yaml. Upstream: bakesample_storage_pathandoutput_dirintoinf_*(generate_inference_configs.py:327); asample_storage_path-only fix is insufficient (output_dir is Nonestill trips the guard).Related
run_nameis appended to theinference/dir (generate.py:62) but notevaluation_results/(evaluate.py:756), which compounds the confusion when locating results.Environment
NVIDIA A100 80GB PCIe · driver 565.57.01 · CUDA 12.7 · repo branch
dev@916eaae· UV runtime.