From 0fecdbbd02189c6a5df450bc5f7609d0a68e9522 Mon Sep 17 00:00:00 2001 From: Jammy2211 Date: Thu, 28 May 2026 11:23:56 +0100 Subject: [PATCH] fix: skip latents on A100 search submits MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The first PR-29 submission of the searches/nautilus/imaging/mge A100 run completed Nautilus convergence (log_Z=+31690.50, 65500 evals, 11m40s) but errored in the post-fit latent-variable computation: ValueError: magzero must be passed to the Analysis via kwargs to compute the 'total_lens_flux_mujy' latent. autolens_profiling/config/latent.yaml enables this latent for the latent/imaging/ profiling scripts, but the search-profiling scripts' AnalysisImaging doesn't carry magzero. PYAUTO_SKIP_LATENTS=1 bypasses latent computation entirely — latents are not part of the search profile, and the search itself ran to convergence. Co-Authored-By: Claude Opus 4.7 (1M context) --- hpc/batch_gpu/submit_imaging_mge_a100_hst_fp64 | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/hpc/batch_gpu/submit_imaging_mge_a100_hst_fp64 b/hpc/batch_gpu/submit_imaging_mge_a100_hst_fp64 index 5f0f3ee..e827571 100755 --- a/hpc/batch_gpu/submit_imaging_mge_a100_hst_fp64 +++ b/hpc/batch_gpu/submit_imaging_mge_a100_hst_fp64 @@ -30,6 +30,10 @@ export XLA_PYTHON_CLIENT_PREALLOCATE=false export JAX_ENABLE_X64=True export NUMBA_CACHE_DIR=/tmp/numba_cache export MPLCONFIGDIR=/tmp/matplotlib +# Skip latent-variable computation: autolens_profiling/config/latent.yaml +# enables total_lens_flux_mujy which needs `magzero=` on the Analysis. +# Latents aren't part of the search profile, so we bypass entirely. +export PYAUTO_SKIP_LATENTS=1 nvidia-smi