From 657176c2093f8955152161dce9f84b81da2eb1d3 Mon Sep 17 00:00:00 2001 From: Robrecht Cannoodt Date: Fri, 21 Aug 2026 08:17:18 +0200 Subject: [PATCH 1/2] give fadvi a gpu label and raise the walltime of the methods that timed out --- CHANGELOG.md | 4 ++++ src/methods/batchelor_mnn_correct/config.vsh.yaml | 2 +- src/methods/fadvi/config.vsh.yaml | 2 +- src/methods/mnnpy/config.vsh.yaml | 2 +- src/methods/scalex/config.vsh.yaml | 2 +- src/methods/uce/config.vsh.yaml | 2 +- 6 files changed, 9 insertions(+), 5 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index f172d2335..d0d2cf366 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -57,6 +57,10 @@ * Fix `methods/scgpt_zeroshot` and `methods/scgpt_finetuned` failing to build: stop installing `flash-attn`. Both scripts pass `use_fast_transformer=False`, so it was never used. Behind it sat three more pins with no python 3.12 wheels, now `numpy<2`, `torchtext==0.17.2` and `transformers==4.36.2`. +* Give `methods/fadvi` a `gpu` label. Without one it was scheduled on the CPU partition with no GPU attached, so it + trained on the CPU and hit its walltime on every dataset. +* Raise `methods/uce`, `methods/mnnpy`, `methods/batchelor_mnn_correct` and `methods/scalex` to `veryhightime`. + All four completed on the smallest dataset and were killed at exactly 8 h on the rest. * Split Scanorama into two methods/scores - Split Scanorama into embedding (integrate) and count-correction (correct) modes, instead of running both together. diff --git a/src/methods/batchelor_mnn_correct/config.vsh.yaml b/src/methods/batchelor_mnn_correct/config.vsh.yaml index 6280faabf..7fb62b341 100644 --- a/src/methods/batchelor_mnn_correct/config.vsh.yaml +++ b/src/methods/batchelor_mnn_correct/config.vsh.yaml @@ -30,4 +30,4 @@ runners: - type: executable - type: nextflow directives: - label: [lowcpu, highmem, hightime] + label: [lowcpu, highmem, veryhightime] diff --git a/src/methods/fadvi/config.vsh.yaml b/src/methods/fadvi/config.vsh.yaml index 80a8e7e0c..dcfaa34cd 100644 --- a/src/methods/fadvi/config.vsh.yaml +++ b/src/methods/fadvi/config.vsh.yaml @@ -97,4 +97,4 @@ runners: # Allows turning the component into a Nextflow module / pipeline. - type: nextflow directives: - label: [midtime,midmem,midcpu] + label: [hightime, midmem, midcpu, gpu] diff --git a/src/methods/mnnpy/config.vsh.yaml b/src/methods/mnnpy/config.vsh.yaml index 433b57808..de9e247e0 100644 --- a/src/methods/mnnpy/config.vsh.yaml +++ b/src/methods/mnnpy/config.vsh.yaml @@ -57,4 +57,4 @@ runners: - type: executable - type: nextflow directives: - label: [hightime, lowcpu, highmem] + label: [veryhightime, lowcpu, highmem] diff --git a/src/methods/scalex/config.vsh.yaml b/src/methods/scalex/config.vsh.yaml index 59c1aed46..c008056e2 100644 --- a/src/methods/scalex/config.vsh.yaml +++ b/src/methods/scalex/config.vsh.yaml @@ -33,4 +33,4 @@ runners: - type: executable - type: nextflow directives: - label: [lowmem, lowcpu, hightime] + label: [lowmem, midcpu, veryhightime] diff --git a/src/methods/uce/config.vsh.yaml b/src/methods/uce/config.vsh.yaml index 216498003..2b831ab2b 100644 --- a/src/methods/uce/config.vsh.yaml +++ b/src/methods/uce/config.vsh.yaml @@ -43,4 +43,4 @@ runners: - type: executable - type: nextflow directives: - label: [hightime, veryhighmem, midcpu, biggpu] + label: [veryhightime, veryhighmem, midcpu, biggpu] From 1ac7c787a42617a64427f465600a5d3940c4052c Mon Sep 17 00:00:00 2001 From: Robrecht Cannoodt Date: Fri, 21 Aug 2026 09:08:59 +0200 Subject: [PATCH 2/2] keep methods at hightime --- CHANGELOG.md | 3 +-- src/methods/batchelor_mnn_correct/config.vsh.yaml | 2 +- src/methods/mnnpy/config.vsh.yaml | 2 +- src/methods/scalex/config.vsh.yaml | 2 +- src/methods/uce/config.vsh.yaml | 2 +- 5 files changed, 5 insertions(+), 6 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index d0d2cf366..3e8411d56 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -59,8 +59,7 @@ wheels, now `numpy<2`, `torchtext==0.17.2` and `transformers==4.36.2`. * Give `methods/fadvi` a `gpu` label. Without one it was scheduled on the CPU partition with no GPU attached, so it trained on the CPU and hit its walltime on every dataset. -* Raise `methods/uce`, `methods/mnnpy`, `methods/batchelor_mnn_correct` and `methods/scalex` to `veryhightime`. - All four completed on the smallest dataset and were killed at exactly 8 h on the rest. +* Give `methods/scalex` a `midcpu` label instead of `lowcpu`. It is CPU-bound only because its engine has no CUDA. * Split Scanorama into two methods/scores - Split Scanorama into embedding (integrate) and count-correction (correct) modes, instead of running both together. diff --git a/src/methods/batchelor_mnn_correct/config.vsh.yaml b/src/methods/batchelor_mnn_correct/config.vsh.yaml index 7fb62b341..6280faabf 100644 --- a/src/methods/batchelor_mnn_correct/config.vsh.yaml +++ b/src/methods/batchelor_mnn_correct/config.vsh.yaml @@ -30,4 +30,4 @@ runners: - type: executable - type: nextflow directives: - label: [lowcpu, highmem, veryhightime] + label: [lowcpu, highmem, hightime] diff --git a/src/methods/mnnpy/config.vsh.yaml b/src/methods/mnnpy/config.vsh.yaml index de9e247e0..433b57808 100644 --- a/src/methods/mnnpy/config.vsh.yaml +++ b/src/methods/mnnpy/config.vsh.yaml @@ -57,4 +57,4 @@ runners: - type: executable - type: nextflow directives: - label: [veryhightime, lowcpu, highmem] + label: [hightime, lowcpu, highmem] diff --git a/src/methods/scalex/config.vsh.yaml b/src/methods/scalex/config.vsh.yaml index c008056e2..fb890d6be 100644 --- a/src/methods/scalex/config.vsh.yaml +++ b/src/methods/scalex/config.vsh.yaml @@ -33,4 +33,4 @@ runners: - type: executable - type: nextflow directives: - label: [lowmem, midcpu, veryhightime] + label: [lowmem, midcpu, hightime] diff --git a/src/methods/uce/config.vsh.yaml b/src/methods/uce/config.vsh.yaml index 2b831ab2b..216498003 100644 --- a/src/methods/uce/config.vsh.yaml +++ b/src/methods/uce/config.vsh.yaml @@ -43,4 +43,4 @@ runners: - type: executable - type: nextflow directives: - label: [veryhightime, veryhighmem, midcpu, biggpu] + label: [hightime, veryhighmem, midcpu, biggpu]