Size SLURM memory from input sequence length#44

Open

DimaMolod wants to merge 8 commits into

mainfrom

feature/length-aware-memory-requests

Commits on May 23, 2026

feat: size SLURM memory from input sequence length

DimaMolod
and
claude
committed
feat: backend-specific memory defaults (AF2 vs AF3)

DimaMolod
and
claude
committed
fix: keep length-aware inference sizing correct with precomputed features

DimaMolod
and
claude
committed
feat: length filtering (issue #33 + backend total caps), CI, and follow-ups

DimaMolod
and
claude
committed
fix: don't cache missing-file length reads (collapsed sizing to base)

DimaMolod
and
claude
committed
feat: length-based GPU model selection for structure_inference

DimaMolod
and
claude
committed
refactor: VRAM node-exclusion GPU routing (use the whole pool)

DimaMolod
and
claude
committed
docs: note length filtering applies to local/workstation runs too

DimaMolod
and
claude
committed