Skip to content

docs: refresh for v3.5 (remove deleted ensemble, fix version floors, document aimnet/extras/registry models)#95

Open
isayev wants to merge 2 commits into
mainfrom
docs/v3.5-refresh
Open

docs: refresh for v3.5 (remove deleted ensemble, fix version floors, document aimnet/extras/registry models)#95
isayev wants to merge 2 commits into
mainfrom
docs/v3.5-refresh

Conversation

@isayev

@isayev isayev commented Jun 12, 2026

Copy link
Copy Markdown
Contributor

Workstream W7 of the tech-debt roadmap. Docs-only — no code change.

The 8-model AIMNET ensemble was removed in 3.5 (use_ensemble / AUTO3D_USE_ENSEMBLE are now no-ops; there is no --use-ensemble flag), but the docs still presented it as a working "high accuracy" mode across five pages. This refresh:

  • Removes the deleted ensemble from advanced_usage, hpc, troubleshooting, cli, migration (kept only as clearly-labeled "removed/no-op" notes in migration.rst). Replaces ensemble guidance with "a single registry model is used; pick aimnet2-2025/-nse/-pd for different chemistry".
  • Fixes version floors: Python 3.10 → 3.11, PyTorch 2.1.0 → 2.8.
  • Documents that aimnet is a core dependency serving AIMNet2, the [ani]/[ase] extras (pip install "Auto3D[ani]"), and first-use model auto-download to ~/.cache/aimnet (AIMNET_CACHE_DIR).
  • Adds registry model names + custom-model-path to the engine tables (usage, cli, advanced_usage, quickstart, drug_discovery) and parameters.yaml.

Left unchanged (already accurate): README, api.rst, custom_nnp.rst, integrations.rst, citation.rst, docs/legacy-v2/**.

Test Plan

  • grep -rn "AUTO3D_USE_ENSEMBLE\|--use-ensemble" docs/source/ → only the two intentional "removed/no-op" notes in migration.rst.
  • Read the Docs build (the existing docs workflow runs on this PR).

Independent of #93/#94 (docs vs code); merges cleanly alongside them.

isayev added 2 commits June 11, 2026 22:45
…sion floors, document aimnet/extras/registry models

The 8-model AIMNET ensemble was removed in 3.5 (use_ensemble / AUTO3D_USE_ENSEMBLE
are no-ops; there is no --use-ensemble flag), but the docs still presented it as a
working 'high accuracy' mode across advanced_usage, hpc, troubleshooting, cli, and
migration. Remove those, replace ensemble guidance with 'a single registry model is
used; pick aimnet2-2025/-nse/-pd for different chemistry'. Also:
- Fix version floors: Python 3.10 -> 3.11, PyTorch 2.1.0 -> 2.8.
- Document that aimnet is a core dependency serving AIMNet2, the [ani]/[ase] extras,
  and first-use model auto-download to ~/.cache/aimnet (AIMNET_CACHE_DIR).
- Add registry model names + custom-model-path to the engine tables (usage, cli,
  advanced_usage, quickstart, drug_discovery) and parameters.yaml.

Docs-only; no code change. README/api/custom_nnp/integrations/citation/legacy-v2
were already accurate and left unchanged.
The RTD build pinned Python 3.10, but Auto3D requires-python is >=3.11, so RTD's
'pip install .' step failed on every build (main included). Match the floor.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant