fix(runs): make ACTL sampleworks image self-contained

Author: xraymemory

.actlignore

@@ -0,0 +1,13 @@
+# Keep ACTL sync focused on source. Large data/results should live under
+# /mnt/diffuse-shared or the pod home PVC, not in the synced checkout.
+.pixi/
+grid_search_results/
+outputs/
+data/
+initial_dataset_40*/
+checkpoints/
+release_data/
+*.ckpt
+*.pt
+*.tar.gz
+*.tgz

Dockerfile

@@ -7,8 +7,8 @@
 # Build:
 #   docker build -t sampleworks .
 #
-# CI builds pull checkpoints automatically from Docker Hub via:
-#   COPY --from=diffuseproject/sampleworks-checkpoints:latest
+# CI builds pull checkpoints automatically from Harbor via:
+#   COPY --from=harbor.astera.sh/library/sampleworks-checkpoints:latest
 # No checkpoint files are needed in the build context or on the CI runner.
 #
 # To rebuild the checkpoints base image (only needed when checkpoints change):
@@ -56,7 +56,7 @@
 #   /checkpoints/protenix_base_default_v0.5.0.pt     - Protenix model (~1.4GB)
 #
 # Checkpoints base image:
-#   All checkpoints live in diffuseproject/sampleworks-checkpoints:latest on Docker Hub.
+#   All checkpoints live in harbor.astera.sh/library/sampleworks-checkpoints:latest.
 #   To rebuild that image, see /data/users/diffuse/checkpoint-build/ on the GPU server.
 
 # ============================================================================
@@ -108,7 +108,7 @@ RUN chmod +x /usr/local/bin/entrypoint.sh
 # ============================================================================
 # Checkpoints (~10 GB) rarely change, so this layer is placed before pixi
 # installs to stay cached even when dependencies update.
-COPY --from=diffuseproject/sampleworks-checkpoints:latest /checkpoints/ /checkpoints/
+COPY --from=harbor.astera.sh/library/sampleworks-checkpoints:latest /checkpoints/ /checkpoints/
 
 # ============================================================================
 # Install all three environments: boltz, protenix, rf3
@@ -129,6 +129,12 @@ RUN pixi run -e boltz python -c "\
 from sampleworks.core.forward_models.xray.real_space_density_deps.ops import dilate_atom_centric; \
 print('CUDA extensions compiled successfully')" || echo "CUDA extension pre-compilation skipped (no GPU during build)"
 
+COPY run_all_models.sh ./
+RUN chmod +x /app/run_all_models.sh \
+    && printf '#!/usr/bin/env bash\nexec /app/run_all_models.sh "$@"\n' > /usr/local/bin/run_all_models.sh \
+    && chmod +x /usr/local/bin/run_all_models.sh \
+    && printf '\n# ACTL scientist workflow: land in the baked Sampleworks app.\nif [[ $- == *i* ]] && [ -z "${SAMPLEWORKS_NO_AUTO_CD:-}" ] && [ -d /app ]; then\n    cd /app\nfi\n' >> /root/.bashrc
+
 # Set default checkpoint paths via environment variables
 ENV BOLTZ1_CHECKPOINT=/checkpoints/boltz1_conf.ckpt \
     BOLTZ2_CHECKPOINT=/checkpoints/boltz2_conf.ckpt \

README.md

@@ -152,35 +152,50 @@ Output layout: `grid_search_results/<protein>/<model>[_<method>]/<scaler>/ens<N>
 Instructions for running evaluation and metrics scripts are coming soon.
 
 
-## Preset experiments (`sampleworks-runs`)
+## ACTL preset experiments (`run_all_models.sh` / `sampleworks-runs`)
 
-For canonical multi-model/multi-GPU sweeps, the `sampleworks-runs` CLI orchestrates parallel `run_grid_search.py` jobs from a single TOML preset. Each preset declares its jobs (model, pixi env, GPU assignment, args); the runner launches them in parallel, tees per-job logs, and aggregates exit codes.
+For canonical multi-model/multi-GPU sweeps, `sampleworks-runs` orchestrates parallel `run_grid_search.py` jobs from a single TOML preset. Each preset declares the model, pixi env, GPU assignment, output subdir, and CLI args. The runner launches jobs in parallel, tees per-job logs, and aggregates exit codes.
 
-**Pod-side prerequisite.** Bundled presets reference the canonical `/data/inputs`, `/data/results`, and `/root/.sampleworks` paths set up by the ACTL pod-init script. On a fresh sampleworks pod, run once per session:
+On ACTL, start the pod with the prebuilt image and shared storage, then run one command inside the pod shell:
 
 ```bash
-bash /mnt/diffuse-shared/raw/sampleworks/actl_setup_sampleworks_paths.sh
+actl pod up sampleworks-pr236 --profile 8x --image sampleworks --storage shared --pvc-size 200Gi --mount diffuse-shared --yes
+
+# inside the ACTL pod shell
+# the sampleworks image drops interactive shells in /app
+run_all_models.sh --dry-run   # inspect commands first
+run_all_models.sh             # run /app/src/sampleworks/runs/presets/all_models.toml
 ```
 
-That creates symlinks pointing the canonical paths at the shared mount (and namespaces `/data/results` by hostname or `$SAMPLEWORKS_ACTL_RUN_NAME`). Overrides via env var (`DATA_DIR=...`) or CLI (`--set defaults.DATA_DIR=...`) work without the symlinks.
+The wrapper keeps the TOML preset as the source of truth. It only supplies ACTL-friendly defaults:
+
+- `DATA_DIR=/mnt/diffuse-shared/raw/sampleworks/initial_dataset_40_occ_sweeps`
+- `RESULTS_DIR=/mnt/diffuse-shared/results/sampleworks/<pod>/all_models`
+- `MSA_CACHE_DIR=/mnt/diffuse-shared/cache/sampleworks/msa`
+- `PYTHONPATH=/app/src`, using the copy baked into the sampleworks image
+- direct `/app/.pixi/envs/<env>/bin/python` execution, so it reuses the environments baked into the sampleworks image without refreshing pixi caches
+- `/tmp` pixi/uv caches for any missing environment preparation, avoiding shared-storage Git cache issues
+
+Common commands:
 
 ```bash
-pixi run -e rf3 sampleworks-runs --list                          # bundled presets
-pixi run -e rf3 sampleworks-runs rf3_partial                     # run a preset
-pixi run -e rf3 sampleworks-runs rf3_partial --show              # inspect resolved values
-pixi run -e rf3 sampleworks-runs rf3_partial --dry-run           # print pixi run commands, don't execute
-pixi run -e rf3 sampleworks-runs all_models --only rf3,protenix  # subset jobs
-
-# Override any value without editing the TOML:
-pixi run -e rf3 sampleworks-runs rf3_partial \
-    --set jobs.rf3.gpus=7 \
+run_all_models.sh --list                         # bundled presets
+run_all_models.sh all_models --show              # inspect resolved values
+run_all_models.sh all_models --only rf3,protenix # subset jobs
+run_all_models.sh rf3_partial                    # run a smaller preset
+
+# Override paths or parameters without editing TOML:
+DATA_DIR=/mnt/diffuse-shared/raw/sampleworks/my_dataset run_all_models.sh rf3_partial
+run_all_models.sh rf3_partial \
+    --set jobs.rf3.gpus=0 \
     --set jobs.rf3.args.gradient-weights="0.0 0.01 0.02"
 ```
 
-Bundled presets live in `src/sampleworks/runs/presets/*.toml`. Add a new preset by dropping a `.toml` file alongside them or pointing at any path:
+Bundled presets live in `src/sampleworks/runs/presets/*.toml`. You can also copy one, edit it, and run it by path:
 
 ```bash
-sampleworks-runs ./my_experiment.toml
+cp src/sampleworks/runs/presets/all_models.toml my_experiment.toml
+run_all_models.sh ./my_experiment.toml
 ```
 
 Env-var defaults (`DATA_DIR`, `RESULTS_DIR`, `MSA_CACHE_DIR`, `PROTEINS_CSV`) declared per preset are filled from the process environment when set, otherwise from the preset's `[defaults]` block.