Tomography reconstruction multinode optimization

config.yml

@@ -46,6 +46,7 @@ globus:
       uri: beegfs.als.lbl.gov
       uuid: d33b5d6e-1603-414e-93cb-bcb732b7914a
       name: bl733-beegfs-data
+
     # 8.3.2 ENDPOINTS
 
     spot832:
@@ -148,8 +149,8 @@ harbor_images832:
   multires_image: tomorecon_nersc_mpi_hdf5@sha256:cc098a2cfb6b1632ea872a202c66cb7566908da066fd8f8c123b92fa95c2a43c
 
 ghcr_images832:
-  recon_image: ghcr.io/als-computing/microct:master
-  multires_image: ghcr.io/als-computing/microct:master
+  recon_image: ghcr.io/als-computing/microct@sha256:1fdfb786726ee03301d624319e3d16702045072f38e2b0cca9d6237e5ab3f5ff
+  multires_image: ghcr.io/als-computing/microct@sha256:1fdfb786726ee03301d624319e3d16702045072f38e2b0cca9d6237e5ab3f5ff
 
 prefect:
   deployments:
@@ -158,3 +159,88 @@ prefect:
 
 scicat:
   jobs_api_url: https://dataportal.als.lbl.gov/api/ingest/jobs
+
+hpc_submission_settings832:
+  nersc_reconstruction:
+    # ── SLURM resource allocation ─────────────────────────────────────────────
+    qos: realtime
+    account: als
+    reservation: ""
+    num_nodes: 4
+    cpus-per-task: 128
+    walltime: "0:30:00"
+  nersc_multiresolution:
+    # ── SLURM resource allocation ─────────────────────────────────────────────
+    qos: realtime
+    account: als
+    reservation: ""
+    cpus-per-task: 128
+    walltime: "0:15:00"
+  nersc_segmentation_sam3:
+    # ── SLURM resource allocation ─────────────────────────────────────────────
+    qos: regular
+    account: als
+    constraint: gpu
+    reservation: ""
+    num_nodes: 4
+    ntasks-per-node: 1
+    gpus-per-node: 4
+    cpus-per-task: 128
+    walltime: "00:59:00"
+    # ── Inference parameters ──────────────────────────────────────────────────
+    script_name: "src/inference_v6.py"
+    batch_size: 1
+    patch_size: 400
+    confidence:
+    - 0.5
+    overlap: 0.25
+    prompts:
+    - "Phloem Fibers"
+    - "Hydrated Xylem vessels"
+    - "Air-based Pith cells"
+    - "Dehydrated Xylem vessels"
+    # ── Paths ─────────────────────────────────────────────────────────────────
+    cfs_path: /global/cfs/cdirs/als/data_mover/8.3.2
+    conda_env_path: /global/cfs/cdirs/als/data_mover/8.3.2/envs/sam3-py311
+    seg_scripts_dir: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/inference_latest/forge_feb_seg_model_demo/
+    checkpoints_dir: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/sam3_finetune/sam3/
+    bpe_path: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/sam3_finetune/sam3/bpe_simple_vocab_16e6.txt.gz
+    original_checkpoint_path: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/sam3_finetune/sam3/sam3.pt
+    finetuned_checkpoint_path: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/sam3_finetune/sam3/checkpoint_v6.pt
+  nersc_segmentation_dinov3:
+    # ── SLURM resource allocation ─────────────────────────────────────────────
+    qos: regular
+    account: als
+    constraint: gpu
+    reservation: ""
+    num_nodes: 4
+    ntasks-per-node: 1
+    nproc_per_node: 4
+    gpus-per-node: 4
+    cpus-per-task: 128
+    walltime: "00:59:00"
+    # ── Inference parameters ──────────────────────────────────────────────────
+    script_name: "src.inference_dino_v1"
+    batch_size: 4
+    # ── Paths ─────────────────────────────────────────────────────────────────
+    cfs_path: /global/cfs/cdirs/als/data_mover/8.3.2
+    conda_env_path: /global/cfs/cdirs/als/data_mover/8.3.2/envs/dino_demo
+    seg_scripts_dir: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/inference_v5_multiseg/forge_feb_seg_model_demo/
+    dino_checkpoint_path: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/dino/best.ckpt
+  nersc_combine_segmentations:
+    # ── SLURM resource allocation ─────────────────────────────────────────────
+    qos: regular
+    account: als
+    constraint: cpu
+    reservation: ""
+    num_nodes: 4
+    ntasks: 128
+    cpus-per-task: 1
+    walltime: "01:00:00"
+    # ── Combination parameters ────────────────────────────────────────────────
+    script_name: "src.combine_sam_dino_v3"
+    dilate_px: 5
+    # ── Paths ─────────────────────────────────────────────────────────────────
+    cfs_path: /global/cfs/cdirs/als/data_mover/8.3.2
+    conda_env_path: /global/cfs/cdirs/als/data_mover/8.3.2/envs/dino_demo
+    seg_scripts_dir: /global/cfs/cdirs/als/data_mover/8.3.2/tomography_segmentation_scripts/inference_latest/forge_feb_seg_model_demo