Fix RhoData trailing-newline bug, add GHA summary, improve WandB naming

- Fix `RhoData`: `strip()` + filter blank lines in filelist parsing
  (was treating trailing newlines as empty sample IDs → `.CHGCAR`)
- Add GHA summary step to Modal benchmark workflow (config table +
  results, matching EC2 benchmark format)
- WandB run names include dataset/samples/timestamp for local runs,
  GHA run number for CI runs
- Print parseable `BENCHMARK_*` output for GHA step summary

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: ryan-williams

.github/workflows/gpu-benchmark-modal.yml

@@ -49,6 +49,7 @@ jobs:
         run: uv pip install --system modal
 
       - name: Run benchmark
+        id: bench
         env:
           GPU: ${{ inputs.gpu || 'L4' }}
           EPOCHS: ${{ inputs.epochs || '5' }}
@@ -58,11 +59,57 @@ jobs:
           DATASET: ${{ inputs.dataset || 's3' }}
           WANDB_PROJECT: ${{ inputs.wandb_project || 'elf-net-ci' }}
         run: |
-          modal run modal/benchmark.py \
+          OUTPUT=$(modal run modal/benchmark.py \
             --gpu "$GPU" \
             --epochs "$EPOCHS" \
             --channels "$CHANNELS" \
             --residual-blocks "$RESIDUAL_BLOCKS" \
             --samples "$SAMPLES" \
             --dataset "$DATASET" \
-            --wandb-project "$WANDB_PROJECT"
+            --wandb-project "$WANDB_PROJECT" 2>&1 | tee /dev/stderr)
+
+          # Parse metrics from output
+          for key in VAL_LOSS TRAIN_LOSS WALLCLOCK GPU DATASET SAMPLES WANDB_URL; do
+            val=$(echo "$OUTPUT" | grep -oP "BENCHMARK_${key}=\K.*" || true)
+            if [ -n "$val" ]; then
+              echo "BENCHMARK_${key}=${val}" >> "$GITHUB_OUTPUT"
+            fi
+          done
+
+      - name: Summary
+        env:
+          EPOCHS: ${{ inputs.epochs || '5' }}
+          CHANNELS: ${{ inputs.channels || '32' }}
+          RESIDUAL_BLOCKS: ${{ inputs.residual_blocks || '16' }}
+          VAL_LOSS: ${{ steps.bench.outputs.BENCHMARK_VAL_LOSS }}
+          TRAIN_LOSS: ${{ steps.bench.outputs.BENCHMARK_TRAIN_LOSS }}
+          WALLCLOCK: ${{ steps.bench.outputs.BENCHMARK_WALLCLOCK }}
+          GPU: ${{ steps.bench.outputs.BENCHMARK_GPU }}
+          DATASET: ${{ steps.bench.outputs.BENCHMARK_DATASET }}
+          SAMPLES: ${{ steps.bench.outputs.BENCHMARK_SAMPLES }}
+          WANDB_URL: ${{ steps.bench.outputs.BENCHMARK_WANDB_URL }}
+          WANDB_PROJECT: ${{ inputs.wandb_project || 'elf-net-ci' }}
+        run: |
+          echo "## GPU Benchmark Results (Modal)" >> "$GITHUB_STEP_SUMMARY"
+          echo "" >> "$GITHUB_STEP_SUMMARY"
+          echo "### Configuration" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Parameter | Value |" >> "$GITHUB_STEP_SUMMARY"
+          echo "|-----------|-------|" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Platform | Modal ($GPU) |" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Epochs | $EPOCHS |" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Channels | $CHANNELS |" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Residual Blocks | $RESIDUAL_BLOCKS |" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Dataset | $DATASET ($SAMPLES samples) |" >> "$GITHUB_STEP_SUMMARY"
+          COMMIT_LINK="[\`${GITHUB_SHA::7}\`](${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/commit/${GITHUB_SHA})"
+          echo "| Commit | $COMMIT_LINK |" >> "$GITHUB_STEP_SUMMARY"
+          if [ -n "$WANDB_URL" ]; then
+            WANDB_RUN_ID=$(echo "$WANDB_URL" | grep -oP '[^/]+$')
+            echo "| WandB | [$WANDB_PROJECT](https://wandb.ai/PrinceOA/$WANDB_PROJECT) run [$WANDB_RUN_ID]($WANDB_URL) |" >> "$GITHUB_STEP_SUMMARY"
+          fi
+          echo "" >> "$GITHUB_STEP_SUMMARY"
+          echo "### Results" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Metric | Value |" >> "$GITHUB_STEP_SUMMARY"
+          echo "|--------|-------|" >> "$GITHUB_STEP_SUMMARY"
+          echo "| val_loss | $VAL_LOSS |" >> "$GITHUB_STEP_SUMMARY"
+          echo "| train_loss | $TRAIN_LOSS |" >> "$GITHUB_STEP_SUMMARY"
+          echo "| Wallclock | ${WALLCLOCK}s |" >> "$GITHUB_STEP_SUMMARY"

modal/benchmark.py

@@ -112,7 +112,6 @@ def run_benchmark(
     # Build filelist from files on disk (S3 set has no mp_filelist.txt)
     data_dir = ds_root / input_dir
     all_ids = sorted(p.stem for p in data_dir.glob("*.CHGCAR"))
-    all_ids = [s for s in all_ids if s]  # filter blanks
     log.info("Dataset %r: %d total samples in %s", dataset, len(all_ids), data_dir)
 
     # Filter by file size (avoid OOM on large grids)
@@ -176,7 +175,7 @@ def run_benchmark(
             label_link.symlink_to(ds_root / "label")
         log.info("Using volume directly (no local copy)")
 
-    Path(data_root, "mp_filelist.txt").write_text("\n".join(subset))
+    Path(data_root, "mp_filelist.txt").write_text("\n".join(subset) + "\n")
 
     # Always use gradient checkpointing (32ch/16blk needs it even for ≤25MB files)
     use_grad_ckpt = True
@@ -195,11 +194,14 @@ def run_benchmark(
 
     # Set WandB run name and env vars for platform tagging
     import os
+    import time
 
     os.environ["INSTANCE_TYPE"] = f"modal-{gpu_type}"
-    # Set workflow-like name so WandB run name is descriptive
+    # Set workflow-like name so WandB run name is descriptive.
+    # GHA sets GITHUB_RUN_NUMBER; for local runs, use timestamp.
     os.environ["GITHUB_WORKFLOW"] = "Modal Benchmark"
-    os.environ.setdefault("GITHUB_RUN_NUMBER", f"{dataset}-{samples}s")
+    if "GITHUB_RUN_NUMBER" not in os.environ:
+        os.environ["GITHUB_RUN_NUMBER"] = time.strftime("%y%m%d-%H%M")
 
     results = run_training(
         channels=channels,
@@ -263,3 +265,14 @@ def main(
         results["wallclock_s"],
         gpu,
     )
+
+    # Print parseable output for GHA summary
+    wandb_url = results.get("wandb_run_url") or ""
+    print(f"BENCHMARK_VAL_LOSS={results['final_val_loss']:.6f}")  # noqa: T201
+    print(f"BENCHMARK_TRAIN_LOSS={results['final_train_loss']:.6f}")  # noqa: T201
+    print(f"BENCHMARK_WALLCLOCK={results['wallclock_s']:.0f}")  # noqa: T201
+    print(f"BENCHMARK_GPU={gpu}")  # noqa: T201
+    print(f"BENCHMARK_DATASET={dataset}")  # noqa: T201
+    print(f"BENCHMARK_SAMPLES={samples}")  # noqa: T201
+    if wandb_url:
+        print(f"BENCHMARK_WANDB_URL={wandb_url}")  # noqa: T201

src/electrai/dataloader/dataset.py

@@ -103,7 +103,7 @@ def __init__(self, datapath: str, precision: str, augmentation: bool, **kwargs):
         if isinstance(datapath, str) and Path(datapath).is_file():
             with Path(datapath).open() as f:
                 lines = f.readlines()
-            member_list = [line.replace("\n", "") for line in lines]
+            member_list = [line.strip() for line in lines if line.strip()]
         else:
             raise ValueError("No filename found.")