Re-render workflow with dataset download_items + dawson-standard CPU runner

.riahub/workflows/train.yaml

@@ -15,7 +15,7 @@ permissions:
 
 jobs:
   WavesFM-Training:
-    runs-on: "ubuntu-latest"
+    runs-on: "dawson-standard"
     env:
       WAVESFM_TASK: "rml"
       WAVESFM_EPOCHS: "3"
@@ -27,9 +27,7 @@ jobs:
       # downstream step uses the env var, no hard-coded paths.
       WAVESFM_REPO_DIR: "/opt/wavesfm/repo"
       WAVESFM_ADAPTED_DATA: "/opt/wavesfm/adapted_data.h5"
-      # Override the model-download base URL to logan's internal LAN IP.
+      # Override model-download base URL to internal LAN IP (external proxy LFS unreachable from runner)
-      # External riahub.ai LFS batch endpoint is unreachable from runner;
-      # internal IP works directly.
       RIAHUB_BASE_URL: "http://192.168.0.170:3000"
     steps:
       - name: Display basic runner info
@@ -183,6 +181,123 @@ jobs:
             exit 1
           fi
 
+      - name: "Checkout Dataset (qoherent/icc-demo/icc_canary_2026_05_28-v1.0.0.h5)"
+        shell: bash
+        timeout-minutes: 10
+        env:
+          RIAHUB_USER: ${{ secrets.QMBDEMO_USER }}
+          RIAHUB_TOKEN: ${{ secrets.QMBDEMO_TOKEN }}
+          GIT_TERMINAL_PROMPT: "0"
+        run: |
+          set -euo pipefail
+          DEFAULT_BASE_URL='https://riahub.ai'
+          BASE_URL_SOURCE=${RIAHUB_BASE_URL:-$DEFAULT_BASE_URL}
+          BASE_URL_SOURCE="${BASE_URL_SOURCE%/}"
+
+          build_base_candidates() {
+            local raw="$1"
+            if [[ "$raw" =~ ^https?:// ]]; then
+              echo "$raw"
+              if [[ "$raw" == http://* ]]; then
+                echo "https://${raw#http://}"
+              elif [[ "$raw" == https://* ]]; then
+                echo "http://${raw#https://}"
+              fi
+              return
+            fi
+            echo "https://$raw"
+            echo "http://$raw"
+          }
+
+          AUTH_HEADER=""
+          if [[ -n "${RIAHUB_USER:-}" && -n "${RIAHUB_TOKEN:-}" ]]; then
+            AUTH_HEADER=$(printf 'Authorization: basic %s' \
+              "$(printf '%s:%s' "$RIAHUB_USER" "$RIAHUB_TOKEN" | base64 | tr -d '\n')")
+          fi
+          # ``sudo env GIT_TERMINAL_PROMPT=0`` propagates the env var across
+          # sudo's default ``env_reset`` boundary; a bare ``sudo git`` would
+          # see an empty env on most distros' default sudoers, so the
+          # step-level ``env:`` block's GIT_TERMINAL_PROMPT=0 would NOT
+          # actually reach git child processes. Without it, git falls back
+          # to opening ``/dev/tty`` (the PTY allocated by act_runner) and
+          # prompting for credentials on a 401, hanging until timeout.
+          git_auth() {
+            if [[ -n "$AUTH_HEADER" ]]; then
+              sudo env GIT_TERMINAL_PROMPT=0 git -c "http.extraheader=$AUTH_HEADER" "$@"
+            else
+              sudo env GIT_TERMINAL_PROMPT=0 git "$@"
+            fi
+          }
+
+          REPO_PATH='/qoherent/icc-demo.git'
+          DEST_ROOT='/opt/qmb/riahub/dataset/qoherent/icc-demo/9f87fa9fe2badd314ad81379064e236ea494e89d'
+          sudo mkdir -p "$(dirname "$DEST_ROOT")"
+          if ! command -v git-lfs >/dev/null 2>&1; then
+            sudo apt-get update -y
+            sudo apt-get install -y git-lfs
+          fi
+          mapfile -t BASE_CANDIDATES < <(build_base_candidates "$BASE_URL_SOURCE")
+          MATERIALIZED=0
+          for base in "${BASE_CANDIDATES[@]}"; do
+            base="${base%/}"
+            REPO_URL="${base}${REPO_PATH}"
+            echo "Fetching dataset from $REPO_URL"
+            sudo rm -rf "$DEST_ROOT"
+            sudo mkdir -p "$DEST_ROOT"
+            sudo git -C "$DEST_ROOT" init || continue
+            sudo git -C "$DEST_ROOT" remote add origin "$REPO_URL" || continue
+            # See ``_render_model_checkout`` for the rationale on skipping
+            # ``git lfs install --local`` — short version: the smudge
+            # filter it would register tries its own credential lookup
+            # during ``git checkout FETCH_HEAD`` and hangs forever on
+            # /dev/tty when the repo is internal/private. We rely on
+            # the explicit ``git lfs fetch`` (with auth) +
+            # ``git lfs checkout`` (local) pair below instead.
+            sudo git -C "$DEST_ROOT" sparse-checkout init --no-cone || continue
+            sudo git -C "$DEST_ROOT" sparse-checkout set --no-cone -- \
+              'icc_canary_2026_05_28-v1.0.0.h5' || continue
+            if ! git_auth -C "$DEST_ROOT" fetch --depth=1 origin '9f87fa9fe2badd314ad81379064e236ea494e89d'; then
+              continue
+            fi
+            # See ``_render_model_checkout`` for the rationale on
+            # ``GIT_LFS_SKIP_SMUDGE=1`` — short version: the runner has
+            # the LFS smudge filter installed system-wide
+            # (``/etc/gitconfig``), so checkout fires it and the filter's
+            # credential helper hangs on /dev/tty for internal repos.
+            # Skipping smudge here lets the explicit ``git lfs fetch``
+            # below handle materialization with proper auth.
+            if ! sudo env GIT_TERMINAL_PROMPT=0 GIT_LFS_SKIP_SMUDGE=1 \
+                git -C "$DEST_ROOT" -c advice.detachedHead=false checkout FETCH_HEAD; then
+              continue
+            fi
+            if ! git_auth -C "$DEST_ROOT" lfs fetch origin --include='icc_canary_2026_05_28-v1.0.0.h5' --exclude=""; then
+              echo "LFS fetch failed for candidate $base, trying next" >&2
+              continue
+            fi
+            if ! sudo git -C "$DEST_ROOT" lfs checkout; then
+              echo "LFS checkout failed for candidate $base, trying next" >&2
+              continue
+            fi
+            POINTER_FOUND=0
+            _LFS_REL_PATH='icc_canary_2026_05_28-v1.0.0.h5'
+            if [[ "$(sudo head -c 9 "$DEST_ROOT/$_LFS_REL_PATH" 2>/dev/null || true)" == "version h" ]]; then
+              echo "LFS materialization left a pointer at $DEST_ROOT/$_LFS_REL_PATH for candidate $base, trying next" >&2
+              POINTER_FOUND=1
+            fi
+            if [[ "$POINTER_FOUND" -ne 0 ]]; then
+              continue
+            fi
+            MATERIALIZED=1
+            break
+          done
+          if [[ "$MATERIALIZED" -ne 1 ]]; then
+            echo "Failed to materialize dataset using base URL candidates derived from: $BASE_URL_SOURCE" >&2
+            if [[ -z "$AUTH_HEADER" ]]; then
+              echo "  (set QMBDEMO_USER+QMBDEMO_TOKEN repo secrets for internal/private repos)" >&2
+            fi
+            exit 1
+          fi
+
       - name: Clone WavesFM
         shell: bash
         run: |
@@ -375,4 +490,4 @@ jobs:
             ${{ env.WAVESFM_OUTPUT_DIR }}/best.pth
             ${{ env.WAVESFM_OUTPUT_DIR }}/log.txt
           if-no-files-found: warn
-# re-triggered after org transfer at 2026-05-28T05:53:45+00:00
+# committed at 2026-05-28T06:36:42.903115+00:00