Download scripts/push_to_hf.sh from openenvforge/3ambench: direct link, hf CLI and curl.
- Browser
- Download file 8.4 kB
-
https://huggingface.co/datasets/openenvforge/3ambench/resolve/main/scripts/push_to_hf.sh
- Command line
-
hf download hf://datasets/openenvforge/3ambench/scripts/push_to_hf.sh
-
curl -L -o push_to_hf.sh https://huggingface.co/datasets/openenvforge/3ambench/resolve/main/scripts/push_to_hf.sh
8.4 kB
| # Stage (and, only with CONFIRM=1, push) the 3amBench dataset and the OpenEnv Space to Hugging Face. | |
| # NOT executed by the build. Usage: | |
| # HF_REPO=<you>/3ambench scripts/push_to_hf.sh # dry run: stages into dist/ (or $STAGE), uploads nothing | |
| # HF_REPO=<you>/3ambench CONFIRM=1 scripts/push_to_hf.sh # upload + tag $VERSION (default v0.2.0) | |
| # SPACE_REPO=<you>/3ambench-env CONFIRM=1 scripts/push_to_hf.sh # also push the OpenEnv Space | |
| # The expert tasks are staged from EXPERT_DIR (default dist/v02-candidates, written by `alertforge expert`) next to | |
| # the core tasks, with their reference policies in partial/ and null/. The expert manifest and gate log are not | |
| # staged, and the card ids in each expert task's tests/spec.json are stripped here, so the task files don't name the | |
| # defect catalog (the ids are not secret: src/alertforge/expert/ defines them). | |
| # Everything else is staged from what git would commit (tracked files plus untracked files that .gitignore and | |
| # .git/info/exclude don't exclude), so local-only work in the tree is never uploaded. | |
| set -euo pipefail | |
| : "${HF_REPO:?set HF_REPO=<hf-user-or-org>/3ambench}" | |
| ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)" | |
| VERSION="${VERSION:-v0.2.0}" | |
| EXPERT_DIR="${EXPERT_DIR:-$ROOT/dist/v02-candidates}" | |
| STAGE="${STAGE:-$ROOT/dist}" # where the staged copies go (dist/ is gitignored) | |
| DS="$STAGE/hf-dataset" | |
| SP="$STAGE/hf-space" | |
| [ -d "$EXPERT_DIR/tasks" ] || { echo "refusing: no expert tasks in $EXPERT_DIR/tasks (build them or set EXPERT_DIR)"; exit 1; } | |
| git -C "$ROOT" rev-parse --git-dir >/dev/null 2>&1 || { echo "refusing: $ROOT is not a git checkout (staging follows git)"; exit 1; } | |
| rm -rf "$DS" "$SP"; mkdir -p "$DS" "$SP" | |
| stage() { # stage DEST PATH...: copy what git would commit under PATH... into DEST, keeping relative paths and modes | |
| local dest="$1"; shift | |
| local list; list="$(mktemp)" | |
| git -C "$ROOT" ls-files -z --cached --others --exclude-standard -- "$@" > "$list" | |
| [ -s "$list" ] || { rm -f "$list"; echo "refusing: nothing to stage for $*"; exit 1; } | |
| (cd "$ROOT" && tar --null -T "$list" -cf -) | tar -xf - -C "$dest" | |
| rm -f "$list" | |
| } | |
| # ---- dataset repo: tasks/ + registry.json + manifest + card (the layout `harbor run --repo` expects) | |
| stage "$DS" README.md LICENSE NOTICE.md registry.json manifest.jsonl .gitignore | |
| stage "$DS" tasks partial null skill workflows openenv scripts tests docs # the card references all of these | |
| # generator: same relative layout as the source tree (generate.py imports ../src, world.py reads ../../workflows) | |
| stage "$DS" src pyproject.toml generator/generate.py | |
| [ -d "$ROOT/assets" ] && stage "$DS" assets | |
| # expert tier: tasks, partial and null policies, one afx-* directory each, next to the core ones | |
| WANT="$(find "$EXPERT_DIR/tasks" -mindepth 1 -maxdepth 1 -type d -name 'afx-*' | wc -l | tr -d ' ')" | |
| for kind in tasks partial null; do | |
| n=0 | |
| for d in "$EXPERT_DIR/$kind"/afx-*/; do | |
| if [ -f "$d/solve.sh" ] || [ -f "$d/task.toml" ]; then cp -R "${d%/}" "$DS/$kind/"; n=$((n + 1)); fi | |
| done | |
| [ "$n" -gt 0 ] && [ "$n" -eq "$WANT" ] || { echo "refusing: $n of $WANT afx-* directories usable in $EXPERT_DIR/$kind"; exit 1; } | |
| done | |
| find "$DS" \( -name '__pycache__' -o -name '.pytest_cache' -o -name '*.egg-info' \) -type d -prune -exec rm -rf {} + | |
| find "$DS" \( -name '.DS_Store' -o -name '*.pyc' \) -delete | |
| printf '*.png filter=lfs diff=lfs merge=lfs -text\n*.parquet filter=lfs diff=lfs merge=lfs -text\n' > "$DS/.gitattributes" | |
| if find "$DS" -type l | grep -q .; then echo "refusing: symlinks in staging (Harbor rejects them)"; exit 1; fi | |
| # card ids: drop `defect` from each expert requirement, written as the generator writes spec.json, and re-sum it | |
| python3 - "$DS/tasks" <<'PY' || { echo "refusing: could not strip the card ids"; exit 1; } | |
| import glob, hashlib, json, os, sys | |
| for spec_p in sorted(glob.glob(os.path.join(sys.argv[1], "afx-*", "tests", "spec.json"))): | |
| with open(spec_p) as fh: | |
| spec = json.load(fh) | |
| if not any("defect" in r for r in spec.get("requirements", [])): | |
| continue | |
| for r in spec["requirements"]: | |
| r.pop("defect", None) | |
| data = (json.dumps(spec, separators=(",", ":")) + "\n").encode() | |
| with open(spec_p, "wb") as fh: | |
| fh.write(data) | |
| sums_p = os.path.join(os.path.dirname(spec_p), "checksums.sha256") | |
| with open(sums_p) as fh: | |
| lines = fh.read().splitlines() | |
| hits = [i for i, ln in enumerate(lines) if ln.endswith(" spec.json")] | |
| if len(hits) != 1: | |
| sys.exit(f"{sums_p}: expected one spec.json line, found {len(hits)}") | |
| lines[hits[0]] = f"{hashlib.sha256(data).hexdigest()} spec.json" | |
| with open(sums_p, "w") as fh: | |
| fh.write("\n".join(lines) + "\n") | |
| PY | |
| if grep -l '"defect"' "$DS"/tasks/afx-*/tests/spec.json; then echo "refusing: card ids left in the specs above"; exit 1; fi | |
| if command -v sha256sum >/dev/null; then SHA=(sha256sum -c --quiet); else SHA=(shasum -a 256 -c --quiet); fi | |
| for t in "$DS"/tasks/afx-*/; do | |
| (cd "$t/tests" && "${SHA[@]}" checksums.sha256 >/dev/null) || { echo "refusing: checksum mismatch in ${t}tests"; exit 1; } | |
| done | |
| # every task the registry names has to be staged | |
| python3 - "$DS" <<'PY' || { echo "refusing: registry.json names tasks that are not staged"; exit 1; } | |
| import json, os, sys | |
| ds, missing = sys.argv[1], 0 | |
| for d in json.load(open(os.path.join(ds, "registry.json"))): | |
| gone = [t["path"] for t in d["tasks"] if not os.path.isfile(os.path.join(ds, t["path"], "task.toml"))] | |
| missing += len(gone) | |
| print(f" {d['name']}@{d['version']}: {len(d['tasks'])} tasks" + (f", not staged: {gone[:3]}" if gone else "")) | |
| sys.exit(1 if missing else 0) | |
| PY | |
| DEV="$( (grep -l '^version = ".*-dev"' "$DS"/tasks/afx-*/task.toml || true) | wc -l | tr -d ' ')" | |
| [ "$DEV" = "0" ] || echo "warning: $DEV expert task.toml files still carry a -dev version; rebuild them with the release version" | |
| # ---- OpenEnv Space: env package at the root, generator + released tasks vendored for the Docker build | |
| mkdir -p "$SP/.env-src" "$SP/alertforge_src" | |
| stage "$SP/.env-src" openenv/alertforge_env && cp -R "$SP/.env-src/openenv/alertforge_env/." "$SP"/ && rm -rf "$SP/.env-src" | |
| stage "$SP/alertforge_src" src workflows pyproject.toml README.md | |
| stage "$SP" tasks | |
| sed -e 's#^COPY pyproject.toml README.md /app/alertforge/#COPY alertforge_src /app/alertforge#' \ | |
| -e '/^COPY src \/app\/alertforge\/src/d' -e '/^COPY workflows \/app\/alertforge\/workflows/d' \ | |
| -e 's#^COPY openenv/alertforge_env /app/env/alertforge_env#COPY . /app/env/alertforge_env#' \ | |
| "$ROOT/openenv/alertforge_env/server/Dockerfile" > "$SP/Dockerfile" | |
| find "$SP" \( -name '__pycache__' -o -name '*.egg-info' \) -type d -prune -exec rm -rf {} + | |
| python3 "$ROOT/scripts/redact.py" --check "$DS" && python3 "$ROOT/scripts/redact.py" --check "$SP" || { echo "refusing: secrets or local paths in staging"; exit 1; } | |
| CORE_N="$(find "$DS/tasks" -maxdepth 2 -path '*/af-[0-9]*/task.toml' | wc -l | tr -d ' ')" | |
| EXPERT_N="$(find "$DS/tasks" -maxdepth 2 -path '*/afx-*/task.toml' | wc -l | tr -d ' ')" | |
| echo "staged dataset at $DS ($CORE_N core + $EXPERT_N expert tasks) and Space at $SP" | |
| if [ "${CONFIRM:-0}" != "1" ]; then | |
| echo "dry run only. Re-run with CONFIRM=1 to upload to https://huggingface.co/datasets/$HF_REPO"; exit 0 | |
| fi | |
| [ "$DEV" = "0" ] || { echo "refusing to upload: $DEV expert tasks carry a -dev version"; exit 1; } | |
| hf repo create "$HF_REPO" --repo-type dataset || true | |
| hf upload "$HF_REPO" "$DS" . --repo-type dataset --commit-message "3amBench $VERSION" | |
| hf repo tag create "$HF_REPO" "$VERSION" --repo-type dataset | |
| # The Hub tag v0.1.1 was created on a revision that still held the v0.1.0 task files, and the core tasks change in | |
| # 0.2.0 (reserved .example domains), so it is not re-pointed: delete it by hand once this upload is verified | |
| # (CHANGELOG 0.2.0, "Hugging Face tags"). | |
| echo "then: hf repo tag delete $HF_REPO v0.1.1 --repo-type dataset -y" | |
| if [ -n "${SPACE_REPO:-}" ]; then | |
| (cd "$SP" && openenv push --repo-id "$SPACE_REPO") | |
| fi | |
| # clean-cache verification, as a user would run it: | |
| echo "verify: rm -rf ~/.cache/harbor/tasks && uvx harbor run --repo https://huggingface.co/datasets/$HF_REPO@$VERSION -d 3ambench@${VERSION#v} -a oracle -n 4" | |
| echo "verify: uvx harbor run --repo https://huggingface.co/datasets/$HF_REPO@$VERSION -d 3ambench-expert-lb@${VERSION#v} -a oracle -n 4" | |