Michael Rabinovich Cursor commited on
Commit
900c7c1
·
1 Parent(s): 00d6e0c

step 6 (A2): switch space to sdk:docker, install cadgenbench from private LeForge via GH_PAT build secret

Browse files

The HF Spaces Gradio SDK can't substitute Space secrets into requirements.txt
at build time (only at runtime), and embedding the literal token in
requirements.txt is a no-go (would land in git history of the Space).

Switching to sdk: docker collapses two open Step 6 sub-decisions into one
small change:

(A2) private GitHub dep install -- handled by `RUN --mount=type=secret`
in the Dockerfile; the PAT (Space secret GH_PAT) is mounted at
/run/secrets/GH_PAT for that one RUN and never lands in any layer.

(D) Playwright/Chromium for STEP -> PNG rendering -- handled by
`playwright install --with-deps chromium` in the same Dockerfile.

Layer ordering optimises rebuild speed: OS deps + requirements.txt deps
+ Chromium come early (stable), cadgenbench from git comes mid (bumps
on each commit), app.py + results.jsonl come last (changes most).

Front-matter: sdk: gradio -> sdk: docker, drop sdk_version + app_file
(Docker SDK uses Dockerfile CMD), add app_port: 7860. .dockerignore
keeps legacy/'s 19 MB out of the build context.

GH_PAT set on the Space via HfApi().add_space_secret(...) -- fine-grained
GitHub PAT, read-only Contents on MichaelRabinovich/LeForge, 1-year
expiry. Mirrors at /tmp/gh_pat locally for `docker build --secret`.

Co-authored-by: Cursor <cursoragent@cursor.com>

Files changed (3) hide show
  1. .dockerignore +11 -0
  2. Dockerfile +69 -0
  3. README.md +2 -3
.dockerignore ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Build context exclusions for `docker build` (HF Space + local).
2
+ # COPY directives in the Dockerfile are explicit so this is purely a
3
+ # build-speed optimisation: prevents sending 19 MB of legacy HTML and
4
+ # any local caches to the daemon on every build.
5
+ .git
6
+ .gitattributes
7
+ .gitignore
8
+ __pycache__
9
+ *.pyc
10
+ .DS_Store
11
+ legacy/
Dockerfile ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # syntax=docker/dockerfile:1.7
2
+ #
3
+ # Hugging Face Space (sdk: docker) for the CADGenBench leaderboard.
4
+ #
5
+ # HF builds this server-side on each git push. Local smoke test:
6
+ #
7
+ # docker buildx build --platform linux/amd64 \
8
+ # --secret id=GH_PAT,src=/tmp/gh_pat \
9
+ # -t cadgenbench-space-test .
10
+ #
11
+ # The GH_PAT mount is a fine-grained GitHub PAT, read-only on
12
+ # MichaelRabinovich/LeForge, used only by the `pip install cadgenbench`
13
+ # step below. It never lands in an image layer, env var, or disk file
14
+ # in the final image. The same value lives on the Space as a Settings
15
+ # Secret (mirrors /tmp/gh_pat locally).
16
+
17
+ FROM python:3.12-slim-bookworm
18
+
19
+ ENV PYTHONUNBUFFERED=1 \
20
+ PYTHONDONTWRITEBYTECODE=1 \
21
+ PIP_DISABLE_PIP_VERSION_CHECK=1 \
22
+ PLAYWRIGHT_BROWSERS_PATH=/opt/ms-playwright \
23
+ GRADIO_SERVER_NAME=0.0.0.0 \
24
+ GRADIO_SERVER_PORT=7860
25
+
26
+ # OS deps:
27
+ # git, ca-certificates -> pip install from git+https://
28
+ # libgl1, libglib2.0-0, libsm6,
29
+ # libxext6, libxrender1, -> OCP / build123d / Pillow runtime
30
+ # libgomp1, libfontconfig1
31
+ # Chromium's own runtime libs are added by `playwright install --with-deps`
32
+ # below, so they aren't repeated here.
33
+ RUN apt-get update && apt-get install -y --no-install-recommends \
34
+ git ca-certificates \
35
+ libgl1 libglib2.0-0 libsm6 libxext6 libxrender1 libgomp1 libfontconfig1 \
36
+ && rm -rf /var/lib/apt/lists/*
37
+
38
+ # Space-side Python deps (gradio, pandas, huggingface_hub, datasets).
39
+ COPY requirements.txt /tmp/requirements.txt
40
+ RUN pip install --no-cache-dir -r /tmp/requirements.txt \
41
+ && rm /tmp/requirements.txt
42
+
43
+ # Playwright + headless Chromium for cadgenbench's STEP -> PNG renderer
44
+ # (cadgenbench.common.viewer wraps tcv-screenshots -> playwright). Browsers
45
+ # land in /opt/ms-playwright (world-readable) so the non-root runtime user
46
+ # below can find them via PLAYWRIGHT_BROWSERS_PATH. ~200 MB layer.
47
+ RUN pip install --no-cache-dir playwright \
48
+ && playwright install --with-deps chromium
49
+
50
+ # cadgenbench from the private GitHub repo, pinned to a commit. The PAT
51
+ # mount is visible only inside this single RUN: not embedded in any layer,
52
+ # not exposed as env, not written to disk after the layer commits. Bumping
53
+ # CADGENBENCH_SHA is the one-line path to picking up a new cadgenbench.
54
+ ARG CADGENBENCH_SHA=d7e0468
55
+ RUN --mount=type=secret,id=GH_PAT,mode=0400,required=true \
56
+ pip install --no-cache-dir \
57
+ "cadgenbench @ git+https://$(cat /run/secrets/GH_PAT)@github.com/MichaelRabinovich/LeForge.git@${CADGENBENCH_SHA}"
58
+
59
+ # Drop privileges. HF Spaces conventionally run as uid 1000 with
60
+ # WORKDIR /home/user/app.
61
+ RUN useradd -m -u 1000 user \
62
+ && mkdir -p /home/user/app \
63
+ && chown -R user:user /home/user/app
64
+ USER user
65
+ WORKDIR /home/user/app
66
+ COPY --chown=user:user app.py results.jsonl ./
67
+
68
+ EXPOSE 7860
69
+ CMD ["python", "app.py"]
README.md CHANGED
@@ -3,9 +3,8 @@ title: CADGenBench Leaderboard
3
  emoji: 🔧
4
  colorFrom: indigo
5
  colorTo: pink
6
- sdk: gradio
7
- sdk_version: 6.14.0
8
- app_file: app.py
9
  pinned: true
10
  short_description: Leaderboard for AI-driven CAD generation
11
  ---
 
3
  emoji: 🔧
4
  colorFrom: indigo
5
  colorTo: pink
6
+ sdk: docker
7
+ app_port: 7860
 
8
  pinned: true
9
  short_description: Leaderboard for AI-driven CAD generation
10
  ---