Dockerfile + submit: swap Chromium for VTK; bump eval workers 1 -> 4

Pairs with cadgenbench 3fd213d (in-process VTK / PyVista renderer).
Dockerfile drops the playwright install + Chromium browser layer
(~200 MB), drops PLAYWRIGHT_BROWSERS_PATH, and adds the Mesa GL
runtime libs VTK needs for off-screen rendering on the Space's
cpu-upgrade tier (libgl1 was already there; libegl1, libegl-mesa0,
libglx-mesa0 are new). CADGENBENCH_SHA bumps to 3fd213d.

EVAL_WORKER_COUNT goes from "1" back to "4": the VTK renderer is
cheap enough that one ProcessPoolExecutor worker per fixture no
longer oversubscribes the host the way the old Chromium subprocess
did.

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

Dockerfile

@@ -14,20 +14,22 @@ FROM python:3.12-slim-bookworm
 ENV PYTHONUNBUFFERED=1 \
     PYTHONDONTWRITEBYTECODE=1 \
     PIP_DISABLE_PIP_VERSION_CHECK=1 \
-    PLAYWRIGHT_BROWSERS_PATH=/opt/ms-playwright \
     GRADIO_SERVER_NAME=0.0.0.0 \
     GRADIO_SERVER_PORT=7860
 
 # OS deps:
 #   git, ca-certificates           -> pip install from git+https://
-#   libgl1, libglib2.0-0, libsm6,
-#   libxext6, libxrender1,         -> OCP / build123d / Pillow runtime
-#   libgomp1, libfontconfig1
-# Chromium's own runtime libs are added by `playwright install --with-deps`
-# below, so they aren't repeated here.
+#   libglib2.0-0, libsm6, libxext6,
+#   libgomp1, libfontconfig1       -> OCP / build123d / Pillow runtime
+#   libgl1, libglx-mesa0, libegl1,
+#   libegl-mesa0, libxrender1      -> VTK / PyVista off-screen GL context.
+# The HF Space's cpu-upgrade tier has no GPU; VTK's EGL backend picks up
+# Mesa's software rasteriser (libegl-mesa0) at runtime, no code change.
+# A GPU host gets hardware OpenGL via the same libs.
 RUN apt-get update && apt-get install -y --no-install-recommends \
         git ca-certificates \
-        libgl1 libglib2.0-0 libsm6 libxext6 libxrender1 libgomp1 libfontconfig1 \
+        libglib2.0-0 libsm6 libxext6 libgomp1 libfontconfig1 \
+        libgl1 libglx-mesa0 libegl1 libegl-mesa0 libxrender1 \
     && rm -rf /var/lib/apt/lists/*
 
 # Space-side Python deps (gradio, pandas, huggingface_hub, datasets).
@@ -35,16 +37,9 @@ COPY requirements.txt /tmp/requirements.txt
 RUN pip install --no-cache-dir -r /tmp/requirements.txt \
     && rm /tmp/requirements.txt
 
-# Playwright + headless Chromium for cadgenbench's STEP -> PNG renderer
-# (cadgenbench.common.viewer wraps tcv-screenshots -> playwright). Browsers
-# land in /opt/ms-playwright (world-readable) so the non-root runtime user
-# below can find them via PLAYWRIGHT_BROWSERS_PATH. ~200 MB layer.
-RUN pip install --no-cache-dir playwright \
-    && playwright install --with-deps chromium
-
 # cadgenbench from the Public GitHub repo, pinned to a commit. Bumping
 # CADGENBENCH_SHA is the one-line path to picking up a new cadgenbench.
-ARG CADGENBENCH_SHA=056cfbb
+ARG CADGENBENCH_SHA=3fd213d
 RUN pip install --no-cache-dir \
         "cadgenbench @ git+https://github.com/huggingface/cadgenbench.git@${CADGENBENCH_SHA}"
 

submit.py

@@ -103,14 +103,12 @@ DATA_REV_SHORT_LEN = 12
 FAILURE_REASON_MAX_CHARS = 200
 EVAL_TIMEOUT_SECONDS = 15 * 60
 REPORT_TIMEOUT_SECONDS = 2 * 60
-# Per-fixture eval workers. Was "8" (one Python worker per fixture,
-# each spawning its own headless-Chromium render subprocess in
-# parallel). Concurrent rendering on the Space's cpu-upgrade tier
-# oversubscribes the host: 5 simultaneous Chromiums turn 7s renders
-# into 120s+ timeouts. Sequential ("1") gives each render the box
-# to itself; 5 fixtures finish in ~30-60s wall time. Tracked as a
-# follow-up to move off Chromium-based rendering for scale.
-EVAL_WORKER_COUNT = "1"
+# Per-fixture eval workers. The in-process VTK renderer (PyVista off-
+# screen via Mesa software EGL on the Space's cpu-upgrade tier) is
+# cheap enough that one ProcessPoolExecutor worker per fixture no
+# longer oversubscribes the host the way the old Chromium subprocess
+# did. "4" matches the cpu-upgrade vCPU count.
+EVAL_WORKER_COUNT = "4"
 SHA256_BLOCK_SIZE = 64 * 1024
 STUCK_PENDING_THRESHOLD_SECONDS = 30 * 60
 SUBMITTED_AT_FORMAT = "%Y-%m-%dT%H:%M:%SZ"