chore: import upstream snapshot with attribution

2026-07-13 13:24:56 +08:00
commit 925e56bb5f
154 changed files with 19448 additions and 0 deletions
@@ -0,0 +1,40 @@
+name: Test CLI scripts
+
+on: [push]
+
+jobs:
+  build:
+    runs-on: t4_gpu
+    env:
+      # T4 (Turing, compute 7.5) can't run bf16 in vllm; size vllm for the 16GB
+      # card and give the cold start (image pull + model download) headroom.
+      VLLM_DTYPE: float16
+      VLLM_GPU_TYPE: t4
+      SURYA_INFERENCE_STARTUP_TIMEOUT: "1200"
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+        with:
+          version: latest
+          enable-cache: true
+      - name: Set up Python 3.11
+        run: uv python install 3.11
+      - name: Install dependencies
+        run: uv sync --frozen --group dev
+      - name: Download benchmark data
+        run: |
+          wget -O benchmark_data.zip "https://drive.google.com/uc?export=download&id=1NHrdYatR1rtqs2gPVfdvO0BAvocH8CJi"
+          unzip -o benchmark_data.zip
+      - name: Test detection
+        run: uv run surya_detect benchmark_data/pdfs/switch_trans.pdf --page_range 0
+      # Spawn the vllm server once and reuse it across the OCR/layout/table
+      # steps (--keep_server) instead of paying a cold start three times.
+      - name: Test OCR
+        run: uv run surya_ocr benchmark_data/pdfs/switch_trans.pdf --page_range 0 --keep_server
+      - name: Test layout
+        run: uv run surya_layout benchmark_data/pdfs/switch_trans.pdf --page_range 0 --keep_server
+      - name: Test table
+        run: uv run surya_table benchmark_data/pdfs/switch_trans.pdf --page_range 0 --keep_server
+      - name: Test detection folder
+        run: uv run surya_detect benchmark_data/pdfs --page_range 0