Merge pull request #33 from synacktraa/main

feat: use uv for running scripts in venvs

benchmarking/README.md

@@ -9,7 +9,7 @@ COHERE_API_KEY=
 
 ### Fetch evals from supabase
   ```
-  python fetch_evals.py
+  uv run fetch_evals.py
   ```
 
 ### Generating results
@@ -22,27 +22,27 @@ COHERE_API_KEY=
 
 - Anthropic Grounding Locatr
   ```
-  python grounding_locatr.py anthropic
+  uv run grounding_locatr.py anthropic
   ```
 
 - OS Atlas Grounding Locatr
   ```
-  python grounding_locatr.py os_atlas
+  uv run grounding_locatr.py os_atlas
   ```
 
 ### Comparing results with evals
 
 - Original Locatr
   ```
-  python compare.py original
+  uv run compare.py original
   ```
 
 - Anthropic Grounding Locatr
   ```
-  python compare.py anthropic
+  uv run compare.py anthropic
   ```
 
 - OS Atlas Grounding Locatr
   ```
-  python compare.py os_atlas
+  uv run compare.py os_atlas
   ```

benchmarking/compare.py

@@ -1,6 +1,7 @@
+import sys
 from pathlib import Path
+
 import yaml
-import sys
 from playwright.sync_api import sync_playwright
 
 if len(sys.argv) == 2:

benchmarking/fetch_evals.py

@@ -1,9 +1,10 @@
 import os
-import yaml
 import re
 from urllib.parse import urlparse, urlunparse
-from supabase import create_client
+
+import yaml
 from postgrest.exceptions import APIError
+from supabase import create_client
 
 # Initialize Supabase client
 supabase= create_client(os.environ["SUPABASE_URL"], os.environ["SUPABASE_KEY"])

benchmarking/grounding_locatr.py

@@ -1,22 +1,20 @@
-from playwright.sync_api import sync_playwright
-from pathlib import Path
+import logging
 import re
 import sys
-import yaml
+from base64 import standard_b64encode
 from dataclasses import dataclass, asdict
-from typing import Literal
 from pathlib import Path
-from base64 import standard_b64encode
 from textwrap import dedent
-from json_repair import repair_json
-import logging
+from typing import Literal
 
+import yaml
 from anthropic import Anthropic
-
 from dotenv import load_dotenv
-
 from gradio_client import Client, handle_file
+from json_repair import repair_json
 from PIL import Image, ImageDraw
+from playwright.sync_api import sync_playwright
+
 
 load_dotenv()
 

benchmarking/pyproject.toml

@@ -0,0 +1,16 @@
+[project]
+name = "benchmarking"
+version = "0.1.0"
+description = "Add your description here"
+readme = "README.md"
+requires-python = ">=3.12.0"
+dependencies = [
+    "anthropic>=0.45.2",
+    "gradio-client>=1.7.0",
+    "json-repair>=0.37.0",
+    "pillow>=11.1.0",
+    "playwright>=1.50.0",
+    "python-dotenv>=1.0.1",
+    "pyyaml>=6.0.2",
+    "supabase>=2.13.0",
+]

pyproject.toml

@@ -63,4 +63,7 @@ exclude_lines = [
     "if __name__ == .__main__.:",
     "pass",
     "raise ImportError",
-]
+]
+
+[tool.uv.workspace]
+members = ["benchmarking"]