Update SpatialBench pipeline

.env.example

@@ -0,0 +1,9 @@
+# SpatialBench — API Keys
+# Copy this file to .env and fill in your keys.
+# The .env file is read automatically by run_experiments.py and app.py.
+# On HuggingFace Spaces, set these as Space Secrets instead.
+
+GEMINI_API_KEY=your_gemini_key_here
+OPENAI_API_KEY=your_openai_key_here
+ANTHROPIC_API_KEY=your_anthropic_key_here
+DEEPSEEK_API_KEY=your_deepseek_key_here

app.py

@@ -376,6 +376,9 @@ Key findings:
 """
 
 CSS = """
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=IBM+Plex+Mono:wght@400;500&display=swap');
+*, body, .gradio-container { font-family: 'Inter', ui-sans-serif, system-ui, sans-serif !important; }
+code, pre, .monospace { font-family: 'IBM Plex Mono', ui-monospace, monospace !important; }
 .leaderboard-table { font-size: 0.9em; }
 .status-badge-running { color: #2196F3; font-weight: bold; }
 .status-badge-completed { color: #4CAF50; font-weight: bold; }

pipeline/job_monitor.py

@@ -103,10 +103,8 @@ def _query_slurm(job_ids: list[str]) -> dict[str, JobStatus]:
         return {}
 
 
-def submit_sbatch(script_text: str, script_path: str) -> str | None:
-    """Write script_text to script_path, submit via sbatch, return job_id."""
-    with open(script_path, "w") as f:
-        f.write(script_text)
+def submit_sbatch(script_path: str) -> str | None:
+    """Submit an existing sbatch script at script_path, return job_id."""
     try:
         result = subprocess.run(
             ["sbatch", script_path],

requirements.txt

@@ -22,11 +22,5 @@ google-generativeai>=0.5.0
 
 # (DeepSeek uses the OpenAI-compatible client — no extra package needed)
 
-# Sentence embeddings for reasoning quality analysis
-sentence-transformers>=2.6.0
-
-# ROUGE for reasoning quality analysis
-rouge-score>=0.1.2
-
 # Environment variable loading
 python-dotenv>=1.0.0

run_experiments.py

@@ -98,7 +98,7 @@ def run_slurm(jobs: list[ExperimentJob], monitor: JobMonitor, dry_run: bool) ->
             tmp.write(script_text)
             script_path = tmp.name
 
-        job_id = submit_sbatch(script_text, script_path)
+        job_id = submit_sbatch(script_path)
         if job_id:
             monitor.add(
                 job_id=job_id,

upload_to_hf.sh

@@ -0,0 +1,119 @@
+#!/usr/bin/env bash
+# upload_to_hf.sh
+# ----------------
+# Upload the pipeline/ folder to a HuggingFace Space.
+#
+# Usage:
+#   ./upload_to_hf.sh [HF_SPACE]
+#
+# Examples:
+#   ./upload_to_hf.sh your-username/SpatialBench
+#   HF_TOKEN=hf_xxx ./upload_to_hf.sh your-username/SpatialBench
+#
+# The script uploads only the files needed to run the Space:
+#   app.py, pipeline/, configs/, requirements.txt, README.md
+# It never uploads .env, __pycache__, or result/log directories.
+#
+# Requirements:
+#   pip install huggingface_hub
+
+set -euo pipefail
+
+# ---------------------------------------------------------------------------
+# Configuration
+# ---------------------------------------------------------------------------
+
+HF_SPACE="${1:-}"
+if [[ -z "$HF_SPACE" ]]; then
+    echo "Usage: $0 <your-username/SpaceName>"
+    echo "Example: $0 weijiang/SpatialBench"
+    exit 1
+fi
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# ---------------------------------------------------------------------------
+# Auth
+# ---------------------------------------------------------------------------
+
+if [[ -z "${HF_TOKEN:-}" ]]; then
+    # Try reading from huggingface-cli login cache
+    HF_TOKEN_FILE="${HOME}/.cache/huggingface/token"
+    if [[ -f "$HF_TOKEN_FILE" ]]; then
+        HF_TOKEN="$(cat "$HF_TOKEN_FILE")"
+        export HF_TOKEN
+    fi
+fi
+
+if [[ -z "${HF_TOKEN:-}" ]]; then
+    echo "No HuggingFace token found."
+    echo "Either:"
+    echo "  1. Run:  huggingface-cli login"
+    echo "  2. Set:  export HF_TOKEN=hf_xxx"
+    exit 1
+fi
+
+echo "Uploading to HuggingFace Space: $HF_SPACE"
+echo "Source directory: $SCRIPT_DIR"
+echo ""
+
+# ---------------------------------------------------------------------------
+# Upload via Python (huggingface_hub)
+# ---------------------------------------------------------------------------
+
+python - <<PYEOF
+import os
+import sys
+from pathlib import Path
+from huggingface_hub import HfApi, upload_folder
+
+api = HfApi(token=os.environ["HF_TOKEN"])
+space_id = "$HF_SPACE"
+local_dir = Path("$SCRIPT_DIR")
+
+# Ensure the Space exists (type=gradio); if it already exists this is a no-op.
+try:
+    api.create_repo(
+        repo_id=space_id,
+        repo_type="space",
+        space_sdk="gradio",
+        exist_ok=True,
+        private=True,
+    )
+    print(f"Space ready: https://huggingface.co/spaces/{space_id}")
+except Exception as e:
+    print(f"[WARN] Could not create/verify Space: {e}")
+    print("Proceeding with upload anyway...")
+
+# Files/dirs to ignore (never upload secrets, cache, or large result dirs)
+IGNORE_PATTERNS = [
+    ".env",
+    "*.pyc",
+    "__pycache__",
+    "*.egg-info",
+    ".git",
+    ".gitignore",
+    # Large result/log directories that live outside pipeline/ anyway
+    "spatial-reasoning-results*",
+    "spatial_reasoning_logs",
+    "llm-maze-evaluation-results",
+    "eval_llm_logs",
+]
+
+print(f"\nUploading files from: {local_dir}")
+print("Ignoring: " + ", ".join(IGNORE_PATTERNS))
+print("")
+
+url = upload_folder(
+    repo_id=space_id,
+    repo_type="space",
+    folder_path=str(local_dir),
+    path_in_repo=".",
+    ignore_patterns=IGNORE_PATTERNS,
+    commit_message="Update SpatialBench pipeline",
+    token=os.environ["HF_TOKEN"],
+)
+
+print(f"\nUpload complete!")
+print(f"Space URL: https://huggingface.co/spaces/{space_id}")
+PYEOF