Initial deployment: GLiNER Large entity extractor with 70 labels

README.md

@@ -0,0 +1,62 @@
+---
+title: T9 Oracle Entity Extractor
+emoji: 🔬
+colorFrom: blue
+colorTo: green
+sdk: gradio
+sdk_version: 4.44.0
+app_file: app.py
+pinned: false
+license: mit
+hardware: t4-small
+---
+
+# T9 Oracle Entity Extractor
+
+**Zero-shot NER for Medical Device Technical Documentation**
+
+This Space provides entity extraction using GLiNER Large (1.7GB) for technical documentation in the medical device domain.
+
+## Features
+
+- **70 Entity Labels** across 10 tiers
+- **Zero-shot learning** - no training required
+- **Medical device focus** - optimized for endoscope equipment, parts, specifications
+- **Hardware:** NVIDIA T4 GPU for fast inference
+
+## Entity Types
+
+### Tier 1: Critical Identifiers
+- part_number, component_name, manufacturer, model_number
+
+### Tier 2: Specifications
+- pressure, temperature, voltage, current, material, dimensions, flow_rate, power
+
+### Tier 3: Standards & Compliance
+- standard_reference (ISO, ASTM, EN, IEC, ANSI), certification, compliance
+
+### Tier 4-10: Additional Labels
+- Thread standards, geometry, documentation, operational parameters, manufacturing IDs, medical device specific, visual elements, quality & maintenance
+
+## API Usage
+
+```python
+from gradio_client import Client
+
+client = Client("YOUR_USERNAME/t9-oracle-gliner-entity-extractor")
+
+text = "Part Number: A70002-2, Material: SS316L, Pressure: 60 psi"
+result = client.predict(text, api_name="/extract")
+print(result)
+```
+
+## Configuration
+
+- **Model:** urchade/gliner_large-v2.1
+- **GPU:** NVIDIA T4 (16GB VRAM)
+- **Cost:** $0.60/hour (Persistent)
+- **Max input:** 10,000 characters per request
+
+## Project
+
+Part of the T9 Oracle Knowledge Base Extraction System for Auto Sink medical device documentation.

app.py

@@ -0,0 +1,170 @@
+"""
+T9 Oracle GLiNER Entity Extractor - HF Space Deployment
+Gradio API endpoint for zero-shot NER with 70 medical device labels
+
+Deployed on: Persistent T4 GPU
+Model: urchade/gliner_large-v2.1 (1.7GB)
+Cost: $0.60/hour
+"""
+
+import gradio as gr
+import json
+import logging
+from typing import List, Dict
+from gliner import GLiNER
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+# 70 APPROVED ENTITY LABELS (from T9 configuration)
+ENTITY_LABELS = [
+    # Tier 1: Critical Identifiers (4)
+    "part_number", "component_name", "manufacturer", "model_number",
+    # Tier 2: Specifications & Measurements (13)
+    "pressure", "temperature", "voltage", "current", "material",
+    "dimension", "weight", "volume", "flow_rate", "power",
+    "diameter", "length", "thickness",
+    # Tier 3: Standards & Compliance (4)
+    "standard_reference", "certification", "compliance", "safety_class",
+    # Tier 4: Geometry & Mechanical (11)
+    "thread_standard", "pipe_size", "tubing_size", "connector_type",
+    "surface_finish", "surface_treatment", "width", "height",
+    "tolerance", "hardness", "torque",
+    # Tier 5: Documentation (7)
+    "diagram_reference", "drawing_number", "procedure_number",
+    "test_protocol", "revision", "sku_number", "part_label",
+    # Tier 6: Operational Parameters (9)
+    "accuracy", "speed", "frequency", "resistance",
+    "operating_temperature", "supply_voltage", "response_time",
+    "duty_cycle", "operating_range",
+    # Tier 7: Manufacturing (8)
+    "operator_id", "tool_number", "gauge_id", "fixture_number",
+    "machine_id", "lot_number", "serial_number", "batch_id",
+    # Tier 8: Medical Device (7)
+    "medical_device", "scope_manufacturer", "channel_type",
+    "port_type", "hub_type", "color_code", "leak_test",
+    # Tier 9: Visual Elements (2)
+    "diagram_type", "technical_annotation",
+    # Tier 10: Quality & Maintenance (7)
+    "calibration_interval", "service_interval", "mtbf",
+    "warranty", "expiration_date", "production_date", "inspection_report"
+]
+
+# Load GLiNER model (runs once on Space startup)
+logger.info("Loading GLiNER Large model (1.7GB)...")
+model = GLiNER.from_pretrained("urchade/gliner_large-v2.1")
+logger.info(f"✓ GLiNER loaded with {len(ENTITY_LABELS)} labels")
+
+
+def extract_entities(text: str, max_length: int = 10000) -> str:
+    """
+    Extract entities from text using GLiNER zero-shot NER
+    
+    Args:
+        text: Input text (max 10,000 characters recommended)
+        max_length: Maximum text length per prediction
+        
+    Returns:
+        JSON string with extracted entities
+    """
+    if not text or not text.strip():
+        return json.dumps({"entities": [], "error": "Empty text provided"})
+    
+    # Truncate if too long
+    if len(text) > max_length:
+        logger.warning(f"Text truncated from {len(text)} to {max_length} chars")
+        text = text[:max_length]
+    
+    try:
+        # GLiNER prediction
+        predictions = model.predict_entities(text, ENTITY_LABELS)
+        
+        # Format output
+        entities = []
+        for pred in predictions:
+            entities.append({
+                "text": pred.get("text", ""),
+                "label": pred.get("label", ""),
+                "start": pred.get("start", 0),
+                "end": pred.get("end", 0),
+                "score": float(pred.get("score", 0.0))
+            })
+        
+        logger.info(f"Extracted {len(entities)} entities from {len(text)} chars")
+        
+        return json.dumps({
+            "entities": entities,
+            "input_length": len(text),
+            "entity_count": len(entities),
+            "labels_used": len(ENTITY_LABELS)
+        }, indent=2)
+        
+    except Exception as e:
+        logger.error(f"Extraction failed: {e}")
+        return json.dumps({"entities": [], "error": str(e)})
+
+
+def batch_extract(text_batch: str) -> str:
+    """
+    Extract entities from multiple texts (newline-separated)
+    
+    Args:
+        text_batch: Multiple texts separated by double newlines
+        
+    Returns:
+        JSON string with results for each text
+    """
+    texts = [t.strip() for t in text_batch.split("\n\n") if t.strip()]
+    
+    results = []
+    for i, text in enumerate(texts):
+        result_json = extract_entities(text)
+        result = json.loads(result_json)
+        result["text_index"] = i
+        results.append(result)
+    
+    return json.dumps({"results": results, "batch_size": len(texts)}, indent=2)
+
+
+# Create Gradio interface
+demo = gr.Interface(
+    fn=extract_entities,
+    inputs=[
+        gr.Textbox(
+            lines=10,
+            placeholder="Enter technical text here (max 10,000 chars)...",
+            label="Input Text"
+        )
+    ],
+    outputs=gr.JSON(label="Extracted Entities"),
+    title="T9 Oracle Entity Extractor (GLiNER Large)",
+    description=f"""
+    **Zero-shot NER for Medical Device Technical Documentation**
+    
+    Extracts **{len(ENTITY_LABELS)} entity types** across 10 tiers:
+    - Part numbers, dimensions, materials, standards
+    - Electrical specs, pressure, temperature, flow rates
+    - Thread standards, tolerances, surface treatments
+    - Medical device specific (scopes, channels, colors)
+    - Quality & maintenance data
+    
+    **Model:** GLiNER Large v2.1 (1.7GB)  
+    **Hardware:** NVIDIA T4 GPU (16GB VRAM)  
+    **Max input:** 10,000 characters per request
+    """,
+    examples=[
+        ["Part Number: A70002-2, Material: SS316L, Pressure: 60 psi, Thread: 1/4\" NPT"],
+        ["Standard: ISO 1179-2, ASTM A112, Temperature: -40 to 85°C, Dimension: 6mm x 35mm"],
+        ["Manufacturer: Olympus, Channel: Biopsy, Color: Orange Tubing, Serial: SN-123456"]
+    ],
+    api_name="extract",  # Important: enables API access
+    allow_flagging="never"
+)
+
+# Launch with API enabled
+if __name__ == "__main__":
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=False  # HF Spaces handles sharing
+    )

requirements.txt

@@ -0,0 +1,8 @@
+# T9 Oracle GLiNER Entity Extractor - HF Space Dependencies
+# Approved configuration from /tmp/entity_extraction_final_configuration.md
+
+gliner==0.2.8
+torch>=2.0.0
+transformers>=4.30.0
+gradio>=4.0.0
+Pillow>=10.0.0