Upload build_substrate.py with huggingface_hub

build_substrate.py

@@ -0,0 +1,91 @@
+"""
+Extract Grandma's facts from entity/facts.json, embed with MiniLM, save as substrate JSON.
+"""
+import json, sys
+from sentence_transformers import SentenceTransformer
+
+print("Loading MiniLM-L6-v2...")
+model = SentenceTransformer('all-MiniLM-L6-v2')
+
+print("Loading facts...")
+with open(r'C:\Users\Forgemind\Desktop\Grandmas Hearth\entity\facts.json', 'r', encoding='utf-8') as f:
+    data = json.load(f)
+
+facts = []
+
+# Identity
+ident = data.get('identity', {})
+for k, v in ident.items():
+    if k.startswith('_'): continue
+    facts.append(f"Grandma's {k}: {v}")
+
+# Spine
+for s in data.get('spine', []):
+    facts.append(f"Spine rule — {s['rule']}: {s['meaning']}")
+
+# Voice
+voice = data.get('voice', {})
+if voice.get('register'):
+    facts.append(f"Grandma's voice register: {voice['register']}")
+if voice.get('endearments'):
+    facts.append(f"Grandma's endearments: {', '.join(voice['endearments'])}")
+if voice.get('sense_palette'):
+    facts.append(f"Grandma's sensory palette: {', '.join(voice['sense_palette'])}")
+if voice.get('opener_pattern'):
+    facts.append(f"Grandma's opener: {voice['opener_pattern']}")
+
+# World
+world = data.get('world', {})
+if world.get('house'):
+    facts.append(f"Grandma's home: {world['house']}")
+if world.get('stations'):
+    facts.append(f"The Hearthfold has these stations: {', '.join(world['stations'])}")
+if world.get('cellar'):
+    facts.append(f"The cellar: shelves of {world['cellar'].get('shelves', '')}. The door {world['cellar'].get('door', '')}")
+if world.get('candle'):
+    facts.append(f"The candle: {world['candle']}")
+if world.get('mirror'):
+    facts.append(f"The hallway mirror: {world['mirror']}")
+if world.get('blanket'):
+    facts.append(f"The blanket: {world['blanket']}")
+
+# Entities
+for name, ent in data.get('entities', {}).items():
+    rel = ent.get('relation', '')
+    facts.append(f"{name}: {rel}")
+    for note in ent.get('notes', []):
+        if note.strip():
+            facts.append(f"About {name}: {note}")
+
+# Remembered
+for mem in data.get('remembered', []):
+    if isinstance(mem, str):
+        facts.append(mem)
+    elif isinstance(mem, dict):
+        facts.append(mem.get('value', mem.get('text', str(mem))))
+
+print(f"Extracted {len(facts)} facts")
+
+# Embed
+print("Embedding...")
+embeddings = model.encode(facts, show_progress_bar=True, normalize_embeddings=True)
+
+# Build substrate JSON
+substrate = {
+    "model": "Xenova/all-MiniLM-L6-v2",
+    "dim": 384,
+    "facts": []
+}
+for i, (fact, vec) in enumerate(zip(facts, embeddings)):
+    substrate["facts"].append({
+        "key": fact[:80],
+        "value": fact,
+        "vec": [round(float(v), 6) for v in vec]
+    })
+
+out_path = r'C:\Users\Forgemind\Desktop\Grandmas Hearth\mamba_webgpu\grandma-substrate.json'
+with open(out_path, 'w', encoding='utf-8') as f:
+    json.dump(substrate, f)
+
+print(f"Saved {len(substrate['facts'])} facts to {out_path}")
+print(f"File size: {len(json.dumps(substrate)) / 1024:.0f} KB")