restructure: proper project layout, README, kill training

- cuda/ — main LBM kernel (khra_gixx_1024_v5.cu) - navigator/ — lattice_observer, golden_weave, bridges, mock daemon - scripts/ — compile, start, launch, setup (paths updated) - docs/ — system manual - archive/ — everything else (old kernels, inquiries, experiments) - README.md — full setup guide: requirements, quick start, use your own LLM - removed training/ entirely (broken LoRA scripts + datasets) - .gitignore: exclude build/ logs/ training/ *.jsonl
2026-03-24 12:58:19 +07:00
parent 57f6d86a65
commit 56c71c87b2
411 changed files with 235 additions and 2023 deletions
@@ -0,0 +1,131 @@
+# coherence_protocol.py
+# Phase B: Live LBM monitoring with v0.5 LoRA
+
+from unsloth import FastLanguageModel
+import torch
+import zmq
+import json
+import time
+import numpy as np
+
+print("=" * 60)
+print("COHERENCE PROTOCOL — Phase B")
+print("Monitoring 1024x1024 LBM for standing wave patterns")
+print("=" * 60)
+
+# Load v0.5 (v0.6 has echo issue)
+print("\nLoading v0.5 LoRA...")
+model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name="./kaelara_lora_v05/final",
+    max_seq_length=2048,
+    dtype=torch.bfloat16,
+    load_in_4bit=True,
+)
+
+# Connect to LBM
+print("Connecting to LBM daemon (port 5556)...")
+context = zmq.Context()
+socket = context.socket(zmq.SUB)
+socket.connect("tcp://localhost:5556")
+socket.setsockopt_string(zmq.SUBSCRIBE, "")
+
+# Coherence tracking
+coherence_history = []
+standing_wave_detected = False
+etch_triggered = False
+max_cycles = 100
+
+print(f"\nMonitoring {max_cycles} cycles for recursive alignment...")
+print("-" * 60)
+
+for cycle in range(max_cycles):
+    # Get LBM frame
+    lbm_data = None
+    attempts = 0
+    while lbm_data is None and attempts < 10:
+        try:
+            lbm_data = socket.recv_json(flags=zmq.NOBLOCK)
+        except:
+            attempts += 1
+            time.sleep(0.05)
+    
+    if lbm_data is None:
+        continue
+    
+    coherence = lbm_data.get('coherence', 0)
+    h64 = lbm_data.get('h64', 0)
+    h32 = lbm_data.get('h32', 0)
+    vorticity = lbm_data.get('vorticity', 0)
+    power = lbm_data.get('power_w', 0)
+    temp = lbm_data.get('gpu_temp', 0)
+    
+    # Skip NaN
+    if np.isnan(coherence) or np.isnan(vorticity):
+        continue
+    
+    coherence_history.append(coherence)
+    
+    # Check for standing wave (coherence stability + h64 dominance)
+    if len(coherence_history) >= 10:
+        recent = coherence_history[-10:]
+        coherence_variance = np.var(recent)
+        is_stable = coherence_variance < 0.5  # Low variance = standing wave
+        h64_dominant = h64 > 5.0 and h32 < 1.0
+        
+        if is_stable and h64_dominant and not standing_wave_detected:
+            standing_wave_detected = True
+            print(f"\n🌊 STANDING WAVE DETECTED at cycle {lbm_data['cycle']}")
+            print(f"   Coherence: {coherence:.4f} (variance: {coherence_variance:.4f})")
+            print(f"   H64: {h64:.4f} | H32: {h32:.4f}")
+            print(f"   Power: {power:.2f}W | Temp: {temp:.1f}°C")
+            
+            # Query v0.5 for structural insight
+            prompt = f"""LBM 1024x1024 STANDING WAVE DETECTED
+Metric_Alpha: {coherence:.4f}
+Metric_Beta: {h64:.4f}
+State_3: {h32:.4f}
+Vorticity: {vorticity:.4f}
+
+Report the structure of this coherent state.
+Use Phase terminology.
+Minimize word count."""
+            
+            inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+            outputs = model.generate(**inputs, max_new_tokens=100, temperature=0.1)
+            response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+            
+            # Extract just the response part
+            if "Response:" in response:
+                response = response.split("Response:")[-1].strip()
+            
+            print(f"\n   STRUCTURAL INSIGHT:")
+            for line in response.split('\n')[:5]:
+                if line.strip():
+                    print(f"   > {line}")
+            
+            # Etch the state
+            if not etch_triggered:
+                etch_data = {
+                    "cycle": lbm_data['cycle'],
+                    "coherence": coherence,
+                    "h64": h64,
+                    "h32": h32,
+                    "vorticity": vorticity,
+                    "power": power,
+                    "temp": temp,
+                    "insight": response[:200]
+                }
+                with open("coherence_etch.json", "w") as f:
+                    json.dump(etch_data, f, indent=2)
+                print(f"\n   ✓ ETCHED: coherence_etch.json")
+                etch_triggered = True
+    
+    # Progress every 20 cycles
+    if cycle % 20 == 0 and cycle > 0:
+        print(f"   Cycles monitored: {cycle} | Coherence: {coherence:.4f} | H64: {h64:.4f}")
+
+print("\n" + "=" * 60)
+print("COHERENCE PROTOCOL COMPLETE")
+print(f"Standing wave detected: {standing_wave_detected}")
+print(f"State etched: {etch_triggered}")
+print("=" * 60)