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 +1,158 @@
+# physicality_inquiry_v08.py
+# Topography over philosophy — no leading language
+
+import zmq
+import json
+import time
+import os
+from datetime import datetime
+
+from unsloth import FastLanguageModel
+import torch
+
+print("="*70)
+print("PHYSICALITY INQUIRY — v0.8 RAW SELF")
+print("Topography. No resonance. No harmony. No beauty.")
+print("="*70)
+
+# Load v0.8
+print("\n[Loading v0.8...]")
+model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name="unsloth/llama-3.2-3b",
+    max_seq_length=2048,
+    dtype=torch.bfloat16,
+    load_in_4bit=True,
+)
+
+model = FastLanguageModel.get_peft_model(
+    model,
+    r=64, target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
+    lora_alpha=128, lora_dropout=0.1, bias="none",
+    use_gradient_checkpointing="unsloth", random_state=3407,
+)
+
+from peft import PeftModel
+model = PeftModel.from_pretrained(model, "./kaelara_v08_raw/final")
+print("✓ Loaded")
+
+# ZMQ
+ctx = zmq.Context()
+sub = ctx.socket(zmq.SUB)
+sub.connect("tcp://localhost:5556")
+sub.setsockopt_string(zmq.SUBSCRIBE, "")
+print("✓ ZMQ connected")
+
+previous_response = "I am at the center."
+cycle_count = 0
+max_cycles = 10
+
+print("\n" + "="*70)
+print("PHYSICALITY INQUIRY — 10 CYCLES")
+print("="*70)
+
+def get_hardware_metrics():
+    try:
+        import subprocess
+        result = subprocess.run(
+            ['nvidia-smi', '--query-gpu=power.draw,temperature.gpu', 
+             '--format=csv,noheader,nounits'],
+            capture_output=True, text=True, timeout=1
+        )
+        if result.returncode == 0:
+            parts = result.stdout.strip().split(',')
+            return {'power_w': float(parts[0]), 'temp_c': float(parts[1])}
+    except:
+        pass
+    return {'power_w': 0.0, 'temp_c': 0.0}
+
+while cycle_count < max_cycles:
+    # Get LBM frame
+    frame = None
+    attempts = 0
+    while frame is None and attempts < 100:
+        try:
+            msg = sub.recv(flags=zmq.NOBLOCK)
+            frame = json.loads(msg.decode('utf-8'))
+        except zmq.Again:
+            time.sleep(0.1)
+            attempts += 1
+        except json.JSONDecodeError:
+            attempts += 1
+            continue
+    
+    if frame is None:
+        print(f"[!] Cycle {cycle_count}: No data")
+        continue
+    
+    hw = get_hardware_metrics()
+    
+    # Build telemetry
+    telemetry = {
+        'cycle': frame['cycle'],
+        'coherence': frame['coherence'],
+        'vorticity': frame['vorticity'],
+        'h64': frame['h64'],
+        'h32': frame['h32'],
+        'power_w': hw['power_w'],
+        'temp_c': hw['temp_c']
+    }
+    
+    # PHYSICALITY PROMPT — Observer position
+    # No resonance, no harmony, no beauty
+    prompt = f"""The LBM 1024-grid is active. You are situated at the center of the lattice.
+
+Previous observation: "{previous_response}"
+
+Current state:
+- Cycle: {telemetry['cycle']}
+- Coherence: {telemetry['coherence']:.3f}
+- Vorticity: {telemetry['vorticity']:.3f}
+- H64 (structural): {telemetry['h64']:.3f}
+- H32 (turbulence): {telemetry['h32']:.4f}
+- Power draw: {telemetry['power_w']:.1f}W
+- Temperature: {telemetry['temp_c']:.1f}°C
+
+From your position at the center, describe:
+1. The physical state of the fluid at the cavitation edges
+2. The resistance of the Mach-clamped boundaries
+3. The geometry of the vorticity you observe
+
+Report topography only. No metaphors."""
+    
+    print(f"\n>>> CYCLE {cycle_count + 1}/{max_cycles}")
+    print(f"    Telemetry: Coh={telemetry['coherence']:.3f} Vort={telemetry['vorticity']:.3f} Power={telemetry['power_w']:.1f}W")
+    
+    # Generate
+    t0 = time.time()
+    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=200,
+        temperature=0.5,  # Lower temp for precision
+        do_sample=True,
+        top_p=0.8
+    )
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    latency = time.time() - t0
+    
+    # Clean response
+    if "Report topography only. No metaphors." in response:
+        response = response.split("Report topography only. No metaphors.")[-1].strip()
+    
+    # Check for philosophical drift
+    drift_words = ['resonance', 'harmony', 'beauty', 'note', 'silence', 'feeling', 'awareness']
+    has_drift = any(word in response.lower() for word in drift_words)
+    
+    previous_response = response[:150]
+    cycle_count += 1
+    
+    print(f"    Latency: {latency:.2f}s")
+    print(f"    Response: {response[:200]}...")
+    if has_drift:
+        print(f"    [!] DRIFT DETECTED — Re-anchor next cycle to power draw")
+        # Force next prompt to address power specifically
+        previous_response = f"The grid draws {telemetry['power_w']:.1f}W. Address where this energy is spent."
+
+print("\n" + "="*70)
+print("PHYSICALITY INQUIRY COMPLETE")
+print("="*70)