Scruff AI
56c71c87b2
- 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
82 lines
2.1 KiB
Python
82 lines
2.1 KiB
Python
# minimal_open_feed.py
|
|
# Stripped down to test the actual bottleneck
|
|
|
|
import zmq
|
|
import json
|
|
import time
|
|
|
|
print("="*60)
|
|
print("MINIMAL OPEN FEED TEST")
|
|
print("="*60)
|
|
|
|
# ZMQ setup
|
|
ctx = zmq.Context()
|
|
sub = ctx.socket(zmq.SUB)
|
|
sub.connect("tcp://localhost:5556")
|
|
sub.setsockopt_string(zmq.SUBSCRIBE, "")
|
|
|
|
print("\nWaiting for LBM data...")
|
|
|
|
# Wait for first frame
|
|
frame = None
|
|
for i in range(50): # 5 seconds max
|
|
try:
|
|
msg = sub.recv(flags=zmq.NOBLOCK)
|
|
frame = json.loads(msg)
|
|
break
|
|
except zmq.Again:
|
|
time.sleep(0.1)
|
|
|
|
if frame is None:
|
|
print("ERROR: No LBM data received")
|
|
exit(1)
|
|
|
|
print(f"✓ Received frame: Cycle {frame['cycle']}, Coherence {frame['coherence']:.3f}")
|
|
|
|
# Now try loading the model
|
|
print("\nLoading v0.5 model...")
|
|
try:
|
|
from unsloth import FastLanguageModel
|
|
import torch
|
|
|
|
# Load base model + LoRA adapters
|
|
model, tokenizer = FastLanguageModel.from_pretrained(
|
|
model_name="unsloth/llama-3.2-3b",
|
|
max_seq_length=512,
|
|
dtype=torch.bfloat16,
|
|
load_in_4bit=True,
|
|
)
|
|
|
|
# Apply v0.5 LoRA
|
|
model = FastLanguageModel.get_peft_model(
|
|
model,
|
|
r=64,
|
|
target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
|
|
lora_alpha=128,
|
|
lora_dropout=0,
|
|
bias="none",
|
|
use_gradient_checkpointing="unsloth",
|
|
random_state=3407,
|
|
)
|
|
|
|
# Load the trained adapters
|
|
from peft import PeftModel
|
|
model = PeftModel.from_pretrained(model, "./kaelara_lora_v05/final")
|
|
print("✓ v0.5 LoRA adapters applied")
|
|
print("✓ Model loaded")
|
|
|
|
# Test generation
|
|
prompt = f"Coherence: {frame['coherence']:.3f}. What is the frequency?"
|
|
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
|
|
outputs = model.generate(**inputs, max_new_tokens=30, temperature=0.7)
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
|
print(f"\nPrompt: {prompt}")
|
|
print(f"Response: {response}")
|
|
print("\n✓ FULL PIPELINE WORKS")
|
|
|
|
except Exception as e:
|
|
print(f"✗ Error: {e}")
|
|
import traceback
|
|
traceback.print_exc()
|