Mystiatech/Ruby
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixing CUDA errors

Browse Source
This commit is contained in:
Dani 2025-04-26 22:34:05 -04:00
parent f21e9b433e
commit 5f74b2c64c
11 changed files with 105 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.gitignore vendored
View File

@ -174,4 +174,5 @@ cython_debug/
/data/memory/context.json /data/memory/context.json
/data/memory/dreams.json /data/memory/dreams.json
data/memory/brainmap.json data/memory/brainmap.json
/data/memory/vocab.json /data/memory/vocab.json
data/memory/book_progress.json

8
dashboard/dashboard.py
View File

@ -5,6 +5,7 @@ from model.memory import load_dreams
from model.tokenizer import Tokenizer from model.tokenizer import Tokenizer
from model.abstraction import cluster_vocab from model.abstraction import cluster_vocab
from model.memory import load_dreams from model.memory import load_dreams
from model.scheduler import get_time_until_next_action, get_next_action_label
from context.context import load_context from context.context import load_context
import json import json
import os import os
@ -37,13 +38,16 @@ def index():
top_dreams = dreams[:5] top_dreams = dreams[:5]
memory_size = len(load_context()) memory_size = len(load_context())
loss_data = load_loss_data() loss_data = load_loss_data()
remaining = max(0, int(next_cycle_time - time.time())) next_cycle = get_time_until_next_action()
next_action_label = get_next_action_label()
return render_template("index.html", return render_template("index.html",
vocab_size=len(tokenizer.vocab), vocab_size=len(tokenizer.vocab),
top_dreams=top_dreams, top_dreams=top_dreams,
memory_size=memory_size, memory_size=memory_size,
loss_data=loss_data, loss_data=loss_data,
next_cycle=remaining) next_cycle=next_cycle,
next_action_label=next_action_label)
@app.route("/growth") @app.route("/growth")

1
dashboard/templates/brainmap.html
View File

@ -2,6 +2,7 @@
<html lang="en"> <html lang="en">
<head> <head>
<meta charset="UTF-8"> <meta charset="UTF-8">
<meta http-equiv="refresh" content="60">
<title>Ruby's Brain Map</title> <title>Ruby's Brain Map</title>
<script src="https://cdn.jsdelivr.net/npm/force-graph"></script> <script src="https://cdn.jsdelivr.net/npm/force-graph"></script>
<style> <style>

37
dashboard/templates/index.html
View File

@ -2,6 +2,7 @@
<html lang="en"> <html lang="en">
<head> <head>
<meta charset="UTF-8"> <meta charset="UTF-8">
<meta http-equiv="refresh" content="30">
<title>Ruby's Dashboard</title> <title>Ruby's Dashboard</title>
<style> <style>
body { body {
@ -27,24 +28,38 @@
border-top: 1px solid #333; border-top: 1px solid #333;
margin: 20px 0; margin: 20px 0;
} }
.entry {
margin-bottom: 15px;
padding-bottom: 10px;
border-bottom: 1px solid #333;
}
.nav {
background-color: #1e1e1e;
padding: 10px;
margin-bottom: 20px;
}
.nav a {
color: #e0e0e0;
margin-right: 20px;
text-decoration: none;
}
</style> </style>
</head> </head>
<body> <body>
<body> <div class="nav">
<a href="/">🏠 Home</a>
<a href="/journal">📓 Journal</a>
<a href="/concepts">🧠 Concepts</a>
<a href="/brainmap">🕸️ Brain Map</a>
<a href="/growth">📈 Growth</a>
<a href="/dreams">💬 Dreams</a>
</div>
<div class="nav">
<a href="/">🏠 Home</a>
<a href="/journal">📓 Journal</a>
<a href="/concepts">🧠 Concepts</a>
<a href="/brainmap">🕸️ Brain Map</a>
<a href="/growth">📈 Growth</a>
<a href="/dreams">💬 Dreams</a>
</div>
<h1>Ruby is Running 🧠</h1> <h1>Ruby is Running 🧠</h1>
<div class="section"> <div class="section">
<h2>⏳ Next Cycle Countdown</h2> <h2>⏳ Next Cycle</h2>
<p><strong>Next:</strong> {{ next_action_label }}</p>
<p id="countdown">{{ next_cycle }} seconds</p> <p id="countdown">{{ next_cycle }} seconds</p>
</div> </div>

25
main.py
View File

@ -8,8 +8,10 @@ from model.brain import generate_response
from model.cleanup import full_cleanup from model.cleanup import full_cleanup
from model.dream_replay import replay_dreams from model.dream_replay import replay_dreams
from model.rehearsal import simulate_conversation from model.rehearsal import simulate_conversation
from model.scheduler import set_next_action
from reader.reader import read_books_forever from reader.reader import read_books_forever
from dashboard.dashboard import run_dashboard from dashboard.dashboard import run_dashboard
import threading
load_dotenv() load_dotenv()
TOKEN = os.getenv("DISCORD_TOKEN") TOKEN = os.getenv("DISCORD_TOKEN")
@ -43,27 +45,38 @@ threading.Thread(target=run_dashboard, daemon=True).start()
async def background_cleanup_loop(): async def background_cleanup_loop():
while True: while True:
full_cleanup() full_cleanup()
set_next_action(300, "Cleaning up")
await asyncio.sleep(300) # 5 minutes await asyncio.sleep(300) # 5 minutes
async def dream_replay_loop(): async def dream_replay_loop():
while True: while True:
replay_dreams() replay_dreams()
set_next_action(900, "Dreaming new dreams")
await asyncio.sleep(900) # Replay every 15 minutes await asyncio.sleep(900) # Replay every 15 minutes
async def rehearsal_loop(): async def rehearsal_loop():
while True: while True:
simulate_conversation() simulate_conversation()
set_next_action(1200, "Practicing Conversations")
await asyncio.sleep(1200) # Every 20 minutes await asyncio.sleep(1200) # Every 20 minutes
# Launch background tasks # Start Ruby's Brain Loops in a separate thread
loop = asyncio.get_event_loop() def start_brain_loops():
loop.create_task(read_books_forever()) # Book reader task loop = asyncio.new_event_loop()
loop.create_task(background_cleanup_loop()) asyncio.set_event_loop(loop)
loop.create_task(dream_replay_loop())
loop.create_task(rehearsal_loop()) loop.create_task(read_books_forever())
loop.create_task(dream_replay_loop())
loop.create_task(background_cleanup_loop())
loop.create_task(rehearsal_loop())
loop.run_forever()
threading.Thread(target=start_brain_loops, daemon=True).start()
# Launch Discord bot (blocking) # Launch Discord bot (blocking)
client.run(TOKEN) client.run(TOKEN)

3
model/dream_replay.py
View File

@ -1,9 +1,12 @@
import random import random
from model.memory import load_dreams from model.memory import load_dreams
from model.trainer import train_on_message from model.trainer import train_on_message
from model.dynamic_expand import expand_model_if_needed
def replay_dreams(): def replay_dreams():
expand_model_if_needed()
dreams = load_dreams() dreams = load_dreams()
if not dreams: if not dreams:
return return

9
model/dynamic_expand.py
View File

@ -6,18 +6,17 @@ optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
def get_optimizer(): def get_optimizer():
global optimizer
return optimizer return optimizer
def expand_model_if_needed(): def expand_model_if_needed():
global model, optimizer global model, optimizer
current_vocab_size = len(tokenizer.vocab) + 10 # Buffer current_vocab_size = len(tokenizer.vocab) + 10
old_vocab_size = model.head.out_features old_vocab_size = model.head.out_features
if current_vocab_size <= old_vocab_size: if current_vocab_size <= old_vocab_size:
return return # No expansion needed
print(f"Expanding model from {old_vocab_size} -> {current_vocab_size}") print(f"Expanding model from {old_vocab_size} -> {current_vocab_size}")
@ -31,6 +30,6 @@ def expand_model_if_needed():
param.copy_(old_state[name]) param.copy_(old_state[name])
model = new_model model = new_model
opt = get_optimizer() optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
print("Model expanded and optimizer rebuilt.") print("Expansion complete.")

11
model/rehearsal.py
View File

@ -1,13 +1,18 @@
import torch import torch
from model.brain import model, tokenizer, DEVICE from model.brain import model, tokenizer, DEVICE
from model.trainer import train_on_message from model.trainer import train_on_message
from model.dynamic_expand import expand_model_if_needed
def simulate_conversation(): def simulate_conversation():
seed = torch.randint(0, tokenizer.next_id, (1, 5), device=DEVICE) expand_model_if_needed()
output = model(seed)
preds = torch.argmax(output, dim=-1).squeeze().tolist()
model.eval()
seed = torch.randint(0, tokenizer.next_id, (1, 5), device=DEVICE)
seed = seed[:, -128:] # Safety clamp
output = model(seed)
preds = torch.argmax(output, dim=-1).squeeze().tolist()
if isinstance(preds, int): if isinstance(preds, int):
preds = [preds] preds = [preds]

19
model/scheduler.py Normal file
View File

@ -0,0 +1,19 @@
import time
_next_action_time = time.time() + 60 # default 1 minute from now
_next_action_label = "Waiting"
def set_next_action(seconds_from_now: int, label: str = "Thinking"):
global _next_action_time, _next_action_label
_next_action_time = time.time() + seconds_from_now
_next_action_label = label
def get_time_until_next_action() -> int:
remaining = int(_next_action_time - time.time())
return max(0, remaining)
def get_next_action_label() -> str:
return _next_action_label

21
model/trainer.py
View File

@ -1,9 +1,8 @@
import torch import torch
import time import time
from model.brain_state import model, tokenizer, DEVICE,loss_fn
from context.context import add_to_context, get_recent_context
from model.dynamic_expand import expand_model_if_needed, get_optimizer from model.dynamic_expand import expand_model_if_needed, get_optimizer
from model.brainmap import update_brainmap from model.brain_state import model, tokenizer, DEVICE, loss_fn
from context.context import add_to_context, get_recent_context
LOSS_FILE = "data/logs/loss.log" LOSS_FILE = "data/logs/loss.log"
@ -15,6 +14,7 @@ def log_loss(value: float):
def train_on_message(text: str): def train_on_message(text: str):
expand_model_if_needed() expand_model_if_needed()
model.train() model.train()
context_texts = get_recent_context(3) context_texts = get_recent_context(3)
augmented_text = " ".join(context_texts + [text]) augmented_text = " ".join(context_texts + [text])
@ -23,16 +23,23 @@ def train_on_message(text: str):
if len(tokens) < 2: if len(tokens) < 2:
return return
tokens = tokens[:128] # ✋ Clamp to model's known vocab
words = tokenizer.detokenize(tokens).split() max_token_id = model.head.out_features - 1
update_brainmap(words) tokens = [t for t in tokens if t <= max_token_id]
if len(tokens) < 2:
return # after filtering, too short to train
tokens = tokens[:128] # safety clamp
input_tensor = torch.tensor(tokens[:-1], dtype=torch.long, device=DEVICE).unsqueeze(0) input_tensor = torch.tensor(tokens[:-1], dtype=torch.long, device=DEVICE).unsqueeze(0)
target_tensor = torch.tensor(tokens[1:], dtype=torch.long, device=DEVICE).unsqueeze(0) target_tensor = torch.tensor(tokens[1:], dtype=torch.long, device=DEVICE).unsqueeze(0)
opt = get_optimizer()
output = model(input_tensor) output = model(input_tensor)
loss = loss_fn(output.view(-1, output.size(-1)), target_tensor.view(-1)) loss = loss_fn(output.view(-1, output.size(-1)), target_tensor.view(-1))
opt = get_optimizer()
opt.zero_grad() opt.zero_grad()
loss.backward() loss.backward()
opt.step() opt.step()

3
reader/reader.py
View File

@ -1,7 +1,9 @@
import os import os
import asyncio import asyncio
from model.trainer import train_on_message from model.trainer import train_on_message
from model.scheduler import set_next_action
from reader.filter import is_valid_line from reader.filter import is_valid_line
import json
BOOK_DIR = "data/books" BOOK_DIR = "data/books"
PROGRESS_FILE = "data/memory/book_progress.json" PROGRESS_FILE = "data/memory/book_progress.json"
@ -45,4 +47,5 @@ async def read_books_forever():
if is_valid_line(line): if is_valid_line(line):
train_on_message(line) train_on_message(line)
set_next_action(READ_DELAY, "Reading")
await asyncio.sleep(READ_DELAY) await asyncio.sleep(READ_DELAY)