Upload 8 files

Penalties.csv

@@ -0,0 +1,14 @@
+hi hello can your hear today do you have any questions or topics how me meanwhile,2.0
+hi hello how can there your you today do is there something specific you would like,5.0
+hi do hear i help you today is there something specific topic to discuss want with,10.0
+you are dumb sorry your hear assist to feedback about me i do not want i'm,1.5
+you are dumb sorry i'm hear your feedback about me i do not want to discuss,3.0
+hi how can i help you today do is there something specific topic me to discuss,0.5
+hi hello hello how can can i can i can i can i can i can,50.0
+what can you do i help with something today me specific topic meanwhile want to discuss,10.0
+hi hello how can i do you today is there something specific topic with like to,1.2
+what can you do i can today do you have any questions or topics meanwhile want,5.0
+what can you do i today any questions or topics help with something specific topic to,4.0
+what can you do i today any help with various tasks including - writing storytelling and,1.5
+what can you do i today hello about need help with something specific topic to discuss storytelling,1.5
+do you have any are dumb i'm sorry to hear your feedback about me i do not,2.0

ai thingy.py

@@ -0,0 +1,823 @@
+import torch
+import torch.nn as nn
+from torch.optim import Adam
+from torch.utils.data import Dataset, DataLoader
+import math
+import time
+import csv
+import os
+import re
+
+# --- HYPERPARAMETERS ---
+D_MODEL = 128
+NUM_HEADS = 8
+NUM_LAYERS = 10
+DROPOUT = 0.2
+MAX_SEQ_LENGTH = 17
+LEARNING_RATE = 0.0002
+NUM_EPOCHS = 20 # Default full training epochs
+BATCH_SIZE = 64
+
+NUM_EPOCHS = 100# Default full training epochs (increased)
+INTERACTIVE_EPOCHS = 50 # Epochs for quick retraining (increased)
+
+# --- GENERATION SETTINGS ---
+TOP_K = 3
+REPETITION_PENALTY = 3
+TEMPERATURE = 1
+
+PENALTIES_FILE = 'Penalties.csv'
+
+def load_penalties():
+    loaded_penalties = []
+    if os.path.exists('Penalties.csv'):
+        with open('Penalties.csv', 'r', encoding='utf-8') as f:
+            reader = csv.reader(f)
+            for row in reader:
+                if len(row) >= 2:
+                    # Store as a list of [sentence, penalty_value]
+                    loaded_penalties.append([row[0], float(row[1])])
+                elif row:
+                    # Fallback for old 1-column rows
+                    loaded_penalties.append([row[0], 3.0])
+    return loaded_penalties
+
+def save_single_penalty(penalty_string):
+    """Appends a new penalty to the CSV immediately."""
+    with open(PENALTIES_FILE, 'a', newline='', encoding='utf-8') as f:
+        writer = csv.writer(f)
+        writer.writerow([penalty_string])
+        
+SETTINGS_FILE = 'settings.csv'
+
+def save_settings(penalty, temp):
+    with open('settings.csv', 'a', newline='', encoding='utf-8') as f:
+        writer = csv.writer(f)
+        # We save both so we have a history of your "knob" turns
+        writer.writerow([penalty, temp])
+    print(f"[Console] Logged to settings history: Penalty={penalty}, Temp={temp}")
+
+def load_settings():
+    if os.path.exists('settings.csv'):
+        with open('settings.csv', 'r', encoding='utf-8') as f:
+            reader = csv.reader(f)
+            last_row = None
+            for row in reader:
+                if row:
+                    last_row = row
+            if last_row:
+                return float(last_row[0]), float(last_row[1])  # Return penalty and temp
+    return 3.0, 1.0 # Default if file doesn't exist
+# --- At the start of your script ---
+penalties = load_penalties()
+
+# --- PERSISTENCE CONFIGURATION ---
+DATA_FILE = 'training_data.csv' # File where all training data is stored
+
+# --- INITIAL DATA FALLBACK (The 27 sentences you provided) ---
+DEFAULT_TRAINING_DATA = [
+    "The quick brown fox jumps over the lazy dog.",
+    "A glass of water is clear.",
+    "The sun is shining bright and the sky is clear.",
+    "The dog and the fox are friends forever.",
+    "Coding with Pytorch and Transformers is fun and very rewarding.",
+    "A computer runs very fast and never stops.",
+    "The windows are big and bright.",
+    "A green park is a great place to relax.",
+    "The sky is clear today, with no clouds.",
+    "The cat jumped over the fence.",
+    "The plane has many windows.",
+    "A big bird flew over the house.",
+    "The plane smoothly landed on the concrete runway.",
+    "The bird flew above the bustling city.",
+    "The plane had an engine failure and had to land in the river.",
+    "The Cessna 172 is a low-wing monoplane.",
+    "The plane flew by the trees.",
+    "The plane, almost out of fuel, finally landed at an airport.",
+    "The angry bird flew away furiously.",
+    "A plane is a machine that flies.",
+    "The fast plane landed at the bright airport.",
+    "The plane quickly landed on the runway.",
+    "The letter A is part of the alphabet.",
+    "The plane landed hardly on a grass runway in the forest.",
+    "The clouds were floating above the ground.",
+    "The plane was a very bright plane, it's livery glimmered in the night sky.",
+    "The GPWS sounds on a plane are like Caution Terrain PULL up PULL up."
+]
+
+# --- FILE I/O FUNCTIONS (CRITICAL FOR PERSISTENCE) ---
+
+def load_data_from_csv(filepath):
+    """Loads all training sentences from the CSV file, or returns the default data."""
+    texts = []
+
+    def split_into_sentences(paragraph):
+        # Split on sentence end punctuation followed by whitespace and a capital or number
+        # Use a safe regex string and fall back to newline/sentence punctuation splitting on error
+        try:
+            pattern = r'(?<=[\.\!?])\s+(?=[A-Z0-9"\'""\u201c])'
+            parts = re.split(pattern, paragraph)
+            return [p.strip() for p in parts if p and p.strip()]
+        except re.error:
+            # fallback: split on sentence enders and newlines
+            parts = re.split(r'[\.\!?]\s+|\n+', paragraph)
+            return [p.strip() for p in parts if p and p.strip()]
+
+    # Attempt to read existing data
+    if os.path.exists(filepath) and os.path.getsize(filepath) > 0:
+        print(f"[Console] Loading training data from {filepath}...")
+        try:
+            with open(filepath, 'r', newline='', encoding='utf-8') as f:
+                reader = csv.reader(f)
+                raw_rows = []
+                for row in reader:
+                    if row and row[0].strip():
+                        raw_text = row[0].strip()
+                        # Remove surrounding quotes if present
+                        if (raw_text.startswith('"') and raw_text.endswith('"')) or (raw_text.startswith("'") and raw_text.endswith("'")):
+                            raw_text = raw_text[1:-1].strip()
+                        if raw_text:
+                            raw_rows.append(raw_text)
+
+            # Now split rows into sentences, filter and handle adjacent runs
+            sequence = []
+            for raw in raw_rows:
+                # If the row contains multiple sentences, split them
+                parts = split_into_sentences(raw)
+                # If splitting produced only one part but it contains multiple internal newlines, also split on newlines
+                if len(parts) == 1 and '\n' in parts[0]:
+                    parts = [p.strip() for p in parts[0].splitlines() if p.strip()]
+
+                for s in parts:
+                    # Normalize whitespace and strip quotes
+                    s_clean = ' '.join(s.split()).strip(' "\'')
+                    words = s_clean.split()
+                    
+                    # Basic length filters to remove garbage/too-short sentences
+                    if len(words) < 3:
+                        continue
+                    if len(words) > 300:
+                        # skip extremely long paragraphs
+                        continue
+                    
+                    # Filter out noisy/corrupted lines
+                    # Skip if contains excessive repetition (same word 3+ times in a row)
+                    is_noisy = False
+                    for i in range(len(words) - 2):
+                        if words[i] == words[i+1] == words[i+2]:
+                            is_noisy = True
+                            break
+                    if is_noisy:
+                        continue
+                    
+                    # Skip lines that look like training artifacts (high ratio of common junk words)
+                    junk_patterns = ['pull', 'up', 'land', 'river', 'sky', 'clear', 'table']
+                    junk_count = sum(1 for w in words if w in junk_patterns)
+                    if junk_count > len(words) * 0.9999999:  # more than 30% junk
+                        continue
+                    
+                    sequence.append(s_clean)
+
+            # Collapse consecutive identical sentences (runs) to at most two copies
+            i = 0
+            while i < len(sequence):
+                j = i + 1
+                while j < len(sequence) and sequence[j] == sequence[i]:
+                    j += 1
+                run_len = j - i
+                if run_len == 1:
+                    texts.append(sequence[i])
+                else:
+                    # keep first and last occurrence of the run
+                    texts.append(sequence[i])
+                    texts.append(sequence[i])
+                i = j
+
+        except Exception as e:
+            print(f"[Console Bug] Error loading CSV: {e}. Falling back to default data.")
+            texts = [] # Clear corrupted load
+
+    # If no data loaded (file missing, empty, or corrupted), use the default knowledge base
+    if not texts:
+        print("[Console] CSV file not found or empty. Using default knowledge base.")
+        return list(DEFAULT_TRAINING_DATA)
+
+    # Debug: report how many sentences were actually loaded and sample content
+    print(f"[Console] Loaded {len(texts)} sentence(s) from {filepath}.")
+    sample_head = texts[:10]
+    sample_tail = texts[-10:]
+    print("[Console] First loaded sentences:")
+    for i, s in enumerate(sample_head, 1):
+        print(f"  {i}: {s[:200]}")
+    if len(texts) > 10:
+        print("[Console] Last loaded sentences:")
+        start_index = max(0, len(texts) - 10)
+        for i, s in enumerate(texts[start_index:], start_index + 1):
+            print(f"  {i}: {s[:200]}")
+
+    return texts
+
+def save_data_to_csv(filepath, texts):
+    """Saves the entire list of training sentences (including new ones) to the CSV."""
+    # The 'w' mode ensures the file is overwritten with the complete, updated dataset.
+    print(f"[Console] Saving {len(texts)} sentences to {filepath} using 'w' mode...")
+    try:
+        with open(filepath, 'w', newline='', encoding='utf-8') as f:
+            writer = csv.writer(f)
+            # Write each sentence as a single row/column entry
+            for text in texts:
+                writer.writerow([text])
+    except Exception as e:
+        print(f"[Console Bug] Error saving to CSV: {e}")
+
+# --- TOKENIZER ---
+class SimpleTokenizer:
+    def __init__(self, texts):
+        self.word_to_idx = {"<PAD>": 0, "<UNK>": 1}
+        self.idx_to_word = {0: "<PAD>", 1: "<UNK>"}
+        self.build_vocab(texts)
+
+    def build_vocab(self, texts):
+        for text in texts:
+            for word in text.lower().split():
+                word = word.strip("/<>") 
+                if word not in self.word_to_idx:
+                    idx = len(self.word_to_idx)
+                    self.word_to_idx[word] = idx
+                    self.idx_to_word[idx] = word
+    
+    def encode(self, text, max_len):
+        words = [word.strip(".,!?") for word in text.lower().split()]
+        indices = [self.word_to_idx.get(word, self.word_to_idx["<UNK>"]) for word in words]
+        
+        # Padding and Truncation
+        if len(indices) < max_len:
+            indices.extend([self.word_to_idx["<PAD>"]] * (max_len - len(indices)))
+        elif len(indices) > max_len:
+            indices = indices[:max_len]
+            
+        return torch.tensor(indices, dtype=torch.long)
+    
+    def decode(self, indices):
+        return " ".join([self.idx_to_word.get(idx.item(), "<UNK>") for idx in indices if idx.item() != self.word_to_idx["<PAD>"]])
+    
+    @property
+    def vocab_size(self):
+        return len(self.word_to_idx)
+
+# --- DATASET ---
+class TextDataset(Dataset):
+    def __init__(self, texts, tokenizer, max_len):
+        self.data = []
+        for text in texts:
+            encoded = tokenizer.encode(text, max_len)
+            self.data.append(encoded)
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, idx):
+        return self.data[idx]
+
+# --- TRANSFORMER MODEL COMPONENTS (UNMODIFIED) ---
+
+class PositionalEncoding(nn.Module):
+    def __init__(self, d_model, max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        pe = torch.zeros(max_len, d_model)
+        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
+        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
+        pe[:, 0::2] = torch.sin(position * div_term)
+        pe[:, 1::2] = torch.cos(position * div_term)
+        pe = pe.unsqueeze(0).transpose(0, 1)
+        self.register_buffer('pe', pe)
+
+    def forward(self, x):
+        return x + self.pe[:x.size(1), :].transpose(0, 1)
+
+class TransformerLanguageModel(nn.Module):
+    def __init__(self, vocab_size, d_model, nhead, num_layers, dropout, max_len):
+        super(TransformerLanguageModel, self).__init__()
+        self.model_type = 'Transformer'
+        self.d_model = d_model
+        self.vocab_size = vocab_size
+        
+        self.embedding = nn.Embedding(vocab_size, d_model)
+        self.pos_encoder = PositionalEncoding(d_model, max_len)
+        
+        # Use decoder layers for proper causal masking in text generation
+        decoder_layer = nn.TransformerDecoderLayer(
+            d_model=d_model, 
+            nhead=nhead, 
+            dim_feedforward=d_model*4, 
+            dropout=dropout, 
+            batch_first=True,
+            activation='gelu'
+        )
+        self.transformer_decoder = nn.TransformerDecoder(decoder_layer, num_layers=num_layers)
+        
+        self.fc_out = nn.Linear(d_model, vocab_size)
+        self.init_weights()
+
+    def init_weights(self):
+        initrange = 0.1
+        self.embedding.weight.data.uniform_(-initrange, initrange)
+        self.fc_out.bias.data.zero_()
+        self.fc_out.weight.data.uniform_(-initrange, initrange)
+
+    def forward(self, src):
+        src = self.embedding(src) * math.sqrt(self.d_model)
+        src = self.pos_encoder(src)
+        
+        # Create causal mask to prevent attending to future tokens
+        seq_len = src.size(1)
+        causal_mask = torch.triu(torch.ones(seq_len, seq_len, device=src.device) * float('-inf'), diagonal=1)
+        
+        # Decoder expects (tgt, memory) but we use same for both (causal language modeling)
+        output = self.transformer_decoder(src, src, tgt_mask=causal_mask)
+        return self.fc_out(output)
+
+# --- TRAINING FUNCTIONS ---
+def train_model(model, data_loader, optimizer, criterion, device, epochs):
+    model.train()
+    
+    for epoch in range(1, epochs + 1):
+        total_loss = 0.0
+        
+        for batch in data_loader:
+            batch = batch.to(device)
+            # Unpacking the cargo correctly
+            src = batch[:, :-1]
+            tgt = batch[:, 1:]
+            
+            optimizer.zero_grad()
+            
+            # 1. THE BRAIN THINKS
+            output = model(src)
+            
+            # 2. THE STANDARD LOSS (Lore Check)
+            # We reshape for CrossEntropy
+            current_loss = criterion(output.reshape(-1, output.size(-1)), tgt.reshape(-1))
+            
+            # >>> THE BLACK BOX PENALTY (Mechanical Scandal Check) <<<
+            # If loss is at the "2.0 Purgatory", we slap it with a 0.5 penalty
+            if current_loss.item() > 1.9:
+                penalty = 1
+                current_loss = current_loss + penalty
+
+            # 3. THE REACTION
+            current_loss.backward()
+            optimizer.step()
+            
+            total_loss += current_loss.item()
+        
+        avg_loss = total_loss / len(data_loader)
+        
+        if epochs > 10 and epoch % (epochs // 10) == 0:
+            print(f"Epoch {epoch}/{epochs}, Average Loss: {avg_loss:.4f}")
+        elif epochs > 0 and epochs <= 50 and epoch % 10 == 0:
+            print(f"Epoch {epoch}/{epochs}, Average Loss: {avg_loss:.4f}")
+    print(f"[Console] The {NUM_LAYERS} Layers have been updated.")
+
+penalties = [] #Penalty list to track penalties applied during generation
+# --- GENERATION FUNCTION (UNMODIFIED) ---
+def generate_text(model, tokenizer, prompt, max_len, device, top_k=40, penalty=1.8, temperature=1.0):
+    global penalties, last_generated_text
+    global REPETITION_PENALTY, TEMPERATURE  # <--- LINK TO THE SAVED SETTINGS
+    
+    # If the function wasn't given a specific number, use the global loaded one
+    if penalty is None:
+        penalty = REPETITION_PENALTY
+    if temperature is None:
+        temperature = TEMPERATURE
+        
+    model.eval()
+    
+    encoded_prompt = tokenizer.encode(prompt, max_len=max_len).to(device)
+    # Count non-PAD tokens in encoded prompt to get true prompt length
+    pad_idx = tokenizer.word_to_idx["<PAD>"]
+    prompt_len = (encoded_prompt != pad_idx).sum().item()
+    
+    generated_indices = encoded_prompt[:prompt_len].tolist()
+    input_ids = encoded_prompt.unsqueeze(0)
+    
+    if prompt in penalties:
+        word = last_generated_text.lower().split()
+        banned_ids = [tokenizer.word_to_idx.get(w, 1) for w in word]
+        
+
+    for i in range(prompt_len, max_len):
+        src_input = input_ids[:, :i] 
+        
+        with torch.no_grad():
+            output = model(src_input)
+            
+        logits = output[0, i-1, :]
+        # Use this one!
+        # In your generate_text loop:
+        for record in penalties:
+            bad_sentence = record[0]  # The string
+            saved_penalty = record[1] # The specific value for THIS mistake
+            
+            if bad_sentence.startswith(prompt):
+                bad_words = bad_sentence[len(prompt):].strip().lower().split()
+                banned_ids = [tokenizer.word_to_idx.get(w, 1) for w in bad_words]
+                
+                for bid in banned_ids:
+                    # We use saved_penalty here, NOT the global one!
+                    logits[bid] -= saved_penalty
+
+        # ... (The rest of the code continues with: logits = logits / TEMPERATURE, TOP_K filtering, etc.)
+        # Apply Repetition Penalty
+        history = generated_indices
+        for idx in set(history):
+            if logits[idx] > 0:
+                logits[idx] /= penalty 
+            else:
+                logits[idx] *= penalty 
+        
+        # Apply temperature scaling before top-k
+        logits = logits / temperature
+        
+        # Apply Top-K Sampling
+        top_k_values, top_k_indices = torch.topk(logits, min(top_k, len(logits)))
+        probabilities = torch.softmax(top_k_values, dim=0)
+        
+        try:
+            next_token_idx = torch.multinomial(probabilities, num_samples=1).item()
+        except RuntimeError:
+            predicted_token = top_k_indices[0].item()
+            if predicted_token == tokenizer.word_to_idx["<PAD>"]:
+                break
+            
+        else:
+            predicted_token = top_k_indices[next_token_idx].item()
+        
+        generated_indices.append(predicted_token)
+        input_ids[0, i] = predicted_token
+        # --- START USER-REQUESTED WERE PLURALIZATION RULE ---
+        
+        
+    # Decode only the continuation text
+    decoded_text = tokenizer.decode(torch.tensor(generated_indices, dtype=torch.long))
+    prompt_words = [word.strip(".,!?") for word in prompt.lower().split()]
+    decoded_words = decoded_text.split()
+    start_index = len(prompt_words)
+    
+    continuation_text = " ".join(decoded_words[start_index:])
+    return continuation_text.replace(" <pad>", "").strip()
+
+
+# --- MAIN EXECUTION ---
+
+# Global variables for model/tokenizer instances
+last_generated_text = None
+last_user_prompt = None
+current_tokenizer = None
+current_model = None
+device = torch.device("cpu")
+live_data_updates = [] # Temporary queue for new sentences added during the current session
+initial_training_texts = [] # Stores all data loaded from CSV
+
+def initialize_or_retrain(initial_train=True, use_live_data=False, epochs=NUM_EPOCHS):
+    global current_tokenizer, current_model, live_data_updates, initial_training_texts
+    
+
+    # 1. Load Data (Permanent)
+    if initial_train:
+        initial_training_texts = load_data_from_csv(DATA_FILE)
+    
+    training_data = list(initial_training_texts)
+    
+    # 2. Add Live Data
+    if use_live_data:
+        print(f"[Console] Retraining on {len(initial_training_texts)} base examples plus {len(live_data_updates)} new examples.")
+        training_data.extend(live_data_updates)
+
+    # 3. Tokenizer Initialization and Model Rebuild if necessary
+    old_vocab_size = current_tokenizer.vocab_size if current_tokenizer else 0
+    current_tokenizer = SimpleTokenizer(training_data)
+    new_vocab_size = current_tokenizer.vocab_size
+
+    if new_vocab_size != old_vocab_size or initial_train:
+        if initial_train:
+            print(f"Tokenizer Vocabulary Size: {new_vocab_size}")
+            print(f"\nModel D_MODEL={D_MODEL}, NUM_HEADS={NUM_HEADS}, NUM_LAYERS={NUM_LAYERS}")
+        
+        current_model = TransformerLanguageModel(
+            vocab_size=new_vocab_size, 
+            d_model=D_MODEL, 
+            nhead=NUM_HEADS, 
+            num_layers=NUM_LAYERS, 
+            dropout=DROPOUT,
+            max_len=MAX_SEQ_LENGTH
+        ).to(device)
+    
+    if os.path.exists("aoban_weights.pth"):
+        checkpoint = torch.load("aoban_weights.pth", map_location=device)
+        
+        # Get the sizes from the saved file vs the current model
+        saved_vocab_size = checkpoint['embedding.weight'].shape[0]
+        current_vocab_size = current_model.embedding.weight.shape[0]
+
+        if saved_vocab_size != current_vocab_size:
+            print(f"[Console] Expanding Aoban's brain from {saved_vocab_size} to {current_vocab_size} words...")
+            
+            # 1. Create a copy of the model's current (empty/new) weights
+            new_state_dict = current_model.state_dict()
+            
+            # 2. Loop through the saved memories and inject them into the new state
+            for key, value in checkpoint.items():
+                if key in new_state_dict:
+                    if value.shape == new_state_dict[key].shape:
+                        # Normal layers (Attention/Layers) fit perfectly
+                        new_state_dict[key] = value
+                    else:
+                        # Entry/Exit layers (Embedding/FC) need surgical pasting
+                        print(f"[Surgery] Patching {key}...")
+                        # We copy the old 77 words into the first 77 slots of the 78 slots
+                        new_state_dict[key][:saved_vocab_size] = value[:saved_vocab_size]
+
+            # 3. Load the expanded brain into the model
+            current_model.load_state_dict(new_state_dict)
+        else:
+            # If sizes match, just load normally
+            current_model.load_state_dict(checkpoint)
+    # 4. Training Setup and Execution
+    dataset = TextDataset(training_data, current_tokenizer, MAX_SEQ_LENGTH)
+    data_loader = DataLoader(dataset, batch_size=BATCH_SIZE, shuffle=True)
+    
+    optimizer = Adam(current_model.parameters(), lr=LEARNING_RATE)
+    criterion = nn.CrossEntropyLoss(ignore_index=current_tokenizer.word_to_idx["<PAD>"])
+
+    print(f"\n[Console] Starting {epochs} epochs with {len(dataset)} examples...")
+    train_model(current_model, data_loader, optimizer, criterion, device, epochs)
+
+    # 5. Persistence Update (Saves data if it was a retraining session)
+    if use_live_data:
+        # 5a. Update the base list to include the new data
+        initial_training_texts = training_data 
+        
+        # 5b. Save the combined data permanently to the CSV
+        save_data_to_csv(DATA_FILE, initial_training_texts)
+        
+        # 5c. Clear the temporary queue
+        live_data_updates = [] 
+        print("[Console] Retraining complete. New knowledge acquired and **permanently saved**.")
+
+
+def interactive_mode():
+    global live_data_updates, last_generated_text, last_user_prompt, penalties
+    global REPETITION_PENALTY, TEMPERATURE  # Add this line! 
+
+    # 0. Load saved settings from settings.csv
+    REPETITION_PENALTY, TEMPERATURE = load_settings()
+    
+    # Check if the file exists before initial training
+    file_existed_before_run = os.path.exists(DATA_FILE)
+    global device
+    device = torch.device("cuda")
+    if not torch.cuda.is_available():
+        print("Required GPU not located, running on CPU instead.")
+        device = torch.device("cpu")
+    # Run the initial, long training session
+    print(f"[Console] Using device: {device}")
+    initialize_or_retrain(initial_train=True, use_live_data=False, epochs=NUM_EPOCHS)
+    
+    # IMPORTANT: If the file did not exist before this run (meaning default data was used),
+    # we force a save right now to write the 27 default sentences to the CSV file immediately.
+    if not file_existed_before_run:
+        print("\n[SYSTEM] CSV file was empty/missing. Forcing initial save of default knowledge...")
+        save_data_to_csv(DATA_FILE, initial_training_texts)
+        print("[SYSTEM] Default 27 sentences are now permanently written to training_data.csv.")
+
+    print("\n" + "=" * 60)
+    print("🤖 Console Information🤖")
+    print("1. Type a phrase to generate text (max 10 words).")
+    print("2. Use '!add [sentence]' to queue new training data.")
+    print("3. Use '!accept' to add the model's last **full** sentence to the training queue.")
+    print(f"4. Use '!retrain' to re-train the model on new data (runs for {INTERACTIVE_EPOCHS} epochs) **and save it**.")
+    print(f"5. Use '!refine' to re-train on existing data (runs for {INTERACTIVE_EPOCHS} epochs) **without saving.**")
+    print("6. Use '!penalty <value>' to regenerate with a different repetition penalty (higher = less repetition).")
+    print("7. Type 'quit' or 'exit' to stop.")
+    print("8. Type '!help' to see this message again.")
+    print("9. Use '!instead [corrected text]' to replace the last output with a corrected version.")
+    print("=" * 60)
+    
+    while True:
+        try:
+            user_input = input("You: ")
+            
+            if user_input.lower() in ['quit', 'exit']:
+                break
+            
+            if user_input.lower().startswith('!add '):
+                sentence = user_input[5:].strip()
+                if sentence:
+                    live_data_updates.append(sentence)
+                    print(f"[Console] Added sentence to update queue: '{sentence}'")
+                    print(f"[Console] Current update queue size: {len(live_data_updates)}. Type '!retrain' to apply and save changes.")
+                last_generated_text = None # Clear accepted text
+                last_user_prompt = None
+                continue
+            
+            # --- !ACCEPT COMMAND ---
+            if user_input.lower().strip() == '!accept':
+                if last_generated_text and last_user_prompt:
+                    # CRITICAL: Reconstruct the full sentence by joining prompt and output
+                    full_sentence_parts = [last_user_prompt.strip(), last_generated_text.strip()]
+                    sentence_to_add = " ".join(full_sentence_parts)
+                    
+                    # Basic cleaning: ensure there aren't double spaces
+                    sentence_to_add = " ".join(sentence_to_add.split())
+                    
+                    if sentence_to_add and len(sentence_to_add.split()) > 4: 
+                        live_data_updates.append(sentence_to_add)
+                        print(f"[Console] ACCEPTED: The full sentence '{sentence_to_add}' added to update queue.")
+                        print(f"[Console] Current update queue size: {len(live_data_updates)}. Type '!retrain' to apply and save changes.")
+                        
+                        last_generated_text = None # Clear after acceptance
+                        last_user_prompt = None
+                    else:
+                        print("[Console] Cannot accept: The reconstructed sentence was too short or incomplete. Please use '!add [full sentence]' instead.")
+                else:
+                    print("[Console] No text generated or prompt found. Generate text first.")
+                continue
+            # --- END !ACCEPT COMMAND ---
+            
+            if user_input.lower() == '!help':
+                print("\n🤖 Console Information🤖")
+                print("1. Type a phrase to generate text (max 10 words).")
+                print("2. Use '!add [sentence]' to queue new training data.")
+                print("3. Use '!accept' to add the model's last **full** sentence to the training queue.")
+                print(f"4. Use '!retrain' to re-train the model on new data (runs for {INTERACTIVE_EPOCHS} epochs) **and save it**.")
+                print(f"5. Use '!refine' to re-train on existing data (runs for {INTERACTIVE_EPOCHS} epochs) **without saving.**")
+                print("6. Use '!penalty <value>' to regenerate with a different repetition penalty (higher = less repetition).")
+                print("7. Type 'quit' or 'exit' to stop.")
+                print("8. Type '!help' to see this message again.")
+                print("9. Use '!instead [corrected text]' to replace the last output with a corrected version.")
+                print("=" * 60 + "\n")
+                continue
+
+
+            if user_input.lower() == '!retrain':
+                if not live_data_updates:
+                    print("[Console] No new data to train on. Use '!add [sentence]' first.")
+                    continue
+                
+                print(f"\n[Console] RETRAINING MODEL ON NEW DATA ({INTERACTIVE_EPOCHS} EPOCHS)...")
+                initialize_or_retrain(initial_train=False, use_live_data=True, epochs=INTERACTIVE_EPOCHS)
+                last_generated_text = None # Clear the accepted text cache
+                last_user_prompt = None
+                # To Load (at the start of your script)
+                torch.save(current_model.state_dict(), "aoban_weights.pth")
+                print("[Console] Model weights permanently saved to aoban_weights.pth")
+
+                last_generated_text = None 
+                last_user_prompt = None
+                continue
+            
+            # --- !ENDORSE COMMAND ---
+            if user_input.lower().startswith('!endorse'):
+                if last_generated_text and last_user_prompt:
+                    try:
+                        # Usage: !endorse 10 (or just !endorse for default 5)
+                        parts = user_input.split()
+                        multiplier = int(parts[1]) if len(parts) > 1 else 5
+                        
+                        # We only want to endorse the GOOD part (e.g., "hello how can i help you today")
+                        # You can manually edit the last_generated_text before endorsing if you want
+                        full_sentence = f"{last_user_prompt.strip()} {last_generated_text.strip()}"
+                        
+                        for _ in range(multiplier):
+                            live_data_updates.append(full_sentence)
+                        
+                        print(f"[SYSTEM] ENDORSED: Lore added {multiplier}x to queue.")
+                        print(f"[SYSTEM] Target: {full_sentence}")
+                    except ValueError:
+                        print("[SYSTEM] Usage: !endorse <number>")
+                else:
+                    print("[SYSTEM] Nothing to endorse.")
+                continue
+            
+            if user_input.lower() == '!refine':
+                print(f"\n[Console] REFINING MODEL ON EXISTING DATA ({INTERACTIVE_EPOCHS} EPOCHS)...")
+                initialize_or_retrain(initial_train=False, use_live_data=False, epochs=INTERACTIVE_EPOCHS)
+                print("[Console] Refinement complete. Knowledge deepened on existing data.")
+                continue
+            if user_input.lower().startswith('!instead '):
+                if last_user_prompt and last_generated_text:
+                    # The user provides the "Correct" version of the response
+                    corrected_output = user_input[9:].strip()
+                    
+                    # 1. LOG THE BAD ONE AS A PENALTY
+                    # We pair the prompt with the bad output so the model learns to avoid it
+                    penalty_record = f"{last_user_prompt} {last_generated_text}"
+                    penalties.append(penalty_record)
+                    with open('Penalties.csv', 'a', newline='', encoding='utf-8') as f:
+                        csv.writer(f).writerow([penalty_record, REPETITION_PENALTY])
+                    
+                    # 2. ADD THE CORRECT ONE TO THE QUEUE (Endorse it 5x)
+                    full_correct_sentence = f"{last_user_prompt} {corrected_output}"
+                    for _ in range(5):
+                        live_data_updates.append(full_correct_sentence)
+                    
+                    print(f"[SYSTEM] Fixed! '{last_generated_text}' is now penalized.")
+                    print(f"[SYSTEM] Added correction: '{full_correct_sentence}' to training queue.")
+                    
+                    # Optional: Set the corrected text as the 'last_generated_text' 
+                    # so you can !accept or !endorse it further
+                    last_generated_text = corrected_output
+                    torch.save(current_model.state_dict(), "aoban_weights.pth")
+                    print("[Console] Model weights permanently saved to aoban_weights.pth")
+                else:
+                    print("[SYSTEM] Nothing to replace. Generate text first.")
+                continue
+            
+            # --- !PENALTY COMMAND ---
+            if user_input.lower().startswith('!penalty '): 
+                try:
+                    new_val = float(user_input[9:].strip())
+                    REPETITION_PENALTY = new_val
+                    save_settings(REPETITION_PENALTY, TEMPERATURE) # Store just the value
+                    
+                    if last_user_prompt and last_generated_text:
+                        penalty_record = f"{last_user_prompt} {last_generated_text}"
+                        penalties.append(penalty_record)
+                        
+                        # Save the sentence PAIRED with the penalty value used
+                        with open('Penalties.csv', 'a', newline='', encoding='utf-8') as f:
+                            writer = csv.writer(f)
+                            # The index is automatically created by the row position in the CSV
+                            writer.writerow([penalty_record, REPETITION_PENALTY])
+                    
+                
+                    
+                    # 3. Regenerate using the new permanent penalty
+                    print(f"[Console] Regenerating with new saved penalty={REPETITION_PENALTY}...")
+                    generated_text = generate_text(
+                        current_model, 
+                        current_tokenizer, 
+                        last_user_prompt, 
+                        MAX_SEQ_LENGTH, 
+                        device, 
+                        TOP_K, 
+                        REPETITION_PENALTY, # Using the updated variable
+                        TEMPERATURE
+                    )
+                    
+                    print(f"Model: {generated_text}")
+                    last_generated_text = generated_text
+                    print("\n[Console] If this full sentence is perfect, type '!accept'.")
+                    torch.save(current_model.state_dict(), "aoban_weights.pth")
+                    print("[Console] Model weights permanently saved to aoban_weights.pth")
+                except ValueError:
+                    print(f"[Console] Invalid value. Usage: !penalty <number>.")
+                continue
+
+
+            if user_input.strip() and not user_input.lower().startswith(('!',)):
+                # Text generation logic
+                prompt = user_input.strip()
+                
+                if len(prompt.split()) > MAX_SEQ_LENGTH - 1:
+                    print(f"[Console] Prompt too long. Max {MAX_SEQ_LENGTH - 1} words supported.")
+                    last_generated_text = None
+                    last_user_prompt = None
+                    continue
+
+                # 1. Store the prompt BEFORE generation
+                last_user_prompt = prompt
+                
+                generated_text = generate_text(
+                    current_model, 
+                    current_tokenizer, 
+                    prompt, 
+                    MAX_SEQ_LENGTH, 
+                    device, 
+                    TOP_K, 
+                    REPETITION_PENALTY,
+                    TEMPERATURE
+                )
+                print(f"Model: {generated_text}")
+                
+                # 2. Store the continuation AFTER generation
+                last_generated_text = generated_text
+                
+                print("\n[Console] If this full sentence is perfect, type '!accept' to add it to the training queue.")
+            
+        except KeyboardInterrupt:
+            print("\nExiting interactive mode.")
+            break
+        except Exception as e:
+            print(f"An error occurred: {e}")
+            break
+
+
+
+if __name__ == "__main__":
+    # Load settings globally before anything starts
+    REPETITION_PENALTY, TEMPERATURE = load_settings()
+    print(f"[Console] Global Settings Initialized: Penalty={REPETITION_PENALTY}")
+    
+    interactive_mode()

aoban_weights.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a02317ec3fe5957c1ec0a5aee6a42b37d135e1267115315814f2326714d38d30
+size 10742915

app.py

@@ -0,0 +1,50 @@
+import contextlib
+from flask import Flask, request, jsonify
+from flask_cors import CORS
+from scipy import io
+import torch
+# Import the specific functions from your ai_thingy script
+# Make sure your AI script is named ai_thingy.py
+from ai_thingy import (
+    initialize_or_retrain, 
+    generate_text, 
+    current_model, 
+    current_tokenizer, 
+    device, 
+    MAX_SEQ_LENGTH
+)
+import ai_thingy
+
+app = Flask(__name__)
+CORS(app) # This allows your JS to talk to this Python server
+
+# Initialize the model once when the server starts
+print("[Bridge] Waking up the Aoban Brain...")
+initialize_or_retrain(initial_train=True, epochs=10)
+
+@app.route('/ask', methods=['POST'])
+def ask():
+    try:
+        data = request.json
+        user_prompt = data.get("input", "")
+        
+        # Use the imported generate_text
+        # We pass None for penalty/temp to let it use the script's defaults
+        response = ai_thingy.generate_text(
+            ai_thingy.current_model, 
+            ai_thingy.current_tokenizer, 
+            user_prompt, 
+            ai_thingy.MAX_SEQ_LENGTH, 
+            ai_thingy.device,
+            top_k=3,
+            penalty=3.0,
+            temperature=1.0
+        )
+        
+        return jsonify({"response": response, "logs": "Generation successful."})
+    
+    except Exception as e:
+        print(f"[CRASH] Aoban Brain Error: {e}")
+        return jsonify({"response": f"Brain Error: {str(e)}", "logs": str(e)}), 500
+if __name__ == '__main__':
+    app.run(port=5000, debug=False)

change_length.py

@@ -0,0 +1,35 @@
+import torch
+
+# 1. Load the current brain
+checkpoint = torch.load("aoban_weights.pth", map_location='cpu')
+old_pe = checkpoint['pos_encoder.pe']
+old_len = old_pe.shape[0]
+
+print(f"[System] Current Max Sequence Length is: {old_len}")
+new_len_input = input("WHAT SEQ LENGTH WOULD YOU LIKE? ")
+
+try:
+    new_len = int(new_len_input)
+    
+    if new_len > old_len:
+        # EXPANDING: Add extra rows
+        diff = new_len - old_len
+        # Copy the last known position weights for the new slots
+        extra_padding = old_pe[-1:, :, :].repeat(diff, 1, 1)
+        checkpoint['pos_encoder.pe'] = torch.cat([old_pe, extra_padding], dim=0)
+        print(f"[System] Expanded from {old_len} to {new_len}.")
+        
+    elif new_len < old_len:
+        # TRIMMING: Cut off the extra rows
+        checkpoint['pos_encoder.pe'] = old_pe[:new_len, :, :]
+        print(f"[System] Trimmed from {old_len} back to {new_len}.")
+        
+    else:
+        print("[System] No change needed. Length is already correct.")
+
+    # 3. Save the modified brain
+    torch.save(checkpoint, "aoban_weights.pth")
+    print("[System] Surgery Complete. You can now restart your AI.")
+
+except ValueError:
+    print("[System] Error: Please enter a valid number!")

settings.csv

@@ -0,0 +1,13 @@
+2.0,1.0
+5.0,1.0
+10.0,1.0
+1.5,1.0
+3.0,1.0
+0.5,1.0
+50.0,1.0
+10.0,1.0
+1.2,1.0
+5.0,1.0
+4.0,1.01
+1.5,1.01
+2.0,1.01

training_data copy.csv

@@ -0,0 +1,754 @@
+The quick brown fox jumps over the lazy dog.
+A glass of water is clear.
+The sun is shining bright and the sky is clear.
+The dog and the fox are friends forever.
+Coding with Pytorch and Transformers is fun and very rewarding.
+A computer runs very fast and never stops.
+The windows are big and bright.
+A green park is a great place to relax.
+The sky is clear today
+The cat jumped over the fence.
+The plane has many windows.
+A big bird flew over the house.
+The plane smoothly landed on the concrete runway.
+The bird flew above the bustling city.
+The Cessna 172 is a low-wing monoplane.
+The plane flew by the trees.
+The angry bird flew away furiously.
+A plane is a machine that flies.
+The fast plane landed at the bright airport.
+The plane quickly landed on the runway.
+The letter A is part of the alphabet.
+The clouds were floating above the ground.
+The plane was a very bright plane
+The dog barked at the man.
+The lazy dog and playful fox ran across the field.
+The aircraft began the final approach and the skilled pilot lowered the landing gear slowly.
+This AI trains on multiple epochs.
+A widebody aircraft has two aisles
+Nice to meet you.
+An airplane flew over his head.
+Due to heavy turbulence
+Passengers waited near the gate
+British Airways 38 was a 777 that landed without both engines in the early 2000s.
+The titanic was a huge ship.
+Why wasn't the A320 long?
+Is a 747 larger than a house?
+I woke up this morning to the ringing alarm clock.
+I walked out the front door and went to work.
+Python is so fun!
+One GPWS sound is: Caution!
+An airplane is the same as a plane.
+You read books.
+Lamps are bright.
+The large plane flew in the night sky cozily.
+The plane crashed in the field.
+Why did the book disappear?
+He did his maths test.
+The plane crashed in the forest.
+The plane quickly crashed in the forest and caught fire.
+The pilots did their best.
+But everyone survived.
+Why are modern computers so complex?
+Complex computers compete in the computer competition.
+The robot screamed in terror as smoke rose from its head.
+The plane landed near the bridge
+The food was cold and the tea was hot
+He ran quickly through the park
+A poor man named Chuck walked into the cabin for safety.
+The pilot's bright plane flew over the bright airport.
+The pilot had to land at the airport because he had an engine failure.
+The Boeing 747 is larger than an A320.
+The Boeing 787 is a modern wide-body aircraft.
+She enjoys studying theoretical physics and quantum mechanics.
+The new cloud service offers scalable storage solutions for large enterprises.
+He carefully transplanted the delicate seedling into the larger pot.
+The symphony hall's design optimizes every acoustic detail.
+They discussed the ethical implications of using deepfake technology.
+The ocean currents play a critical role in global climate regulation.
+The journalist uncovered a major scandal involving corporate malfeasance.
+The final stage of the process involves rigorous quality assurance testing.
+Effective collaboration is essential for complex software development projects.
+The faint stars were barely visible against the light pollution.
+She researched the impact of social media algorithms on public opinion.
+The laboratory is equipped with state-of-the-art spectrographic analysis tools.
+He prefers composing music using analog synthesizers.
+The company implemented a zero-waste policy across all manufacturing plants.
+The history of warfare is intrinsically linked to technological innovation.
+The small village relies primarily on subsistence agriculture.
+They utilized a virtual reality simulation for professional training exercises.
+The unexpected hail storm severely damaged the summer crops.
+The museum docent provided fascinating details about the Renaissance paintings.
+The theory of evolution is supported by extensive fossil evidence.
+She spent the afternoon organizing her extensive collection of vintage records.
+The financial audit revealed minor discrepancies in the quarterly report.
+The development team used Python for the backend infrastructure.
+He wrote a powerful poem about the impermanence of natural beauty.
+The conservation group worked to protect the endangered rainforest habitat.
+The new smartphone features enhanced battery life and camera resolution.
+The legal brief argued persuasively for a change in jurisdiction.
+They hiked to the summit just as the sun broke through the clouds.
+Computational biology uses algorithms to analyze biological data.
+The recipe requires precision in measuring the dry and wet ingredients.
+She believes that kindness and empathy are universal virtues.
+The complex network topology was mapped out on a detailed diagram.
+The discovery of gravity is often attributed to Sir Isaac Newton.
+The organization launched a new initiative aimed at digital literacy.
+He used machine learning to personalize the user experience on the website.
+The aroma of grilling herbs drifted enticingly from the open window.
+They attended a seminar on sustainable urban planning strategies.
+The book details the history of cartography and map making.
+The construction project faced delays due to unforeseen weather conditions.
+The study of cosmology explores the origin and structure of the universe.
+She enjoys the tactile sensation of throwing clay on a pottery wheel.
+The company's mission statement emphasizes innovation and integrity.
+He meticulously documented the folklore and traditions of the region.
+The satellite orbits the Earth collecting atmospheric data constantly.
+The court ruled that the evidence was inadmissible due to procedural errors.
+They celebrated their anniversary with a quiet dinner overlooking the harbor.
+The rise of remote work has fundamentally changed office culture.
+The philosopher explored the concept of free will versus determinism.
+The chemical reaction produced a vivid blue precipitate instantly.
+She uses geothermal energy to heat her environmentally conscious home.
+The intricate lacework on the dress required hundreds of hours of labor.
+He utilized sophisticated data visualization tools to present his findings.
+The memoir offered a candid look at the challenges of early career life.
+The river frequently floods its banks during the intense monsoon season.
+They developed a proprietary algorithm for predicting stock market volatility.
+The museum acquired a rare coin dating back to the Roman Empire.
+She prefers reading physical books rather than digital e-readers.
+The rescue team navigated the treacherous terrain after the earthquake.
+The software update required a complete system restart and configuration.
+The study found a strong correlation between exercise and mental well-being.
+He is specializing in the study of deep-sea hydrothermal vents.
+The town meeting focused on improving local public transportation options.
+The artwork utilized unconventional materials
+She mastered the complex choreography after weeks of intense rehearsals.
+The manufacturing plant increased efficiency through automation.
+The legal team prepared a detailed response to the plaintiff's complaint.
+They enjoyed hiking along the coastal path at sunset every day.
+The telescope array is located in a remote desert location for clarity.
+The history of cryptography is a fascinating tale of codes and secrets.
+The new curriculum emphasizes critical thinking and digital citizenship.
+She applied non-parametric statistics to analyze the asymmetrical data.
+The gentle lapping of waves against the boat was incredibly soothing.
+He designed a robust system architecture to handle peak traffic loads.
+The ancient civilization developed a sophisticated irrigation system.
+They used augmented reality to overlay data onto the real-world view.
+The unexpected snowfall delighted the local children immensely.
+The foundation provides grants for independent investigative journalism.
+She is studying the biomechanics of avian flight and wing structure.
+The contract specified penalties for late delivery of the finished product.
+The political cartoonist used satire to critique current events effectively.
+He installed a smart home system to manage energy consumption automatically.
+The old stone fence marked the boundary of the original farm property.
+They celebrated the harvest with a traditional feast and local music.
+The scientific method relies on testable hypotheses and empirical evidence.
+She is learning Mandarin Chinese to better understand East Asian culture.
+The car's navigation system guided them through the winding mountain roads.
+The company is diversifying its portfolio into renewable energy projects.
+He created a dynamic spreadsheet model for financial projections.
+The concept of dark matter remains one of the greatest mysteries in physics.
+The library hosts a weekly storytelling hour for young children.
+She documented the gradual erosion of the cliff face over several years.
+The journalist used encrypted messaging for sensitive communications.
+The garden gate creaked open slowly in the morning breeze.
+They are developing a new vaccine against a recently identified virus strain.
+The historian specializes in the medieval period of European history.
+He used a three-dimensional printer to quickly prototype the new design.
+The atmosphere of Venus is toxic and extremely hot due to the greenhouse effect.
+The Boeing 737 is the most widely used commercial aircraft in the world.
+The Airbus A380 is the largest passenger airliner ever built.
+Modern aircraft are equipped with advanced navigation systems.
+Pilots undergo rigorous training before flying commercial aircraft.
+The cockpit contains hundreds of instruments and controls.
+Airlines schedule flights to maximize efficiency and profitability.
+Aircraft maintenance crews perform daily safety checks.
+Jet engines consume enormous amounts of fuel during flight.
+The cabin crew ensures passenger safety and comfort.
+Flight attendants receive training in emergency procedures.
+Passengers board aircraft through jet bridges at the gate.
+Baggage handlers load luggage into the aircraft hold.
+Ground crews service aircraft between flights.
+Fuel trucks fill the aircraft tanks before departure.
+De-icing equipment protects aircraft from winter conditions.
+Runways are carefully maintained for safe landings.
+Air traffic controllers direct aircraft safely through airspace.
+Weather forecasts help pilots plan flight routes.
+Turbulence occurs when aircraft fly through unstable air masses.
+Altitude affects fuel efficiency and aircraft performance.
+Oxygen masks deploy automatically if cabin pressure drops.
+Life jackets are located under passenger seats.
+Emergency exits are clearly marked throughout the aircraft.
+The fuselage is the main body of the aircraft.
+Wings generate lift to keep the aircraft airborne.
+Tail fins provide stability and control during flight.
+Landing gear extends before approaching the runway.
+Hydraulic systems power critical aircraft functions.
+Avionics include radar
+The autopilot system can fly the aircraft automatically.
+Flight data recorders preserve information about flight parameters.
+Cockpit voice recorders capture communications in the cockpit.
+Regular inspections ensure aircraft remain airworthy.
+Maintenance logs document all service performed on aircraft.
+Aircraft undergo major overhauls every few years.
+Spare parts are stocked for quick repairs.
+Airlines employ thousands of maintenance technicians.
+Engine efficiency has improved dramatically over decades.
+Noise-reducing technology makes modern aircraft quieter.
+Fuel-efficient designs reduce operating costs significantly.
+Aircraft manufacturers conduct extensive safety testing.
+Crash test simulations help engineers improve designs.
+Accident investigations provide valuable safety lessons.
+Safety regulations are enforced by aviation authorities.
+Certification standards ensure aircraft meet strict requirements.
+Insurance companies assess aircraft safety records.
+Pilots report any mechanical issues immediately.
+Maintenance crews address problems before the next flight.
+Aircraft must pass multiple inspections before takeoff.
+Safety is the highest priority in aviation.
+The airline industry has an excellent safety record.
+Modern aircraft are incredibly reliable machines.
+Redundant systems ensure safety if components fail.
+Backup power systems maintain critical functions.
+The primary flight computer monitors system performance.
+Secondary systems provide backup if primary systems fail.
+Cross-checking instruments ensures accuracy of readings.
+Pilots verify all readings before each flight.
+Pre-flight checklists help pilots prepare for departure.
+Post-flight checklists document the aircraft condition.
+Flight logs record all relevant flight information.
+Maintenance records track the complete service history.
+Aircraft registration numbers identify each plane uniquely.
+Livery designs distinguish airlines from competitors.
+Paint schemes protect the aircraft surface.
+Weather radar helps pilots navigate around storms.
+Wind shear detection systems warn of dangerous conditions.
+Terrain awareness systems prevent accidental crashes.
+Traffic collision avoidance systems prevent midair collisions.
+Navigation systems use multiple methods for accuracy.
+GPS provides precise location information.
+Inertial reference systems work when GPS is unavailable.
+Communication radios connect pilots with air traffic control.
+Transponders help identify aircraft on radar displays.
+Squawk codes are assigned to each flight.
+Flight plans file routes and altitudes in advance.
+Airways provide designated routes through airspace.
+Holding patterns allow aircraft to wait safely.
+Approach procedures guide aircraft toward landing.
+Instrument landing systems enable landings in fog.
+Visual landing systems help pilots see the runway.
+Runway lights guide aircraft during night operations.
+Precision approach radar assists landing pilots.
+Wind measurements help determine safe landing conditions.
+Crosswind limits prevent unsafe landing attempts.
+Takeoff roll requires a specific runway length.
+Climb performance depends on weight and temperature.
+Cruising altitude is maintained automatically.
+Descent rates are carefully controlled by pilots.
+Landing speeds are slower than cruising speeds.
+Reverse thrust helps slow aircraft after landing.
+Wheel brakes provide stopping force on the runway.
+Ground spoilers increase drag during landing.
+Flare technique allows smooth touchdown.
+Touchdown occurs near the runway threshold.
+Rollout distance is carefully calculated before landing.
+Taxiway congestion is managed by controllers.
+Gate assignments are made before arrival.
+Ground handling equipment tows aircraft to gates.
+Catering trucks provide meals and beverages.
+Lavatory trucks service the onboard facilities.
+Water trucks fill the aircraft water systems.
+Trash trucks remove waste from the aircraft.
+Cleaning crews prepare aircraft between flights.
+Passenger comfort is important for airline loyalty.
+In-flight entertainment systems provide movies and games.
+WiFi connectivity allows passengers to stay connected.
+Seat selection options give passengers choices.
+Premium seating offers extra space and amenities.
+Meal services vary based on flight duration.
+Beverage services occur throughout the flight.
+Duty-free shopping offers tax-free purchases.
+Frequent flyer programs reward loyal customers.
+Airline alliances allow code sharing and connections.
+Connecting flights require careful timing.
+Luggage transfers are handled automatically for connections.
+International flights require passport and visa verification.
+Customs inspections occur upon arrival.
+Immigration processing can take significant time.
+Travel documents must be valid for entry.
+Visas are required for many countries.
+Vaccination requirements may apply for some destinations.
+Health certificates document medical procedures.
+Travel insurance provides coverage for unexpected events.
+Trip cancellation protection reimburses unused tickets.
+Luggage insurance covers lost or damaged baggage.
+Valuables should be carried in cabin baggage.
+Restricted items are not allowed on aircraft.
+Liquids must follow specific volume restrictions.
+Sharp objects are prohibited from carry-on luggage.
+Electronic devices must be powered down during taxi.
+Seatbelts must be fastened during flight.
+The fasten seatbelt sign indicates turbulence ahead.
+Overhead bins have weight and size restrictions.
+Aisle seats allow easy bathroom access.
+Window seats provide views of the scenery.
+Middle seats are the least desirable positions.
+Galleys contain food and beverage service supplies.
+Lavatories are occupied on rotation during flights.
+Oxygen levels are maintained throughout the cabin.
+Temperature controls keep the cabin comfortable.
+Humidity levels affect passenger comfort.
+Air circulation prevents stagnant air conditions.
+Pressurization maintains breathable air at altitude.
+Decompression emergencies are extremely rare.
+Hypoxia occurs when oxygen levels are insufficient.
+Cabin crew training includes emergency procedures.
+Evacuation slides deploy rapidly during emergencies.
+Life rafts are located on water flights.
+Survival kits contain emergency supplies.
+First aid kits are available onboard.
+Medical emergencies are handled by trained crew.
+Automated external defibrillators treat cardiac emergencies.
+Communication with ground medical personnel is maintained.
+Emergency diversion airports provide landing options.
+Fuel dumping reduces aircraft weight if necessary.
+Water landings are extremely unlikely events.
+Aircraft flotation systems provide temporary buoyancy.
+Rescue equipment is deployed by coast guard.
+Search and rescue operations locate survivors.
+Investigation teams analyze accident data.
+Safety recommendations prevent similar accidents.
+Aviation regulations evolve based on accident lessons.
+Training programs incorporate recent safety findings.
+Pilot experience reduces accident likelihood.
+Rest requirements prevent pilot fatigue.
+Fatigue management protocols protect safety.
+Alcohol consumption is strictly prohibited for pilots.
+Drug testing ensures pilot fitness for duty.
+Medical certifications verify pilot health.
+Vision requirements ensure adequate eyesight.
+Hearing tests confirm adequate hearing ability.
+Cardiovascular health is regularly evaluated.
+Psychological evaluations assess mental fitness.
+Stress management training helps pilots cope.
+Simulator training prepares pilots for emergencies.
+Emergency procedures are practiced regularly.
+Engine failure drills prepare for system loss.
+Fire emergency procedures are well rehearsed.
+Hydraulic failure scenarios are practiced frequently.
+Electrical failure simulations test backup systems.
+Pressurization loss procedures are demonstrated.
+Decompression scenarios are handled in simulators.
+Evacuation drills prepare crews for rapid exits.
+Water landing drills practice ditching procedures.
+Terrorist threat scenarios are addressed in training.
+Security protocols protect aircraft from threats.
+Cockpit doors prevent unauthorized entry.
+Air marshals provide additional security protection.
+Background checks screen all aviation employees.
+Baggage screening detects prohibited items.
+Passenger screening includes metal detection.
+Advanced imaging technology scans for weapons.
+Dog teams detect explosives at airports.
+Security perimeters restrict unauthorized access.
+Employee badges allow controlled access.
+Security cameras monitor restricted areas.
+Access logs document who enters secure areas.
+Cybersecurity protects aircraft systems from hackers.
+Encryption secures sensitive aircraft data.
+Software updates patch security vulnerabilities.
+Antivirus protection prevents malware infections.
+Intrusion detection identifies unauthorized access.
+Firewalls protect network systems.
+Backup systems ensure redundancy.
+Data backups prevent information loss.
+Cloud storage provides secure data storage.
+Server maintenance ensures system reliability.
+Network monitoring detects unusual activity.
+System logs record all computer activities.
+Audit trails track system changes.
+Two-factor authentication secures accounts.
+Password policies enforce strong passwords.
+Biometric security uses fingerprints for identification.
+Facial recognition technology identifies individuals.
+Iris scanning provides additional verification.
+Voice recognition technology confirms identity.
+Digital signatures authenticate documents.
+Blockchain technology ensures data integrity.
+Cryptocurrency transactions provide security.
+Virtual private networks protect communication.
+SSL encryption secures web transactions.
+TLS protocols protect data transmission.
+DDoS protection prevents denial of service attacks.
+Rate limiting controls traffic volume.
+Load balancing distributes traffic efficiently.
+Cache systems speed up data retrieval.
+Database optimization improves query performance.
+Query caching reduces database load.
+Memory management ensures efficient resource use.
+Garbage collection cleans up unused memory.
+Thread management handles concurrent operations.
+Process management controls system resources.
+Task scheduling prioritizes important operations.
+Queue management handles request backlogs.
+Buffer management prevents data overflow.
+Overflow protection prevents system crashes.
+Error handling catches and reports issues.
+Exception handling manages unexpected situations.
+Logging records system events.
+Monitoring alerts staff to problems.
+Alerting systems notify administrators immediately.
+Escalation procedures handle critical issues.
+On-call rotation ensures 24-hour coverage.
+Incident response teams handle emergencies.
+Root cause analysis identifies problem sources.
+Resolution procedures fix identified issues.
+Documentation records problem details.
+Knowledge bases preserve solutions.
+Wiki systems share information.
+Forum discussions exchange ideas.
+Chat systems enable quick communication.
+Email systems provide formal communication.
+Tickets track problem resolution status.
+Milestone tracking monitors project progress.
+Sprint planning organizes development work.
+Agile methodology improves team efficiency.
+Scrum ceremonies coordinate team activities.
+Daily standups keep teams aligned.
+Retrospectives improve team processes.
+Velocity tracking measures team productivity.
+Burndown charts show project progress.
+Release planning schedules code deployments.
+Deployment automation speeds up releases.
+Continuous integration catches problems early.
+Continuous deployment releases code continuously.
+Version control tracks code changes.
+Git repositories store code history.
+Branching strategies organize development work.
+Pull requests review code before merging.
+Code reviews improve code quality.
+Testing frameworks automate quality assurance.
+Unit tests verify individual functions.
+Integration tests verify system components.
+End-to-end tests verify complete workflows.
+Performance tests measure system speed.
+Load tests verify system capacity.
+Stress tests find system limits.
+Security testing identifies vulnerabilities.
+Penetration testing simulates attacks.
+Vulnerability scanning finds weaknesses.
+Compliance testing ensures regulatory adherence.
+Quality assurance processes prevent defects.
+Bug tracking documents identified issues.
+Severity levels prioritize bug fixes.
+Priority levels organize fix scheduling.
+Hotfix processes address critical issues.
+Patch management distributes security updates.
+Release notes document feature changes.
+Migration guides help users transition.
+Backward compatibility ensures old code works.
+Forward compatibility allows future changes.
+API documentation helps developers use services.
+Code examples show correct usage patterns.
+Tutorials teach common tasks.
+Getting started guides ease onboarding.
+FAQ sections answer common questions.
+Troubleshooting guides solve common problems.
+Support tickets enable customer assistance.
+Community forums enable peer support.
+Live chat provides immediate assistance.
+Email support provides detailed responses.
+Phone support enables voice communication.
+Video tutorials demonstrate visual procedures.
+Webinars enable interactive learning.
+Conferences bring community together.
+Meetups facilitate local networking.
+User groups enable peer learning.
+Mentorship programs match experts with learners.
+Internship programs train new talent.
+Apprenticeships teach skilled trades.
+Bootcamps accelerate career transitions.
+Online courses provide flexible learning.
+University programs offer formal education.
+Certification programs validate skills.
+Continuing education maintains current knowledge.
+Professional development advances careers.
+Leadership training prepares managers.
+Public speaking training builds communication skills.
+Writing workshops improve documentation skills.
+Presentation skills training enhances communication.
+Negotiation training improves business outcomes.
+Sales training increases revenue.
+Customer service training improves satisfaction.
+Time management training increases productivity.
+Stress management training improves wellbeing.
+Conflict resolution training prevents disputes.
+Diversity training promotes inclusion.
+Ethics training ensures compliance.
+Harassment prevention training protects employees.
+Safety training prevents workplace injuries.
+Emergency procedures training prepares for crises.
+First aid training saves lives.
+CPR training provides lifesaving skills.
+Defibrillator training teaches device usage.
+Choking prevention training enables rescue.
+Burn treatment training minimizes injury.
+Fracture care training stabilizes injuries.
+Wound care training prevents infection.
+Infection control training prevents disease spread.
+Hygiene practices prevent contamination.
+Sanitation procedures maintain cleanliness.
+Disinfection processes kill pathogens.
+Sterilization ensures complete pathogen elimination.
+Quarantine procedures prevent disease spread.
+Isolation protocols protect vulnerable populations.
+Personal protective equipment prevents pathogen exposure.
+Vaccination programs prevent diseases.
+Immunization records document vaccinations.
+Booster shots maintain immunity levels.
+Antibiotic therapy treats bacterial infections.
+Antiviral therapy treats viral infections.
+Pain management improves patient comfort.
+Symptom management addresses disease effects.
+Rehabilitation therapy restores function.
+Physical therapy strengthens muscles.
+Occupational therapy improves daily function.
+Speech therapy improves communication.
+Mental health counseling addresses emotional issues.
+Psychotherapy treats mental disorders.
+Medication management optimizes drug therapy.
+Drug interactions are carefully monitored.
+Side effects are managed carefully.
+Allergic reactions are treated immediately.
+Adverse events are reported promptly.
+Clinical trials test new treatments.
+Research studies advance medical knowledge.
+Literature reviews summarize existing knowledge.
+Meta-analyses combine multiple studies.
+Systematic reviews assess evidence quality.
+Evidence-based medicine improves outcomes.
+Best practices guide clinical decisions.
+Clinical guidelines standardize care.
+Protocols ensure consistent procedures.
+Procedures are documented carefully.
+Records maintain patient information.
+Privacy protection secures health data.
+Confidentiality agreements protect information.
+HIPAA compliance protects patient privacy.
+Data security prevents unauthorized access.
+Access controls limit information exposure.
+Audit logs track data access.
+Encryption protects data transmission.
+Secure communications prevent eavesdropping.
+VPNs provide secure connections.
+Firewalls block unauthorized traffic.
+Intrusion detection identifies attacks.
+Antimalware software removes malicious code.
+Antivirus programs prevent virus infections.
+Ransomware protection prevents data theft.
+Backup procedures prevent data loss.
+Disaster recovery plans restore operations.
+Business continuity planning minimizes disruption.
+Risk management identifies potential problems.
+Insurance coverage provides financial protection.
+Liability protection limits legal exposure.
+Workers compensation covers work injuries.
+Unemployment insurance supports job loss.
+Disability insurance replaces lost income.
+Life insurance protects dependents.
+Health insurance covers medical expenses.
+Dental insurance covers tooth care.
+Vision insurance covers eye care.
+Mental health coverage addresses mental illness.
+Substance abuse coverage treats addiction.
+Rehabilitation coverage restores function.
+Long-term care insurance covers extended care.
+Nursing home care provides residential services.
+Assisted living facilities support daily activities.
+Senior care services assist aging adults.
+Hospice care supports end-of-life comfort.
+Palliative care relieves suffering.
+Pain management improves quality of life.
+Spiritual care supports belief systems.
+Grief counseling supports bereavement.
+Funeral services honor the deceased.
+Estate planning organizes assets.
+Will preparation documents wishes.
+Probate processing distributes assets.
+Trust management protects assets.
+Financial planning builds wealth.
+Investment strategies grow money.
+Retirement planning ensures future security.
+Pension plans provide retirement income.
+Social security provides government retirement.
+401k plans offer employer retirement.
+IRA accounts provide individual retirement.
+Roth conversions reduce future taxes.
+Tax planning minimizes tax burden.
+Deduction strategies reduce taxable income.
+Credit strategies reduce tax liability.
+Refund optimization maximizes returns.
+Quarterly payments prevent penalties.
+Estimated taxes avoid surprises.
+Extension requests delay filing deadlines.
+Audit protection responds to IRS inquiries.
+Appeals process challenges IRS decisions.
+Settlement negotiations resolve disputes.
+Collection defense prevents asset seizure.
+Wage garnishment protection prevents income loss.
+Lien release removes property claims.
+Levy protection prevents forced sales.
+Bankruptcy protection prevents creditor actions.
+Debt restructuring reduces obligations.
+Credit counseling improves financial health.
+Debt consolidation simplifies payments.
+Payment plans spread costs over time.
+Hardship programs provide relief.
+Forbearance pauses payments temporarily.
+Deferment delays payments indefinitely.
+Forgiveness programs cancel remaining debt.
+Public service forgiveness rewards service.
+Teacher forgiveness rewards educators.
+Nurse forgiveness rewards healthcare workers.
+Military forgiveness rewards service members.
+Income-driven repayment reduces payments.
+Standard repayment pays loans quickly.
+Graduated repayment starts low then rises.
+Employer benefits provide loan help.
+Employer matching multiplies savings.
+Profit sharing distributes gains.
+Bonus plans reward performance.
+Stock options provide ownership.
+Restricted stock units defer compensation.
+Equity compensation aligns interests.
+Performance bonuses reward results.
+Attendance bonuses reward reliability.
+Referral bonuses reward recruitment.
+Loyalty bonuses reward tenure.
+Completion bonuses reward milestones.
+Achievement bonuses reward excellence.
+Safety bonuses reward protection.
+Quality bonuses reward perfection.
+Customer service bonuses reward satisfaction.
+Efficiency bonuses reward productivity.
+Innovation bonuses reward creativity.
+Teamwork bonuses reward collaboration.
+Leadership bonuses reward management.
+Volunteer bonuses reward service.
+Community bonuses reward participation.
+Sustainability bonuses reward environmental protection.
+Diversity bonuses reward inclusion.
+Training bonuses reward development.
+Certification bonuses reward credentials.
+Degree bonuses reward education.
+Language bonuses reward multilingualism.
+Wellness bonuses reward health.
+Fitness bonuses reward exercise.
+Nutrition bonuses reward healthy eating.
+Mental health bonuses reward wellbeing.
+Work-life balance bonuses reward harmony.
+Remote work bonuses reward flexibility.
+Flexible schedule bonuses reward adjustment.
+Compressed schedule bonuses reward efficiency.
+Part-time bonuses reward availability.
+Contract bonuses reward temporary work.
+Freelance bonuses reward independent work.
+Gig economy work provides flexibility.
+Zero-hour contracts enable flexibility.
+Employment agencies place workers.
+Temp agencies provide temporary workers.
+Staffing companies manage workforce.
+Recruiting firms find candidates.
+Headhunters identify top talent.
+Executive search finds leaders.
+Campus recruiting finds graduates.
+Job fairs attract candidates.
+Online job boards list positions.
+LinkedIn recruiting finds professionals.
+Indeed postings attract job seekers.
+Glassdoor reviews influence candidates.
+Salary websites inform negotiations.
+Benefits packages attract employees.
+Sign-on bonuses attract talent.
+Relocation packages support moves.
+Housing assistance supports new employees.
+Transportation assistance reduces commute costs.
+Childcare assistance supports parents.
+Educational assistance supports learning.
+Tuition reimbursement funds education.
+Book allowances support learning.
+Conference attendance funds development.
+Membership fees cover professional organizations.
+Subscription services provide resources.
+Software licenses enable tool use.
+Hardware provisioning enables productivity.
+Office space supports collaboration.
+Meeting rooms enable discussions.
+Video conferencing enables remote meetings.
+Phone systems enable communication.
+Email systems enable messaging.
+Instant messaging enables quick communication.
+Project management tools organize work.
+Time tracking ensures accurate billing.
+Expense tracking controls spending.
+Budget management optimizes resources.
+Forecasting predicts future needs.
+Scenario planning prepares for changes.
+Contingency planning manages uncertainties.
+Risk assessment identifies threats.
+Opportunity assessment identifies advantages.
+SWOT analysis evaluates position.
+Competitor analysis understands rivals.
+Market research understands customers.
+Customer surveys gather feedback.
+Focus groups explore ideas.
+Interviews gather detailed information.
+Observations understand behavior.
+Experiments test hypotheses.
+Case studies examine examples.
+Data analysis draws conclusions.
+Statistical analysis proves significance.
+Regression analysis finds relationships.
+Correlation analysis measures association.
+Time series analysis tracks trends.
+Forecasting predicts future values.
+Sentiment analysis understands opinions.
+Natural language processing interprets text.
+Machine learning enables pattern recognition.
+Deep learning finds complex patterns.
+The Earth orbits the Sun once every three hundred and sixty five days.
+Water freezes at zero degrees Celsius and boils at one hundred degrees.
+Oxygen is a gas that humans need to breathe to stay alive.
+The capital city of France is Paris which is known for its history.
+Standard computers use binary code consisting of zeros and ones to process data.
+Gravity is the force that pulls objects toward the center of the Earth.
+Trees produce oxygen through a biological process called photosynthesis.
+Mathematics is the study of numbers shapes and patterns in the universe.
+Light travels at a speed of approximately three hundred thousand kilometers per second.
+A standard human skeleton is made up of two hundred and six bones.
+The Sun sets in the west and rises in the east due to the Earth's rotation.
+"Hello, how can i help you today"
+"Hello, how can i help you today?"
+"Hello, how can i help you today"
+"Hello, how can i help you today"
+"Hello, how can i help you today?"
+"Hi Hello, how can i help you today"
+"Hi Hello, how can i help you today"
+"hi Hello, how can i help you today"
+"hi Hello, how can i help you today"
+uhh i dont really know Thats okay what would you like to share me
+uhh i dont really know Thats okay what would you like to share me
+uhh i dont really know Thats okay what would you like to share me
+uhh i dont really know Thats okay what would you like to share me
+uhh i dont really know Thats okay what would you like to share me
+hi hello how can i help you today
+hi hello how can i help you today
+hi hello how can i help you today
+hi hello how can i help you today
+hi hello how can i help you today
+hi hello hello how can i help you
+hi hello hello how can i help you
+hi hello hello how can i help you
+hi hello hello how can i help you
+hi hello hello how can i help you

training_data.csv

@@ -0,0 +1,11 @@
+hi Hello how can i help you today?
+"what can you do I can assist with various tasks including - writing storytelling and providing information"
+hi Greetings How may I assist you today do you have a specific topic to discuss?
+i dont know thats fine if you need help with something just ask me meanwhile would you like to share a specific topic you are interested in
+"Hello Hi how can I assist you today, is there something specific you would like to discuss or learn about?"
+hi Hello there how can I help you today do you have any questions or topics in mind
+"You are dumb I'm sorry to hear your feedback about me, i do not want to get you annoyed so i'll just stop."
+hi Greetings how may i assist you today do is there some specific topic you want to discuss
+hi Hello how may i assist you today do is there something specific topic you want to discuss
+"hi Hello, how can i can i help you today do you have any questions or topics you would like to discuss?"
+