LLMs-from-scratch/ch07/01_main-chapter-code/exercise_experiments.py

# Copyright (c) Sebastian Raschka under Apache License 2.0 (see LICENSE.txt).
# Source for "Build a Large Language Model From Scratch"
#   - https://www.manning.com/books/build-a-large-language-model-from-scratch
# Code: https://github.com/rasbt/LLMs-from-scratch
#
# Code to run the exercises; see exercise-solutions.ipynb for more information

from functools import partial
from importlib.metadata import version
import json
import math
import os
import re
import time
import urllib

import matplotlib.pyplot as plt
import tiktoken
import torch
from torch.utils.data import Dataset, DataLoader
from tqdm import tqdm

# Import from local files in this folder
from gpt_download import download_and_load_gpt2
from previous_chapters import (
    calc_loss_loader,
    generate,
    GPTModel,
    load_weights_into_gpt,
    text_to_token_ids,
    train_model_simple,
    token_ids_to_text
)


class InstructionDataset(Dataset):
    def __init__(self, data, tokenizer):
        self.data = data

        # Pre-tokenize texts
        self.encoded_texts = []
        for entry in data:
            instruction_plus_input = format_input(entry)
            response_text = f"\n\n### Response:\n{entry['output']}"
            full_text = instruction_plus_input + response_text
            self.encoded_texts.append(
                tokenizer.encode(full_text)
            )

    def __getitem__(self, index):
        return self.encoded_texts[index]

    def __len__(self):
        return len(self.data)


class InstructionDatasetWithMasking(Dataset):
    def __init__(self, data, tokenizer):
        self.data = data

        # New: Separate list for instruction lengths
        self.instruction_lengths = []
        self.encoded_texts = []

        for entry in data:
            instruction_plus_input = format_input(entry)
            response_text = f"\n\n### Response:\n{entry['output']}"
            full_text = instruction_plus_input + response_text

            self.encoded_texts.append(
                tokenizer.encode(full_text)
            )

            # New: collect instruction lengths
            instruction_length = len(tokenizer.encode(instruction_plus_input))
            self.instruction_lengths.append(instruction_length)

    def __getitem__(self, index):
        # New: return both instruction lengths and texts separately
        return self.instruction_lengths[index], self.encoded_texts[index]

    def __len__(self):
        return len(self.data)


class InstructionDatasetPhi(Dataset):
    def __init__(self, data, tokenizer):
        self.data = data

        # Pre-tokenize texts
        self.encoded_texts = []
        for entry in data:

            ###################################################################
            # NEW: Use `format_input_phi` and adjust the response text template
            instruction_plus_input = format_input_phi(entry)
            response_text = f"\n<|assistant|>:\n{entry['output']}"
            ###################################################################
            full_text = instruction_plus_input + response_text
            self.encoded_texts.append(
                tokenizer.encode(full_text)
            )

    def __getitem__(self, index):
        return self.encoded_texts[index]

    def __len__(self):
        return len(self.data)


class LinearWithLoRA(torch.nn.Module):
    def __init__(self, linear, rank, alpha):
        super().__init__()
        self.linear = linear
        self.lora = LoRALayer(
            linear.in_features, linear.out_features, rank, alpha
        )

    def forward(self, x):
        return self.linear(x) + self.lora(x)


class LoRALayer(torch.nn.Module):
    def __init__(self, in_dim, out_dim, rank, alpha):
        super().__init__()
        self.A = torch.nn.Parameter(torch.empty(in_dim, rank))
        torch.nn.init.kaiming_uniform_(self.A, a=math.sqrt(5))  # similar to standard weight initialization
        self.B = torch.nn.Parameter(torch.zeros(rank, out_dim))
        self.alpha = alpha

    def forward(self, x):
        x = self.alpha * (x @ self.A @ self.B)
        return x


def replace_linear_with_lora(model, rank, alpha):
    for name, module in model.named_children():
        if isinstance(module, torch.nn.Linear):
            # Replace the Linear layer with LinearWithLoRA
            setattr(model, name, LinearWithLoRA(module, rank, alpha))
        else:
            # Recursively apply the same function to child modules
            replace_linear_with_lora(module, rank, alpha)


def custom_collate_fn(
    batch,
    pad_token_id=50256,
    ignore_index=-100,
    allowed_max_length=None,
    device="cpu"
):
    # Find the longest sequence in the batch
    batch_max_length = max(len(item)+1 for item in batch)

    # Pad and prepare inputs and targets
    inputs_lst, targets_lst = [], []

    for item in batch:
        new_item = item.copy()
        # Add an <|endoftext|> token
        new_item += [pad_token_id]
        # Pad sequences to max_length
        padded = new_item + [pad_token_id] * (batch_max_length - len(new_item))
        inputs = torch.tensor(padded[:-1])  # Truncate the last token for inputs
        targets = torch.tensor(padded[1:])  # Shift +1 to the right for targets

        # New: Replace all but the first padding tokens in targets by ignore_index
        mask = targets == pad_token_id
        indices = torch.nonzero(mask).squeeze()
        if indices.numel() > 1:
            targets[indices[1:]] = ignore_index

        # New: Optionally truncate to maximum sequence length
        if allowed_max_length is not None:
            inputs = inputs[:allowed_max_length]
            targets = targets[:allowed_max_length]

        inputs_lst.append(inputs)
        targets_lst.append(targets)

    # Convert list of inputs and targets to tensors and transfer to target device
    inputs_tensor = torch.stack(inputs_lst).to(device)
    targets_tensor = torch.stack(targets_lst).to(device)

    return inputs_tensor, targets_tensor


def custom_collate_with_masking_fn(
    batch,
    pad_token_id=50256,
    ignore_index=-100,
    allowed_max_length=None,
    device="cpu"
):
    # Find the longest sequence in the batch
    batch_max_length = max(len(item)+1 for instruction_length, item in batch)   # New: batch is now a tuple

    # Pad and prepare inputs and targets
    inputs_lst, targets_lst = [], []

    for instruction_length, item in batch:  # New: batch is now a tuple
        new_item = item.copy()
        # Add an <|endoftext|> token
        new_item += [pad_token_id]
        # Pad sequences to max_length
        padded = new_item + [pad_token_id] * (batch_max_length - len(new_item))
        inputs = torch.tensor(padded[:-1])  # Truncate the last token for inputs
        targets = torch.tensor(padded[1:])  # Shift +1 to the right for targets

        # Replace all but the first padding tokens in targets by ignore_index
        mask = targets == pad_token_id
        indices = torch.nonzero(mask).squeeze()
        if indices.numel() > 1:
            targets[indices[1:]] = ignore_index

        # New: Mask all input and instruction tokens in the targets
        targets[:instruction_length-1] = -100

        # Optionally truncate to maximum sequence length
        if allowed_max_length is not None:
            inputs = inputs[:allowed_max_length]
            targets = targets[:allowed_max_length]

        inputs_lst.append(inputs)
        targets_lst.append(targets)

    # Convert list of inputs and targets to tensors and transfer to target device
    inputs_tensor = torch.stack(inputs_lst).to(device)
    targets_tensor = torch.stack(targets_lst).to(device)

    return inputs_tensor, targets_tensor


def download_and_load_file(file_path, url):

    if not os.path.exists(file_path):
        with urllib.request.urlopen(url) as response:
            text_data = response.read().decode("utf-8")
        with open(file_path, "w", encoding="utf-8") as file:
            file.write(text_data)
    else:
        with open(file_path, "r", encoding="utf-8") as file:
            text_data = file.read()

    with open(file_path, "r") as file:
        data = json.load(file)

    return data


def format_input_phi(entry):
    instruction_text = (
        f"<|user|>\n{entry['instruction']}"
    )

    input_text = f"\n{entry['input']}" if entry["input"] else ""

    return instruction_text + input_text


def format_input(entry):
    instruction_text = (
        f"Below is an instruction that describes a task. "
        f"Write a response that appropriately completes the request."
        f"\n\n### Instruction:\n{entry['instruction']}"
    )

    input_text = f"\n\n### Input:\n{entry['input']}" if entry["input"] else ""

    return instruction_text + input_text


def plot_losses(epochs_seen, tokens_seen, train_losses, val_losses, plot_name):
    fig, ax1 = plt.subplots(figsize=(12, 6))

    # Plot training and validation loss against epochs
    ax1.plot(epochs_seen, train_losses, label="Training loss")
    ax1.plot(epochs_seen, val_losses, linestyle="-.", label="Validation loss")
    ax1.set_xlabel("Epochs")
    ax1.set_ylabel("Loss")
    ax1.legend(loc="upper right")

    # Create a second x-axis for tokens seen
    ax2 = ax1.twiny()  # Create a second x-axis that shares the same y-axis
    ax2.plot(tokens_seen, train_losses, alpha=0)  # Invisible plot for aligning ticks
    ax2.set_xlabel("Tokens seen")

    fig.tight_layout()  # Adjust layout to make room
    print(f"Plot saved as {plot_name}")
    plt.savefig(plot_name)
    # plt.show()


def main(mask_instructions=False, alpaca52k=False, phi3_prompt=False, lora=False):
    #######################################
    # Print package versions
    #######################################
    print()
    pkgs = [
        "matplotlib",  # Plotting library
        "tiktoken",    # Tokenizer
        "torch",       # Deep learning library
        "tqdm",        # Progress bar
        "tensorflow",  # For OpenAI's pretrained weights
    ]
    for p in pkgs:
        print(f"{p} version: {version(p)}")
    print(50*"-")

    #######################################
    # Download and prepare dataset
    #######################################
    file_path = "instruction-data.json"

    if alpaca52k:
        url = "https://raw.githubusercontent.com/tatsu-lab/stanford_alpaca/main/alpaca_data.json"
    else:
        url = "https://raw.githubusercontent.com/rasbt/LLMs-from-scratch/main/ch07/01_main-chapter-code/instruction-data.json"
    data = download_and_load_file(file_path, url)

    train_portion = int(len(data) * 0.85)  # 85% for training
    test_portion = int(len(data) * 0.1)    # 10% for testing

    train_data = data[:train_portion]
    test_data = data[train_portion:train_portion + test_portion]
    val_data = data[train_portion + test_portion:]

    print("Training set length:", len(train_data))
    print("Validation set length:", len(val_data))
    print("Test set length:", len(test_data))
    print(50*"-")

    tokenizer = tiktoken.get_encoding("gpt2")
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print("Device:", device)
    print(50*"-")

    if alpaca52k:
        allowed_max_length = 512
    else:
        allowed_max_length = 1024

    if mask_instructions and phi3_prompt:
        raise ValueError("Simultaneous support for instruction masking and the Phi-3 prompt template has not been implemented, yet.")

    if mask_instructions:
        customized_collate_fn = partial(custom_collate_with_masking_fn, device=device, allowed_max_length=allowed_max_length)
        CustomDataset = InstructionDatasetWithMasking
    elif phi3_prompt:
        customized_collate_fn = partial(custom_collate_fn, device=device, allowed_max_length=allowed_max_length)
        CustomDataset = InstructionDatasetPhi
    else:
        customized_collate_fn = partial(custom_collate_fn, device=device, allowed_max_length=allowed_max_length)
        CustomDataset = InstructionDataset

    num_workers = 0

    if alpaca52k:
        batch_size = 4
    else:
        batch_size = 8

    torch.manual_seed(123)

    train_dataset = CustomDataset(train_data, tokenizer)
    train_loader = DataLoader(
        train_dataset,
        batch_size=batch_size,
        collate_fn=customized_collate_fn,
        shuffle=True,
        drop_last=True,
        num_workers=num_workers
    )

    val_dataset = CustomDataset(val_data, tokenizer)
    val_loader = DataLoader(
        val_dataset,
        batch_size=batch_size,
        collate_fn=customized_collate_fn,
        shuffle=False,
        drop_last=False,
        num_workers=num_workers
    )

    #######################################
    # Load pretrained model
    #######################################
    BASE_CONFIG = {
        "vocab_size": 50257,     # Vocabulary size
        "context_length": 1024,  # Context length
        "drop_rate": 0.0,        # Dropout rate
        "qkv_bias": True         # Query-key-value bias
    }

    model_configs = {
        "gpt2-small (124M)": {"emb_dim": 768, "n_layers": 12, "n_heads": 12},
        "gpt2-medium (355M)": {"emb_dim": 1024, "n_layers": 24, "n_heads": 16},
        "gpt2-large (774M)": {"emb_dim": 1280, "n_layers": 36, "n_heads": 20},
        "gpt2-xl (1558M)": {"emb_dim": 1600, "n_layers": 48, "n_heads": 25},
    }

    CHOOSE_MODEL = "gpt2-medium (355M)"

    BASE_CONFIG.update(model_configs[CHOOSE_MODEL])

    model_size = CHOOSE_MODEL.split(" ")[-1].lstrip("(").rstrip(")")
    settings, params = download_and_load_gpt2(model_size=model_size, models_dir="gpt2")

    model = GPTModel(BASE_CONFIG)
    load_weights_into_gpt(model, params)
    model.eval()
    model.to(device)

    print("Loaded model:", CHOOSE_MODEL)
    print(50*"-")

    if lora:
        total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
        print(f"Total trainable parameters before: {total_params:,}")

        for param in model.parameters():
            param.requires_grad = False

        total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
        print(f"Total trainable parameters after: {total_params:,}")
        replace_linear_with_lora(model, rank=16, alpha=16)

        total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
        print(f"Total trainable LoRA parameters: {total_params:,}")
        model.to(device)

    #######################################
    # Finetuning the model
    #######################################
    print("Initial losses")
    with torch.no_grad():
        train_loss = calc_loss_loader(train_loader, model, device, num_batches=5)
        val_loss = calc_loss_loader(val_loader, model, device, num_batches=5)

    print("   Training loss:", train_loss)
    print("   Validation loss:", val_loss)

    start_time = time.time()

    num_epochs = 2
    optimizer = torch.optim.AdamW(model.parameters(), lr=0.00005, weight_decay=0.1)

    torch.manual_seed(123)

    start_context = format_input_phi(val_data[0]) if phi3_prompt else format_input(val_data[0])

    train_losses, val_losses, tokens_seen = train_model_simple(
        model, train_loader, val_loader, optimizer, device,
        num_epochs=num_epochs, eval_freq=5, eval_iter=5,
        start_context=start_context, tokenizer=tokenizer
    )

    end_time = time.time()
    execution_time_minutes = (end_time - start_time) / 60
    print(f"Training completed in {execution_time_minutes:.2f} minutes.")

    epochs_tensor = torch.linspace(0, num_epochs, len(train_losses))

    plot_name = "loss-plot.pdf"
    if mask_instructions:
        plot_name = plot_name.replace(".pdf", "-mask-instructions.pdf")
    if alpaca52k:
        plot_name = plot_name.replace(".pdf", "-alpaca52k.pdf")
    if phi3_prompt:
        plot_name = plot_name.replace(".pdf", "-phi3-prompt.pdf")
    if lora:
        plot_name = plot_name.replace(".pdf", "-lora.pdf")
    if not any([mask_instructions, alpaca52k, phi3_prompt, lora]):
        plot_name = plot_name.replace(".pdf", "-baseline.pdf")

    plot_losses(epochs_tensor, tokens_seen, train_losses, val_losses, plot_name)
    print(50*"-")

    #######################################
    # Saving results
    #######################################
    print("Generating responses")
    for i, entry in tqdm(enumerate(test_data), total=len(test_data)):

        input_text = format_input_phi(entry) if phi3_prompt else format_input(entry)

        token_ids = generate(
            model=model,
            idx=text_to_token_ids(input_text, tokenizer).to(device),
            max_new_tokens=256,
            context_size=BASE_CONFIG["context_length"],
            eos_id=50256
        )
        generated_text = token_ids_to_text(token_ids, tokenizer)

        if phi3_prompt:
            response_text = generated_text[len(input_text):].replace("<|assistant|>:", "").strip()
        else:
            response_text = generated_text[len(input_text):].replace("### Response:", "").strip()

        test_data[i]["model_response"] = response_text

    test_data_path = "instruction-data-with-response.json"
    file_name = f"{re.sub(r'[ ()]', '', CHOOSE_MODEL) }-sft.pth"

    if mask_instructions:
        test_data_path = test_data_path.replace(".json", "-mask-instructions.json")
        file_name = file_name.replace(".pth", "-mask-instructions.pth")
    if alpaca52k:
        test_data_path = test_data_path.replace(".json", "-alpaca52k.json")
        file_name = file_name.replace(".pth", "-alpaca52k.pth")
    if phi3_prompt:
        test_data_path = test_data_path.replace(".json", "-phi3-prompt.json")
        file_name = file_name.replace(".pth", "-phi3-prompt.pth")
    if lora:
        test_data_path = test_data_path.replace(".json", "-lora.json")
        file_name = file_name.replace(".pth", "-lora.pth")
    if not any([mask_instructions, alpaca52k, phi3_prompt, lora]):
        test_data_path = test_data_path.replace(".json", "-baseline.json")
        file_name = file_name.replace(".pth", "-baseline.pth")

    with open(test_data_path, "w") as file:
        json.dump(test_data, file, indent=4)  # "indent" for pretty-printing
    print(f"Responses saved as {test_data_path}")

    torch.save(model.state_dict(), file_name)
    print(f"Model saved as {file_name}")


if __name__ == "__main__":

    import argparse

    parser = argparse.ArgumentParser(
        description="Instruction finetune a GPT model"
    )
    options = {"baseline", "mask_instructions", "alpaca_52k", "phi3_prompt", "lora"}
    parser.add_argument(
        "--exercise_solution",
        type=str,
        default="last_block",
        help=(
            f"Which experiment to run. Options: {options}."
        )
    )
    args = parser.parse_args()

    if args.exercise_solution == "baseline":
        main()
    elif args.exercise_solution == "mask_instructions":
        main(mask_instructions=True)
    elif args.exercise_solution == "alpaca_52k":
        main(alpaca52k=True)
    elif args.exercise_solution == "phi3_prompt":
        main(phi3_prompt=True)
    elif args.exercise_solution == "lora":
        main(lora=True)
    else:
        raise ValueError(f"{args.exercise_solution} is not a valid --args.exercise_solution option. Options: {options}")