model.py

# GPT-3 Paper 
import os
import math
import time
import inspect
from dataclasses import dataclass
import torch
import torch.nn as nn
import tiktoken
from torch.nn import functional as F


class CausalSelfAttention(nn.Module):

    def __init__(self, config):
        super().__init__()

        #assertion to ensure the embedding dimension is divisible by the number of heads (important for reshaping later).
        assert config.n_embd % config.n_head == 0

        # key, query, value projections for all heads, but in a batch. Each vector has the same dimension (C) as the input embedding.
        self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd)

        
        # output projection find the meaning?
        self.c_proj = nn.Linear(config.n_embd, config.n_embd)
        self.c_proj.NANGPT_SCALE_INIT = 1



        # regularization
        self.n_head = config.n_head
        self.n_embd = config.n_embd
        self.register_buffer("bias", torch.tril(torch.ones(config.block_size, config.block_size)).view(1, 1, config.block_size, config.block_size))

    def forward(self, x):
        # x is tokenised version of input.txt
        B, T, C = x.size() # batch size, sequence length, embedding dimensionality (n_embd)
        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
        # nh is "number of heads", hs is "head size", and C (number of channels) = nh * hs
        # e.g. in GPT-2 (124M), n_head=12, hs=64, so nh*hs=C=768 channels in the Transformer
        qkv = self.c_attn(x)
        q, k, v = qkv.split(self.n_embd, dim=2)
        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)

        # att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
        # att = att.masked_fill(self.bias[:, :, :T, :T] == 0, float('-inf'))# find what is it???

        # att = F.softmax(att, dim=-1)
        # y = att @ v # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
        


        ## This function combines the dot product, scaling, and softmax operations into a single step.
        y = F.scaled_dot_product_attention(q, k, v, is_causal = True) # Flash attention

        y = y.transpose(1, 2).contiguous().view(B, T, C) # re-assemble all head outputs side by side
        # output projection
        y = self.c_proj(y)
        return y


class MLP(nn.Module):
# MLP (Multi-Layer Perceptron)
## This class implements a simple multi-layer perceptron (MLP) sub-module. 
## It's often used within transformers for non-linear transformations.

    def __init__(self, config):
        #sqeeze and expand
        super().__init__()
        #c_fc: Projects the input (x) to a dimension four times larger than the embedding dimension (n_embd).
        self.c_fc    = nn.Linear(config.n_embd, 4 * config.n_embd)

        # GELU (Gaussian Error Linear Unit) activation function for non-linearity. 
        #Here, an approximate version using tanh is used.
        self.gelu    = nn.GELU(approximate='tanh')

        # Projects the output back to the original embedding dimension (n_embd).
        self.c_proj  = nn.Linear(4 * config.n_embd, config.n_embd)
        self.c_proj.NANOGPT_SCALE_INIT = 1

    def forward(self, x):

      #Takes the input (x).
      # Applies the linear layer (c_fc), followed by the GELU activation.
      # Applies the final projection layer (c_proj).
      # Returns the transformed output.
        x = self.c_fc(x)
        x = self.gelu(x)
        x = self.c_proj(x)
        return x

class Block(nn.Module):
  # This class combines the CausalSelfAttention layer (explained previously) and the MLP layer to form a single transformer block. 
  # The input is processed through the attention layer, followed by layer normalization and an MLP, and 
  # then again with layer normalization.

    def __init__(self, config):
        super().__init__()
        
        #ln_1: A layer normalization layer applied before the causal self-attention.
        #attn: An instance of the CausalSelfAttention class (explained previously).
        #mlp: An instance of the MLP class (explained previously).
        
        self.ln_1 = nn.LayerNorm(config.n_embd)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = nn.LayerNorm(config.n_embd)
        self.mlp = MLP(config)

    def forward(self, x):
      # Takes the input (x).
      # Performs a residual connection with the output from the causal self-attention layer (attn), preceded by layer normalization (ln_1).
      # Performs another residual connection with the output from the MLP layer (mlp), preceded by layer normalization (ln_2).
      # Returns the final output after the second residual connection.
        x = x + self.attn(self.ln_1(x))
        x = x + self.mlp(self.ln_2(x))
        return x


@dataclass
class GPTConfig:
    block_size: int = 1024 # max sequence length
    vocab_size: int = 50304 # number of tokens: 50,000 BPE merges + 256 bytes tokens + 1 <|endoftext|> token
    n_layer: int = 12 # number of layers
    n_head: int = 12 # number of heads
    n_embd: int = 768 # embedding dimension


class GPT(nn.Module):

    def __init__(self, config):
        super().__init__()
        self.config = config



        # Creates a transformer module dictionary containing several key components:
        #wte: Word token embedding layer (nn.Embedding). Maps each word index to its corresponding embedding vector.
        #wpe: Positional embedding layer (nn.Embedding). Adds positional information to the word embeddings.
        #h: A module list containing multiple Block instances (explained earlier). These are the core processing units of the transformer.
        #ln_f: Final layer normalization layer (nn.LayerNorm) applied to the output of the transformer blocks.

        self.transformer = nn.ModuleDict(dict(
            wte = nn.Embedding(config.vocab_size, config.n_embd),
            wpe = nn.Embedding(config.block_size, config.n_embd),
            h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
            ln_f = nn.LayerNorm(config.n_embd),
        ))


        #Creates the language modeling head (lm_head), a linear layer that projects the final hidden state from the 
        #transformer to the vocabulary size, predicting the next word in the sequence.
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)

        # weight sharing Implements weight sharing between the word token embedding layer (wte) 
        #and the language modeling head (lm_head). This reduces the number of parameters and encourages 
        #the model to learn a meaningful representation for words that can be used for both embedding and prediction.
        self.transformer.wte.weight = self.lm_head.weight

        # weight initialization
        #Initializes the weights of the model using a custom function (_init_weights).
        self.apply(self._init_weights)

    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            std = 0.02
            if hasattr(module, 'NANGPT_SCALE_INIT'):
                std *= (2 * self.config.n_layer) ** -0.5
            torch.nn.init.normal_(module.weight, mean = 0.0, std = std)
            if module.bias is not None:
                torch.nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            torch.nn.init.normal_(module.weight, mean=0.0, std = 0.02)



    def forward(self, idx, targets=None):
        # idx is of shape (B, T)
        B, T = idx.size()
        assert T <= self.config.block_size, f"Cannot forward sequence of length {T}, block size is only {self.config.block_size}"
        # forward the token and posisition embeddings
        pos = torch.arange(0, T, dtype=torch.long, device=idx.device) # shape (T)
        pos_emb = self.transformer.wpe(pos) # position embeddings of shape (T, n_embd)
        tok_emb = self.transformer.wte(idx) # token embeddings of shape (B, T, n_embd)
        x = tok_emb + pos_emb
        # forward the blocks of the transformer
        for block in self.transformer.h:
            x = block(x)
        # forward the final layernorm and the classifier
        x = self.transformer.ln_f(x)
        logits = self.lm_head(x) # (B, T, vocab_size)
        loss = None
        if targets is not None:
            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1))
        return logits, loss

    @classmethod
    def from_pretrained(cls, model_type):
        """Loads pretrained GPT-2 model weights from huggingface"""
        assert model_type in {'gpt2', 'gpt2-medium', 'gpt2-large', 'gpt2-xl'}
        from transformers import GPT2LMHeadModel
        print("loading weights from pretrained gpt: %s" % model_type)

        # n_layer, n_head and n_embd are determined from model_type
        config_args = {
            'gpt2':         dict(n_layer=12, n_head=12, n_embd=768),  # 124M params
            'gpt2-medium':  dict(n_layer=24, n_head=16, n_embd=1024), # 350M params
            'gpt2-large':   dict(n_layer=36, n_head=20, n_embd=1280), # 774M params
            'gpt2-xl':      dict(n_layer=48, n_head=25, n_embd=1600), # 1558M params
        }[model_type]
        config_args['vocab_size'] = 50257 # always 50257 for GPT model checkpoints
        config_args['block_size'] = 1024 # always 1024 for GPT model checkpoints
        # create a from-scratch initialized minGPT model
        config = GPTConfig(**config_args)
        model = GPT(config)
        sd = model.state_dict()
        sd_keys = sd.keys()
        sd_keys = [k for k in sd_keys if not k.endswith('.attn.bias')] # discard this mask / buffer, not a param

        # init a huggingface/transformers model
        model_hf = GPT2LMHeadModel.from_pretrained(model_type)
        sd_hf = model_hf.state_dict()

        # copy while ensuring all of the parameters are aligned and match in names and shapes
        sd_keys_hf = sd_hf.keys()
        sd_keys_hf = [k for k in sd_keys_hf if not k.endswith('.attn.masked_bias')] # ignore these, just a buffer
        sd_keys_hf = [k for k in sd_keys_hf if not k.endswith('.attn.bias')] # same, just the mask (buffer)
        transposed = ['attn.c_attn.weight', 'attn.c_proj.weight', 'mlp.c_fc.weight', 'mlp.c_proj.weight']
        # basically the openai checkpoints use a "Conv1D" module, but we only want to use a vanilla Linear
        # this means that we have to transpose these weights when we import them
        assert len(sd_keys_hf) == len(sd_keys), f"mismatched keys: {len(sd_keys_hf)} != {len(sd_keys)}"
        for k in sd_keys_hf:
            if any(k.endswith(w) for w in transposed):
                # special treatment for the Conv1D weights we need to transpose
                assert sd_hf[k].shape[::-1] == sd[k].shape
                with torch.no_grad():
                    sd[k].copy_(sd_hf[k].t())
            else:
                # vanilla copy over the other parameters
                assert sd_hf[k].shape == sd[k].shape
                with torch.no_grad():
                    sd[k].copy_(sd_hf[k])

        return model

    def configure_optimizers(self, weight_decay, learning_rate, device_type):
        # start with all of the candidate parameters (that require grad)
        param_dict = {pn: p for pn, p in self.named_parameters()}
        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
        # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
        # i.e. all weight tensors in matmuls + embeddings decay, all biases and layernorms don't.
        decay_params = [p for n, p in param_dict.items() if p.dim() >= 2]
        nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2]
        optim_groups = [
            {'params': decay_params, 'weight_decay': weight_decay},
            {'params': nodecay_params, 'weight_decay': 0.0}
        ]
        num_decay_params = sum(p.numel() for p in decay_params)
        num_nodecay_params = sum(p.numel() for p in nodecay_params)
        
        print(f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters")
        print(f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters")
        # Create AdamW optimizer and use the fused version if it is available
        fused_available = 'fused' in inspect.signature(torch.optim.AdamW).parameters
        use_fused = fused_available and device_type == "cuda"
        
        print(f"using fused AdamW: {use_fused}")
        optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=(0.9, 0.95), eps=1e-8, fused=use_fused)
        return optimizer

# model = GPT.from_pretrained('gpt2')

device = 'cpu'
if torch.cuda.is_available():
    device = 'cuda'
elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
    device = "mps"
print(f"using device: {device}")

# SEED
torch.manual_seed(1337)
if torch.cuda.is_available():
    torch.cuda.manual_seed(1337)

# STOP
# num_return_sequences = 5
# max_length = 30



import tiktoken

class DataLoaderLite:
    def __init__(self, B, T):
        self.B = B
        self.T = T

        # at init load tokens from disk and store them in memory
        with open('input.txt', 'r') as f:
            text = f.read()
        enc = tiktoken.get_encoding('gpt2') 
        tokens = enc.encode(text)
        self.tokens = torch.tensor(tokens)
        print(f'loaded {len(self.tokens)} tokens')
        print(f'1 epoch = {len(self.tokens) // (B * T)} batches')

        # state
        self.current_position = 0
    
    def next_batch(self):
        B, T = self.B, self.T
        buf = self.tokens[self.current_position: self.current_position + B * T + 1]
        x = (buf[:-1]).view(B, T) # inputs
        y = (buf[1:]).view(B, T) # targets
        # advance the position in the tensor
        self.current_position += B*T
        # if loading the next batch would be out of bounds, reset
        if self.current_position + (B * T + 1) > len(self.tokens):
            self.current_position = 0
        return x, y

# CHANGES IN CURRENT CODE
torch.set_float32_matmul_precision('high')
model = GPT(GPTConfig())
model.to(device)
# model = torch.compile(model)

# CODE UPDATE HERE
max_lr = 6e-4 
min_lr = max_lr * 0.1
# warmup_steps = 100
# # max_steps = 50

def get_lr(it,warmup_steps, max_steps):
    if it < warmup_steps:
        return max_lr * (it + 1) / warmup_steps
    if it > max_steps:
        return min_lr
    decay_ratio = (it - warmup_steps) / (max_steps - warmup_steps)
    assert 0 <= decay_ratio <=1
    coeff = 0.5 * (1.0 + math.cos(math.pi * decay_ratio))
    return min_lr + coeff * (max_lr - min_lr)


# NEW CODE
import time
train_loader = DataLoaderLite(B = 8, T = 512)

# train_loader = DataLoaderLite(B = B, T = T)
x, y = train_loader.next_batch()
x.shape, y.shape

def run_train (max_steps = 50 ,warmup_steps = 100, PATH = "/content/drive/MyDrive/S21/gpt_124M.pth" ):
  # optimizer = torch.optim.AdamW(model.parameters(), lr = 3e-4, betas=(0.9, 0.95), eps=1e-8)
  optimizer = model.configure_optimizers(weight_decay=0.1, learning_rate=6e-4, device_type=device)
  for step in range(max_steps):
      t0 = time.time()
      x, y = train_loader.next_batch()
      x, y = x.to(device), y.to(device)
      optimizer.zero_grad()
      # NEW CODE ADDED HERE
      with torch.autocast(device_type=device, dtype=torch.bfloat16):
          logits, loss = model(x, y) 
      loss.backward()
      norm = torch.nn.utils.clip_grad_norm(model.parameters(), 1.0)
      # NEW CODE
      lr = get_lr(step, max_steps = 50 ,warmup_steps = 100)
      for param_group in optimizer.param_groups:
          param_group['lr'] = lr
          
      optimizer.step()
      torch.cuda.synchronize() 
      t1 = time.time()
      dt = (t1 - t0) * 1000
      tokens_per_sec = (train_loader.B * train_loader.T) / (t1 - t0)
      print(f'step{step} | loss: {loss.item()} | dt: {dt:.2f}ms | tok/sec: {tokens_per_sec: .2f} | norm: {norm:.2f}')
  print(loss)
  torch.save(model.state_dict(), PATH)
  return model

def load_fromsaved(PATH = "/content/drive/MyDrive/S21/gpt_124M.pth" ):

  # Create a new GPT model instance
  model = GPT(GPTConfig())
  model.to(device)

  # Load the saved weights into the model
  model.load_state_dict(torch.load(PATH))
  

  # Print confirmation message
  print("Loaded model weights from:", PATH)
  model.to(device)

  return model 


def gen_text(model,start_tokens, max_length=100, num_return_sequences=10 ):
  """
  Generates text using the loaded GPT model.

  Args:
      model: The GPT model to use for generation.
      start_tokens (optional): A list of token IDs to use as the starting prompt.
      max_length: The maximum length of the generated text.
      num_return_sequences: The number of text sequences to generate.

  Returns:
      None
  """
  decoded_texts = ''
  enc = tiktoken.get_encoding('gpt2') 
  tokens = enc.encode(start_tokens)
  tokens = torch.tensor(tokens, dtype= torch.long) # (8,) #check tiktoken app
  tokens = tokens.unsqueeze(0).repeat(num_return_sequences, 1) # (5, 8)
  x = tokens.to(device)

  # Set random seeds for consistent generation across runs
  torch.manual_seed(42)
  torch.cuda.manual_seed(42)
  generated_text = ""
  while x.size(1) < max_length:
        # forward the model to get the logits
        with torch.no_grad():
            logits = model(x)[0] # (B, T, vocab_size)
            # take the logits at the last position
            logits = logits[:, -1, :] # (B, vocab_size)
            # get the probabilities
            probs = F.softmax(logits, dim=-1)
            # do top-k sampling of 50 (huggingface pipeline default)
            # topk_probs here becomes (5, 50), topk_indices is (5, 50)
            topk_probs, topk_indices = torch.topk(probs, 50, dim=-1)
            # select a token from the top-k probabilities
            # note: multinomial does not demand the input to sum to 1
            ix = torch.multinomial(topk_probs, 1) # (B, 1)
            # gather the corresponding indices
            xcol = torch.gather(topk_indices, -1, ix) # (B, 1)
            # append to the sequence
            x = torch.cat((x, xcol), dim=1)

  # print the generated text
  for i in range(num_return_sequences):
        tokens = x[i, :max_length].tolist()
        decoded = enc.decode(tokens)
        print(">", decoded)
        generated_text += decoded
  return generated_text