import torch import inspect import torch.nn as nn from dataclasses import dataclass from torch.nn import functional as F # Model Architecture ================================================================================================================ @dataclass class GPTConfig: block_size: int = 1024 # max sequence length vocab_size: int = 50257 # 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 in the multihead attention n_embd: int = 768 # embedding dimension dropout: float = 0.1 # FLASH ATTENTION # Flash attention is a kernel fusion operation of the attention operation. # It was found out manually. It cannot be found by compilers Because it requires an algorithmic rewrite of how attention is implemented. # Though it performs more operations, it is faster than regular attention because it is mindful of the memory hierarchy and has high AI. # It avoids read and write operations. It never materializes the large NxN attention matrix which reduces AI. # It relies on the online softmax trick which incrementally calculates softmax without having to materialize the inputs to the softmax. # This is a combination of attention and multi-head attention. # There are 1024 tokens in a sequence each emitting 3 vectors - Q, K, V. class CausalSelfAttention(nn.Module): def __init__(self, config): super().__init__() assert config.n_embd % config.n_head == 0 # Key, Query and value Projections for all heads, but in a batch self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd) # Output projection self.c_proj = nn.Linear(config.n_embd, config.n_embd) self.c_proj.NANOGPT_SCALE_INIT = 1 # Regularization self.n_head = config.n_head self.n_embd = config.n_embd # not really a 'bias', more of a mask, but following the OpenAI/HF naming though 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): B, T, C = x.size() # (batch_size, sequence_length, 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) # Attention (materializes the large (T,T) matrix for all the queries and keys) # att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1))) # att = att.masked_fill(self.bias[:,:,:T,:T] == 0, float('-inf')) # Only looks at previous tokens # att = F.softmax(att, dim=-1) # y = att @ v # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs) y = F.scaled_dot_product_attention(q, k, v, is_causal=True) 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): def __init__(self, config): super().__init__() self.c_fc = nn.Linear(config.n_embd, 4 * config.n_embd) self.gelu = nn.GELU() self.c_proj = nn.Linear(4 * config.n_embd, config.n_embd) self.c_proj.NANOGPT_SCALE_INIT = 1 self.dropout = nn.Dropout(config.dropout) def forward(self, x): x = self.c_fc(x) x = self.gelu(x) x = self.c_proj(x) x = self.dropout(x) return x # In the GPT-3 paper, the LayerNorm layers are applied # before the linear and attention layers. class Block(nn.Module): def __init__(self, config): super().__init__() 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): x = x + self.attn(self.ln_1(x)) x = x + self.mlp(self.ln_2(x)) return x # A final Layernorm is added before the final linear head. class GPT(nn.Module): def __init__(self, config): super().__init__() self.config = config self.transformer = nn.ModuleDict(dict( wte = nn.Embedding(config.vocab_size, config.n_embd), # Embedding wpe = nn.Embedding(config.block_size, config.n_embd), # Position embeddings h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]), # Transformer blocks ln_f = nn.LayerNorm(config.n_embd), # Final layer norm (GPT3) )) self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False) # Weight Sharing Scheme self.transformer.wte.weight = self.lm_head.weight # init params self.apply(self._init_weights) # Weight Initialization def _init_weights(self, module): if isinstance(module, nn.Linear): std = 0.02 if hasattr(module, 'NANOGPT_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: # Flatten out multidiemntsional input for cross entropy. 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 # The parameters are divided into decay and nondecay params. # It is common to not decay bias and 1 dimensional tensors. def configure_optimizers(self, weight_decay, learning_rate, device, master_process): # 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] # Embeddings and weights in matmul nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2] # 1D tensors like LayerNorms, biases 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) if master_process: 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 'cuda' in device # Fuses the kernels used in the updation of parameters to make it faster if master_process: print(f"Using fused AdamW: {use_fused} \n") optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=(0.9, 0.95), eps=1e-8, fused=use_fused) return optimizer