Upload gpt2_train_per.py

scripts/gpt2_train_per.py

@@ -1 +1,366 @@
-# This file is intentionally left blank.
+# This file is intentionally left blank.# Solving for residual std scaling issue
+import os
+import math
+import time
+import inspect
+from dataclasses import dataclass
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from torch.nn.utils import clip_grad_norm_
+from torch.utils.checkpoint import checkpoint # Moved this import to the top
+
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:64"
+
+class CausalSelfAttention(nn.Module):
+
+    def __init__(self, config):
+        super().__init__()
+        assert config.n_embd % config.n_head == 0
+        # key, query, 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.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):
+        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'))
+        att = F.softmax(att, dim=-1)
+        y = att @ v # (B, nh, T, T) x (B, nh, T, hs)
+
+        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(approximate='tanh')
+        self.c_proj  = nn.Linear(4 * config.n_embd, config.n_embd)
+        self.c_proj.NANOGPT_SCALE_INIT = 1
+
+    def forward(self, x):
+        x = self.c_fc(x)
+        x = self.gelu(x)
+        x = self.c_proj(x)
+        return x
+
+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)
+
+    # In forward of Block:
+    def forward(self, x):
+        def _forward_block(x):
+            x = x + self.attn(self.ln_1(x))
+            x = x + self.mlp(self.ln_2(x))
+            return x
+        return checkpoint(_forward_block, x)
+
+
+@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 = 6 # number of layers (reduced from 12)
+    n_head: int = 6 # number of heads (reduced from 12)
+    n_embd: int = 384 # embedding dimension (reduced from 768)
+
+
+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),
+            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),
+        ))
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
+
+        # weight sharing
+        self.transformer.wte.weight = self.lm_head.weight
+
+        # weight initialization
+        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
+
+    # Device setup same as before
+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 for reproducibility
+torch.manual_seed(42)
+if torch.cuda.is_available():
+    torch.cuda.manual_seed(42)
+
+# Hyperparameters
+B, T = 8,128  # batch size and sequence length (8192 tokens per batch)
+max_iters = 2000
+warmup_iters = 200
+base_lr = 3e-4
+final_lr = 1e-5
+grad_clip = 1.0
+patience = 20  # early stopping patience
+num_val_batches = 10
+accum_steps = 4  # effectively batch 32 by accumulation
+
+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
+# Load full tokens
+with open('input.txt', 'r') as f:
+    text = f.read()
+enc = tiktoken.get_encoding('gpt2')
+tokens = torch.tensor(enc.encode(text))
+
+# Simple 90/10 train-val split to avoid data leakage
+num_train_tokens = int(0.9 * len(tokens))
+train_tokens = tokens[:num_train_tokens]
+val_tokens = tokens[num_train_tokens:]
+
+# Create data loaders pointing to split tokens
+train_loader = DataLoaderLite(B, T)
+train_loader.tokens = train_tokens
+train_loader.current_position = 0
+
+val_loader = DataLoaderLite(B, T)
+val_loader.tokens = val_tokens
+val_loader.current_position = 0
+
+# Clear CUDA cache before model initialization
+if torch.cuda.is_available():
+    torch.cuda.empty_cache()
+
+# Initialize model
+model = GPT(GPTConfig())
+model.to(device)
+
+# Optimizer
+optimizer = torch.optim.AdamW(model.parameters(), lr=base_lr)
+
+# Learning rate schedule: linear warmup + cosine decay
+def get_lr(step):
+    if step < warmup_iters:
+        return base_lr * step / warmup_iters
+    progress = (step - warmup_iters) / (max_iters - warmup_iters)
+    return final_lr + 0.5 * (base_lr - final_lr) * (1 + math.cos(math.pi * progress))
+
+best_val_loss = float('inf')
+no_improve_steps = 0
+
+model.train()
+
+from torch.amp import GradScaler, autocast
+scaler = GradScaler('cuda')
+
+for step in range(max_iters):
+    optimizer.zero_grad()
+    for _ in range(accum_steps):
+        x, y = train_loader.next_batch()
+        x, y = x.to(device), y.to(device)
+
+        with autocast('cuda'):
+            logits, loss = model(x, y)
+            loss = loss / accum_steps  # scale loss
+
+        scaler.scale(loss).backward()
+
+    # Gradient clipping and optimizer step
+    scaler.unscale_(optimizer)
+    clip_grad_norm_(model.parameters(), grad_clip)
+    scaler.step(optimizer)
+    scaler.update()
+
+    torch.cuda.empty_cache()
+
+  # Validation and logs every N steps (adjust for accum steps)
+    if step % (100 // accum_steps) == 0 or step == max_iters - 1:
+        model.eval()
+        val_losses = []
+        with torch.no_grad():
+            for _ in range(num_val_batches):
+                xv, yv = val_loader.next_batch()
+                xv, yv = xv.to(device), yv.to(device)
+                _, val_loss = model(xv, yv)
+                val_losses.append(val_loss.item())
+        avg_val_loss = sum(val_losses) / len(val_losses)
+
+        lr = get_lr(step) # Assign the learning rate
+        print(f"Step {step}: train loss {loss.item():.5f}, val loss {avg_val_loss:.5f}, lr {lr:.6f}")
+
+        # Early stopping and checkpoint saving
+        if avg_val_loss < best_val_loss:
+            best_val_loss = avg_val_loss
+            no_improve_steps = 0
+            torch.save(model.state_dict(), 'best_model.pt')
+            print("Checkpoint saved.")
+        else:
+            no_improve_steps += 1
+            if no_improve_steps >= patience:
+                print("Early stopping triggered.")
+                break
+        model.train()
+
+# Load best model for sampling/generation
+model.load_state_dict(torch.load('best_model.pt'))
+model.eval()
+
+# Sampling/generation code (unchanged from original)
+num_return_sequences = 5
+max_length = 30
+x = val_loader.next_batch()[0][:num_return_sequences].to(device)  # start from some validation tokens
+
+while x.size(1) < max_length:
+    with torch.no_grad():
+        logits = model(x)[0]
+        logits = logits[:, -1, :]
+        probs = F.softmax(logits, dim=-1)
+        topk_probs, topk_indices = torch.topk(probs, 50, dim=-1)
+        ix = torch.multinomial(topk_probs, 1)
+        xcol = torch.gather(topk_indices, -1, ix)
+        x = torch.cat((x, xcol), dim=1)
+
+for i in range(num_return_sequences):
+    tokens = x[i, :max_length].tolist()
+    decoded = enc.decode(tokens)
+    print(">", decoded)