Upload Nano-GPT 42M (99,000 iterations, val_loss=3.0801)

LICENSE

@@ -0,0 +1 @@
+LICENSE placeholder.

README.md

@@ -0,0 +1,3 @@
+# Model Files Only
+
+README content will be added manually.

best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93e8051f16d9e53421d857f5d8939eec72f9d014f21f8ecb6cbdbdfc07f43265
+size 510910761

config.json

@@ -0,0 +1,9 @@
+{
+  "vocab_size": 32000,
+  "n_layers": 8,
+  "n_heads": 8,
+  "n_embd": 512,
+  "block_size": 512,
+  "dropout": 0.1,
+  "bias": true
+}

config.py

@@ -0,0 +1,73 @@
+"""
+Nano-GPT Configuration  
+L4-SAFE: Reduced memory usage
+"""
+
+import torch
+from dataclasses import dataclass
+
+@dataclass
+class NanoGPTConfig:
+    # Model Architecture
+    vocab_size: int = 32000
+    n_layers: int = 8          # REDUCED from 12
+    n_heads: int = 8           # REDUCED from 12
+    n_embd: int = 512          # REDUCED from 768
+    block_size: int = 512      # REDUCED from 1024 (KEY!)
+    dropout: float = 0.1
+    bias: bool = True
+    
+    # Training Hyperparameters
+    batch_size: int = 16       # REDUCED from 32 (KEY!)
+    gradient_accumulation_steps: int = 8  # INCREASED from 4
+    learning_rate: float = 3e-4
+    max_iters: int = 100000
+    weight_decay: float = 0.1
+    beta1: float = 0.9
+    beta2: float = 0.95
+    grad_clip: float = 1.0
+    
+    # Learning Rate Scheduling
+    decay_lr: bool = True
+    warmup_iters: int = 2000
+    lr_decay_iters: int = 100000
+    min_lr: float = 3e-5
+    
+    # Evaluation & Logging
+    eval_interval: int = 1000
+    eval_iters: int = 100      # REDUCED from 200
+    log_interval: int = 100
+    
+    # Checkpointing
+    save_interval: int = 5000
+    checkpoint_dir: str = "checkpoints"
+    
+    # Data
+    dataset_mix: dict = None
+    
+    # Hardware
+    device: str = 'cuda' if torch.cuda.is_available() else 'cpu'
+    dtype: str = 'bfloat16'
+    compile: bool = False      # DISABLED torch.compile (uses more memory!)
+    
+    # Reproducibility
+    seed: int = 42
+    
+    def __post_init__(self):
+        if self.dataset_mix is None:
+            self.dataset_mix = {
+                'fineweb': 1.0
+            }
+    
+    @property
+    def n_params(self):
+        return (2 * self.vocab_size * self.n_embd + 
+                12 * self.n_layers * self.n_embd * self.n_embd) / 1e6
+
+config = NanoGPTConfig()
+
+if __name__ == "__main__":
+    print(f"Model size: ~{config.n_params:.1f}M parameters")
+    print(f"Sequence length: {config.block_size}")
+    print(f"Batch size: {config.batch_size}")
+    print(f"Effective batch: {config.batch_size * config.gradient_accumulation_steps}")

model.py

@@ -0,0 +1,171 @@
+"""
+Nano-GPT: GPT-2 style decoder-only transformer
+From scratch implementation
+"""
+
+import math
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from config import config
+
+class CausalSelfAttention(nn.Module):
+    """Multi-head causal self-attention"""
+    
+    def __init__(self, config):
+        super().__init__()
+        assert config.n_embd % config.n_heads == 0
+        
+        # Key, Query, Value for all heads
+        self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd, bias=config.bias)
+        # Output projection
+        self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
+        
+        # Regularization
+        self.attn_dropout = nn.Dropout(config.dropout)
+        self.resid_dropout = nn.Dropout(config.dropout)
+        
+        self.n_heads = config.n_heads
+        self.n_embd = config.n_embd
+        self.dropout = config.dropout
+        
+        # Causal mask
+        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, sequence, embedding
+        
+        # Calculate Q, K, V
+        q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
+        k = k.view(B, T, self.n_heads, C // self.n_heads).transpose(1, 2)
+        q = q.view(B, T, self.n_heads, C // self.n_heads).transpose(1, 2)
+        v = v.view(B, T, self.n_heads, C // self.n_heads).transpose(1, 2)
+        
+        # Attention
+        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)
+        att = self.attn_dropout(att)
+        y = att @ v
+        
+        # Reassemble heads
+        y = y.transpose(1, 2).contiguous().view(B, T, C)
+        
+        # Output projection
+        y = self.resid_dropout(self.c_proj(y))
+        return y
+
+class MLP(nn.Module):
+    """Feed-forward network"""
+    
+    def __init__(self, config):
+        super().__init__()
+        self.c_fc = nn.Linear(config.n_embd, 4 * config.n_embd, bias=config.bias)
+        self.gelu = nn.GELU()
+        self.c_proj = nn.Linear(4 * config.n_embd, config.n_embd, bias=config.bias)
+        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
+
+class Block(nn.Module):
+    """Transformer block"""
+    
+    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
+
+class NanoGPT(nn.Module):
+    """Nano-GPT Model"""
+    
+    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),
+            drop = nn.Dropout(config.dropout),
+            h = nn.ModuleList([Block(config) for _ in range(config.n_layers)]),
+            ln_f = nn.LayerNorm(config.n_embd),
+        ))
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
+        
+        # Weight tying
+        self.transformer.wte.weight = self.lm_head.weight
+        
+        # Initialize weights
+        self.apply(self._init_weights)
+        for pn, p in self.named_parameters():
+            if pn.endswith('c_proj.weight'):
+                torch.nn.init.normal_(p, mean=0.0, std=0.02/math.sqrt(2 * config.n_layers))
+        
+        print(f"Number of parameters: {self.get_num_params()/1e6:.2f}M")
+    
+    def get_num_params(self):
+        return sum(p.numel() for p in self.parameters())
+    
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
+            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):
+        device = idx.device
+        b, t = idx.size()
+        assert t <= self.config.block_size
+        
+        # Embeddings
+        pos = torch.arange(0, t, dtype=torch.long, device=device)
+        tok_emb = self.transformer.wte(idx)
+        pos_emb = self.transformer.wpe(pos)
+        x = self.transformer.drop(tok_emb + pos_emb)
+        
+        # Transformer blocks
+        for block in self.transformer.h:
+            x = block(x)
+        x = self.transformer.ln_f(x)
+        
+        # Language model head
+        if targets is not None:
+            logits = self.lm_head(x)
+            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-1)
+        else:
+            logits = self.lm_head(x[:, [-1], :])
+            loss = None
+        
+        return logits, loss
+    
+    @torch.no_grad()
+    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=None):
+        """Generate text"""
+        for _ in range(max_new_tokens):
+            idx_cond = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:]
+            logits, _ = self(idx_cond)
+            logits = logits[:, -1, :] / temperature
+            
+            if top_k is not None:
+                v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
+                logits[logits < v[:, [-1]]] = -float('Inf')
+            
+            probs = F.softmax(logits, dim=-1)
+            idx_next = torch.multinomial(probs, num_samples=1)
+            idx = torch.cat((idx, idx_next), dim=1)
+        
+        return idx