models/micro_gpt_tf.py

"""
MicroGPT-TF: GPT with a TensorFlow-inspired design pattern, implemented in pure PyTorch.
Uses Pre-LN (LayerNorm before sublayers) and a slightly different MLP structure
reminiscent of TF Transformer implementations.
"""

import math
import torch
import torch.nn as nn
import torch.nn.functional as F


class LayerNorm(nn.Module):
    """Standard LayerNorm (TF-style uses gamma/beta naming)."""
    def __init__(self, dim, eps=1e-6):
        super().__init__()
        self.gamma = nn.Parameter(torch.ones(dim))
        self.beta = nn.Parameter(torch.zeros(dim))
        self.eps = eps

    def forward(self, x):
        return F.layer_norm(x, x.shape[-1:], self.gamma, self.beta, self.eps)


class CausalSelfAttention(nn.Module):
    def __init__(self, n_embd, n_head, block_size, dropout):
        super().__init__()
        assert n_embd % n_head == 0
        self.n_head = n_head
        self.head_dim = n_embd // n_head
        # TF-style: single dense layer for QKV combined, then split
        self.qkv = nn.Linear(n_embd, n_embd * 3, bias=False)
        self.proj = nn.Linear(n_embd, n_embd, bias=False)
        self.attn_drop = nn.Dropout(dropout)
        self.resid_drop = nn.Dropout(dropout)
        self.register_buffer('mask', torch.tril(torch.ones(block_size, block_size)).view(1, 1, block_size, block_size))

    def forward(self, x):
        B, T, C = x.shape
        qkv = self.qkv(x).reshape(B, T, 3, self.n_head, self.head_dim).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]
        att = (q @ k.transpose(-2, -1)) / math.sqrt(self.head_dim)
        att = att.masked_fill(self.mask[:, :, :T, :T] == 0, float('-inf'))
        att = F.softmax(att, dim=-1)
        att = self.attn_drop(att)
        y = att @ v
        y = y.transpose(1, 2).contiguous().view(B, T, C)
        return self.resid_drop(self.proj(y))


class MLP(nn.Module):
    """TF-inspired MLP with bias and ReLU (instead of GELU)."""
    def __init__(self, n_embd, dropout):
        super().__init__()
        self.fc1 = nn.Linear(n_embd, 4 * n_embd, bias=True)
        self.fc2 = nn.Linear(4 * n_embd, n_embd, bias=True)
        self.drop = nn.Dropout(dropout)

    def forward(self, x):
        x = F.relu(self.fc1(x))
        x = self.fc2(x)
        return self.drop(x)


class Block(nn.Module):
    """TF-style Pre-LN transformer block."""
    def __init__(self, n_embd, n_head, block_size, dropout):
        super().__init__()
        self.ln1 = LayerNorm(n_embd)
        self.attn = CausalSelfAttention(n_embd, n_head, block_size, dropout)
        self.ln2 = LayerNorm(n_embd)
        self.mlp = MLP(n_embd, dropout)

    def forward(self, x):
        x = x + self.attn(self.ln1(x))
        x = x + self.mlp(self.ln2(x))
        return x


class MicroGPT_TF(nn.Module):
    """
    GPT with TF-inspired design:
    - Pre-LN (LayerNorm before sublayers)
    - Combined QKV projection
    - ReLU activation instead of GELU
    - Bias in linear layers
    """

    def __init__(self, vocab_size, block_size, n_layer=2, n_head=4, n_embd=128, dropout=0.1):
        super().__init__()
        self.block_size = block_size
        self.wte = nn.Embedding(vocab_size, n_embd)
        self.wpe = nn.Embedding(block_size, n_embd)
        self.blocks = nn.ModuleList([Block(n_embd, n_head, block_size, dropout) for _ in range(n_layer)])
        self.ln_f = LayerNorm(n_embd)
        self.lm_head = nn.Linear(n_embd, vocab_size, bias=True)
        self.apply(self._init_weights)

    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            nn.init.normal_(module.weight, mean=0.0, std=0.02)
            if module.bias is not None:
                nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            nn.init.normal_(module.weight, mean=0.0, std=0.02)

    def forward(self, idx, targets=None):
        B, T = idx.shape
        if T > self.block_size:
            raise ValueError(f'block size exceeded: {T} > {self.block_size}')
        pos = torch.arange(T, device=idx.device)
        x = self.wte(idx) + self.wpe(pos)
        for block in self.blocks:
            x = block(x)
        x = self.ln_f(x)
        logits = self.lm_head(x)
        loss = None
        if targets is not None:
            loss = F.cross_entropy(logits.reshape(-1, logits.size(-1)), targets.reshape(-1))
        return logits, loss

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=40):
        self.eval()
        for _ in range(max_new_tokens):
            idx_cond = idx[:, -self.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