model.py

import torch
import torch.nn as nn
import torch.nn.functional as F
import math
from dataclasses import dataclass

@dataclass
class ModelConfig:
    """Configuration matching SmolLM2-135M"""
    vocab_size: int = 49152
    hidden_size: int = 576
    num_hidden_layers: int = 30
    num_attention_heads: int = 9
    intermediate_size: int = 1536
    max_position_embeddings: int = 2048
    layer_norm_eps: float = 1e-5
    hidden_dropout_prob: float = 0.1
    attention_dropout_prob: float = 0.1
    
    @property
    def head_dim(self):
        return self.hidden_size // self.num_attention_heads


class RotaryEmbedding(nn.Module):
    """Rotary Position Embedding (RoPE)"""
    def __init__(self, dim, max_position_embeddings=2048, base=10000):
        super().__init__()
        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
        self.register_buffer("inv_freq", inv_freq)
        self.max_seq_len_cached = max_position_embeddings
        
        t = torch.arange(self.max_seq_len_cached, dtype=self.inv_freq.dtype)
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        self.register_buffer("cos_cached", emb.cos(), persistent=False)
        self.register_buffer("sin_cached", emb.sin(), persistent=False)
    
    def forward(self, x, seq_len):
        return (
            self.cos_cached[:seq_len, ...],
            self.sin_cached[:seq_len, ...],
        )


def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(q, k, cos, sin):
    """Apply rotary position embedding to query and key tensors."""
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed


class MultiHeadAttention(nn.Module):
    """Multi-head attention with RoPE"""
    def __init__(self, config: ModelConfig):
        super().__init__()
        self.num_heads = config.num_attention_heads
        self.head_dim = config.head_dim
        self.hidden_size = config.hidden_size
        
        self.q_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
        self.k_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
        self.v_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
        self.o_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
        
        self.rotary_emb = RotaryEmbedding(self.head_dim, config.max_position_embeddings)
        self.dropout = nn.Dropout(config.attention_dropout_prob)
    
    def forward(self, hidden_states, attention_mask=None):
        batch_size, seq_len, _ = hidden_states.shape
        
        # Project to Q, K, V
        q = self.q_proj(hidden_states)
        k = self.k_proj(hidden_states)
        v = self.v_proj(hidden_states)
        
        # Reshape for multi-head attention
        q = q.view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
        k = k.view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
        v = v.view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
        
        # Apply rotary embeddings
        cos, sin = self.rotary_emb(v, seq_len)
        q, k = apply_rotary_pos_emb(q, k, cos, sin)
        
        # Attention scores
        attn_weights = torch.matmul(q, k.transpose(-2, -1)) / math.sqrt(self.head_dim)
        
        if attention_mask is not None:
            attn_weights = attn_weights + attention_mask
        
        attn_weights = F.softmax(attn_weights, dim=-1)
        attn_weights = self.dropout(attn_weights)
        
        # Apply attention to values
        attn_output = torch.matmul(attn_weights, v)
        
        # Reshape and project
        attn_output = attn_output.transpose(1, 2).contiguous()
        attn_output = attn_output.view(batch_size, seq_len, self.hidden_size)
        attn_output = self.o_proj(attn_output)
        
        return attn_output


class MLP(nn.Module):
    """Feed-forward network"""
    def __init__(self, config: ModelConfig):
        super().__init__()
        self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
        self.up_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
        self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
    
    def forward(self, x):
        # SwiGLU activation
        gate = F.silu(self.gate_proj(x))
        up = self.up_proj(x)
        return self.dropout(self.down_proj(gate * up))


class TransformerBlock(nn.Module):
    """Single transformer block"""
    def __init__(self, config: ModelConfig):
        super().__init__()
        self.attention = MultiHeadAttention(config)
        self.mlp = MLP(config)
        self.input_layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.post_attention_layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
    
    def forward(self, hidden_states, attention_mask=None):
        # Pre-norm architecture
        residual = hidden_states
        hidden_states = self.input_layernorm(hidden_states)
        hidden_states = self.attention(hidden_states, attention_mask)
        hidden_states = residual + hidden_states
        
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states
        
        return hidden_states


class CustomSmolLM(nn.Module):
    """Custom implementation mimicking SmolLM2-135M"""
    def __init__(self, config: ModelConfig):
        super().__init__()
        self.config = config
        
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
        self.layers = nn.ModuleList([
            TransformerBlock(config) for _ in range(config.num_hidden_layers)
        ])
        self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        
        # Tie weights
        self.lm_head.weight = self.embed_tokens.weight
        
        self.apply(self._init_weights)
    
    def _init_weights(self, module):
        std = 0.02
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
    
    def forward(self, input_ids, attention_mask=None, labels=None):
        batch_size, seq_len = input_ids.shape
        
        # Create causal mask
        if attention_mask is None:
            causal_mask = torch.triu(
                torch.full((seq_len, seq_len), float('-inf'), device=input_ids.device),
                diagonal=1
            )
            causal_mask = causal_mask.unsqueeze(0).unsqueeze(0)
        else:
            causal_mask = None  # Simplified for this example
        
        # Embed tokens
        hidden_states = self.embed_tokens(input_ids)
        
        # Pass through transformer blocks
        for layer in self.layers:
            hidden_states = layer(hidden_states, causal_mask)
        
        hidden_states = self.norm(hidden_states)
        logits = self.lm_head(hidden_states)
        
        loss = None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss = F.cross_entropy(
                shift_logits.view(-1, self.config.vocab_size),
                shift_labels.view(-1)
            )
        
        return {'loss': loss, 'logits': logits}