model.py

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

class RMSNorm(nn.Module):
    def __init__(self, dim, eps=1e-5):
        super().__init__()
        self.eps = eps
        self.weight = nn.Parameter(torch.ones(dim))

    def forward(self, x):
        mean_square = (x.pow(2).mean(-1, keepdim=True))
        x = x * torch.rsqrt(mean_square + self.eps)
        return self.weight * x

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):
    # q, k: [bsz, heads, seq_len, head_dim]
    # cos, sin: [seq_len, head_dim] -> unsqueeze to [1, 1, seq_len, head_dim]
    cos = cos.unsqueeze(0).unsqueeze(0)
    sin = sin.unsqueeze(0).unsqueeze(0)

    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed

class MLP(nn.Module):
    def __init__(self, config):
        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.act_fn = nn.SiLU()

    def forward(self, x):
        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))

class Attention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = config.hidden_size // config.num_attention_heads
        self.num_key_value_heads = config.num_key_value_heads
        self.num_key_value_groups = self.num_heads // self.num_key_value_heads

        self.q_proj = nn.Linear(config.hidden_size, self.num_heads * self.head_dim, bias=False)
        self.k_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
        self.v_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
        self.o_proj = nn.Linear(self.num_heads * self.head_dim, config.hidden_size, bias=False)

    def forward(self, x, cos, sin, mask=None):
        bsz, seq_len, _ = x.shape
        q = self.q_proj(x).view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
        k = self.k_proj(x).view(bsz, seq_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
        v = self.v_proj(x).view(bsz, seq_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)

        q, k = apply_rotary_pos_emb(q, k, cos, sin)

        k = k.repeat_interleave(self.num_key_value_groups, dim=1)
        v = v.repeat_interleave(self.num_key_value_groups, dim=1)

        attn_weights = torch.matmul(q, k.transpose(2, 3)) / math.sqrt(self.head_dim)
        if mask is not None:
            attn_weights = attn_weights + mask

        attn_weights = F.softmax(attn_weights, dim=-1)
        output = torch.matmul(attn_weights, v)
        output = output.transpose(1, 2).contiguous().view(bsz, seq_len, -1)
        return self.o_proj(output)

class Block(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.self_attn = Attention(config)
        self.mlp = MLP(config)
        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(self, x, cos, sin, mask=None):
        h = x + self.self_attn(self.input_layernorm(x), cos, sin, mask)
        out = h + self.mlp(self.post_attention_layernorm(h))
        return out

class SmolLM2(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
        self.layers = nn.ModuleList([Block(config) for _ in range(config.num_hidden_layers)])
        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        # RoPE setup
        self.head_dim = config.hidden_size // config.num_attention_heads
        self.rope_theta = getattr(config, "rope_theta", 10000.0)
        self.inv_freq = 1.0 / (self.rope_theta ** (torch.arange(0, self.head_dim, 2).float() / self.head_dim))
        self.max_pos = config.max_position_embeddings * 2
        self._set_cos_sin_cache(self.max_pos)

    def _set_cos_sin_cache(self, seq_len):
        t = torch.arange(seq_len, dtype=torch.float32)
        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, input_ids):
        bsz, seq_len = input_ids.shape
        x = self.embed_tokens(input_ids)

        if self.cos_cached.device != x.device or self.cos_cached.shape[0] < seq_len:
            self.inv_freq = self.inv_freq.to(x.device)
            self._set_cos_sin_cache(max(seq_len, 2048))

        cos = self.cos_cached[:seq_len].to(dtype=x.dtype)
        sin = self.sin_cached[:seq_len].to(dtype=x.dtype)

        mask = None
        if seq_len > 1:
            mask = torch.full((seq_len, seq_len), float("-inf"), device=input_ids.device)
            mask = torch.triu(mask, diagonal=1)

        for layer in self.layers:
            x = layer(x, cos, sin, mask)

        x = self.norm(x)
        logits = self.lm_head(x)
        return logits