modeling_modern_llm.py

from typing import Optional

import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PreTrainedModel

from .configuration_modern_llm import ModernLLMConfig


class RMSNorm(nn.Module):
    def __init__(self, dim, eps=1e-6):
        super().__init__()
        self.eps = eps
        self.weight = nn.Parameter(torch.ones(dim))

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


class RotaryEmbedding(nn.Module):
    def __init__(self, dim, max_position_embeddings=2048, base=1000000.0):
        super().__init__()
        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)

    def forward(self, x, seq_len):
        t = torch.arange(seq_len, device=x.device, dtype=self.inv_freq.dtype)
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        return emb.cos(), emb.sin()


def rotate_half(x):
    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):
    cos = cos.unsqueeze(0).unsqueeze(2).to(q.dtype)
    sin = sin.unsqueeze(0).unsqueeze(2).to(q.dtype)
    return (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)


class SwiGLU(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)

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


class GroupedQueryAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.num_heads = config.num_attention_heads
        self.head_dim = config.hidden_size // config.num_attention_heads
        self.num_kv_heads = config.num_key_value_heads
        self.num_kv_groups = self.num_heads // self.num_kv_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_kv_heads * self.head_dim, bias=False)
        self.v_proj = nn.Linear(config.hidden_size, self.num_kv_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, rot_cos, rot_sin):
        batch_size, seq_len, _ = x.shape
        q = self.q_proj(x).view(batch_size, seq_len, self.num_heads, self.head_dim)
        k = self.k_proj(x).view(batch_size, seq_len, self.num_kv_heads, self.head_dim)
        v = self.v_proj(x).view(batch_size, seq_len, self.num_kv_heads, self.head_dim)
        q, k = apply_rotary_pos_emb(q, k, rot_cos, rot_sin)
        k = k.repeat_interleave(self.num_kv_groups, dim=2)
        v = v.repeat_interleave(self.num_kv_groups, dim=2)
        q, k, v = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2)
        out = F.scaled_dot_product_attention(q, k, v, is_causal=True)
        out = out.transpose(1, 2).contiguous().view(batch_size, seq_len, -1)
        return self.o_proj(out)


class TransformerBlock(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.self_attn = GroupedQueryAttention(config)
        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.mlp = SwiGLU(config)

    def forward(self, x, rot_cos, rot_sin):
        x = x + self.self_attn(self.input_layernorm(x), rot_cos, rot_sin)
        x = x + self.mlp(self.post_attention_layernorm(x))
        return x


class ModernLLMForCausalLM(PreTrainedModel):
    config_class = ModernLLMConfig

    def __init__(self, config):
        super().__init__(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 = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        self.rotary_emb = RotaryEmbedding(config.hidden_size // config.num_attention_heads, config.max_position_embeddings, config.rope_theta)
        self.post_init()

    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, labels=None, **kwargs):
        _, seq_len = input_ids.shape
        x = self.embed_tokens(input_ids)
        cos, sin = self.rotary_emb(x, seq_len)
        for layer in self.layers:
            x = layer(x, cos, sin)
        x = self.norm(x)
        logits = self.lm_head(x)
        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}