modeling_base_loop.py

import math
from dataclasses import dataclass
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PreTrainedModel, PretrainedConfig
from transformers.modeling_outputs import CausalLMOutputWithCrossAttentions

class CausalSelfAttention(nn.Module):

    def __init__(self, config):
        super().__init__()
        assert config.n_embd % config.n_head == 0
        # key, query, value projections for all heads, but in a batch
        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_head = config.n_head
        self.n_embd = config.n_embd
        self.dropout = config.dropout
        # flash attention make GPU go brrrrr but support is only in PyTorch >= 2.0
        self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')
        if not self.flash:
            print("WARNING: using slow attention. Flash Attention requires PyTorch >= 2.0")
            # causal mask to ensure that attention is only applied to the left in the input sequence
            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 size, sequence length, embedding dimensionality (n_embd)

        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
        q, k, v  = self.c_attn(x).split(self.n_embd, dim=2)
        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)

        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
        if self.flash:
            # efficient attention using Flash Attention CUDA kernels
            y = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=self.dropout if self.training else 0, is_causal=True)
        else:
            # manual implementation of 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 # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
        y = y.transpose(1, 2).contiguous().view(B, T, C) # re-assemble all head outputs side by side

        # output projection
        y = self.resid_dropout(self.c_proj(y))
        return y

class MLP(nn.Module):

    def __init__(self, config):
        super().__init__()
        self.c_fc    = nn.Linear(config.n_embd, config.intermediate_dim, bias=config.bias)
        self.gelu    = nn.GELU()
        self.c_proj  = nn.Linear(config.intermediate_dim, 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):

    def __init__(self, config):
        super().__init__()
        self.norm_1 = nn.RMSNorm(config.n_embd)
        self.attn = CausalSelfAttention(config)
        self.norm_2 = nn.RMSNorm(config.n_embd)
        self.mlp = MLP(config)

    def forward(self, x):
        x = x + self.attn(self.norm_1(x))
        x = x + self.mlp(self.norm_2(x))
        return x

class SharedBlock(nn.Module):
    def __init__(self, depth, config):
        super().__init__()
        self.blocks = nn.ModuleList([
            Block(config) for _ in range(depth)
        ])

    def forward(self, x):
        for block in self.blocks:
            x = block(x)
        return x

@dataclass
class GPTConfig(PretrainedConfig):
    model_type: str = 'base_loop'
    block_size: int = 1024
    vocab_size: int = 50304 # GPT-2 vocab_size of 50257, padded up to nearest multiple of 64 for efficiency
    n_layer: int = 3
    n_head: int = 32
    n_embd: int = 2048
    dropout: float = 0.0
    bias: bool = False # True: bias in Linears and LayerNorms, like GPT-2. False: a bit better and faster
    intermediate_dim: int = 5120

    def __init__(self, **kwargs):
        super().__init__(**kwargs)

class GPT(nn.Module):

    def __init__(self, config):
        super().__init__()
        assert config.vocab_size is not None
        assert config.block_size is not None
        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 = SharedBlock(config.n_layer, config),
            norm_f = nn.RMSNorm(config.n_embd),
        ))
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
        # with weight tying when using torch.compile() some warnings get generated:
        # "UserWarning: functional_call was passed multiple values for tied weights.
        # This behavior is deprecated and will be an error in future versions"
        # not 100% sure what this is, so far seems to be harmless. TODO investigate
        self.transformer.wte.weight = self.lm_head.weight # https://paperswithcode.com/method/weight-tying

        # init all weights
        self.apply(self._init_weights)
        # apply special scaled init to the residual projections, per GPT-2 paper
        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_layer))

        # report number of parameters
        print("number of parameters: %.2fM" % (self.get_num_params()/1e6,))

    def get_num_params(self, non_embedding=True):
        """
        Return the number of parameters in the model.
        For non-embedding count (default), the position embeddings get subtracted.
        The token embeddings would too, except due to the parameter sharing these
        params are actually used as weights in the final layer, so we include them.
        """
        n_params = sum(p.numel() for p in self.parameters())
        if non_embedding:
            n_params -= self.transformer.wpe.weight.numel()
        return n_params

    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, steps=8, **kwargs):
        device = idx.device
        b, t = idx.size()
        assert t <= self.config.block_size, f"Cannot forward sequence of length {t}, block size is only {self.config.block_size}"
        pos = torch.arange(0, t, dtype=torch.long, device=device) # shape (t)

        # forward the GPT model itself
        tok_emb = self.transformer.wte(idx) # token embeddings of shape (b, t, n_embd)
        pos_emb = self.transformer.wpe(pos) # position embeddings of shape (t, n_embd)
        x = self.transformer.drop(tok_emb + pos_emb)
        for _ in range(steps):
            x = self.transformer.h(x)

        x = self.transformer.norm_f(x)

        logits = self.lm_head(x)
        
        loss = None
        if targets is not None:
            loss = F.cross_entropy(
                logits.view(-1, logits.size(-1)),
                targets.view(-1),
                ignore_index=-1,
            )

        return logits, loss

# ---- HF wrapper -------------------------------------------------------------

from transformers.generation.utils import GenerationMixin

class Base_Loop_GPTForCausalLM(PreTrainedModel, GenerationMixin):
    config_class = GPTConfig
    main_input_name = "input_ids"
    _tied_weights_keys = ["gpt.transformer.wte.weight", "gpt.lm_head.weight"]

    def __init__(self, config: GPTConfig, **kwargs):
        super().__init__(config)
        self.gpt = GPT(config)
        self.post_init()

    # expose embeddings/heads for HF utilities
    def get_input_embeddings(self):
        return self.gpt.transformer.wte

    def set_input_embeddings(self, new_emb):
        self.gpt.transformer.wte = new_emb
        self.gpt.lm_head.weight = new_emb.weight  # keep tied

    def get_output_embeddings(self):
        return self.gpt.lm_head

    def set_output_embeddings(self, new_out):
        self.gpt.lm_head = new_out

    def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **kwargs):
        return {
            "input_ids": input_ids,
            "attention_mask": attention_mask,
            # no labels during generation
        }

    def forward(self, input_ids=None, attention_mask=None, labels=None, **kwargs):
        # pass the mask all the way through
        logits, loss = self.gpt(
            input_ids, targets=labels, attention_mask=attention_mask
        )
        return CausalLMOutputWithCrossAttentions(loss=loss, logits=logits)