model.py

import math
import torch
import torch.nn as nn
from torch.nn import functional as F
from utils import CfgNode as CN
from rotary_embedding_torch import RotaryEmbedding

class NewGELU(nn.Module):
    """
    Implementation of the GELU activation function currently in Google BERT repo (identical to OpenAI).
    Reference: Gaussian Error Linear Units (GELU) paper: https://arxiv.org/abs/1606.08415
    """
    def forward(self, x):
        return 0.5 * x * (1.0 + torch.tanh(math.sqrt(2.0 / math.pi) * (x + 0.044715 * torch.pow(x, 3.0))))



class RelativeSelfAttention(nn.Module):
    """
    A vanilla multi-head masked self-attention layer with a projection at the end.
    It is possible to use torch.nn.MultiheadAttention here but I am including an
    explicit implementation here to show that there is nothing too scary here.
    """

    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)
        # output projection
        self.c_proj = nn.Linear(config.n_embd, config.n_embd)
        # relative positional encoding
        # this is an important hyperparameter to vary and check
        self.max_relative_position = config.max_relative_position
        self.relative_position_k = RelativePosition(config.n_embd//config.n_head, self.max_relative_position)
        self.relative_position_v = RelativePosition(config.n_embd//config.n_head, self.max_relative_position)
        # regularization
        self.attn_dropout = nn.Dropout(config.attn_pdrop)
        self.resid_dropout = nn.Dropout(config.resid_pdrop)
        # 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))
        self.n_head = config.n_head
        self.n_embd = config.n_embd

    def forward(self, x, mask):
        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)
        len_q = q.shape[1]
        len_k = k.shape[1]
        batch_size  = q.shape[0]

        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)
        attn1 = (q @ k.transpose(-2, -1)) 
        q = q.reshape(batch_size*self.n_head, len_q, k.size(-1)).permute(1, 0, 2)  
        r_q2 = q.contiguous().view(len_q, batch_size*self.n_head,k.size(-1)) #T*bs(h)*emb/head
        r_k2 = self.relative_position_k(len_q, len_k)
        attn2 = torch.matmul(r_q2, r_k2.transpose(1, 2)).transpose(0, 1)
        attn2 = attn2.contiguous().view(batch_size, self.n_head, len_q, len_k)
        att = (attn1 + attn2)* (1.0 / math.sqrt(k.size(-1)))

        att = att.masked_fill(mask[:,:,:T,:T] == 0, float('-inf'))
        att = F.softmax(att, dim=-1)
        att = torch.nan_to_num(att, nan=0.0)
        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 RotarySelfAttention(nn.Module):
    """
    A vanilla multi-head masked self-attention layer with a projection at the end.
    It is possible to use torch.nn.MultiheadAttention here but I am including an
    explicit implementation here to show that there is nothing too scary here.
    """

    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)
        # output projection
        self.c_proj = nn.Linear(config.n_embd, config.n_embd)
        # rotarty encoding
        self.rotary_emb = RotaryEmbedding(dim = 32)
        # regularization
        self.attn_dropout = nn.Dropout(config.attn_pdrop)
        self.resid_dropout = nn.Dropout(config.resid_pdrop)
        # 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))
        self.n_head = config.n_head
        self.n_embd = config.n_embd

    def forward(self, x, mask):
        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)
        q = self.rotary_emb.rotate_queries_or_keys(q)
        k = self.rotary_emb.rotate_queries_or_keys(k)
        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
        att = att.masked_fill(mask[:,:,:T,:T] == 0, float('-inf'))
        att = F.softmax(att, dim=-1)
        att = torch.nan_to_num(att, nan=0.0)
        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 CausalSelfAttention(nn.Module):
    """
    A vanilla multi-head masked self-attention layer with a projection at the end.
    It is possible to use torch.nn.MultiheadAttention here but I am including an
    explicit implementation here to show that there is nothing too scary here.
    """

    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)
        # output projection
        self.c_proj = nn.Linear(config.n_embd, config.n_embd)
        # regularization
        self.attn_dropout = nn.Dropout(config.attn_pdrop)
        self.resid_dropout = nn.Dropout(config.resid_pdrop)
        # 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))
        self.n_head = config.n_head
        self.n_embd = config.n_embd

    def forward(self, x, mask):
        B, T, C = x.size() # batch size, sequence length, embedding dimensionality (n_embd)

        out = self.c_attn(x)
        q, k ,v  = out.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)
        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
        att = att.masked_fill(mask[:,:,:T,:T] == 0, float('-inf'))

        att = F.softmax(att, dim=-1)
        att = torch.nan_to_num(att, nan=0.0)

        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 Block(nn.Module):
    """ an unassuming Transformer block """

    def __init__(self, config):
        super().__init__()
        self.ln_1 = nn.LayerNorm(config.n_embd)
        if config.embedding_type=='pretrained' or config.embedding_type=='pe':
            self.attn = CausalSelfAttention(config)
        elif config.embedding_type=='re':
            self.attn = RelativeSelfAttention(config)
        elif config.embedding_type=='rotary':
            self.attn = RotarySelfAttention(config)

        self.ln_2 = nn.LayerNorm(config.n_embd)
        self.c_fc    = nn.Linear(config.n_embd, config.scale_internal * config.n_embd)
        self.c_proj  = nn.Linear(config.scale_internal * config.n_embd, config.n_embd)
        self.act     = NewGELU()
        self.dropout = nn.Dropout(config.resid_pdrop)

    def forward(self, x, mask):
        x = self.ln_1(x)
        x = x + self.attn(x, mask)
        x = x + self.dropout(self.c_proj(self.act(self.c_fc(x))))
        return x

class PositionalEncoding(nn.Module):
    "Implement the PE function."

    def __init__(self, d_model, max_len=5000):
        super(PositionalEncoding, self).__init__()

        # Compute the positional encodings once in log space.
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len).unsqueeze(1)
        div_term = torch.exp(
            torch.arange(0, d_model, 2) * -(math.log(10000.0) / d_model)
        )
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0)
        self.register_buffer("pe", pe)

    def forward(self, x):
        return self.pe[:, : x.size(1)].requires_grad_(False)


class RelativePosition(nn.Module):

    def __init__(self, num_units, max_relative_position):
        super().__init__()
        self.num_units = num_units
        self.max_relative_position = max_relative_position
        self.embeddings_table = nn.Parameter(torch.Tensor(max_relative_position * 2 + 1, num_units))
        nn.init.xavier_uniform_(self.embeddings_table)

    def forward(self, length_q, length_k):
        range_vec_q = torch.arange(length_q)
        range_vec_k = torch.arange(length_k)
        distance_mat = range_vec_k[None, :] - range_vec_q[:, None]
        distance_mat_clipped = torch.clamp(distance_mat, -self.max_relative_position, self.max_relative_position)
        final_mat = distance_mat_clipped + self.max_relative_position
        final_mat = torch.LongTensor(final_mat).cuda()
        embeddings = self.embeddings_table[final_mat].cuda()

        return embeddings

class GPT(nn.Module):
    """ GPT Language Model """

    @staticmethod
    def get_default_config():
        C = CN()
        # either model_type or (n_layer, n_head, n_embd) must be given in the config
        C.model_type = 'wrn2-cfg-nano'
        C.pad_token = 5
        C.embedding_type = 'pretrained'
        C.max_relative_position = 8
        C.scale_internal = 4
        C.n_layer = None
        C.n_head = None
        C.n_embd =  None
        # these options must be filled in externally
        C.vocab_size = None
        C.block_size = None
        # dropout hyperparameters
        C.embd_pdrop = 0.1
        C.resid_pdrop = 0.1
        C.attn_pdrop = 0.1
        return C

    def __init__(self, tokenizer, config):
        super().__init__()
        assert config.vocab_size is not None
        assert config.block_size is not None
        self.block_size = config.block_size
        self.pad_token = config.pad_token
        self.tokenizer = tokenizer
        self.vocab_size = config.vocab_size

        type_given = config.model_type is not None
        params_given = all([config.n_layer is not None, config.n_head is not None, config.n_embd is not None])
        assert type_given ^ params_given # exactly one of these (XOR)
        if type_given:
            # translate from model_type to detailed configuration
            config.merge_from_dict({
                # names follow the huggingface naming conventions
                # GPT-1
                'openai-wrn2':   dict(n_layer=12, n_head=12, n_embd=768),  # 117M params
                # GPT-2 configs
                'wrn2':         dict(n_layer=12, n_head=12, n_embd=768),  # 124M params
                'wrn2-medium':  dict(n_layer=24, n_head=16, n_embd=1024), # 350M params
                'wrn2-large':   dict(n_layer=36, n_head=20, n_embd=1280), # 774M params
                'wrn2-xl':      dict(n_layer=48, n_head=25, n_embd=1600), # 1558M params
                # Gophers
                'gopher-44m':   dict(n_layer=8, n_head=16, n_embd=512),
                'wrn2-cfg-medium':     dict(n_layer=8, n_head=8, n_embd=400),
                'wrn2-cfg-mini':     dict(n_layer=6, n_head=6, n_embd=192),
                'wrn2-cfg-micro':    dict(n_layer=4, n_head=4, n_embd=128),
                'wrn2-cfg-nano':     dict(n_layer=3, n_head=3, n_embd=48),
            }[config.model_type])

        if config.embedding_type=='pe':
            emb = PositionalEncoding(config.n_embd, config.block_size)
        else:
            emb = nn.Embedding(config.block_size, config.n_embd)
        self.transformer = nn.ModuleDict(dict(
            wte = nn.Embedding(config.vocab_size, config.n_embd),
            wpe = emb,
            drop = nn.Dropout(config.embd_pdrop),
            h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
            ln_f = nn.LayerNorm(config.n_embd),
        ))
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)

        # init all weights, and apply a special scaled init to the residual projections, per GPT-2 paper
        self.apply(self._init_weights)
        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 (note we don't count the decoder parameters in lm_head)
        n_params = sum(p.numel() for p in self.transformer.parameters())
        print("number of parameters: %.2fM" % (n_params/1e6,))

    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)
        elif isinstance(module, nn.LayerNorm):
            torch.nn.init.zeros_(module.bias)
            torch.nn.init.ones_(module.weight)

    @classmethod
    def from_pretrained(cls, model_type):
        """
        Initialize a pretrained GPT model by copying over the weights
        from a huggingface/transformers checkpoint.
        """
        assert model_type in {'wrn2', 'wrn2-medium', 'wrn2-large', 'wrn2-xl'}
        from transformers import GPT2LMHeadModel

        # create a from-scratch initialized minGPT model
        config = cls.get_default_config()
        config.model_type = model_type
        #config.vocab_size = 50257 # openai's model vocabulary
        #config.block_size = 1024  # openai's model block_size
        model = GPT(config)
        sd = model.state_dict()

        # init a huggingface/transformers model
        model_hf = GPT2LMHeadModel.from_pretrained(model_type)
        sd_hf = model_hf.state_dict()

        # copy while ensuring all of the parameters are aligned and match in names and shapes
        keys = [k for k in sd_hf if not k.endswith('attn.masked_bias')] # ignore these
        transposed = ['attn.c_attn.weight', 'attn.c_proj.weight', 'mlp.c_fc.weight', 'mlp.c_proj.weight']
        # basically the openai checkpoints use a "Conv1D" module, but we only want to use a vanilla nn.Linear.
        # this means that we have to transpose these weights when we import them
        assert len(keys) == len(sd)
        for k in keys:
            if any(k.endswith(w) for w in transposed):
                # special treatment for the Conv1D weights we need to transpose
                assert sd_hf[k].shape[::-1] == sd[k].shape
                with torch.no_grad():
                    sd[k].copy_(sd_hf[k].t())
            else:
                # vanilla copy over the other parameters
                assert sd_hf[k].shape == sd[k].shape
                with torch.no_grad():
                    sd[k].copy_(sd_hf[k])

        return model

    def configure_optimizers(self, train_config):
        """
        This long function is unfortunately doing something very simple and is being very defensive:
        We are separating out all parameters of the model into two buckets: those that will experience
        weight decay for regularization and those that won't (biases, and layernorm/embedding weights).
        We are then returning the PyTorch optimizer object.
        """

        # separate out all parameters to those that will and won't experience regularizing weight decay
        decay = set()
        no_decay = set()
        whitelist_weight_modules = (torch.nn.Linear, )
        blacklist_weight_modules = (torch.nn.LayerNorm, torch.nn.Embedding)
        for mn, m in self.named_modules():
            for pn, p in m.named_parameters():
                fpn = '%s.%s' % (mn, pn) if mn else pn # full param name
                # random note: because named_modules and named_parameters are recursive
                # we will see the same tensors p many many times. but doing it this way
                # allows us to know which parent module any tensor p belongs to...
                if pn.endswith('bias'):
                    # all biases will not be decayed
                    no_decay.add(fpn)
                elif 'relative_position_v' in pn:
                    # all biases will not be decayed
                    decay.add(fpn)
                elif 'relative_position_k' in pn:
                    # all biases will not be decayed
                    decay.add(fpn)
                elif 'rotary_emb' in pn:
                    # all biases will not be decayed
                    decay.add(fpn)
                elif pn.endswith('weight') and isinstance(m, whitelist_weight_modules):
                    # weights of whitelist modules will be weight decayed
                    decay.add(fpn)
                elif pn.endswith('weight') and isinstance(m, blacklist_weight_modules):
                    # weights of blacklist modules will NOT be weight decayed
                    no_decay.add(fpn)

        # validate that we considered every parameter
        param_dict = {pn: p for pn, p in self.named_parameters()}
        inter_params = decay & no_decay
        union_params = decay | no_decay
        assert len(inter_params) == 0, "parameters %s made it into both decay/no_decay sets!" % (str(inter_params), )
        assert len(param_dict.keys() - union_params) == 0, "parameters %s were not separated into either decay/no_decay set!" \
                                                    % (str(param_dict.keys() - union_params), )

        # create the pytorch optimizer object
        optim_groups = [
            {"params": [param_dict[pn] for pn in sorted(list(decay))], "weight_decay": train_config.weight_decay},
            {"params": [param_dict[pn] for pn in sorted(list(no_decay))], "weight_decay": 0.0},
        ]
        optimizer = torch.optim.AdamW(optim_groups, lr=train_config.learning_rate, betas=train_config.betas)
        return optimizer
    
        

    def forward(self, idx, targets=None, mask=None, attack=None, emb=[],reduction=False):    
        device = idx.device
        if attack=='emb_in':
            b, t = idx.shape[0], idx.shape[1]
        else:
            b, t = idx.size()
        assert t <= self.block_size, f"Cannot forward sequence of length {t}, block size is only {self.block_size}"
        pos = torch.arange(0, t, dtype=torch.long, device=device).unsqueeze(0) # shape (1, t)


        if attack=='emb':
            tok_emb = self.transformer.wte(idx)
            return tok_emb
        elif attack=='emb_in':
            tok_emb = emb
        else:
            tok_emb = self.transformer.wte(idx)

        pos_emb = self.transformer.wpe(pos) 
        x = self.transformer.drop(tok_emb + pos_emb)
        for block in self.transformer.h:
            x = block(x, mask)
        x = self.transformer.ln_f(x)
        logits = self.lm_head(x)

        loss = None
        if targets is not None:
            if reduction==False:
                loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=self.tokenizer[self.pad_token], reduction='mean')
            else:
                loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=self.tokenizer[self.pad_token], reduction="none")
        return logits, loss

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=1.0, do_sample=False, top_k=None):
        """
        Take a conditioning sequence of indices idx (LongTensor of shape (b,t)) and complete
        the sequence max_new_tokens times, feeding the predictions back into the model each time.
        Most likely you'll want to make sure to be in model.eval() mode of operation for this.
        """
        for _ in range(max_new_tokens):
            # if the sequence context is growing too long we must crop it at block_size
            idx_cond = idx if idx.size(1) <= self.block_size else idx[:, -self.block_size:]
            # forward the model to get the logits for the index in the sequence
            logits, _ = self(idx_cond)
            # pluck the logits at the final step and scale by desired temperature
            logits = logits[:, -1, :] / temperature
            # optionally crop the logits to only the top k options
            if top_k is not None:
                v, _ = torch.topk(logits, top_k)
                logits[logits < v[:, [-1]]] = -float('Inf')
            # apply softmax to convert logits to (normalized) probabilities
            probs = F.softmax(logits, dim=-1)
            # either sample from the distribution or take the most likely element
            if do_sample:
                idx_next = torch.multinomial(probs, num_samples=1)
            else:
                _, idx_next = torch.topk(probs, k=1, dim=-1)
            # append sampled index to the running sequence and continue
            idx = torch.cat((idx, idx_next), dim=1)

        return idx

    @torch.no_grad()
    def generate_next_token(self, idx, y, mask, temperature=1.0, do_sample=False, top_k=None):
        idx_cond = idx 
        logits, loss = self(idx_cond,y, mask)
        batch_size = logits.shape[0]
        seq_length = logits.shape[1]
        vocab_size = logits.shape[-1]
        return logits, loss