model.py

import math
import inspect
from dataclasses import dataclass

import torch
import torch.nn as nn
from torch.nn import functional as F
torch._dynamo.config.suppress_errors = True


class LayerNorm(nn.Module):
    def __init__(self, ndim, bias):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(ndim))
        self.bias = nn.Parameter(torch.zeros(ndim)) if bias else None

    def forward(self, input):
        return F.layer_norm(input, self.weight.shape, self.weight, self.bias, 1e-5)

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
        self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')
        #print('Is the model using flash attention? ', self.flash)
        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, attention_mask=None):
        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
            # print('seq len: ', T)
            y = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=attention_mask, 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, 4 * config.n_embd, bias=config.bias)
        self.gelu    = nn.GELU()
        self.c_proj  = nn.Linear(4 * config.n_embd, 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.ln_1 = LayerNorm(config.n_embd, bias=config.bias)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = LayerNorm(config.n_embd, bias=config.bias)
        self.mlp = MLP(config)

    def forward(self, x):
        x = x + self.attn(self.ln_1(x))
        x = x + self.mlp(self.ln_2(x))
        return x

@dataclass
class GPTConfig:
    block_size: int = 1024
    vocab_size: int = 50304
    n_layer: int = 12
    n_head: int = 12
    n_embd: int = 768
    dropout: float = 0.0
    bias: bool = True

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 = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
            ln_f = LayerNorm(config.n_embd, bias=config.bias),
        ))
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
        self.transformer.wte.weight = self.lm_head.weight

        # 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.kaiming_normal_(module.weight, a=0, mode='fan_in', nonlinearity='relu')
            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, all_logits=False):
        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)
        #print(idx)
        #print(idx.size())
        
        # 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 block in self.transformer.h:
            x = block(x)
        x = self.transformer.ln_f(x)

        if targets is not None:
            # if we are given some desired targets also calculate the loss
            logits = self.lm_head(x)
            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-1)
        elif all_logits:
            #print('beforeLM head: ', x.shape)
            #logits = self.lm_head(x)
            loss = None
            return x, loss
        else:
            # inference-time mini-optimization: only forward the lm_head on the very last position
            logits = self.lm_head(x[:, [-1], :]) # note: using list [-1] to preserve the time dim
            loss = None

        return logits, loss

    def crop_block_size(self, block_size):
        # model surgery to decrease the block size if necessary
        # e.g. we may load the GPT2 pretrained model checkpoint (block size 1024)
        # but want to use a smaller block size for some smaller, simpler model
        assert block_size <= self.config.block_size
        self.config.block_size = block_size
        self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size])
        for block in self.transformer.h:
            if hasattr(block.attn, 'bias'):
                block.attn.bias = block.attn.bias[:,:,:block_size,:block_size]

    @classmethod
    def from_pretrained(cls, model_type, override_args=None):
        assert model_type in {'gpt2', 'gpt2-medium', 'gpt2-large', 'gpt2-xl'}
        override_args = override_args or {} # default to empty dict
        # only dropout can be overridden see more notes below
        assert all(k == 'dropout' for k in override_args)
        from transformers import GPT2LMHeadModel
        print("loading weights from pretrained gpt: %s" % model_type)

        # n_layer, n_head and n_embd are determined from model_type
        config_args = {
            'gpt2':         dict(n_layer=12, n_head=12, n_embd=768),  # 124M params
            'gpt2-medium':  dict(n_layer=24, n_head=16, n_embd=1024), # 350M params
            'gpt2-large':   dict(n_layer=36, n_head=20, n_embd=1280), # 774M params
            'gpt2-xl':      dict(n_layer=48, n_head=25, n_embd=1600), # 1558M params
        }[model_type]
        print("forcing vocab_size=50257, block_size=1024, bias=True")
        config_args['vocab_size'] = 50257 # always 50257 for GPT model checkpoints
        config_args['block_size'] = 1024 # always 1024 for GPT model checkpoints
        config_args['bias'] = True # always True for GPT model checkpoints
        # we can override the dropout rate, if desired
        if 'dropout' in override_args:
            print(f"overriding dropout rate to {override_args['dropout']}")
            config_args['dropout'] = override_args['dropout']
        # create a from-scratch initialized minGPT model
        config = GPTConfig(**config_args)
        model = GPT(config)
        sd = model.state_dict()
        sd_keys = sd.keys()
        sd_keys = [k for k in sd_keys if not k.endswith('.attn.bias')] # discard this mask / buffer, not a param

        # 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
        sd_keys_hf = sd_hf.keys()
        sd_keys_hf = [k for k in sd_keys_hf if not k.endswith('.attn.masked_bias')] # ignore these, just a buffer
        sd_keys_hf = [k for k in sd_keys_hf if not k.endswith('.attn.bias')] # same, just the mask (buffer)
        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 Linear
        # this means that we have to transpose these weights when we import them
        assert len(sd_keys_hf) == len(sd_keys), f"mismatched keys: {len(sd_keys_hf)} != {len(sd_keys)}"
        for k in sd_keys_hf:
            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, weight_decay, learning_rate, betas, device_type):
        # start with all of the candidate parameters
        param_dict = {pn: p for pn, p in self.named_parameters()}
        # filter out those that do not require grad
        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
        # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
        # i.e. all weight tensors in matmuls + embeddings decay, all biases and layernorms don't.
        decay_params = [p for n, p in param_dict.items() if p.dim() >= 2]
        nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2]
        optim_groups = [
            {'params': decay_params, 'weight_decay': weight_decay},
            {'params': nodecay_params, 'weight_decay': 0.0}
        ]
        num_decay_params = sum(p.numel() for p in decay_params)
        num_nodecay_params = sum(p.numel() for p in nodecay_params)
        print(f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters")
        print(f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters")
        # Create AdamW optimizer and use the fused version if it is available
        fused_available = 'fused' in inspect.signature(torch.optim.AdamW).parameters
        use_fused = fused_available and device_type == 'cuda'
        extra_args = dict(fused=True) if use_fused else dict()
        optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=betas, **extra_args)
        print(f"using fused AdamW: {use_fused}")

        return optimizer

    def estimate_mfu(self, fwdbwd_per_iter, dt):
        """ estimate model flops utilization (MFU) in units of A100 bfloat16 peak FLOPS """
        # first estimate the number of flops we do per iteration.
        # see PaLM paper Appendix B as ref: https://arxiv.org/abs/2204.02311
        N = self.get_num_params()
        cfg = self.config
        L, H, Q, T = cfg.n_layer, cfg.n_head, cfg.n_embd//cfg.n_head, cfg.block_size
        flops_per_token = 6*N + 12*L*H*Q*T
        flops_per_fwdbwd = flops_per_token * T
        flops_per_iter = flops_per_fwdbwd * fwdbwd_per_iter
        # express our flops throughput as ratio of A100 bfloat16 peak flops
        flops_achieved = flops_per_iter * (1.0/dt) # per second
        flops_promised = 312e12 # A100 GPU bfloat16 peak flops is 312 TFLOPS
        mfu = flops_achieved / flops_promised
        return mfu

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=None, strategy='sampling', 
                 beam_size=3, eos_token_id=0, repetition_penalty=1.0, return_logits=False):
        assert strategy in ['greedy_search', 'sampling', 'top_k', 'beam_search']

        gen_logits = []

        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.config.block_size else idx[:, -self.config.block_size:]
            # forward the model to get the logits for the index in the sequence
            #print('idx_cond: ', idx_cond)
            logits, _ = self(idx_cond)
            #print('logits: ', logits.shape)
            #print('logits[:, -1, :]: ', logits[:, -1, :].shape)
            # pluck the logits at the final step and scale by desired temperature
            logits = logits[:, -1, :] / temperature
            
            # Apply repetition penalty
            if repetition_penalty != 1.0:
                for i in range(idx.size(0)):
                    for j in range(idx.size(1)):
                        logits[i, idx[i, j]] /= repetition_penalty

            if return_logits:
                gen_logits.append(logits)

            if strategy == 'greedy_search':
                idx_next = torch.argmax(logits, dim=-1, keepdim=True)
            
            elif strategy == 'beam_search':
                # If beam search is the selected strategy

                # First, initialize the sequences and their scores
                beam_seqs = [idx] * beam_size
                beam_scores = torch.zeros((idx.size(0), beam_size), device=idx.device)

                for k in range(max_new_tokens):
                    all_candidates = []

                    for i, seq in enumerate(beam_seqs):
                        # Get next token probabilities
                        idx_cond = seq if seq.size(1) <= self.config.block_size else seq[:, -self.config.block_size:]
                        logits, __ = self(idx_cond)
                        logits = logits[:, -1, :]
                        probs = F.log_softmax(logits, dim=-1)  # Use log probs to avoid numerical instability

                        # Get top sequences for this beam (we could use more than beam_size here for diversity)
                        scores, indices = torch.topk(probs, beam_size)
                        for j in range(beam_size):
                            candidate_seq = torch.cat([seq, indices[:, j:j+1]], dim=1)
                            candidate_score = beam_scores[:, i] + scores[:, j]

                            all_candidates.append((candidate_score, candidate_seq))

                    # Sort all candidates by score
                    all_candidates.sort(key=lambda x: -x[0].mean().item())  # Average score over the batch

                    # Get the top sequences
                    beam_seqs = [all_candidates[i][1] for i in range(beam_size)]
                    beam_scores = torch.stack([all_candidates[i][0] for i in range(beam_size)], dim=1)

                # At the end, choose the sequence with the highest score
                idx = beam_seqs[0]
                return idx if idx[0][0] != eos_token_id else idx[:, 1:]
            
            elif strategy == 'sampling':
                # apply softmax to convert logits to (normalized) probabilities
                probs = F.softmax(logits, dim=-1)
                # sample from the distribution
                idx_next = torch.multinomial(probs, num_samples=1)
            
            elif strategy == 'top_k':
                if top_k is not None:
                    logits, indices = torch.topk(logits, min(top_k, logits.size(-1)))
                    probs = F.softmax(logits, dim=-1)
                    idx_next = torch.multinomial(probs, num_samples=1)
                    idx_next = torch.gather(indices, dim=-1, index=idx_next)
           
            # append sampled index to the running sequence and continue
            if strategy != 'beam_search':
                #gen_tokens.append(idx_next)
                if idx_next == eos_token_id:
                    break
                idx = torch.cat((idx, idx_next), dim=1)

        if return_logits:
            ret1 = idx if idx[0][0] != eos_token_id else idx[:, 1:]
            return (ret1, gen_logits)
        return idx if idx[0][0] != eos_token_id else idx[:, 1:]


    @torch.no_grad()
    def generate_CFG(self, idx, max_new_tokens, temperature=1.0, top_k=1, strategy='top_k', 
                     eos_token_id=0, return_logits=False, guidance_scale = 10):
        gen_logits = []

        idx_cond_init = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:]
        #print('idx_cond_init: ', idx_cond_init)
        idx_not_cond = torch.full_like(idx_cond_init, 0)
        #print('idx_not_cond: ', idx_not_cond)
        #idx_not_cond[:, -1] = 15 # not completely sure about leaving the X token

        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.config.block_size else idx[:, -self.config.block_size:]
            idx_not_cond_crop = idx_not_cond if idx_not_cond.size(1) <= self.config.block_size else idx_not_cond[:, -self.config.block_size:]
            
            # forward the model to get the logits for the index in the sequence
            #print('idx_cond: ', idx_cond)
            logits, _ = self(idx_cond)
            #print('logits: ', logits.shape)
            #print('logits[:, -1, :]: ', logits[:, -1, :].shape)
            # pluck the logits at the final step and scale by desired temperature
            logits = logits[:, -1, :] / temperature
            if return_logits:
                gen_logits.append(logits)

            #print('idx NOT cond: ', idx_not_cond_crop)
            #print('idx_cond: ', idx_cond)
            logits_without_condition, _ = self(idx_not_cond_crop)
            logits_without_condition = logits_without_condition[:, -1, :] / temperature

            probs = F.softmax(logits, dim=-1)
            probs_without_condition = F.softmax(logits_without_condition, dim=-1)
            scores_processed = guidance_scale * (probs - probs_without_condition) + probs_without_condition
            
            #top-k strategy OLD
            #logits, indices = torch.topk(logits, min(top_k, logits.size(-1)))
            #probs = F.softmax(logits, dim=-1)
            #idx_next = torch.multinomial(probs, num_samples=1)
            #idx_next = torch.gather(indices, dim=-1, index=idx_next)
            # Get top-k indices and probabilities from softmaxed scores

            #top-k strategy NEW
            #print('scores_processed: ', scores_processed)
            top_probs, top_indices = torch.topk(scores_processed, min(top_k, scores_processed.size(-1)), dim=-1)
            #print('top_probs: ', top_probs)
            #print('top_indices: ', top_indices)
            # Renormalize the top-k probabilities (since softmax has already been applied)
            top_probs = top_probs / top_probs.sum(dim=-1, keepdim=True)
            #eps = 1e-50
            #top_probs = top_probs / (top_probs.sum(dim=-1, keepdim=True) + eps)
            top_probs = torch.clamp(top_probs, min=0.0)
            
            #print('top_probs: ', top_probs)
            # Sample from the renormalized top-k probabilities
            idx_next = torch.multinomial(top_probs, num_samples=1)
            #print('idx_next: ', idx_next)
            # Map back to the original vocabulary indices
            idx_next = torch.gather(top_indices, dim=-1, index=idx_next)
            
            # append sampled index to the running sequence and continue
            #gen_tokens.append(idx_next)
            if idx_next == eos_token_id:
                break
            idx = torch.cat((idx, idx_next), dim=1)
            idx_not_cond = torch.cat((idx_not_cond, idx_next), dim=1)

        if return_logits:
            ret1 = idx if idx[0][0] != eos_token_id else idx[:, 1:]
            return (ret1, gen_logits)
        return idx if idx[0][0] != eos_token_id else idx[:, 1:]


    @torch.no_grad()
    def extract_embeddings(self, idx, all_logits=True):
        # if the sequence context is growing too long we must crop it at block_size
        idx = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:]
        # forward the model to get the logits
        logits_full, _ = self(idx, all_logits=all_logits)               
        return logits_full