File size: 14,524 Bytes
9fd062d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
import inspect
from dataclasses import dataclass
import torch
import torch.nn as nn
from torch.nn import functional as F
from hellaswag import render_example, iterate_examples
# --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
# From original transformer model gpt2 only have decoder part and also the cross-attention is not used.
# Also there's reshuffling layer-norms and Additional Layer normalization is added right before the soft-max layer.

class CausalSelfAttention(nn.Module): # this class combined the self-attention mechanism and multi-head attention mechanism in one class

    def __init__(self, config):
        super().__init__()

        assert config.n_embd % config.n_head == 0 # n_emb is the embedding size and n_head is the number of heads in the multi-head attention mechanism 
                                                 # (so the embedding size should be divisible by the number of heads)
        self.c_attn = nn.Linear(config.n_embd, 3*config.n_embd) # Linear layer for the query, key and value projections for all heads, but in batch
        self.c_proj = nn.Linear(config.n_embd, config.n_embd) # Linear layer for the final output projection
        self.c_proj.NANOGPT_SCALE_INIT = 1 # Scaling the initialization of the output projection
        
        # Regularization
        self.n_head = config.n_head
        self.n_embd = config.n_embd

        # self.register_buffer("bias", torch.tril(torch.ones(config.block_size, config.block_size)).view(1,1,config.block_size, config.block_size)) # Lower triangular matrix for masking future tokens
    
    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 dimension
        # nh is "number of heads", hs is "head size", and C (number of channels) = nh * hs
        # eg: in GPT-2 (124M), n_head=12, hs=64, so nh*hs = C = 768 channels in Transformer (channels is also called as hidden size)
        qkv = self.c_attn(x) # qkv is the query, key and value projections for all heads
        q,k,v = qkv.split(self.n_embd, dim=2) # Splitting the qkv into query, key and value projections

        k = k.view(B,T,self.n_head, C//self.n_head).transpose(1,2) # Splitting the key into the number of heads and transposing it (B,nh,T,hs)
        q = q.view(B,T,self.n_head, C//self.n_head).transpose(1,2) # Splitting the key into the number of heads and transposing it (B,nh,T,hs)
        v = v.view(B,T,self.n_head, C//self.n_head).transpose(1,2) # Splitting the key into the number of heads and transposing it (B,nh,T,hs)

        # attention (materializes the large (T,T) matrix for all queries and keys)

        # att = ([email protected](-2,-1))*(1.0 / math.sqrt(k.size(-1))) # Multiplying the query and key and scaling it by the square root of the key size
        # att = att.masked_fill(self.bias[:,:,:T,:T]==0, float('-inf')) # Masking the future tokens
        # att = F.softmax(att, dim=-1) # Softmax over the last dimension
        # y = att@v # Multiplying the attention weights with the values (B,nh,T,T) x (B,nh,T,hs) = (B,nh,T,hs)

        # Attention on GPT2: ( matmul + mask + softmax + dropout + matmul ) ==> 15ms
        # Flash Attention: Fused Kernel ==> 2.5ms

        y = F.scaled_dot_product_attention(q, k, v, is_causal=True)

        y = y.transpose(1,2).contiguous().view(B,T,C) # re-assemble all head outputs side by side

        # Output Projection
        y = self.c_proj(y) # Projecting the output to the original size
        return y


class MLP(nn.Module):

    def __init__(self, config):
        super().__init__() # Inheriting from the parent class nn.Module
        self.c_fc = nn.Linear(config.n_embd, 4*config.n_embd) # Fully connected layer for the first part of the MLP which takes the input and projects it to 4 times the size of the input
        self.gelu = nn.GELU(approximate='tanh') # GELU activation function
        self.c_proj = nn.Linear(4*config.n_embd, config.n_embd) # Fully connected layer for the second part of the MLP which projects the output of the previous layer to the original size
        self.c_proj.NANOGPT_SCALE_INIT = 1 # Scaling the initialization of the output projection


    def forward(self,x):
        x = self.c_fc(x)
        x = self.gelu(x)
        x = self.c_proj(x)
        return x


# Block is basically a transformer block which consists of a self-attention mechanism and a feed-forward neural network (decoder part)
class Block(nn.Module):

    def __init__(self,config):
        super().__init__()        
        self.ln_1 = nn.LayerNorm(config.n_embd) # Layer normalization before the self-attention
        self.attn = CausalSelfAttention(config) # Self-attention mechanism
        self.ln_2 = nn.LayerNorm(config.n_embd) # Layer normalization after the self-attention
        self.mlp = MLP(config) # Multi-layer perceptron for each position
    
    # forward pass of the block, the input x is the sequence of embeddings and return is the updated sequence of embeddings
    def forward(self,x):
        x = x + self.attn(self.ln_1(x)) # residual connection followed by self-attention
        # Our text first goes to ln_1, then to the self-attention mechanism, then to ln_2, and finally to the MLP
        x = x + self.mlp(self.ln_2(x)) # residual connection followed by MLP (ffn)
        # In attention 1024 sequence lined up communicated with each other & exchange info.
        # Whereas MLP happens to every single token individually and there's no communication between tokens or exchange of information between tokens.
        return x

@dataclass
class GPTConfig:
    # block_size: int = 256 # maximum sequence length
    # vocab_size: int = 50257 # number of tokens in the vocabulary i.e. 50,000 BPE merges + 256 byte tokens + 1 <|endoftext|> token
    # n_layer: int = 12 # number of transformer layers
    # n_head: int = 12 # number of heads in the multi-head attention mechanism
    # n_embd: int = 768 # embedding dimension of each token

    # # changed the default values of the parameters
    block_size: int = 256 # maximum sequence length
    vocab_size: int = 50257 # number of tokens in the vocabulary i.e. 50,000 BPE merges + 256 byte tokens + 1 <|endoftext|> token
    n_layer: int = 6 # number of transformer layers
    n_head: int = 6 # number of heads in the multi-head attention mechanism
    n_embd: int = 768 # embedding dimension of each token

class GPT(nn.Module): # Kind of skeleton of the model
    
    def __init__(self,config):
        super().__init__()
        self.config = config

        # transformer is the main container and it have further sub-modules like wte, wpe, h, ln_f
        self.transformer = nn.ModuleDict(dict(
            wte = nn.Embedding(config.vocab_size, config.n_embd), # token embedding weights
            wpe = nn.Embedding(config.block_size, config.n_embd), # positional embedding weights
            h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]), # transformer blocks as a list of n_layer (h is hidden layer)
            ln_f = nn.LayerNorm(config.n_embd), # final layer normalization before the softmax
        ))
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias = False) # language model head is a linear layer with vocab_size output

        # Weight sharing scheme
        self.transformer.wte.weight = self.lm_head.weight # weight tying the token embeddings with the pre-softmax linear transformation, using this we saved 40m parameters

        # init parameters
        self.apply(self._init_weights) # initializing the weights of the model
    
    def _init_weights(self, module):
        if isinstance(module, nn.Linear): 
            std = 0.02
            if hasattr(module, 'NANOGPT_SCALE_INIT'):
                std *= (2*self.config.n_layer)**-0.5 # scale by the number of layers
            torch.nn.init.normal_(module.weight, mean=0.0, std = std) # initializing the weights of the linear layer with normal distribution
            if module.bias is not None:
                torch.nn.init.zeros_(module.bias) # initializing the bias of the linear layer with zeros
        elif isinstance(module, nn.Embedding): 
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02) 

    
    def forward(self,idx, targets= None): 
        # idx is of shape [batch_size, sequence_length] (B,T)
        B,T = idx.size() # batch size and sequence length
        assert T<=self.config.block_size ,f"Cannot forward sequence of length {T}, block size is only {self.config.block_size}"

        # forward the token and position embeddings
        pos = torch.arange(0, T, dtype = torch.long, device =idx.device) # tensor of shape [T]
        pos_emb = self.transformer.wpe(pos) # position embeddings of shape (T, n_embd)
        tok_emb = self.transformer.wte(idx) # token embeddings of shape (B,T,n_embd)
        x = tok_emb + pos_emb 

        # forward the blocks of the transformer
        for block in self.transformer.h:
            x = block(x)
        # Forward the final layernorm and the classifier
        x = self.transformer.ln_f(x)
        logits = self.lm_head(x) # (B,T,vocab_size)
        loss = None
        if targets is not None:
            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1)) # Cross-entropy flattens out the 3D (B,T,vocab_size) tensor to 2D 
                                                                                        # (B*T,vocab_size) tensor, It also flattens out the target tensor to 1D tensor
        return logits , loss


    @classmethod
    def from_pretrained(cls, model_type):
        """Load pretrained GPT2 model weights from huggingface"""

        assert model_type in {'gpt2', 'gpt2-medium', 'gpt2-large', 'gpt2-xl'} # Checking if the model type is valid
        
        print("Loading weights from pretrained gpt: %s" %model_type)
        from transformers import GPT2LMHeadModel
        # n_layer, n_head, and n_embd are determined by the model type

        config_args = {
            'gpt2':         dict(n_layer=12, n_head=12, n_embd=768),  # 124M parameters
            'gpt2-medium':  dict(n_layer=24, n_head=16, n_embd=1024), # 350M parameters
            'gpt2-large':   dict(n_layer=36, n_head=20, n_embd=1280), # 774M parameters
            'gpt2-xl':      dict(n_layer=48, n_head=25, n_embd=1600), # 1558M parameters
        }[model_type]

        config_args['vocab_size'] = 50257 # always 50257 for GPT model checkpoints
        config_args['block_size'] = 1024  # always 1024 for GPT model checkpoint

        # create a from-scratch initialized minGPT model
        config = GPTConfig(**config_args)
        model = GPT(config)
        sd = model.state_dict() # state_dict is the model weights
        sd_keys = sd.keys() # keys are the names of the weights
        sd_keys = [k for k in sd_keys if not k.endswith('.attn.bias')] # discard this mask / buffer key, not parameters of the model

        # 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 correctly and matches 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
        # missing in sd_keys: lm_head.weight

        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 # return the model with the pretrained weights

    def configure_optimizers(self, weight_decay, learning_rate, device_type):
        # start with all of the candidate parameters (that require gradients)
        param_dict = {pn: p for pn, p in self.named_parameters()} # named parameters
        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad} # only parameters that require gradients

        # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
        # i.e. all weight tensors in matmuls + embeddings, all biases and layernorm don't.
        decay_params = [p for n, p in param_dict.items() if p.dim() >= 2] # weight tensors in matmuls + embeddings
        nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2] # biases and layernorm
        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)
        if master_process:
            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 # check if fused is available in AdamW
        use_fused = fused_available and device_type == "cuda"
        if master_process:
            print(f"using fused AdamW: {use_fused}")
        optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=(0.9, 0.95), eps=1e-8, fused=use_fused)
        return optimizer