File size: 7,411 Bytes
2e3175f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PreTrainedModel
from transformers.generation import GenerationMixin
from transformers.modeling_outputs import CausalLMOutputWithPast

from .configuration_parchment import ParchmentConfig


class Embeddings(nn.Module):
    def __init__(self, vocab_size, d_model):
        super().__init__()
        self.embeds = nn.Embedding(vocab_size, d_model)
        nn.init.normal_(self.embeds.weight, mean=0, std=d_model ** -0.5)

    def forward(self, token_ids):
        return self.embeds(token_ids)


class RoPE(nn.Module):
    def __init__(self, d_k, max_seq_len, base=10000.0):
        super().__init__()
        self.max_seq_len = max_seq_len
        # persistent=True so inv_freq is saved/loaded via from_pretrained
        inv_freq = 1.0 / (base ** (torch.arange(0, d_k, 2).float() / d_k))
        self.register_buffer("inv_freq", inv_freq, persistent=True)
        self._cos_cache: torch.Tensor | None = None
        self._sin_cache: torch.Tensor | None = None

    def _build_cache(self, device: torch.device, dtype: torch.dtype):
        t = torch.arange(self.max_seq_len, device=device, dtype=torch.float32)
        freqs = torch.outer(t, self.inv_freq.to(device, torch.float32))
        emb = torch.cat([freqs, freqs], dim=-1)
        self._cos_cache = emb.cos()[None, None].to(dtype)
        self._sin_cache = emb.sin()[None, None].to(dtype)

    def rotate_half(self, x):
        x1, x2 = x[..., : x.shape[-1] // 2], x[..., x.shape[-1] // 2 :]
        return torch.cat([-x2, x1], dim=-1)

    def forward(self, q, k):
        if self._cos_cache is None or self._cos_cache.device != q.device or self._cos_cache.dtype != q.dtype:
            self._build_cache(q.device, q.dtype)
        seq = q.shape[2]
        cos = self._cos_cache[:, :, :seq]
        sin = self._sin_cache[:, :, :seq]
        q = (q * cos) + (self.rotate_half(q) * sin)
        k = (k * cos) + (self.rotate_half(k) * sin)
        return q, k


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

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


class MultiHeadAttention(nn.Module):
    def __init__(self, n_heads, d_model, max_seq_len, rope_base):
        super().__init__()
        self.n_heads = n_heads
        self.d_k = d_model // n_heads
        self.d_model = d_model
        self.W_Q = nn.Linear(d_model, d_model, bias=False)
        self.W_K = nn.Linear(d_model, d_model, bias=False)
        self.W_V = nn.Linear(d_model, d_model, bias=False)
        self.W_O = nn.Linear(d_model, d_model, bias=False)
        self.rope = RoPE(self.d_k, max_seq_len, base=rope_base)
        self.W_O.RESIDUAL_SCALE_INIT = True

    def forward(self, x):
        B, T, _ = x.shape
        Q = self.W_Q(x).view(B, T, self.n_heads, self.d_k).permute(0, 2, 1, 3)
        K = self.W_K(x).view(B, T, self.n_heads, self.d_k).permute(0, 2, 1, 3)
        V = self.W_V(x).view(B, T, self.n_heads, self.d_k).permute(0, 2, 1, 3)
        Q, K = self.rope(Q, K)
        out = F.scaled_dot_product_attention(Q, K, V, is_causal=True)
        out = out.permute(0, 2, 1, 3).contiguous().view(B, T, self.d_model)
        return self.W_O(out)


class SwiGLU(nn.Module):
    def __init__(self, d_model):
        super().__init__()
        hidden = int(2 / 3 * 4 * d_model)
        hidden = (hidden + 63) // 64 * 64
        self.w1 = nn.Linear(d_model, hidden, bias=False)
        self.w2 = nn.Linear(hidden, d_model, bias=False)
        self.w3 = nn.Linear(d_model, hidden, bias=False)
        self.w2.RESIDUAL_SCALE_INIT = True

    def forward(self, x):
        return self.w2(F.silu(self.w1(x)) * self.w3(x))


class TransformerBlock(nn.Module):
    def __init__(self, d_model, n_heads, max_seq_len, rope_base):
        super().__init__()
        self.attn = MultiHeadAttention(n_heads, d_model, max_seq_len, rope_base)
        self.ff = SwiGLU(d_model)
        self.norm1 = RMSNorm(d_model)
        self.norm2 = RMSNorm(d_model)

    def forward(self, x):
        x = x + self.attn(self.norm1(x))
        x = x + self.ff(self.norm2(x))
        return x


class ParchmentModel(PreTrainedModel):
    config_class = ParchmentConfig
    base_model_prefix = "model"

    def __init__(self, config: ParchmentConfig):
        super().__init__(config)
        self.embeddings = Embeddings(config.vocab_size, config.d_model)
        self.blocks = nn.ModuleList([
            TransformerBlock(config.d_model, config.n_heads, config.max_seq_len, config.rope_base)
            for _ in range(config.n_layers)
        ])
        self.norm = RMSNorm(config.d_model, eps=config.rms_norm_eps)
        self.post_init()

    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            std = 0.02
            if hasattr(module, "RESIDUAL_SCALE_INIT"):
                std /= math.sqrt(2 * self.config.n_layers)
            nn.init.normal_(module.weight, mean=0.0, std=std)
            if module.bias is not None:
                nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            nn.init.normal_(module.weight, mean=0, std=self.config.d_model ** -0.5)

    def forward(self, input_ids: torch.LongTensor) -> torch.Tensor:
        x = self.embeddings(input_ids)
        for block in self.blocks:
            x = block(x)
        return self.norm(x)


class ParchmentForCausalLM(PreTrainedModel, GenerationMixin):
    config_class = ParchmentConfig
    base_model_prefix = "model"
    _tied_weights_keys = {"lm_head.weight": "model.embeddings.embeds.weight"}
    _keys_to_ignore_on_load_missing = [r"lm_head\.weight", r".*\.rope\."]
    _supports_cache_class = False
    _supports_static_cache = False

    def __init__(self, config: ParchmentConfig):
        super().__init__(config)
        self.model = ParchmentModel(config)
        self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False)
        self.post_init()

    def get_input_embeddings(self):
        return self.model.embeddings.embeds

    def set_input_embeddings(self, value):
        self.model.embeddings.embeds = value

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, value):
        self.lm_head = value

    def _init_weights(self, module):
        self.model._init_weights(module)

    def forward(
        self,
        input_ids: torch.LongTensor,
        attention_mask: torch.Tensor | None = None,
        labels: torch.LongTensor | None = None,
        **kwargs,
    ) -> CausalLMOutputWithPast:
        hidden = self.model(input_ids)
        logits = self.lm_head(hidden)

        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 CausalLMOutputWithPast(loss=loss, logits=logits)

    def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **kwargs):
        return {
            "input_ids": input_ids[:, -self.config.max_seq_len:],
            "attention_mask": attention_mask,
        }