File size: 6,361 Bytes
31dc8dc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn

from functools import lru_cache
from typing import Any


def apply_rotary_emb(
    x: torch.Tensor,
    cos: torch.Tensor,
    sin: torch.Tensor,
) -> torch.Tensor:
    cos = cos.unsqueeze(-2)
    sin = sin.unsqueeze(-2)
    x1, x2 = torch.chunk(x.to(torch.float32), 2, dim=-1)
    y1 = x1 * cos - x2 * sin
    y2 = x2 * cos + x1 * sin
    return torch.cat((y1, y2), dim=-1).to(x.dtype)


class RotaryEmbedding(nn.Module):
    def __init__(
        self,
        head_size: int,
        rotary_dim: int,
        max_position_embeddings: int,
        base: float,
    ) -> None:
        super().__init__()
        self.head_size = head_size
        assert rotary_dim == head_size
        inv_freq = 1.0 / (base ** (torch.arange(0, rotary_dim, 2, dtype=torch.float) / rotary_dim))
        t = torch.arange(max_position_embeddings, dtype=torch.float)
        freqs = torch.einsum("i,j -> ij", t, inv_freq)
        cos = freqs.cos()
        sin = freqs.sin()
        cache = torch.cat((cos, sin), dim=-1)
        self.register_buffer("cos_sin_cache", cache, persistent=False)

    @torch.compile
    def forward(
        self,
        positions: torch.Tensor,
        query: torch.Tensor,
        key: torch.Tensor,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        cos_sin = self.cos_sin_cache[positions]
        cos, sin = cos_sin.chunk(2, dim=-1)
        query_shape = query.shape
        key_shape = key.shape
        if query.dim() == 2:
            q_tokens = query.size(0)
            nheads_q = query_shape[-1] // self.head_size
            query = query.view(q_tokens, nheads_q, self.head_size)
            query = apply_rotary_emb(query, cos, sin).view(query_shape)
        elif query.dim() == 3:
            query = apply_rotary_emb(query, cos, sin)
        else:
            raise ValueError(f"Unsupported query ndim for RotaryEmbedding: {query.dim()}")

        if key.dim() == 2:
            k_tokens = key.size(0)
            nheads_k = key_shape[-1] // self.head_size
            key = key.view(k_tokens, nheads_k, self.head_size)
            key = apply_rotary_emb(key, cos, sin).view(key_shape)
        elif key.dim() == 3:
            key = apply_rotary_emb(key, cos, sin)
        else:
            raise ValueError(f"Unsupported key ndim for RotaryEmbedding: {key.dim()}")
        return query, key


class PartialRotaryEmbedding(nn.Module):
    def __init__(
        self,
        head_size: int,
        rotary_dim: int,
        max_position_embeddings: int,
        base: float,
    ) -> None:
        super().__init__()
        if rotary_dim <= 0 or rotary_dim > head_size or rotary_dim % 2 != 0:
            raise ValueError(f"Invalid rotary_dim={rotary_dim} for head_size={head_size}.")
        self.head_size = head_size
        self.rotary_dim = rotary_dim
        inv_freq = 1.0 / (base ** (torch.arange(0, rotary_dim, 2, dtype=torch.float) / rotary_dim))
        t = torch.arange(max_position_embeddings, dtype=torch.float)
        freqs = torch.einsum("i,j -> ij", t, inv_freq)
        cos = freqs.cos()
        sin = freqs.sin()
        cache = torch.cat((cos, sin), dim=-1)
        self.register_buffer("cos_sin_cache", cache, persistent=False)

    @torch.compile
    def _apply_rope(
        self,
        positions: torch.Tensor,
        x: torch.Tensor,
    ) -> torch.Tensor:
        cos_sin = self.cos_sin_cache[positions]
        cos, sin = cos_sin.chunk(2, dim=-1)
        x_shape = x.shape
        if x.dim() == 2:
            tokens = x.size(0)
            nheads = x_shape[-1] // self.head_size
            x = x.view(tokens, nheads, self.head_size)
            x_rot = x[..., : self.rotary_dim]
            x_pass = x[..., self.rotary_dim :]
            x_rot = apply_rotary_emb(x_rot, cos, sin)
            return torch.cat((x_rot, x_pass), dim=-1).view(x_shape)
        if x.dim() == 3:
            x_rot = x[..., : self.rotary_dim]
            x_pass = x[..., self.rotary_dim :]
            x_rot = apply_rotary_emb(x_rot, cos, sin)
            return torch.cat((x_rot, x_pass), dim=-1)
        raise ValueError(f"Unsupported x ndim for PartialRotaryEmbedding: {x.dim()}")

    def forward(
        self,
        positions: torch.Tensor,
        query: torch.Tensor,
        key: torch.Tensor,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        return self._apply_rope(positions, query), self._apply_rope(positions, key)


def _normalize_rope_scaling(
    rope_scaling: dict[str, Any] | tuple[tuple[str, Any], ...] | None,
) -> tuple[tuple[str, Any], ...] | None:
    if rope_scaling is None:
        return None
    if isinstance(rope_scaling, tuple):
        return rope_scaling
    if not isinstance(rope_scaling, dict):
        raise TypeError(f"rope_scaling must be a dict, tuple, or None, got: {type(rope_scaling)!r}")
    return tuple(sorted(rope_scaling.items()))


def _validate_rope_scaling(
    rope_scaling: tuple[tuple[str, Any], ...] | None,
) -> None:
    if rope_scaling is None:
        return

    rope_scaling_dict = dict(rope_scaling)
    rope_type = rope_scaling_dict.get("rope_type", rope_scaling_dict.get("type", "default"))

    # HF configs may standardize default rope into a dict form. We accept that and
    # treat it identically to rope_scaling=None because Diffulex currently only
    # implements the default rotary embedding path here.
    if rope_type in ("default", None):
        return

    raise NotImplementedError(
        "Diffulex RotaryEmbedding currently supports only the default rope variant, "
        f"got rope_scaling={rope_scaling_dict}."
    )


@lru_cache(1)
def _get_rope_cached(
    head_size: int,
    rotary_dim: int,
    max_position: int,
    base: float,
    rope_scaling: tuple[tuple[str, Any], ...] | None = None,
):
    _validate_rope_scaling(rope_scaling)
    rotary_cls = RotaryEmbedding if rotary_dim == head_size else PartialRotaryEmbedding
    rotary_emb = rotary_cls(head_size, rotary_dim, max_position, base)
    return rotary_emb


def get_rope(
    head_size: int,
    rotary_dim: int,
    max_position: int,
    base: float,
    rope_scaling: dict | tuple[tuple[str, Any], ...] | None = None,
):
    return _get_rope_cached(
        head_size=head_size,
        rotary_dim=rotary_dim,
        max_position=max_position,
        base=base,
        rope_scaling=_normalize_rope_scaling(rope_scaling),
    )