rope.py

# Copyright (c) 2024, NVIDIA Corporation & Affiliates. All rights reserved. 
# 
# This work is made available under the Nvidia Source Code License-NC. 
# To view a copy of this license, visit 
# https://github.com/NVlabs/QLIP/blob/main/LICENSE

# MIT License

# Copyright (c) 2022 BAAI-Vision

# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:

# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.

# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.


from math import pi
import torch
from torch import nn
from einops import rearrange, repeat
import logging


def broadcat(tensors, dim = -1):
    num_tensors = len(tensors)
    shape_lens = set(list(map(lambda t: len(t.shape), tensors)))
    assert len(shape_lens) == 1, 'tensors must all have the same number of dimensions'
    shape_len = list(shape_lens)[0]
    dim = (dim + shape_len) if dim < 0 else dim
    dims = list(zip(*map(lambda t: list(t.shape), tensors)))
    expandable_dims = [(i, val) for i, val in enumerate(dims) if i != dim]
    assert all([*map(lambda t: len(set(t[1])) <= 2, expandable_dims)]), 'invalid dimensions for broadcastable concatentation'
    max_dims = list(map(lambda t: (t[0], max(t[1])), expandable_dims))
    expanded_dims = list(map(lambda t: (t[0], (t[1],) * num_tensors), max_dims))
    expanded_dims.insert(dim, (dim, dims[dim]))
    expandable_shapes = list(zip(*map(lambda t: t[1], expanded_dims)))
    tensors = list(map(lambda t: t[0].expand(*t[1]), zip(tensors, expandable_shapes)))
    return torch.cat(tensors, dim = dim)

def rotate_half(x):
    x = rearrange(x, '... (d r) -> ... d r', r = 2)
    x1, x2 = x.unbind(dim = -1)
    x = torch.stack((-x2, x1), dim = -1)
    return rearrange(x, '... d r -> ... (d r)')


class VisionRotaryEmbeddingFast(nn.Module):
    def __init__(
        self,
        dim,
        pt_seq_len,
        ft_seq_len=None,
        custom_freqs = None,
        freqs_for = 'lang',
        theta = 10000,
        max_freq = 10,
        num_freqs = 1,
        patch_dropout = 0.
    ):
        super().__init__()
        if custom_freqs:
            freqs = custom_freqs
        elif freqs_for == 'lang':
            freqs = 1. / (theta ** (torch.arange(0, dim, 2)[:(dim // 2)].float() / dim))
        elif freqs_for == 'pixel':
            freqs = torch.linspace(1., max_freq / 2, dim // 2) * pi
        elif freqs_for == 'constant':
            freqs = torch.ones(num_freqs).float()
        else:
            raise ValueError(f'unknown modality {freqs_for}')

        if ft_seq_len is None: ft_seq_len = pt_seq_len
        t = torch.arange(ft_seq_len) / ft_seq_len * pt_seq_len

        freqs = torch.einsum('..., f -> ... f', t, freqs)
        freqs = repeat(freqs, '... n -> ... (n r)', r = 2)
        freqs = broadcat((freqs[:, None, :], freqs[None, :, :]), dim = -1)

        freqs_cos = freqs.cos().view(-1, freqs.shape[-1])
        freqs_sin = freqs.sin().view(-1, freqs.shape[-1])

        self.patch_dropout = patch_dropout

        self.register_buffer("freqs_cos", freqs_cos)
        self.register_buffer("freqs_sin", freqs_sin)

        logging.info(f'Shape of rope freq: {self.freqs_cos.shape}')

    def forward(self, t, patch_indices_keep=None):
        if patch_indices_keep is not None:
            batch = t.size()[0]
            batch_indices = torch.arange(batch)
            batch_indices = batch_indices[..., None]

            freqs_cos = repeat(self.freqs_cos, 'i j -> n i m j', n=t.shape[0], m=t.shape[1])
            freqs_sin = repeat(self.freqs_sin, 'i j -> n i m j', n=t.shape[0], m=t.shape[1])

            freqs_cos = freqs_cos[batch_indices, patch_indices_keep]
            freqs_cos = rearrange(freqs_cos, 'n i m j -> n m i j')
            freqs_sin = freqs_sin[batch_indices, patch_indices_keep]
            freqs_sin = rearrange(freqs_sin, 'n i m j -> n m i j')

            return  t * freqs_cos + rotate_half(t) * freqs_sin

        return  t * self.freqs_cos + rotate_half(t) * self.freqs_sin