41 lines
2.1 KiB
Python
41 lines
2.1 KiB
Python
import torch
|
|
from einops import rearrange
|
|
from torch import Tensor
|
|
|
|
|
|
def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor,pe_q = None, attention_mask = None) -> Tensor:
|
|
if pe_q is None:
|
|
q, k = apply_rope(q, k, pe)
|
|
x = torch.nn.functional.scaled_dot_product_attention(q, k, v,attn_mask=attention_mask)
|
|
x = rearrange(x, "B H L D -> B L (H D)")
|
|
return x
|
|
else:
|
|
q, k = apply_rope_qk(q, k, pe_q, pe)
|
|
x = torch.nn.functional.scaled_dot_product_attention(q, k, v,attn_mask=attention_mask)
|
|
x = rearrange(x, "B H L D -> B L (H D)")
|
|
return x
|
|
|
|
def rope(pos: Tensor, dim: int, theta: int) -> Tensor:
|
|
assert dim % 2 == 0
|
|
scale = torch.arange(0, dim, 2, dtype=torch.float64, device=pos.device) / dim # dim =16 + 56 + 56
|
|
omega = 1.0 / (theta**scale) # 64 omega
|
|
out = torch.einsum("...n,d->...nd", pos, omega)
|
|
out = torch.stack([torch.cos(out), -torch.sin(out), torch.sin(out), torch.cos(out)], dim=-1)
|
|
out = rearrange(out, "b n d (i j) -> b n d i j", i=2, j=2)
|
|
return out.float()
|
|
|
|
|
|
def apply_rope(xq: Tensor, xk: Tensor, freqs_cis: Tensor) -> tuple[Tensor, Tensor]:
|
|
xq_ = xq.float().reshape(*xq.shape[:-1], -1, 1, 2)
|
|
xk_ = xk.float().reshape(*xk.shape[:-1], -1, 1, 2)
|
|
xq_out = freqs_cis[:, :, :xq_.shape[2], :, :, 0] * xq_[..., 0] + freqs_cis[:, :, :xq_.shape[2], :, :, 1] * xq_[..., 1]
|
|
xk_out = freqs_cis[:, :, :xk_.shape[2], :, :, 0] * xk_[..., 0] + freqs_cis[:, :, :xk_.shape[2], :, :, 1] * xk_[..., 1]
|
|
return xq_out.reshape(*xq.shape).type_as(xq), xk_out.reshape(*xk.shape).type_as(xk)
|
|
|
|
def apply_rope_qk(xq: Tensor, xk: Tensor, freqs_cis_q: Tensor,freqs_cis_k: Tensor) -> tuple[Tensor, Tensor]:
|
|
xq_ = xq.float().reshape(*xq.shape[:-1], -1, 1, 2)
|
|
xk_ = xk.float().reshape(*xk.shape[:-1], -1, 1, 2)
|
|
xq_out = freqs_cis_q[:, :, :xq_.shape[2], :, :, 0] * xq_[..., 0] + freqs_cis_q[:, :, :xq_.shape[2], :, :, 1] * xq_[..., 1]
|
|
xk_out = freqs_cis_k[:, :, :xk_.shape[2], :, :, 0] * xk_[..., 0] + freqs_cis_k[:, :, :xk_.shape[2], :, :, 1] * xk_[..., 1]
|
|
return xq_out.reshape(*xq.shape).type_as(xq), xk_out.reshape(*xk.shape).type_as(xk)
|