-
-
Notifications
You must be signed in to change notification settings - Fork 41
/
example.py
35 lines (27 loc) · 925 Bytes
/
example.py
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
"""
This script demonstrates the usage of the FlashAttention module from zeta.nn.
"""
import torch
from zeta.nn import FlashAttention
# Set random seed for reproducibility
torch.manual_seed(42)
# Define input tensor shapes
batch_size, num_heads, seq_len_q, d_head = 2, 4, 6, 8
seq_len_kv = 10
# Create random input tensors
q = torch.randn(batch_size, num_heads, seq_len_q, d_head)
k = torch.randn(batch_size, num_heads, seq_len_kv, d_head)
v = torch.randn(batch_size, num_heads, seq_len_kv, d_head)
# Initialize FlashAttention module
attention = FlashAttention(causal=False, dropout=0.1, flash=False)
print("FlashAttention configuration:", attention)
# Perform attention operation
output = attention(q, k, v)
print(f"Output shape: {output.shape}")
# Optional: Add assertion to check expected output shape
assert output.shape == (
batch_size,
num_heads,
seq_len_q,
d_head,
), "Unexpected output shape"