File size: 1,751 Bytes
59d0297 | 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 |
import torch
import torch.nn.functional as F
def _build_sink_window_mask(T, window_size, num_sink_tokens, device):
"""Build causal + sliding window + sink tokens attention mask."""
# Start with causal mask
row = torch.arange(T, device=device).unsqueeze(1)
col = torch.arange(T, device=device).unsqueeze(0)
causal = col <= row
# Add window constraint: attend to [max(0, i - window_size + 1) : i + 1]
window = (row - col) < window_size
# Sink tokens: always attend to first num_sink_tokens positions
sink = col < num_sink_tokens
# Combine: causal AND (within window OR sink token)
mask = causal & (window | sink)
return mask
def flash_attn_func(q, k, v, causal=False, window_size=None, num_sink_tokens=0):
# q: [B, T, H, D], k: [B, T, Hkv, D], v: [B, T, Hkv, D]
B, T, H, D = q.shape
Hkv = k.shape[2]
group = H // Hkv
# Expand KV heads for GQA
if group > 1:
k = k.unsqueeze(3).expand(B, T, Hkv, group, D).reshape(B, T, H, D)
v = v.unsqueeze(3).expand(B, T, Hkv, group, D).reshape(B, T, H, D)
# [B, T, H, D] -> [B, H, T, D] for SDPA
q = q.transpose(1, 2)
k = k.transpose(1, 2)
v = v.transpose(1, 2)
use_window = window_size is not None and window_size != (-1, -1)
if use_window or num_sink_tokens > 0:
ws = window_size[0] if isinstance(window_size, tuple) else (window_size or T)
mask = _build_sink_window_mask(T, ws, num_sink_tokens, q.device)
out = F.scaled_dot_product_attention(q, k, v, attn_mask=mask)
elif causal:
out = F.scaled_dot_product_attention(q, k, v, is_causal=True)
else:
out = F.scaled_dot_product_attention(q, k, v)
return out.transpose(1, 2) # back to [B, T, H, D]
|