admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-06-24 11:47:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update src/lerobot/policies/xvla/modeling_florence2.py

Browse Source 
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Signed-off-by: Maxime Ellerbach <maxime@ellerbach.net>
This commit is contained in:
Maxime Ellerbach
2026-03-27 16:13:14 +01:00
committed by Steven Palma
parent cf7f4fda52
commit 6260ffa159
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/policies/xvla/modeling_florence2.py
+1 -1
View File
@@ -909,7 +909,7 @@ class Florence2FlashAttention2(Florence2Attention):
# TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
# flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignment, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
# Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal("2.10")
self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal("2.1.0")
def _reshape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
return tensor.view(bsz, seq_len, self.num_heads, self.head_dim)