Update src/lerobot/policies/wall_x/qwen_model/qwen2_5_vl_moe.py

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 16:13:23 +01:00
committed by Steven Palma
parent 6260ffa159
commit 0b957bb4b8
@@ -890,7 +890,7 @@ class Qwen2_5_VLFlashAttention2(Qwen2_5_VLAttention):
# 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 forward(
self,