[Bugfix/Feature] Remove Hardcoded Flash Attention in Bagel & Support GQA in SDPA Backend (#3728)

Signed-off-by: Alex Brooks <albrooks@redhat.com>
This commit is contained in:
Alex Brooks authored and GitHub committed 2026-05-29 10:36:15 +08:00
1 parent 593c4c06aa
commit 68a1d23e4d
5 files changed
+151 -451

No files matched your search

File diff suppressed because it is too large. Load diff
@@ -793,13 +793,7 @@ class BagelPipeline(nn.Module, SupportsComponentDiscovery, DiffusionPipelineProf
cfg_renorm_type=gen_params.cfg_renorm_type,
**generation_input,
cfg_text_packed_position_ids=generation_input_cfg_text["cfg_packed_position_ids"],
cfg_text_packed_query_indexes=generation_input_cfg_text["cfg_packed_query_indexes"],
cfg_text_key_values_lens=generation_input_cfg_text["cfg_key_values_lens"],
cfg_text_packed_key_value_indexes=generation_input_cfg_text["cfg_packed_key_value_indexes"],
cfg_img_packed_position_ids=generation_input_cfg_img["cfg_packed_position_ids"],
cfg_img_packed_query_indexes=generation_input_cfg_img["cfg_packed_query_indexes"],
cfg_img_key_values_lens=generation_input_cfg_img["cfg_key_values_lens"],
cfg_img_packed_key_value_indexes=generation_input_cfg_img["cfg_packed_key_value_indexes"],
return_trajectory_latents=req.sampling_params.return_trajectory_latents,
scheduler=self.scheduler,
scheduler_kwargs=self.scheduler_kwargs,