Replace with flash attention#1133

Open

liym27 wants to merge 23 commits intoPaddlePaddle:developfrom

liym27:replace_with_flash_attention

Commits on Jan 20, 2025

Commits on Jan 21, 2025

temporary adjustment dataloader_num_workers=0
jeff41404
committed

Commits on Jan 22, 2025

move ViT in dataloader to support auto_parallel sharding parallel
jeff41404
committed

Commits on Feb 8, 2025

Merge branch 'develop' into verify_auto_parallel_intermediate_api_in_paddlemix
jeff41404
authored

Commits on Feb 10, 2025

update preprocess of image to align loss of hybrid qwen2vl
jeff41404
committed

Commits on Feb 13, 2025

modify the input to 2 fields to meet the requirement of dynamic to static conversion
jeff41404
committed

Commits on Feb 18, 2025

modify model to get input_ids, attention_mask, inputs_embeds, labels from input_ids
jeff41404
committed

Commits on Mar 4, 2025

Commits on Mar 6, 2025

fix vit embed
Xing-lil
committed

Commits on Mar 11, 2025

Commits on Mar 12, 2025

recover modeling_qwen2_vl
jeff41404
committed

Commits on Mar 13, 2025

Commits on Mar 17, 2025