[test+operators] Fuse llama attention to circle attention#217

Closed

glistening wants to merge 21 commits intoSamsung:mainfrom

glistening:attention

Commits on Nov 4, 2025

[test+operators] Fuse attention to circle attention
Sanggyu Lee
committed
Rename model.py to layer.py
Sanggyu Lee
committed
make lint happy by making code ugly
Sanggyu Lee
committed
Fix local-silent but CI-loud lint error
Sanggyu Lee
committed
Add wq,wk,wv,wo and remove_unused_input pass
Sanggyu Lee
committed
Use recording_input in layer.py
Sanggyu Lee
committed
Add prefill.py
Sanggyu Lee
committed
Update layer.py
Sanggyu Lee
committed
Rename model.py to layers.py
Sanggyu Lee
committed
Update input_to_remove comment for prefill.py
Sanggyu Lee
committed
Factor out attention fuser to op_circle_attention.py
Sanggyu Lee
committed
move op_circle_attention.py to onert/op_attention.py
Sanggyu Lee
committed
remove unused import from op_attention.py
Sanggyu Lee
committed
Adjust input prompt size and kv_cache size = 12
Sanggyu Lee
committed
remove @torch.library.impl("circle::attention.llama", "CPU")
Sanggyu Lee
committed
Remove attention_mask and make kv_cache mandatory, not optional
Sanggyu Lee
committed
add decode.py to export LlamaModel decode phase
Sanggyu Lee
committed
Restore attention_mask
Sanggyu Lee
committed
Fix wrong arg order and move layer_idx from inputs to params
Sanggyu Lee
committed
remove layer_idx
Sanggyu Lee
committed
Remove remove_unused_inputs pass
Sanggyu Lee
committed