Liang Su 3cec067795 perf(pi052): optimize flow and full-training paths (#3974) ...

* perf(pi052): optimize equivalent training paths

* fix(pi052): guard FlexAttention backend selection

2026-07-15 17:26:55 +02:00

..

test_pi052_attention_masking.py

style: compact comments in language runtime

2026-07-15 13:52:52 +02:00

test_pi052_bucketed_ce.py

perf(pi052): optimize flow and full-training paths (#3974)

2026-07-15 17:26:55 +02:00

test_pi052_fast_action_loss.py

style: compact comments in language runtime

2026-07-15 13:52:52 +02:00

test_pi052_flex_attention.py

perf(pi052): optimize flow and full-training paths (#3974)

2026-07-15 17:26:55 +02:00

test_pi052_runtime_adapter.py

style: add missing license headers

2026-07-15 13:42:45 +02:00

test_pi052_sdpa_attention.py

refactor(pi052): split pi05_backbone into pi_gemma + modeling_pi052

2026-06-04 20:18:18 +02:00

test_pi052_text_processor.py

refactor(factory): remove PI052 processor overrides

2026-07-15 16:07:14 +02:00

test_pi052_vision_checkpointing.py

perf(pi052): optimize flow and full-training paths (#3974)

2026-07-15 17:26:55 +02:00