huggingface / huggingface/diffusers

process_encoder_hidden_states function is applied on each denoising step

Offen
#9,625 6 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen
stale
Vorherrschende Sprache
Python
Sterne
34.5k
Forks
7.3k
Ø Merge
3 T. 3 Std.
Gemergte PRs (30 T.)
91

Beschreibung

https://github.com/huggingface/diffusers/blob/31058cdaef63ca660a1a045281d156239fba8192/src/diffusers/models/unets/unet_2d_condition.py#L1004

Hey! I noticed that `process_encoder_hidden_states` is applied on each denoising step that influences on an inference performance. This function can be executed only once before running UNet2DConditionModel.forward. The output of `process_encoder_hidden_states` should be combined with `encoder_hidden_states` and be provided to UNet2DConditionModel.forward

Beitragsleitfaden

Beitragsleitfaden öffnen

Rechercherichtung

Start at src/diffusers/models/unets/unet_2d_condition.py around the referenced process_encoder_hidden_states call and inspect how its result enters UNet2DConditionModel.forward. Trace the denoising-step caller to determine where the processed encoder states can be prepared once; done means forward receives the combined states without repeating that processing on every step.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Bewertung

Tech-Stack
python, pytorch
Bereich
machine-learning, performance
Issue-Typ
Refactoring
Schwierigkeit
4/5
Geschätzter Aufwand
3-5 Tage
Aktivitätsstatus
Veraltet
Klarheit
Größtenteils klar
Anfängerfreundlichkeit
35/100

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.