Accueil Explorer Aide
S'inscrire Connexion
david
/
aphrodite-engine
miroir de https://github.com/PygmalionAI/aphrodite-engine
1
0
Fork 0
Fichiers Tickets 0 Wiki
Branche: get_last_latency_hack
Branches Tags
aphrodite-engin...
/
docs
/
pages
/
developer
/
input-processing.md

input-processing.md 1.2 KB
Lien permanent Historique Raw

outline: deep

Input Processing

Each model can override parts of Aphrodite's input processing pipeline via INPUT_REGISTRY and MULTIMODAL_REGISTRY.

Currently, this mechanism is only utilized in multi-modal models for preprocessing multi-modal input data in addition to input prompt, but it can be extended to text-only language models when needed.

Input Processing Pipeline

  1. Input data is passed to AphroditeEngine (or AsyncAphrodite for asynchronous inference).
  2. Tokenize the data if necessary.
  3. Process the inputs using INPUT_REGISTRY.process_input.
  4. Send the processed inputs to ExecutorBase.
  5. Distribute the inputs via WorkerBase to ModelRunnerBase.
  6. If the data contains multi-modal data, convert it into keyword arguments using MULTIMODAL_REGISTRY.map_input.
  7. For example, convert a PIL.Image.Image input to its pixel values for a vision model.