prompt_cache ¶
Best-effort prompt-embedding cache shared by the MLX entrypoints.
Functions:¶
fastvideo.mlx_runtime.prompt_cache.load_prompt_cache ¶
Load a matching cache entry, treating every cache failure as a miss.
Source code in fastvideo/mlx_runtime/prompt_cache.py
fastvideo.mlx_runtime.prompt_cache.save_prompt_cache ¶
Atomically publish an integrity-bound cache entry when possible.
Source code in fastvideo/mlx_runtime/prompt_cache.py
fastvideo.mlx_runtime.prompt_cache.text_encoder_fingerprint ¶
Return a cheap identity for the tokenizer and text-encoder files.