From 0ae878712ee6f889d4c03bd4ef46078e1fcfca2e Mon Sep 17 00:00:00 2001 From: turboderp <11859846+turboderp@users.noreply.github.com> Date: Wed, 25 Jun 2025 23:56:21 +0200 Subject: [PATCH] Exl3: Clear image embedding cache on unload --- backends/exllamav3/model.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/backends/exllamav3/model.py b/backends/exllamav3/model.py index 2cfd196..28df4e2 100644 --- a/backends/exllamav3/model.py +++ b/backends/exllamav3/model.py @@ -24,6 +24,7 @@ from loguru import logger from backends.base_model_container import BaseModelContainer from backends.exllamav3.sampler import ExllamaV3SamplerBuilder +from backends.exllamav3.vision import clear_image_embedding_cache from common.concurrency import iterate_in_threadpool from common.gen_logging import ( log_generation_params, @@ -509,6 +510,9 @@ class ExllamaV3Container(BaseModelContainer): # Wait for other jobs to finish await self.wait_for_jobs(kwargs.get("skip_wait")) + # Clear the image embedding cache + clear_image_embedding_cache() + self.model.unload() self.model = None self.config = None