update

app_high_res.py

@@ -12,7 +12,8 @@ from transformers import AutoProcessor, Idefics2ForConditionalGeneration
 from models.conversation import conv_templates
 from typing import List
 processor = AutoProcessor.from_pretrained("Mantis-VL/mantis-8b-idefics2-video-eval-high-res-35k-mantis-2epoch_4096")
-model = Idefics2ForConditionalGeneration.from_pretrained("Mantis-VL/mantis-8b-idefics2-video-eval-high-res-35k-mantis-2epoch_4096", device_map="auto", torch_dtype=torch.float16)
+model = Idefics2ForConditionalGeneration.from_pretrained("Mantis-VL/mantis-8b-idefics2-video-eval-high-res-35k-mantis-2epoch_4096", torch_dtype=torch.float16)
+model = model.to("cuda") if model.device.type != "cuda" else model
 MAX_NUM_FRAMES = 24
 conv_template = conv_templates["idefics_2"]
 
@@ -61,7 +62,7 @@ all the frames of video are as follows:
 @spaces.GPU
 def generate(text:str, images:List[Image.Image], history: List[dict], **kwargs):
     global processor, model
-    model = model.to("cuda") if model.device.type != "cuda" else model
+    # model = model.to("cuda") if model.device.type != "cuda" else model
     if not images:
         images = None