udpate

app_high_res.py

@@ -11,12 +11,10 @@ import functools
 from transformers import AutoProcessor, Idefics2ForConditionalGeneration
 from models.conversation import conv_templates
 from typing import List
-import subprocess
-subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
 
 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", torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2")
+model = Idefics2ForConditionalGeneration.from_pretrained("Mantis-VL/mantis-8b-idefics2-video-eval-high-res-35k-mantis-2epoch_4096", torch_dtype=torch.bfloat16)
 MAX_NUM_FRAMES = 24
 conv_template = conv_templates["idefics_2"]