Upload app.py

app.py

@@ -6,16 +6,18 @@ from PIL import Image
 import numpy as np
 import transformers
 from typing import Dict, Optional, Sequence, List
-import spaces
+
+import subprocess
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
 import sys
+# sys.path.append('/mnt/lzy/oryx-demo')
 from oryx.conversation import conv_templates, SeparatorStyle
 from oryx.model.builder import load_pretrained_model
 from oryx.utils import disable_torch_init
 from oryx.mm_utils import tokenizer_image_token, get_model_name_from_path, KeywordsStoppingCriteria, process_anyres_video_genli
 from oryx.constants import IGNORE_INDEX, DEFAULT_IMAGE_TOKEN, IMAGE_TOKEN_INDEX
 
-
 model_path = "THUdyh/Oryx-7B"
 model_name = get_model_name_from_path(model_path)
 overwrite_config = {}
@@ -78,7 +80,7 @@ def preprocess_qwen(sources, tokenizer: transformers.PreTrainedTokenizer, has_im
     targets = torch.tensor(targets, dtype=torch.long)
     return input_ids
 
-@spaces.GPU(duration=120)
+
 def oryx_inference(video, text):
     vr = VideoReader(video, ctx=cpu(0))
     total_frame_num = len(vr)