update

app.py

@@ -2,33 +2,35 @@ import gradio as gr
 import os
 import torch
 import numpy as np
+
+os.system('git clone https://github.com/pytorch/fairseq.git; cd fairseq;'
+          'pip install --use-feature=in-tree-build ./; cd ..')
+
 from fairseq import utils, tasks
-from fairseq.utils import checkpoint_utils
+from utils import checkpoint_utils
 from utils.eval_utils import eval_step
 from tasks.mm_tasks.caption import CaptionTask
 from models.ofa import OFAModel
 from PIL import Image
 from torchvision import transforms
 
-
 # Register caption task
-tasks.register_task('caption',CaptionTask)
+tasks.register_task('caption', CaptionTask)
 # turn on cuda if GPU is available
 use_cuda = torch.cuda.is_available()
 # use fp16 only when GPU is available
 use_fp16 = False
 
-os.system('wget https://ofa-silicon.oss-us-west-1.aliyuncs.com/checkpoints/caption_large_best_clean.pt')
-os.system('mkdir -p checkpoints')
-os.system('mv caption_large_best_clean.pt checkpoints/caption.pt')
+os.system('wget https://ofa-silicon.oss-us-west-1.aliyuncs.com/checkpoints/caption_large_best_clean.pt; '
+          'mkdir -p checkpoints; mv caption_large_best_clean.pt checkpoints/caption.pt')
 
 # Load pretrained ckpt & config
 overrides = {"bpe_dir": "utils/BPE", "eval_cider": False, "beam": 5,
              "max_len_b": 16, "no_repeat_ngram_size": 3, "seed": 7}
 models, cfg, task = checkpoint_utils.load_model_ensemble_and_task(
-        utils.split_paths('checkpoints/caption.pt'),
-        arg_overrides=overrides
-    )
+    utils.split_paths('checkpoints/caption.pt'),
+    arg_overrides=overrides
+)
 
 # Move models to GPU
 for model in models:
@@ -109,4 +111,4 @@ def image_caption(inp):
 
 
 io = gr.Interface(fn=image_caption, inputs=gr.inputs.Image(type='pil'), outputs='text')
-io.launch(enable_queue=True)
+io.launch(enable_queue=True)

requirements.txt

@@ -1 +1,4 @@
-fairseq
+ftfy==6.0.3
+tensorboardX==2.4.1
+pycocotools==2.0.4
+pycocoevalcap==1.2