Create app.py

app.py

@@ -0,0 +1,38 @@
+import pathlib
+
+import gradio as gr
+import open_clip
+import torch
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+model, _, transform = open_clip.create_model_and_transforms(
+    "coca_ViT-L-14",
+    pretrained="mscoco_finetuned_laion2B-s13B-b90k"
+)
+model.to(device)
+
+
+def output_generate(image):
+    im = transform(image).unsqueeze(0).to(device)
+    with torch.no_grad(), torch.cuda.amp.autocast():
+        generated = model.generate(im, seq_len=20)
+    return open_clip.decode(generated[0].detach()).split("<end_of_text>")[0].replace("<start_of_text>", "")
+
+
+paths = sorted(pathlib.Path("images").glob("*.jpg"))
+
+iface = gr.Interface(
+    fn=output_generate,
+    inputs=gr.Image(label="Input image", type="pil"),
+    outputs=gr.Text(label="Caption output"),
+    title="CoCa: Contrastive Captioners",
+    description=(
+        """<br> An open source implementation of <strong>CoCa: Contrastive Captioners are Image-Text Foundation Models</strong> <a href=https://arxiv.org/abs/2205.01917>https://arxiv.org/abs/2205.01917.</a>
+        <br> Built using <a href=https://github.com/mlfoundations/open_clip>open_clip</a> with an effort from <a href=https://laion.ai/>LAION</a>. 
+        <br> For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings.<a href="https://huggingface.co/spaces/laion/CoCa?duplicate=true"> <img style="margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>"""
+    ),
+    article="""""",
+    examples=[path.as_posix() for path in paths],
+)
+iface.launch()