diff --git a/.gitattributes b/.gitattributes
index a6344aac8c09253b3b630fb776ae94478aa0275b..81e2bc22e060839d515dd267f600afee9fba252d 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+bee.jpg filter=lfs diff=lfs merge=lfs -text
+bird.jpg filter=lfs diff=lfs merge=lfs -text
diff --git a/README.md b/README.md
index 4b1a77e94b5f088af6311b51993b4f74dc14f5bb..ee5be44ae2b10c3d0e00d6a4ada009c839880ab5 100644
--- a/README.md
+++ b/README.md
@@ -1,4 +1,11 @@
----
-license: apache-2.0
-title: Paligemma Hf
----
+---
+title: Paligemma HF
+emoji: 🤗 
+colorFrom: yellow
+colorTo: green
+sdk: gradio
+sdk_version: 4.20.1
+app_file: app.py
+pinned: false
+license: apache-2.0
+---
diff --git a/app.py b/app.py
new file mode 100644
index 0000000000000000000000000000000000000000..76f0e8044971e4849ba8f4adb7364cb265085946
--- /dev/null
+++ b/app.py
@@ -0,0 +1,327 @@
+import gradio as gr
+import PIL.Image
+import transformers
+from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
+import torch
+import os
+import string
+import functools
+import re
+import flax.linen as nn
+import jax
+import jax.numpy as jnp
+import numpy as np
+
+
+hf_token = os.getenv("HF_TOKEN")
+model_id = "google/paligemma-3b-mix-448"
+COLORS = ['#4285f4', '#db4437', '#f4b400', '#0f9d58', '#e48ef1']
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model = PaliGemmaForConditionalGeneration.from_pretrained(model_id, use_auth_token=hf_token).eval().to(device)
+processor = PaliGemmaProcessor.from_pretrained(model_id)
+
+###### Transformers Inference
+
+def infer(
+    image: PIL.Image.Image,
+    text: str,
+    max_new_tokens: int
+) -> str:
+    inputs = processor(text=text, images=image, return_tensors="pt").to(device)
+    with torch.inference_mode():
+      generated_ids = model.generate(
+          **inputs,
+          max_new_tokens=max_new_tokens,
+          do_sample=False
+      )
+    result = processor.batch_decode(generated_ids, skip_special_tokens=True)
+    return result[0][len(text):]
+
+##### Parse segmentation output tokens into masks
+##### Also returns bounding boxes with their labels
+
+def parse_segmentation(input_image, input_text):
+  out = infer(input_image, input_text, max_new_tokens=100)
+  objs = extract_objs(out.lstrip("\n"), input_image.size[0], input_image.size[1], unique_labels=True)
+  labels = set(obj.get('name') for obj in objs if obj.get('name'))
+  color_map = {l: COLORS[i % len(COLORS)] for i, l in enumerate(labels)}
+  highlighted_text = [(obj['content'], obj.get('name')) for obj in objs]
+  annotated_img = (
+    input_image,
+    [
+        (
+            obj['mask'] if obj.get('mask') is not None else obj['xyxy'],
+            obj['name'] or '',
+        )
+        for obj in objs
+        if 'mask' in obj or 'xyxy' in obj
+    ],
+)
+  has_annotations = bool(annotated_img[1])
+  return annotated_img
+
+
+
+######## Demo
+
+INTRO_TEXT = """## PaliGemma demo\n\n
+| [Github](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/paligemma/README.md) 
+| [Blogpost](https://huggingface.co/blog/paligemma) 
+|\n\n
+PaliGemma is an open vision-language model by Google, inspired by [PaLI-3](https://arxiv.org/abs/2310.09199) and 
+built with open components such as the [SigLIP](https://arxiv.org/abs/2303.15343) 
+vision model and the [Gemma](https://arxiv.org/abs/2403.08295) language model. PaliGemma is designed as a versatile 
+model for transfer to a wide range of vision-language tasks such as image and short video caption, visual question 
+answering, text reading, object detection and object segmentation.
+\n\n
+This space includes models fine-tuned on a mix of downstream tasks, **inferred via 🤗 transformers**.
+See the [Blogpost](https://huggingface.co/blog/paligemma) and 
+[README]((https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/paligemma/README.md)) 
+for detailed information how to use and fine-tune PaliGemma models.
+\n\n
+**This is an experimental research model.** Make sure to add appropriate guardrails when using the model for applications.
+"""
+
+
+with gr.Blocks(css="style.css") as demo:
+  gr.Markdown(INTRO_TEXT)
+  with gr.Tab("Conversation"):
+    with gr.Column():
+        image = gr.Image(type="pil")
+        text_input = gr.Text(label="Input Text")
+
+        text_output = gr.Text(label="Text Output")
+        chat_btn = gr.Button()
+        tokens = gr.Slider(
+            label="Max New Tokens",
+            info="Set to larger for longer generation.",
+            minimum=10,
+            maximum=100,
+            value=20,
+            step=10,
+        )
+
+    chat_inputs = [
+        image,
+        text_input,
+        tokens
+        ]
+    chat_outputs = [
+        text_output
+    ]
+    chat_btn.click(
+        fn=infer,
+        inputs=chat_inputs,
+        outputs=chat_outputs,
+    )
+    
+    examples = [["./bee.jpg", "What is on the flower?"],
+               ["./examples/billard1.jpg", "How many red balls are there?"],
+               ["./examples/bowie.jpg", "Who is this?"],
+               ["./examples/emu.jpg", "What animal is this?"],
+               ["./howto.jpg", "What does this image show?"],
+               ["./examples/password.jpg", "What is the password?"],
+               ["./examples/ulges.jpg", "Who is the author of this book?"]]
+    gr.Markdown("Example images are licensed CC0 by [akolesnikoff@](https://github.com/akolesnikoff), [mbosnjak@](https://github.com/mbosnjak), [maximneumann@](https://github.com/maximneumann) and [merve](https://huggingface.co/merve).")
+
+    gr.Examples(
+        examples=examples,
+        inputs=chat_inputs,
+    )
+  with gr.Tab("Segment/Detect"):
+    image = gr.Image(type="pil")
+    seg_input = gr.Text(label="Entities to Segment/Detect")
+    seg_btn = gr.Button("Submit")
+    annotated_image = gr.AnnotatedImage(label="Output")
+    
+    examples = [["./cats.png", "segment cats"],
+                ["./bee.jpg", "detect bee"],
+               ["./examples/barsik.jpg", "segment cat"],
+               ["./bird.jpg", "segment bird ; bird ; plant"]]
+    gr.Markdown("Example images are licensed CC0 by [akolesnikoff@](https://github.com/akolesnikoff), [mbosnjak@](https://github.com/mbosnjak), [maximneumann@](https://github.com/maximneumann) and [merve](https://huggingface.co/merve).")
+    gr.Examples(
+        examples=examples,
+        inputs=[image, seg_input],
+    )
+
+    seg_inputs = [
+        image,
+        seg_input
+        ]
+    seg_outputs = [
+        annotated_image
+    ]
+    seg_btn.click(
+        fn=parse_segmentation,
+        inputs=seg_inputs,
+        outputs=seg_outputs,
+    )
+
+
+
+
+
+### Postprocessing Utils for Segmentation Tokens
+### Segmentation tokens are passed to another VAE which decodes them to a mask
+
+_MODEL_PATH = 'vae-oid.npz'
+
+_SEGMENT_DETECT_RE = re.compile(
+    r'(.*?)' +
+    r'<loc(\d{4})>' * 4 + r'\s*' +
+    '(?:%s)?' % (r'<seg(\d{3})>' * 16) +
+    r'\s*([^;<>]+)? ?(?:; )?',
+)
+
+
+def _get_params(checkpoint):
+  """Converts PyTorch checkpoint to Flax params."""
+
+  def transp(kernel):
+    return np.transpose(kernel, (2, 3, 1, 0))
+
+  def conv(name):
+    return {
+        'bias': checkpoint[name + '.bias'],
+        'kernel': transp(checkpoint[name + '.weight']),
+    }
+
+  def resblock(name):
+    return {
+        'Conv_0': conv(name + '.0'),
+        'Conv_1': conv(name + '.2'),
+        'Conv_2': conv(name + '.4'),
+    }
+
+  return {
+      '_embeddings': checkpoint['_vq_vae._embedding'],
+      'Conv_0': conv('decoder.0'),
+      'ResBlock_0': resblock('decoder.2.net'),
+      'ResBlock_1': resblock('decoder.3.net'),
+      'ConvTranspose_0': conv('decoder.4'),
+      'ConvTranspose_1': conv('decoder.6'),
+      'ConvTranspose_2': conv('decoder.8'),
+      'ConvTranspose_3': conv('decoder.10'),
+      'Conv_1': conv('decoder.12'),
+  }
+
+
+def _quantized_values_from_codebook_indices(codebook_indices, embeddings):
+  batch_size, num_tokens = codebook_indices.shape
+  assert num_tokens == 16, codebook_indices.shape
+  unused_num_embeddings, embedding_dim = embeddings.shape
+
+  encodings = jnp.take(embeddings, codebook_indices.reshape((-1)), axis=0)
+  encodings = encodings.reshape((batch_size, 4, 4, embedding_dim))
+  return encodings
+
+
+@functools.cache
+def _get_reconstruct_masks():
+  """Reconstructs masks from codebook indices.
+  Returns:
+    A function that expects indices shaped `[B, 16]` of dtype int32, each
+    ranging from 0 to 127 (inclusive), and that returns a decoded masks sized
+    `[B, 64, 64, 1]`, of dtype float32, in range [-1, 1].
+  """
+
+  class ResBlock(nn.Module):
+    features: int
+
+    @nn.compact
+    def __call__(self, x):
+      original_x = x
+      x = nn.Conv(features=self.features, kernel_size=(3, 3), padding=1)(x)
+      x = nn.relu(x)
+      x = nn.Conv(features=self.features, kernel_size=(3, 3), padding=1)(x)
+      x = nn.relu(x)
+      x = nn.Conv(features=self.features, kernel_size=(1, 1), padding=0)(x)
+      return x + original_x
+
+  class Decoder(nn.Module):
+    """Upscales quantized vectors to mask."""
+
+    @nn.compact
+    def __call__(self, x):
+      num_res_blocks = 2
+      dim = 128
+      num_upsample_layers = 4
+
+      x = nn.Conv(features=dim, kernel_size=(1, 1), padding=0)(x)
+      x = nn.relu(x)
+
+      for _ in range(num_res_blocks):
+        x = ResBlock(features=dim)(x)
+
+      for _ in range(num_upsample_layers):
+        x = nn.ConvTranspose(
+            features=dim,
+            kernel_size=(4, 4),
+            strides=(2, 2),
+            padding=2,
+            transpose_kernel=True,
+        )(x)
+        x = nn.relu(x)
+        dim //= 2
+
+      x = nn.Conv(features=1, kernel_size=(1, 1), padding=0)(x)
+
+      return x
+
+  def reconstruct_masks(codebook_indices):
+    quantized = _quantized_values_from_codebook_indices(
+        codebook_indices, params['_embeddings']
+    )
+    return Decoder().apply({'params': params}, quantized)
+
+  with open(_MODEL_PATH, 'rb') as f:
+    params = _get_params(dict(np.load(f)))
+
+  return jax.jit(reconstruct_masks, backend='cpu')
+def extract_objs(text, width, height, unique_labels=False):
+  """Returns objs for a string with "<loc>" and "<seg>" tokens."""
+  objs = []
+  seen = set()
+  while text:
+    m = _SEGMENT_DETECT_RE.match(text)
+    if not m:
+      break
+    print("m", m)
+    gs = list(m.groups())
+    before = gs.pop(0)
+    name = gs.pop()
+    y1, x1, y2, x2 = [int(x) / 1024 for x in gs[:4]]
+    
+    y1, x1, y2, x2 = map(round, (y1*height, x1*width, y2*height, x2*width))
+    seg_indices = gs[4:20]
+    if seg_indices[0] is None:
+      mask = None
+    else:
+      seg_indices = np.array([int(x) for x in seg_indices], dtype=np.int32)
+      m64, = _get_reconstruct_masks()(seg_indices[None])[..., 0]
+      m64 = np.clip(np.array(m64) * 0.5 + 0.5, 0, 1)
+      m64 = PIL.Image.fromarray((m64 * 255).astype('uint8'))
+      mask = np.zeros([height, width])
+      if y2 > y1 and x2 > x1:
+        mask[y1:y2, x1:x2] = np.array(m64.resize([x2 - x1, y2 - y1])) / 255.0
+
+    content = m.group()
+    if before:
+      objs.append(dict(content=before))
+      content = content[len(before):]
+    while unique_labels and name in seen:
+      name = (name or '') + "'"
+    seen.add(name)
+    objs.append(dict(
+        content=content, xyxy=(x1, y1, x2, y2), mask=mask, name=name))
+    text = text[len(before) + len(content):]
+
+  if text:
+    objs.append(dict(content=text))
+
+  return objs
+
+#########
+
+if __name__ == "__main__":
+    demo.queue(max_size=10).launch(debug=True)
\ No newline at end of file
diff --git a/bee.jpg b/bee.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..14721a88bb6ac2b717708297518fff64a28e7fc9
--- /dev/null
+++ b/bee.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b21ba78250f852ca5990063866b1ace6432521d0251bde7f8de783b22c99a6d
+size 5373297
diff --git a/bird.jpg b/bird.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..7b791b686cd4bd2b6e92db234b97798b4593a323
--- /dev/null
+++ b/bird.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f1fb355ccd4b46ec882324a8d786abb0f99fc9695b09b3ab86a0a7009573626
+size 1063853
diff --git a/cats.png b/cats.png
new file mode 100644
index 0000000000000000000000000000000000000000..f9b73cdf324b663d7730d881b72c60695f6bd11b
Binary files /dev/null and b/cats.png differ
diff --git a/examples/barsik.jpg b/examples/barsik.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..55f855f13e882e57272a4eed142c919e907b84b6
Binary files /dev/null and b/examples/barsik.jpg differ
diff --git a/examples/barsik.json b/examples/barsik.json
new file mode 100644
index 0000000000000000000000000000000000000000..6d6f13e76e15985824ab27135a8b62d8b278d0dc
--- /dev/null
+++ b/examples/barsik.json
@@ -0,0 +1,7 @@
+{
+  "name": "barsik",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "segment cat",
+  "license": "CC0 by [maximneumann@](https://github.com/maximneumann)"
+}
\ No newline at end of file
diff --git a/examples/biennale.jpg b/examples/biennale.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..05ba1292a74c2842df4b4341ecaf1a1c1ecbcce0
Binary files /dev/null and b/examples/biennale.jpg differ
diff --git a/examples/biennale.json b/examples/biennale.json
new file mode 100644
index 0000000000000000000000000000000000000000..532ff527f32ad4e5fa1ebd71ebacc14d537370e5
--- /dev/null
+++ b/examples/biennale.json
@@ -0,0 +1,7 @@
+{
+  "name": "biennale",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "In which city is this?",
+  "license": "CC0 by [andsteing@](https://huggingface.co/andsteing)"
+}
\ No newline at end of file
diff --git a/examples/billard1.jpg b/examples/billard1.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..2fbf3c5a9e96df8099640c4d9700fccac7063648
Binary files /dev/null and b/examples/billard1.jpg differ
diff --git a/examples/billard1.json b/examples/billard1.json
new file mode 100644
index 0000000000000000000000000000000000000000..2667d173894c20049779f493091cb00be8205d07
--- /dev/null
+++ b/examples/billard1.json
@@ -0,0 +1,7 @@
+{
+  "name": "billard1",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "How many red balls are there?",
+  "license": "CC0 by [mbosnjak@](https://github.com/mbosnjak)"
+}
\ No newline at end of file
diff --git a/examples/billard2.jpg b/examples/billard2.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..a2a65c4b4c837082190bda4c1ec0d95aae757387
Binary files /dev/null and b/examples/billard2.jpg differ
diff --git a/examples/billard2.json b/examples/billard2.json
new file mode 100644
index 0000000000000000000000000000000000000000..1e66dd97b575f666c962436482fc18ee8682493e
--- /dev/null
+++ b/examples/billard2.json
@@ -0,0 +1,7 @@
+{
+  "name": "billard2",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "How many balls are there?",
+  "license": "CC0 by [mbosnjak@](https://github.com/mbosnjak)"
+}
\ No newline at end of file
diff --git a/examples/bowie.jpg b/examples/bowie.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..a470c3fbcd2af4e81af9de46f6ba26f17db81631
Binary files /dev/null and b/examples/bowie.jpg differ
diff --git a/examples/bowie.json b/examples/bowie.json
new file mode 100644
index 0000000000000000000000000000000000000000..deb4dfd631631946765c9e90fa4555822a453e03
--- /dev/null
+++ b/examples/bowie.json
@@ -0,0 +1,7 @@
+{
+  "name": "bowie",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Who is this?",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/branch.jpg b/examples/branch.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..d95595728b845c0ec2b7ee508473541a48d84290
Binary files /dev/null and b/examples/branch.jpg differ
diff --git a/examples/branch.json b/examples/branch.json
new file mode 100644
index 0000000000000000000000000000000000000000..a86c14f5d3fe2f2d0512ce49fc2ab3b9b6012c61
--- /dev/null
+++ b/examples/branch.json
@@ -0,0 +1,7 @@
+{
+  "name": "branch",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What caused this?",
+  "license": "CC0 by [andsteing@](https://huggingface.co/andsteing)"
+}
\ No newline at end of file
diff --git a/examples/cc_fox.jpg b/examples/cc_fox.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..47c95d0a91241833574ccb19b8a355417a87bc7a
Binary files /dev/null and b/examples/cc_fox.jpg differ
diff --git a/examples/cc_fox.json b/examples/cc_fox.json
new file mode 100644
index 0000000000000000000000000000000000000000..69ee0678e50e701e0167097f4c41ed360f449aed
--- /dev/null
+++ b/examples/cc_fox.json
@@ -0,0 +1,7 @@
+{
+  "name": "cc_fox",
+  "comment": "",
+  "model": "paligemma-3b-mix-448",
+  "prompt": "Which breed is this fox?",
+  "license": "CC0 by [XiaohuaZhai@](https://sites.google.com/view/xzhai)"
+}
diff --git a/examples/cc_landscape.jpg b/examples/cc_landscape.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..0a4b610c4234ffa67305b7952584430f4d953dde
Binary files /dev/null and b/examples/cc_landscape.jpg differ
diff --git a/examples/cc_landscape.json b/examples/cc_landscape.json
new file mode 100644
index 0000000000000000000000000000000000000000..c1a66ec2901cd5108c71a690f23cf6ef51a9fbee
--- /dev/null
+++ b/examples/cc_landscape.json
@@ -0,0 +1,7 @@
+{
+  "name": "cc_landscape",
+  "comment": "",
+  "model": "paligemma-3b-mix-448",
+  "prompt": "What does the image show?",
+  "license": "CC0 by [XiaohuaZhai@](https://sites.google.com/view/xzhai)"
+}
diff --git a/examples/cc_puffin.jpg b/examples/cc_puffin.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..ae6bb3dc676dc40b2854c58d8238ff7770f3072d
Binary files /dev/null and b/examples/cc_puffin.jpg differ
diff --git a/examples/cc_puffin.json b/examples/cc_puffin.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ada0c694b286e2b27ace19084b32950ee56adf7
--- /dev/null
+++ b/examples/cc_puffin.json
@@ -0,0 +1,7 @@
+{
+  "name": "cc_puffin",
+  "comment": "",
+  "model": "paligemma-3b-mix-448",
+  "prompt": "detect puffin in the back; puffin in front",
+  "license": "CC0 by [XiaohuaZhai@](https://sites.google.com/view/xzhai)"
+}
diff --git a/examples/couch.jpg b/examples/couch.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..81800961f0498e46fc06ef525311f0a5d88eb4cb
Binary files /dev/null and b/examples/couch.jpg differ
diff --git a/examples/couch.json b/examples/couch.json
new file mode 100644
index 0000000000000000000000000000000000000000..32f4cba01ded6e629661c4f81ec9125f9af8409e
--- /dev/null
+++ b/examples/couch.json
@@ -0,0 +1,7 @@
+{
+  "name": "couch",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "How many yellow cushions are on the couch?",
+  "license": "CC0"
+}
\ No newline at end of file
diff --git a/examples/couch_.json b/examples/couch_.json
new file mode 100644
index 0000000000000000000000000000000000000000..22a288af099703296a1208279484354f88ed5c20
--- /dev/null
+++ b/examples/couch_.json
@@ -0,0 +1,7 @@
+{
+  "name": "couch",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "How many painting do you see in the image?",
+  "license": "CC0"
+}
\ No newline at end of file
diff --git a/examples/cups.jpg b/examples/cups.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..29fb745612887e7a0d4137a503831fe4dd0841d1
Binary files /dev/null and b/examples/cups.jpg differ
diff --git a/examples/cups.json b/examples/cups.json
new file mode 100644
index 0000000000000000000000000000000000000000..078e3df2986f38350c30eaf2e1e1522a842b7664
--- /dev/null
+++ b/examples/cups.json
@@ -0,0 +1,7 @@
+{
+  "name": "cups",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "how many cups?",
+  "license": "CC0 by [mbosnjak@](https://github.com/mbosnjak)"
+}
\ No newline at end of file
diff --git a/examples/dice.jpg b/examples/dice.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..76d0fbabee3a9aa31d3335a850f25b0c40952d70
Binary files /dev/null and b/examples/dice.jpg differ
diff --git a/examples/dice.json b/examples/dice.json
new file mode 100644
index 0000000000000000000000000000000000000000..a3fb3f9703dd6ea055569fba49b4a96b76df8235
--- /dev/null
+++ b/examples/dice.json
@@ -0,0 +1,7 @@
+{
+  "name": "dice",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "segment dice ; dice",
+  "license": "CC0 by [andresusanopinto@](https://github.com/andresusanopinto)"
+}
\ No newline at end of file
diff --git a/examples/emu.jpg b/examples/emu.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..d298e271a108a89cb89473af26bd202902f8b901
Binary files /dev/null and b/examples/emu.jpg differ
diff --git a/examples/emu.json b/examples/emu.json
new file mode 100644
index 0000000000000000000000000000000000000000..23532eac207641e3d138ceb67f9a051d6d231539
--- /dev/null
+++ b/examples/emu.json
@@ -0,0 +1,7 @@
+{
+  "name": "emu",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What animal is this?",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/fridge.jpg b/examples/fridge.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..dd6af3f32f8b3bad650b2162f1b4628d8e5a26db
Binary files /dev/null and b/examples/fridge.jpg differ
diff --git a/examples/fridge.json b/examples/fridge.json
new file mode 100644
index 0000000000000000000000000000000000000000..c6628d78020b331530c4c6a3726c38c454c4da2f
--- /dev/null
+++ b/examples/fridge.json
@@ -0,0 +1,7 @@
+{
+  "name": "fridge",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Describe the image.",
+  "license": "CC0 by [andresusanopinto@](https://github.com/andresusanopinto)"
+}
\ No newline at end of file
diff --git a/examples/givt.jpg b/examples/givt.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b269c132464ecbd8c7ce8f5464cb6ebf142cc8a1
Binary files /dev/null and b/examples/givt.jpg differ
diff --git a/examples/givt.json b/examples/givt.json
new file mode 100644
index 0000000000000000000000000000000000000000..4e244d55bd0423fd8041accf1ba3d9bb43d494af
--- /dev/null
+++ b/examples/givt.json
@@ -0,0 +1,7 @@
+{
+  "name": "givt",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What does the image show?",
+  "license": "CC-BY [GIVT paper](https://arxiv.org/abs/2312.02116)"
+}
\ No newline at end of file
diff --git a/examples/greenlake.jpg b/examples/greenlake.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..65401579082eebb41a70869d8785b2c84a437476
Binary files /dev/null and b/examples/greenlake.jpg differ
diff --git a/examples/greenlake.json b/examples/greenlake.json
new file mode 100644
index 0000000000000000000000000000000000000000..5de5282b9608ada567cd696e8e4846e0906088da
--- /dev/null
+++ b/examples/greenlake.json
@@ -0,0 +1,7 @@
+{
+  "name": "greenlake",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Describe the image.",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/howto.jpg b/examples/howto.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5f079c6751730ab16008835765e6296c8fcc2d8c
Binary files /dev/null and b/examples/howto.jpg differ
diff --git a/examples/howto.json b/examples/howto.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b44aae0878af6ff9abf1628ccedbb932179d19d
--- /dev/null
+++ b/examples/howto.json
@@ -0,0 +1,7 @@
+{
+  "name": "howto",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What does this image show?",
+  "license": "CC-BY [How to train your ViT?](https://arxiv.org/abs/2106.10270)"
+}
\ No newline at end of file
diff --git a/examples/markers.jpg b/examples/markers.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..756537b93cf074ebfd32a45a7438b46914d335c3
Binary files /dev/null and b/examples/markers.jpg differ
diff --git a/examples/markers.json b/examples/markers.json
new file mode 100644
index 0000000000000000000000000000000000000000..9093a2c9a468dd7995039cae415394f182c35e89
--- /dev/null
+++ b/examples/markers.json
@@ -0,0 +1,7 @@
+{
+  "name": "markers",
+  "comment": "answer en How many cups are there?",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "How many cups are there?",
+  "license": "CC0"
+}
\ No newline at end of file
diff --git a/examples/mcair.jpg b/examples/mcair.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e965dde07d103114690bfc086d24ab3fdb054e65
Binary files /dev/null and b/examples/mcair.jpg differ
diff --git a/examples/mcair.json b/examples/mcair.json
new file mode 100644
index 0000000000000000000000000000000000000000..0f50b7f96253821cb70eb5aac40760d140252ffa
--- /dev/null
+++ b/examples/mcair.json
@@ -0,0 +1,7 @@
+{
+  "name": "mcair",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Can you board this airplane?",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/mcair_.json b/examples/mcair_.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ae3353a0ed5109a9cc9f26dc179ec7a86357b8c
--- /dev/null
+++ b/examples/mcair_.json
@@ -0,0 +1,7 @@
+{
+  "name": "mcair",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Is this a restaurant?",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/minergie.jpg b/examples/minergie.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..8372f189fa3549042e752931b2577fc7384be0fc
Binary files /dev/null and b/examples/minergie.jpg differ
diff --git a/examples/minergie.json b/examples/minergie.json
new file mode 100644
index 0000000000000000000000000000000000000000..cb292ed5e6eafc30ccecb9d7b2569ee370ab3b06
--- /dev/null
+++ b/examples/minergie.json
@@ -0,0 +1,7 @@
+{
+  "name": "minergie",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "ocr",
+  "license": "CC0 by [andsteing@](https://huggingface.co/andsteing)"
+}
\ No newline at end of file
diff --git a/examples/morel.jpg b/examples/morel.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e1498f0c98d6a2a7187ec74b809ca9f8fdc776c2
Binary files /dev/null and b/examples/morel.jpg differ
diff --git a/examples/morel.json b/examples/morel.json
new file mode 100644
index 0000000000000000000000000000000000000000..c4fb09a89a268cae5cbdff0810feea34177da7c8
--- /dev/null
+++ b/examples/morel.json
@@ -0,0 +1,7 @@
+{
+  "name": "morel",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "detect morel",
+  "license": "CC0 by [andsteing@](https://huggingface.co/andsteing)"
+}
\ No newline at end of file
diff --git a/examples/motorcyclists.jpg b/examples/motorcyclists.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..91fdffa020ea0e2ba5ef1a9be7dd68bdb7a081ce
Binary files /dev/null and b/examples/motorcyclists.jpg differ
diff --git a/examples/motorcyclists.json b/examples/motorcyclists.json
new file mode 100644
index 0000000000000000000000000000000000000000..f4a0d1e8b7207ac55ed90d09cbfc68ea065c901c
--- /dev/null
+++ b/examples/motorcyclists.json
@@ -0,0 +1,7 @@
+{
+  "name": "motorcyclists",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What does the image show?",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/parking.jpg b/examples/parking.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..3b3c6d3ebb5057b228f04a85a83184c5a1c8aaba
Binary files /dev/null and b/examples/parking.jpg differ
diff --git a/examples/parking.json b/examples/parking.json
new file mode 100644
index 0000000000000000000000000000000000000000..9964ba3acfadec3e0165377bb182ec416672b49a
--- /dev/null
+++ b/examples/parking.json
@@ -0,0 +1,7 @@
+{
+  "name": "parking",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Describe the image.",
+  "license": "CC0 by [xiaohuazhai@](https://huggingface.co/xiaohuazhai)"
+}
\ No newline at end of file
diff --git a/examples/password.jpg b/examples/password.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c7804dfa42aa3cad23089087bffca604a8c3507e
Binary files /dev/null and b/examples/password.jpg differ
diff --git a/examples/password.json b/examples/password.json
new file mode 100644
index 0000000000000000000000000000000000000000..070f3f8c992a948b177f2cc647a7f9260c0d6c38
--- /dev/null
+++ b/examples/password.json
@@ -0,0 +1,7 @@
+{
+  "name": "password",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What is the password?",
+  "license": "CC0 by [akolesnikoff@](https://github.com/akolesnikoff)"
+}
\ No newline at end of file
diff --git a/examples/preservationhall.jpg b/examples/preservationhall.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..adab242566923b4fa3ac97b179c358b07697f221
Binary files /dev/null and b/examples/preservationhall.jpg differ
diff --git a/examples/preservationhall.json b/examples/preservationhall.json
new file mode 100644
index 0000000000000000000000000000000000000000..6f9be7e1169ea9269ee0e9ebe58d32b424b4c21f
--- /dev/null
+++ b/examples/preservationhall.json
@@ -0,0 +1,7 @@
+{
+  "name": "preservationhall",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Describe the image.",
+  "license": "CC0 by [mitscha@](https://github.com/mitscha)"
+}
\ No newline at end of file
diff --git a/examples/preservationhall_.json b/examples/preservationhall_.json
new file mode 100644
index 0000000000000000000000000000000000000000..5571c5272f91d36f2d67a55aac2d61715e3e5f26
--- /dev/null
+++ b/examples/preservationhall_.json
@@ -0,0 +1,7 @@
+{
+  "name": "preservationhall",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "What's the name of the place?",
+  "license": "CC0 by [mitscha@](https://github.com/mitscha)"
+}
\ No newline at end of file
diff --git a/examples/ulges.jpg b/examples/ulges.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e91d86083e3dbfecc2688735d380a441c0dee227
Binary files /dev/null and b/examples/ulges.jpg differ
diff --git a/examples/ulges.json b/examples/ulges.json
new file mode 100644
index 0000000000000000000000000000000000000000..d22ee5806c716238cd83fcbad2597dcf31dc6e04
--- /dev/null
+++ b/examples/ulges.json
@@ -0,0 +1,7 @@
+{
+  "name": "ulges",
+  "comment": "",
+  "model": "paligemma-3b-mix-224",
+  "prompt": "Who is the author of this book?",
+  "license": "CC0"
+}
\ No newline at end of file
diff --git a/howto.jpg b/howto.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5f079c6751730ab16008835765e6296c8fcc2d8c
Binary files /dev/null and b/howto.jpg differ
diff --git a/password.jpg b/password.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c7804dfa42aa3cad23089087bffca604a8c3507e
Binary files /dev/null and b/password.jpg differ
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e6b184898a1cf71fbc709bffaf3188bca0040aa5
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,5 @@
+git+https://github.com/huggingface/transformers.git@add_palma
+torch
+jax
+flax
+spaces
\ No newline at end of file
diff --git a/vae-oid.npz b/vae-oid.npz
new file mode 100644
index 0000000000000000000000000000000000000000..e30bd245fc0b67df063c5bd49d83c7130bba2637
--- /dev/null
+++ b/vae-oid.npz
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5586010257b8536dddefab65e7755077f21d5672d5674dacf911f73ae95a4447
+size 8479556