Upload 15 files

.gitattributes

@@ -33,3 +33,9 @@ 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
+example_images/2d0fbcc50e88065a040a537b717620e964fb4453314b71d83f3ed3425addcef6.png filter=lfs diff=lfs merge=lfs -text
+example_images/annual_rep_14.png filter=lfs diff=lfs merge=lfs -text
+example_images/annual_rep_15.png filter=lfs diff=lfs merge=lfs -text
+example_images/gazette_de_france.jpg filter=lfs diff=lfs merge=lfs -text
+example_images/paper_3.png filter=lfs diff=lfs merge=lfs -text
+example_images/redhat.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,12 +1,13 @@
 ---
-title: SmolDocling 256M Demo
-emoji: 🖼
-colorFrom: purple
-colorTo: red
+title: SmolVLM
+emoji: 📊
+colorFrom: blue
+colorTo: green
 sdk: gradio
-sdk_version: 5.0.1
+sdk_version: 5.12.0
 app_file: app.py
 pinned: false
+license: apache-2.0
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,154 +1,152 @@
 import gradio as gr
-import numpy as np
-import random
-
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
+from transformers import AutoProcessor, AutoModelForVision2Seq, TextIteratorStreamer
+from transformers.image_utils import load_image
+from threading import Thread
+import re
+import time
 import torch
+import spaces
+import re
+import ast
+import html
+import random
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
-
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
-
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-
-    generator = torch.Generator().manual_seed(seed)
-
-    image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
-        generator=generator,
-    ).images[0]
-
-    return image, seed
-
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
-}
-"""
-
-with gr.Blocks(css=css) as demo:
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
-
-        with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
+from PIL import Image, ImageOps
 
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
+from docling_core.types.doc import DoclingDocument
+from docling_core.types.doc.document import DocTagsDocument
+
+def add_random_padding(image, min_percent=0.1, max_percent=0.10):
+    image = image.convert("RGB")
+
+    width, height = image.size
+
+    pad_w_percent = random.uniform(min_percent, max_percent)
+    pad_h_percent = random.uniform(min_percent, max_percent)
+
+    pad_w = int(width * pad_w_percent)
+    pad_h = int(height * pad_h_percent)
+
+    corner_pixel = image.getpixel((0, 0))  # Top-left corner
+    padded_image = ImageOps.expand(image, border=(pad_w, pad_h, pad_w, pad_h), fill=corner_pixel)
+
+    return padded_image
+
+def normalize_values(text, target_max=500):
+    def normalize_list(values):
+        max_value = max(values) if values else 1
+        return [round((v / max_value) * target_max) for v in values]
+
+    def process_match(match):
+        num_list = ast.literal_eval(match.group(0))
+        normalized = normalize_list(num_list)
+        return "".join([f"<loc_{num}>" for num in normalized])
+
+    pattern = r"\[([\d\.\s,]+)\]"
+    normalized_text = re.sub(pattern, process_match, text)
+    return normalized_text
+
+
+processor = AutoProcessor.from_pretrained("ds4sd/SmolDocling-256M-preview")
+model = AutoModelForVision2Seq.from_pretrained("ds4sd/SmolDocling-256M-preview", 
+        torch_dtype=torch.bfloat16,
+        #_attn_implementation="flash_attention_2"
+        ).to("cuda")
+
+@spaces.GPU
+def model_inference(
+    input_dict, history
+): 
+    text = input_dict["text"]
+    print(input_dict["files"])
+    if len(input_dict["files"]) > 1:
+        if "OTSL" in text or "code" in text:
+            images = [add_random_padding(load_image(image)) for image in input_dict["files"]]
+        else:
+            images = [load_image(image) for image in input_dict["files"]]
+
+    elif len(input_dict["files"]) == 1:
+        if "OTSL" in text or "code" in text:
+            images = [add_random_padding(load_image(input_dict["files"][0]))]
+        else:
+            images = [load_image(input_dict["files"][0])]   
+
+    else:
+        images = []
+
+    if text == "" and not images:
+        gr.Error("Please input a query and optionally image(s).")
+
+    if text == "" and images:
+        gr.Error("Please input a text query along the image(s).")
+
+    if "OCR at text at" in text or "Identify element" in text or "formula" in text:
+        text = normalize_values(text, target_max=500)
+
+    resulting_messages = [
+                {
+                    "role": "user",
+                    "content": [{"type": "image"} for _ in range(len(images))] + [
+                        {"type": "text", "text": text}
+                    ]
+                }
+            ]
+    prompt = processor.apply_chat_template(resulting_messages, add_generation_prompt=True)
+    inputs = processor(text=prompt, images=[images], return_tensors="pt").to('cuda')
+
+    generation_args = {
+        "input_ids": inputs.input_ids,
+        "pixel_values": inputs.pixel_values,
+        "attention_mask": inputs.attention_mask,
+        "num_return_sequences": 1,
+        "no_repeat_ngram_size": 10,
+        "max_new_tokens": 8192,
+    }
+
+    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=False)
+    generation_args = dict(inputs, streamer=streamer, max_new_tokens=8192)
+
+    thread = Thread(target=model.generate, kwargs=generation_args)
+    thread.start()
+
+    yield "..."
+    buffer = ""
+    doctag_output = ""
+
+    for new_text in streamer:
+        if new_text != "<end_of_utterance>":
+            buffer += html.escape(new_text)
+            doctag_output += new_text
+        yield buffer
+
+    if any(tag in doctag_output for tag in ["<doctag>", "<otsl>", "<code>", "<formula>", "<chart>"]):
+        # final_output = buffer
+        # cleaned_output = final_output[len(inputs.input_ids):] if len(final_output) > prompt_length else final_output
+        doc = DoclingDocument(name="Document")
+        if "<chart>" in doctag_output:
+            doctag_output = doctag_output.replace("<chart>", "<otsl>").replace("</chart>", "</otsl>")
+            doctag_output = re.sub(r'(<loc_500>)(?!.*<loc_500>)<[^>]+>', r'\1', doctag_output)
+   
+        doctags_doc = DocTagsDocument.from_doctags_and_image_pairs([doctag_output], images)
+        doc.load_from_doctags(doctags_doc)
+        yield f"**MD Output:**\n\n{doc.export_to_markdown()}"
+
+examples=[[{"text": "Convert this page to docling.", "files": ["example_images/2d0fbcc50e88065a040a537b717620e964fb4453314b71d83f3ed3425addcef6.png"]}],
+          [{"text": "Convert this table to OTSL.", "files": ["example_images/image-2.jpg"]}],
+          [{"text": "Convert code to text.", "files": ["example_images/7666.jpg"]}],
+          [{"text": "Convert formula to latex.", "files": ["example_images/2433.jpg"]}],
+          [{"text": "Convert chart to OTSL.", "files": ["example_images/06236926002285.png"]}],
+          [{"text": "OCR the text in location [47, 531, 167, 565]", "files": ["example_images/s2w_example.png"]}],
+          [{"text": "Extract all section header elements on the page.", "files": ["example_images/paper_3.png"]}],
+          [{"text": "Identify element at location [123, 413, 1059, 1061]", "files": ["example_images/redhat.png"]}],
+          [{"text": "Convert this page to docling.", "files": ["example_images/gazette_de_france.jpg"]}],
+          ]
+
+demo = gr.ChatInterface(fn=model_inference, title="SmolDocling-256M: Ultra-compact VLM for Document Conversion 💫", 
+            description="Play with [ds4sd/SmolDocling-256M-preview](https://huggingface.co/ds4sd/SmolDocling-256M-preview) in this demo. To get started, upload an image and text or try one of the examples. This demo doesn't use history for the chat, so every chat you start is a new conversation.",
+            examples=examples,
+            textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image"], file_count="multiple"), stop_btn="Stop Generation", multimodal=True,
+            cache_examples=False
             )
 
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-
-            with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-            with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
-
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
-    )
-
-if __name__ == "__main__":
-    demo.launch()
+demo.launch(debug=True, share=True)

requirements.txt

@@ -1,6 +1,8 @@
-accelerate
-diffusers
-invisible_watermark
 torch
+accelerate
+huggingface_hub
+gradio
 transformers
-xformers
+spaces
+docling
+docling-core