Update app.py

app.py

@@ -16,7 +16,6 @@ import cv2
 from transformers import (
     Qwen2VLForConditionalGeneration,
     Qwen2_5_VLForConditionalGeneration,
-    Gemma3nForConditionalGeneration,
     AutoModelForImageTextToText,
     AutoProcessor,
     TextIteratorStreamer,
@@ -57,11 +56,12 @@ model_t = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
-# Load olmOCR-7B-0225-preview
-MODEL_ID_O = "google/gemma-3n-E4B-it"
-processor_o = AutoProcessor.from_pretrained(MODEL_ID_O, trust_remote_code=True)
-model_o = AutoModelForImageTextToText.from_pretrained(
-    MODEL_ID_O,
+
+# Load Gemma3n-E4B-it
+MODEL_ID_G = "google/gemma-3n-E4B-it"
+processor_g = AutoProcessor.from_pretrained(MODEL_ID_G, trust_remote_code=True)
+model_g = AutoModelForImageTextToText.from_pretrained(
+    MODEL_ID_G,
     trust_remote_code=True,
     torch_dtype=torch.float16
 ).to(device).eval()
@@ -103,12 +103,12 @@ def generate_image(model_name: str, text: str, image: Image.Image,
     elif model_name == "DREX-062225-7B-exp":
         processor = processor_x
         model = model_x
-    elif model_name == "olmOCR-7B-0225-preview":
-        processor = processor_o
-        model = model_o
     elif model_name == "Typhoon-OCR-3B":
         processor = processor_t
         model = model_t
+    elif model_name == "Gemma3n-E4B-it":
+        processor = processor_g
+        model = model_g
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -117,24 +117,27 @@ def generate_image(model_name: str, text: str, image: Image.Image,
         yield "Please upload an image.", "Please upload an image."
         return
 
-    messages = [{
-        "role": "user",
-        "content": [
-            {"type": "image", "image": image},
-            {"type": "text", "text": text},
-        ]
-    }]
-    prompt_full = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    inputs = processor(
-        text=[prompt_full],
-        images=[image],
+    messages = [{"role": "user", "content": [{"type": "image", "image": image}, {"type": "text", "text": text}]}]
+    inputs = processor.apply_chat_template(
+        messages,
+        tokenize=True,
+        add_generation_prompt=True,
+        return_dict=True,
         return_tensors="pt",
-        padding=True,
         truncation=False,
         max_length=MAX_INPUT_TOKEN_LENGTH
     ).to(device)
     streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-    generation_kwargs = {**inputs, "streamer": streamer, "max_new_tokens": max_new_tokens}
+    generation_kwargs = {
+        **inputs,
+        "streamer": streamer,
+        "max_new_tokens": max_new_tokens,
+        "do_sample": True,
+        "temperature": temperature,
+        "top_p": top_p,
+        "top_k": top_k,
+        "repetition_penalty": repetition_penalty,
+    }
     thread = Thread(target=model.generate, kwargs=generation_kwargs)
     thread.start()
     buffer = ""
@@ -159,12 +162,12 @@ def generate_video(model_name: str, text: str, video_path: str,
     elif model_name == "DREX-062225-7B-exp":
         processor = processor_x
         model = model_x
-    elif model_name == "olmOCR-7B-0225-preview":
-        processor = processor_o
-        model = model_o
     elif model_name == "Typhoon-OCR-3B":
         processor = processor_t
         model = model_t
+    elif model_name == "Gemma3n-E4B-it":
+        processor = processor_g
+        model = model_g
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -174,14 +177,16 @@ def generate_video(model_name: str, text: str, video_path: str,
         return
 
     frames = downsample_video(video_path)
-    messages = [
-        {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
-        {"role": "user", "content": [{"type": "text", "text": text}]}
-    ]
-    for frame in frames:
-        image, timestamp = frame
-        messages[1]["content"].append({"type": "text", "text": f"Frame {timestamp}:"})
-        messages[1]["content"].append({"type": "image", "image": image})
+    content = [{"type": "text", "text": text}]
+    if model_name == "Gemma3n-E4B-it":
+        for frame, _ in frames:
+            content.append({"type": "image", "image": frame})
+    else:
+        for frame in frames:
+            image, timestamp = frame
+            content.append({"type": "text", "text": f"Frame {timestamp}:"})
+            content.append({"type": "image", "image": image})
+    messages = [{"role": "user", "content": content}]
     inputs = processor.apply_chat_template(
         messages,
         tokenize=True,
@@ -276,7 +281,7 @@ with gr.Blocks(css=css, theme=gr.themes.Citrus()) as demo:
                 markdown_output = gr.Markdown(label="Formatted Result (Result.Md)")
 
             model_choice = gr.Radio(
-                choices=["DREX-062225-7B-exp", "olmOCR-7B-0225-preview", "VIREX-062225-7B-exp", "Typhoon-OCR-3B"],
+                choices=["DREX-062225-7B-exp", "VIREX-062225-7B-exp", "Typhoon-OCR-3B", "Gemma3n-E4B-it"],
                 label="Select Model",
                 value="DREX-062225-7B-exp"
             )
@@ -285,7 +290,7 @@ with gr.Blocks(css=css, theme=gr.themes.Citrus()) as demo:
             gr.Markdown("> [DREX-062225-7B-exp](https://huggingface.co/prithivMLmods/DREX-062225-exp): the drex-062225-exp (document retrieval and extraction expert) model is a specialized fine-tuned version of docscopeocr-7b-050425-exp, optimized for document retrieval, content extraction, and analysis recognition. built on top of the qwen2.5-vl architecture.")
             gr.Markdown("> [VIREX-062225-7B-exp](https://huggingface.co/prithivMLmods/VIREX-062225-exp): the virex-062225-exp (video information retrieval and extraction expert - experimental) model is a fine-tuned version of qwen2.5-vl-7b-instruct, specifically optimized for advanced video understanding, image comprehension, sense of reasoning, and natural language decision-making through cot reasoning.")
             gr.Markdown("> [Typhoon-OCR-3B](https://huggingface.co/scb10x/typhoon-ocr-3b): a bilingual document parsing model built specifically for real-world documents in thai and english, inspired by models like olmocr, based on qwen2.5-vl-instruction. this model is intended to be used with a specific prompt only.")
-            gr.Markdown("> [olmOCR-7B-0225](https://huggingface.co/allenai/olmOCR-7B-0225-preview): the olmocr-7b-0225-preview model is based on qwen2-vl-7b, optimized for document-level optical character recognition (ocr), long-context vision-language understanding, and accurate image-to-text conversion with mathematical latex formatting. designed with a focus on high-fidelity visual-textual comprehension.")
+            gr.Markdown("> [Gemma3n-E4B-it](https://huggingface.co/google/gemma-3n-E4B-it): A multimodal model capable of processing images and videos for various tasks.")
             gr.Markdown(">⚠️note: all the models in space are not guaranteed to perform well in video inference use cases.")  
     
     image_submit.click(