Adjust slider values and downsize large images

app.py

@@ -4,9 +4,9 @@ import torch
 import gradio as gr
 from transformers import LlavaNextProcessor, LlavaNextForConditionalGeneration
 
-model_path = "ibm-granite/granite-vision-3.1-2b-preview"
-processor = LlavaNextProcessor.from_pretrained(model_path, use_fast=True)
-model = LlavaNextForConditionalGeneration.from_pretrained(model_path, torch_dtype="auto", device_map="auto")
+model_id = "ibm-granite/granite-vision-3.1-2b-preview"
+processor = LlavaNextProcessor.from_pretrained(model_id, use_fast=True)
+model = LlavaNextForConditionalGeneration.from_pretrained(model_id, torch_dtype="auto", device_map="auto")
 
 def get_text_from_content(content):
     texts = []
@@ -24,6 +24,8 @@ def chat_inference(image, text, temperature, top_p, top_k, max_tokens, conversat
         
     user_content = []
     if image is not None:
+        if image.width > 512 or image.height > 512:
+            image.thumbnail((512, 512))
         user_content.append({"type": "image", "image": image})
     if text and text.strip():
         user_content.append({"type": "text", "text": text.strip()})
@@ -78,16 +80,16 @@ def clear_chat():
     return [], [], "", None
     
 with gr.Blocks(title="Granite Vision 3.1 2B", css="h1 { overflow: hidden; }") as demo:
-    gr.Markdown("# Granite Vision 3.1 2B")
+    gr.Markdown("# [Granite Vision 3.1 2B](https://huggingface.co/ibm-granite/granite-vision-3.1-2b-preview)")
     
     with gr.Row():
         with gr.Column(scale=2):
             image_input = gr.Image(type="pil", label="Upload Image (optional)")
             with gr.Column():
-                temperature_input = gr.Slider(minimum=0.0, maximum=2.0, value=0.2, step=0.01, label="Temperature")
-                top_p_input = gr.Slider(minimum=0.0, maximum=1.0, value=0.95, step=0.01, label="Top p")
-                top_k_input = gr.Slider(minimum=0, maximum=100, value=50, step=1, label="Top k")
-                max_tokens_input = gr.Slider(minimum=10, maximum=300, value=128, step=1, label="Max Tokens")
+                temperature_input = gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Temperature")
+                top_p_input = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.01, label="Top p")
+                top_k_input = gr.Slider(minimum=1, maximum=100, value=50, step=1, label="Top k")
+                max_tokens_input = gr.Slider(minimum=1, maximum=1024, value=512, step=1, label="Max Tokens")
             
         with gr.Column(scale=3):
             chatbot = gr.Chatbot(label="Chat History", elem_id="chatbot", type='messages')

requirements.txt

@@ -2,5 +2,4 @@ torch
 torchvision
 git+https://github.com/huggingface/transformers.git
 gradio
-accelerate
-bitsandbytes
+accelerate