Update app.py

app.py

@@ -1,10 +1,24 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from llava.model.builder import load_pretrained_model
+from llava.mm_utils import process_images, tokenizer_image_token
+from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN
+from llava.conversation import conv_templates
+from PIL import Image
+import copy
+import torch
+import warnings
+import requests
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("AI-Safeguard/Ivy-VL-llava")
+warnings.filterwarnings("ignore")
+
+pretrained = "AI-Safeguard/Ivy-VL-llava"
+model_name = "llava_qwen"
+device = "cuda"
+device_map = "auto"
+
+# Load model, tokenizer, and image processor
+tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map)
+model.eval()
 
 def respond(
     message,
@@ -23,32 +37,51 @@ def respond(
         if val[1]:
             messages.append({"role": "assistant", "content": val[1]})
 
-    messages.append({"role": "user", "content": message})
+    if image:
+        # Process image if provided
+        image_tensor = process_images([image], image_processor, model.config)
+        image_tensor = [_image.to(dtype=torch.float16, device=device) for _image in image_tensor]
+
+        conv_template = "qwen_1_5"
+        question = DEFAULT_IMAGE_TOKEN + "\n" + message
+        conv = copy.deepcopy(conv_templates[conv_template])
+        conv.append_message(conv.roles[0], question)
+        conv.append_message(conv.roles[1], None)
+        prompt_question = conv.get_prompt()
+
+        input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(device)
+        image_sizes = [image.size]
 
-    payload = {
-        "messages": messages,
-        "max_tokens": max_tokens,
-        "temperature": temperature,
-        "top_p": top_p,
-    }
+        cont = model.generate(
+            input_ids,
+            images=image_tensor,
+            image_sizes=image_sizes,
+            do_sample=False,
+            temperature=temperature,
+            max_new_tokens=max_tokens,
+        )
 
-    if image is not None:
-        payload["image"] = image
+        response = tokenizer.batch_decode(cont, skip_special_tokens=True)[0]
+    else:
+        messages.append({"role": "user", "content": message})
 
-    response = ""
+        payload = {
+            "messages": messages,
+            "max_tokens": max_tokens,
+            "temperature": temperature,
+            "top_p": top_p,
+        }
 
-    for message in client.chat_completion(
-        payload,
-        stream=True,
-    ):
-        token = message.choices[0].delta.content
+        response = ""
+        for message in client.chat_completion(
+            payload,
+            stream=True,
+        ):
+            token = message.choices[0].delta.content
+            response += token
 
-        response += token
-        yield response
+    yield response
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[