added embeddings

app.py

@@ -3,11 +3,15 @@ import torch
 from transformers import AutoProcessor, LlavaForConditionalGeneration
 from transformers import BitsAndBytesConfig
 
+from sentence_transformers import SentenceTransformer, util
+
 quantization_config = BitsAndBytesConfig(
     load_in_4bit=True,
     bnb_4bit_compute_dtype=torch.float16
 )
 
+embedder = SentenceTransformer('all-mpnet-base-v2')
+
 model_id = "llava-hf/llava-1.5-7b-hf"
 
 processor = AutoProcessor.from_pretrained(model_id)
@@ -22,13 +26,13 @@ def text_to_image(image, prompt):
     prompt = f'USER: <image>\n{prompt}\nASSISTANT:'
 
     inputs = processor([prompt], images=[image], padding=True, return_tensors="pt").to(model.device)
-    for k, v in inputs.items():
-        print(k, v.shape)
-    print(inputs)
-    output = model.generate(**inputs, max_new_tokens=100)
+    output = model.generate(**inputs, max_new_tokens=500)
     generated_text = processor.batch_decode(output, skip_special_tokens=True)
-    for text in generated_text:
-        return text.split("ASSISTANT:")[-1]
+    text = generated_text.pop()
+    text_output = text.split("ASSISTANT:")[-1]
+    text_embeddings = embedder.encode(text_output)
+
+    return text_output, dict(text_embeddings=text_embeddings)
 
 
 demo = gr.Interface(
@@ -37,7 +41,7 @@ demo = gr.Interface(
         gr.Image(label='Select an image to analyze', type='pil'),
         gr.Textbox(label='Enter Prompt')
     ],
-    outputs=gr.Textbox(label='Maurice says:')
+    outputs=[gr.Textbox(label='Maurice says:'), gr.JSON(label='Embedded text')]
 )
 
 if __name__ == "__main__":