Create Importing_AI

Importing_AI

@@ -0,0 +1,46 @@
+import gradio as gr
+import torch
+from transformers import AutoProcessor, AutoModelForCausalLM, tqdm
+
+# 1. Configuration & Model IDs
+FLORENCE_MODEL_ID = "microsoft/Florence-2-base"
+DOLPHIN_MODEL_ID = "cognitivecomputations/dolphin-2.9.4-qwen2-1.5b" # Or your merged model path
+
+# 2. Load Florence-2 (Vision)
+print("Loading Florence-2...")
+florence_model = AutoModelForCausalLM.from_pretrained(FLORENCE_MODEL_ID, trust_remote_code=True).to("cpu").eval()
+florence_processor = AutoProcessor.from_pretrained(FLORENCE_MODEL_ID, trust_remote_code=True)
+
+# 3. Load Dolphin-Qwen (Reasoning)
+print("Loading Dolphin-Qwen...")
+dolphin_model = AutoModelForCausalLM.from_pretrained(DOLPHIN_MODEL_ID).to("cpu").eval()
+dolphin_tokenizer = AutoProcessor.from_pretrained(DOLPHIN_MODEL_ID)
+
+def process_ui_task(image, prompt):
+    # This function will eventually handle the phased logic:
+    # Phase 1: Florence identifies the button
+    # Phase 3: Dolphin maps it to a Hex Packet
+    
+    # Simple Florence Inference Example
+    inputs = florence_processor(text="<OD>", images=image, return_tensors="pt").to("cpu")
+    generated_ids = florence_model.generate(
+        input_ids=inputs["input_ids"],
+        pixel_values=inputs["pixel_values"],
+        max_new_tokens=1024,
+        do_sample=False,
+        num_beams=3
+    )
+    results = florence_processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
+    
+    return f"Brain Output: {results}"
+
+# 4. Gradio Interface
+interface = gr.Interface(
+    fn=process_ui_task,
+    inputs=[gr.Image(type="pill"), gr.Textbox(label="Instruction")],
+    outputs="text",
+    title="AI Automation Brain (Florence + Dolphin)"
+)
+
+if __name__ == "__main__":
+    interface.launch()