updates to LLM

Phi -4-mini
added updated prompt

summarizer_module/__init__.py

@@ -6,33 +6,45 @@ import torch
 
 device = get_device()
 
-# Use a small local model (e.g., Phi-2)
-MODEL_ID = "microsoft/phi-2"  # Ensure it's downloaded and cached locally
+# Model config: Use phi-2-mini (replace with phi-4-mini when available)
+MODEL_ID = "microsoft/Phi-4-mini-instruct"  # Ensure it's downloaded and cached locally
 
 # Load model and tokenizer
-model = AutoModelForCausalLM.from_pretrained(MODEL_ID).to(device)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    device_map="auto",
+#    load_in_8bit=True, 
+    torch_dtype=torch.float16
+).to(device)
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 summarizer = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
 def summarize_flowchart(flowchart_json):
     """
-    Given a flowchart JSON with 'start' and 'steps', returns a plain English explanation
-    formatted as bullets and sub-bullets.
+    Generates a human-friendly explanation from flowchart JSON.
 
     Args:
-        flowchart_json (dict): Structured representation of flowchart
+        flowchart_json (dict): Contains "start" node and a list of "steps".
 
     Returns:
-        str: Bullet-style natural language summary of the logic
+        str: Bullet-style explanation with proper nesting and flow.
     """
+    # 📄 Prompt optimized for flow comprehension
     prompt = (
-        "Turn the following flowchart into a bullet-point explanation in plain English.\n"
-        "Use bullets for steps and sub-bullets for branches.\n"
-        "\n"
-        f"Flowchart JSON:\n{flowchart_json}\n"
-        "\nExplanation:"
-    )
-
-    result = summarizer(prompt, max_new_tokens=300, do_sample=False)[0]["generated_text"]
-    explanation = result.split("Explanation:")[-1].strip()
+        "You are an expert in visual reasoning and instruction generation.\n"
+        "Convert the following flowchart JSON into a clear, step-by-step summary using bullets.\n"
+        "- Each bullet represents a process step.\n"
+        "- Use indented sub-bullets to explain decision branches (Yes/No).\n"
+        "- Maintain order based on dependencies and parent-child links.\n"
+        "- Avoid repeating the same step more than once.\n"
+        "- Do not include JSON in the output, only human-readable text.\n"
+        "\nFlowchart:\n{flowchart}\n\nBullet Explanation:"
+    ).format(flowchart=json.dumps(flowchart_json, indent=2))
+
+    result = summarizer(prompt, max_new_tokens=400, do_sample=False)[0]["generated_text"]
+    # Extract the portion after the final prompt marker
+    if "Bullet Explanation:" in result:
+        explanation = result.split("Bullet Explanation:")[-1].strip()
+    else:
+        explanation = result.strip()
     return explanation