Update app.py

app.py

@@ -1,142 +1,19 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM
-import torch
-import re
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 
-# Load model
+# -----------------------
+# Load Hugging Face Model
+# -----------------------
 MODEL_ID = "Muhammadidrees/my-gpt-oss"
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
-
-# Load model with proper dtype handling
-try:
-    # First try auto dtype (preserves original model dtype)
-    model = AutoModelForCausalLM.from_pretrained(
-        MODEL_ID, 
-        device_map="auto",
-        torch_dtype="auto",
-        low_cpu_mem_usage=True
-    )
-except Exception as e:
-    print(f"Auto dtype failed: {e}")
-    try:
-        # Try BFloat16 specifically
-        model = AutoModelForCausalLM.from_pretrained(
-            MODEL_ID, 
-            device_map="auto",
-            torch_dtype=torch.bfloat16,
-            low_cpu_mem_usage=True
-        )
-    except Exception as e2:
-        print(f"BFloat16 failed: {e2}")
-        # Final fallback: float32 (works everywhere but slower)
-        model = AutoModelForCausalLM.from_pretrained(
-            MODEL_ID, 
-            device_map="auto",
-            torch_dtype=torch.float32,
-            low_cpu_mem_usage=True
-        )
-
-if tokenizer.pad_token is None:
-    tokenizer.pad_token = tokenizer.eos_token
-    model.config.pad_token_id = tokenizer.eos_token_id
-
-
-def clean_output(text):
-    """Remove reasoning artifacts and clean formatting"""
-    
-    # Remove common reasoning patterns
-    patterns_to_remove = [
-        r"Let's produce.*?(?=\*\*|$)",
-        r"We need to.*?(?=\*\*|$)",
-        r"We must.*?(?=\*\*|$)",
-        r"assistantfinal\*\*",
-        r"Note that.*?(?=\*\*|$)",
-        r"Use concise statements.*?(?=\*\*|$)",
-        r"Provide bullet points.*?(?=\*\*|$)",
-        r"✅ Medical Insights(?!\*\*)",
-    ]
-    
-    for pattern in patterns_to_remove:
-        text = re.sub(pattern, "", text, flags=re.DOTALL | re.IGNORECASE)
-    
-    # Find where actual report starts
-    start_markers = [
-        "**1. Executive Summary**",
-        "**Executive Summary**",
-        "1. Executive Summary",
-        "## 1. Executive Summary"
-    ]
-    
-    for marker in start_markers:
-        if marker in text:
-            idx = text.find(marker)
-            text = text[idx:]
-            break
-    
-    # Fix common formatting issues from the model
-    replacements = {
-        # Fix typos
-        'Albumen': 'Albumin',
-        'Creatinin': 'Creatinine',
-        'Nomal': 'Normal',
-        'Lympho': 'Lymphocytes',
-        'Strea‑ngths': 'Strengths',
-        'Priori­ties': 'Priorities',
-        'Hea​th': 'Health',
-        'Kid­nee': 'Kidney',
-        'Meta‑bolic': 'Metabolic',
-        'Heal­th*': 'Health**',
-        'Per­sone­lized': 'Personalized',
-        'Ac­tion': 'Action',
-        'Pla­n': 'Plan',
-        'Interac­tion': 'Interaction',
-        'Ta­bular': 'Tabular',
-        'Ma­pping': 'Mapping',
-        'Bioma­rker': 'Biomarker',
-        'Val­ue': 'Value',
-        'Sta­tus': 'Status',
-        'Clin­i­cal': 'Clinical',
-        'In­si­ght': 'Insight',
-        'Recom­men­ta­tion': 'Recommendation',
-        'Lon­ge‑ti­mal': 'Longitudinal',
-        'Lon­ge-Term': 'Long-Term',
-        'Insigh ts': 'Insights',
-        'A.I.': 'AI',
-        'Immu­ne': 'Immune',
-        
-        # Fix units
-        'gdL': 'g/dL',
-        'mgl/dL': 'mg/dL',
-        'mg/mL': 'mg/L',
-        'ui/l': 'U/L',
-        'kc/ml': 'K/uL',
-        'fl': 'fL',
-        
-        # Clean up weird unicode characters
-        '­': '',
-        '​': '',
-        '‑': '-',
-        '–': '-',
-        '—': '-',
-        '│': '|',
-        '├─': '|',
-        '•': '-',
-        '‐': '-',
-    }
-    
-    for old, new in replacements.items():
-        text = text.replace(old, new)
-    
-    # Clean up extra whitespace and duplicate asterisks
-    text = re.sub(r'\*{3,}', '**', text)
-    text = re.sub(r'\n{3,}', '\n\n', text)
-    text = re.sub(r' {2,}', ' ', text)
-    text = text.strip()
-    
-    return text
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, device_map="auto")
+pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
 
+# -----------------------
+# Analysis Function
+# -----------------------
 def analyze(
     albumin, creatinine, glucose, crp, mcv, rdw, alp,
     wbc, lymph, age, gender, height, weight
@@ -145,186 +22,116 @@ def analyze(
     try:
         height_m = height / 100
         bmi = round(weight / (height_m ** 2), 2)
-    except:
+    except Exception:
         bmi = "N/A"
 
-    # Determine status for each biomarker
-    def get_status(value, normal_range):
-        low, high = normal_range
-        if value < low:
-            return "Low"
-        elif value > high:
-            return "High"
-        return "Normal"
-
-    albumin_status = get_status(albumin, (3.5, 5.5))
-    creatinine_status = get_status(creatinine, (0.7, 1.3))
-    glucose_status = get_status(glucose, (70, 100))
-    crp_status = get_status(crp, (0, 3))
-    mcv_status = get_status(mcv, (80, 100))
-    rdw_status = get_status(rdw, (11.5, 14.5))
-    alp_status = get_status(alp, (44, 147))
-    wbc_status = get_status(wbc, (4, 11))
-    lymph_status = get_status(lymph, (20, 40))
-
-    # Ultra-direct prompt with explicit instruction to skip reasoning
-    prompt = f"""You are a medical AI. Generate ONLY the final report. Do NOT include any reasoning, planning, or meta-commentary.
-
-OUTPUT ONLY THIS STRUCTURE:
-
-**1. Executive Summary**
-- **Top Priority Issues**: List 2-3 main concerns based on abnormal values
-- **Key Strengths**: List 2-3 positive findings
-
-**2. System-Specific Analysis**
-- **Blood Health** (MCV {mcv} fL, RDW {rdw}%, Lymph {lymph}%, WBC {wbc} K/uL): Brief clinical interpretation
-- **Protein & Liver Health** (Albumin {albumin} g/dL, ALP {alp} U/L): Brief clinical interpretation
-- **Kidney Health** (Creatinine {creatinine} mg/dL): Brief clinical interpretation
-- **Metabolic Health** (Glucose {glucose} mg/dL, CRP {crp} mg/L): Brief clinical interpretation
-
-**3. Personalized Action Plan**
-- **Medical**: Specific tests or consultations needed
-- **Nutrition**: Dietary recommendations and supplements
-- **Lifestyle**: Exercise, hydration, sleep recommendations
-- **Testing**: Follow-up labs (ferritin, Vitamin D, GGT, etc.)
-
-**4. Interaction Alerts**
-Explain how abnormal biomarkers may interact or compound risks.
-
-**5. Tabular Mapping**
-| Biomarker | Value | Status | Clinical Insight | Recommendation |
-|-----------|-------|--------|------------------|----------------|
-| Albumin | {albumin} g/dL | {albumin_status} | [insight] | [action] |
-| Creatinine | {creatinine} mg/dL | {creatinine_status} | [insight] | [action] |
-| Glucose | {glucose} mg/dL | {glucose_status} | [insight] | [action] |
-| CRP | {crp} mg/L | {crp_status} | [insight] | [action] |
-| MCV | {mcv} fL | {mcv_status} | [insight] | [action] |
-| RDW | {rdw}% | {rdw_status} | [insight] | [action] |
-| ALP | {alp} U/L | {alp_status} | [insight] | [action] |
-| WBC | {wbc} K/uL | {wbc_status} | [insight] | [action] |
-| Lymphocytes | {lymph}% | {lymph_status} | [insight] | [action] |
-
-**6. Enhanced AI Insights & Longitudinal Risk**
-- **Subclinical Nutrients**: Predicted Iron, B12, Folate, Copper status
-- **ALP Source**: Likely bone vs liver origin
-- **Immune Trends**: WBC and lymphocyte assessment
-- **Long-term Risk**: Predictive health considerations
-
-Patient: {gender}, Age {age}, BMI {bmi}
-
-START REPORT NOW (no reasoning or commentary):
-
-**1. Executive Summary**"""
+    # Fixed Instruction (no estimation, structured format only)
+    system_prompt = """
+You are a professional AI Medical Assistant.
+You are analyzing patient demographics and Levine biomarker panel values.
+Output MUST strictly follow this structured format:
+
+1. Executive Summary
+   - Top Priority Issues
+   - Key Strengths
+
+2. System-Specific Analysis
+   - Blood Health (MCV, RDW, Lymphocytes, WBC)
+   - Protein & Liver Health (Albumin, ALP)
+   - Kidney Health (Creatinine µmol/L)
+   - Metabolic Health (Glucose mmol/L, CRP)
+   - Other relevant systems
+
+3. Personalized Action Plan
+   - Medical (tests/consults)
+   - Nutrition (diet & supplements)
+   - Lifestyle (hydration, exercise, sleep)
+   - Testing (follow-up labs: ferritin, Vitamin D, GGT)
+
+4. Interaction Alerts
+   - How biomarkers interact (e.g., anemia ↔ infection cycle, ALP with bone/liver origin)
+
+5. Tabular Mapping (Biomarker → Value → Status → AI-Inferred Insight → Client-Friendly Message)
+
+6. Enhanced AI Insights & Longitudinal Risk
+   - Subclinical nutrient predictions (Iron, B12, Folate, Copper)
+   - Elevated ALP interpretation (bone vs liver origin)
+   - WBC & lymphocyte trends for immunity
+   - Predictive longevity risk profile
+"""
+
+    # Patient Data
+    patient_input = f"""
+Patient Profile:
+- Age: {age}
+- Gender: {gender}
+- Height: {height} cm
+- Weight: {weight} kg
+- BMI: {bmi}
+
+Lab Values:
+- Albumin: {albumin} g/dL
+- Creatinine: {creatinine} mg/dL
+- Glucose: {glucose} mg/dL
+- CRP: {crp} mg/L
+- MCV: {mcv} fL
+- RDW: {rdw} %
+- ALP: {alp} U/L
+- WBC: {wbc} K/uL
+- Lymphocytes: {lymph} %
+"""
+
+    prompt = system_prompt + "\n" + patient_input
+
+    # Call LLM
+    result = pipe(
+        prompt,
+        max_new_tokens=1500,
+        do_sample=True,
+        temperature=0.3,
+        top_p=0.9,
+        return_full_text=False
+    )
 
-    try:
-        inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
-        inputs = {k: v.to(model.device) for k, v in inputs.items()}
-        
-        with torch.no_grad():
-            outputs = model.generate(
-                **inputs,
-                max_new_tokens=2800,
-                min_new_tokens=1000,
-                temperature=0.75,
-                top_p=0.92,
-                top_k=40,
-                repetition_penalty=1.2,
-                do_sample=True,
-                early_stopping=False,
-                pad_token_id=tokenizer.pad_token_id,
-                eos_token_id=tokenizer.eos_token_id,
-                no_repeat_ngram_size=4
-            )
-        
-        generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        
-        # Extract the actual report part
-        if "START REPORT NOW" in generated_text:
-            output_text = generated_text.split("START REPORT NOW")[-1]
-        elif "**1. Executive Summary**" in generated_text:
-            parts = generated_text.split("**1. Executive Summary**")
-            # Take the LAST occurrence (the actual output, not from prompt)
-            if len(parts) > 1:
-                output_text = "**1. Executive Summary**" + parts[-1]
-            else:
-                output_text = generated_text
-        else:
-            output_text = generated_text
-        
-        # Clean up the output
-        output_text = clean_output(output_text)
-        
-        # Improved validation with flexible matching
-        required_sections = {
-            "Executive Summary": ["Executive Summary", "Execut­ive Sum", "Executive Sum"],
-            "System-Specific Analysis": ["System-Specific Analysis", "System‑Specific", "Sys‐tem‑Specific"],
-            "Personalized Action Plan": ["Personalized Action Plan", "Per­sone­lized Ac­tion", "Action Plan"],
-            "Interaction Alerts": ["Interaction Alerts", "Interac­tion Alerts"],
-            "Tabular Mapping": ["Tabular Mapping", "Ta­bular Ma­pping", "| Bioma"],
-            "Enhanced AI Insights": ["Enhanced AI Insights", "Enhanced A.I. Insigh", "Lon­ge‑ti"]
-        }
-        
-        missing_sections = []
-        for section, variants in required_sections.items():
-            if not any(variant in output_text for variant in variants):
-                missing_sections.append(section)
-        
-        # Only warn if truly incomplete (very short OR missing most sections)
-        if len(output_text) < 500 or len(missing_sections) >= 4:
-            warning = "\n\n⚠️ **Warning: Incomplete Output**\n\n"
-            warning += f"Generated: {len(output_text)} characters | Missing: {', '.join(missing_sections) if missing_sections else 'None'}\n\n"
-            warning += "**Suggestions:**\n"
-            warning += "- Try regenerating with different parameter values\n"
-            warning += "- Consider using a larger model (Llama-3-8B, Mistral-7B)\n"
-            warning += "- Or use API-based models for guaranteed quality\n"
-            output_text += warning
-        
-        return output_text
-        
-    except Exception as e:
-        return f"❌ **Error**: {str(e)}\n\nPlease verify all inputs and model availability."
+    return result[0]["generated_text"].strip()
 
 
+# -----------------------
 # Gradio Interface
+# -----------------------
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     
     gr.Markdown("""
-    # 🏥 AI Medical Biomarker Analysis
+    # 🏥 AI Medical Biomarker Analysis  
     ### Comprehensive wellness insights from lab values
     """)
     
     with gr.Row():
         with gr.Column(scale=1):
             gr.Markdown("### 👤 Demographics")
-            age = gr.Number(label="Age", value=35)
+            age = gr.Number(label="Age", value=30)
             gender = gr.Dropdown(["Male", "Female"], label="Gender", value="Male")
-            height = gr.Number(label="Height (cm)", value=175)
+            height = gr.Number(label="Height (cm)", value=174)
             weight = gr.Number(label="Weight (kg)", value=75)
             
             gr.Markdown("### 🩸 Blood Panel")
-            wbc = gr.Number(label="WBC (K/uL)", value=7.0)
-            lymph = gr.Number(label="Lymphocytes (%)", value=30)
-            mcv = gr.Number(label="MCV (fL)", value=90)
+            wbc = gr.Number(label="WBC (K/uL)", value=1.0)
+            lymph = gr.Number(label="Lymphocytes (%)", value=2)
+            mcv = gr.Number(label="MCV (fL)", value=88)
             rdw = gr.Number(label="RDW (%)", value=13)
         
         with gr.Column(scale=1):
             gr.Markdown("### 🧬 Chemistry Panel")
-            albumin = gr.Number(label="Albumin (g/dL)", value=4.2)
-            creatinine = gr.Number(label="Creatinine (mg/dL)", value=1.0)
-            glucose = gr.Number(label="Glucose (mg/dL)", value=95)
-            crp = gr.Number(label="CRP (mg/L)", value=1.5)
-            alp = gr.Number(label="ALP (U/L)", value=75)
+            albumin = gr.Number(label="Albumin (g/dL)", value=3.0)
+            creatinine = gr.Number(label="Creatinine (mg/dL)", value=0.9)
+            glucose = gr.Number(label="Glucose (mg/dL)", value=90)
+            crp = gr.Number(label="CRP (mg/L)", value=0.5)
+            alp = gr.Number(label="ALP (U/L)", value=70)
             
-            analyze_btn = gr.Button("🔬 Generate Report", variant="primary", size="lg")
+            analyze_btn = gr.Button("🔬 Generate Report", variant="primary")
     
     gr.Markdown("### 📊 Analysis Output")
-    output = gr.Textbox(
-        label="Medical Report",
-        lines=30,
-        max_lines=50,
-        show_copy_button=True,
-        placeholder="Results will appear here..."
-    )
+    output = gr.Markdown(label="Medical Report")
     
     analyze_btn.click(
         fn=analyze,
@@ -333,6 +140,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         outputs=output
     )
     
-    gr.Markdown("*Disclaimer: For educational purposes only. Consult healthcare professionals.*")
+    gr.Markdown(
+        "*⚠️ Disclaimer: This AI output is for educational purposes only and not a substitute for professional medical advice.*"
+    )
 
-demo.launch()
+# Launch app
+demo.launch()