Update app.py

app.py

@@ -1,9 +1,15 @@
 import os
 import gradio as gr
 from openai import OpenAI
+import json
+import requests
+
+openai_api_key = os.getenv("OPENROUTER_API_KEY")
+openai_base_url = os.getenv("OPENAI_BASE_URL")
+ai_model = os.getenv("AI_MODEL")
 
 # Configure the OpenAI client with your custom API endpoint and API key.
-client = OpenAI(base_url="http://home.mayhew.cloud:1234/v1", api_key="lm-studio")
+client = OpenAI(base_url=openai_base_url, api_key=openai_api_key)
 
 medical_recommendations = "MEDICAL RECOMMENDATIONS:\n\n" + "Birth control options sorted by effectiveness (typical-use rates), with brief pros, cons, and side effects:\n\nHighly Effective Methods (failure rate <1%)\n-Sterilization\n- Prevention rate: >99%\n- Pros: Permanent and low maintenance\n- Cons: Irreversible; requires surgery\n- Side effects: Surgical risks (infection, pain)\n\n-Intrauterine Devices (IUDs) – Hormonal and Copper\n- Prevention rate: >99%\n- Pros: Long-term (3–10 years), low maintenance, reversible\n- Cons: Requires provider insertion; possible initial discomfort\n- Side effects:\n - Hormonal IUD: Initial irregular bleeding\n - Copper IUD: Heavier periods, cramping; rare risk of expulsion or uterine perforation\n\n-Implant (e.g., Nexplanon)\n- Prevention rate: >99%\n- Pros: Lasts up to 3 years, low maintenance, reversible\n- Cons: Requires minor procedure for insertion and removal; may cause irregular bleeding\n- Side effects: Mood changes, headaches, weight gain, pain at insertion site\n\nModerately Effective Methods (failure rate ~1–9%)\n-Injectable (e.g., Depo-Provera)\n- Prevention rate: ~96%\n- Pros: Injection every 3 months; high efficacy when on schedule\n- Cons: Can cause irregular bleeding; fertility may be delayed after stopping\n- Side effects: Weight gain, mood swings, potential bone density loss, injection site reactions\n\n-Oral Contraceptive Pills (combined or progestin-only)\n- Prevention rate: ~91%\n- Pros: Regulates cycles, may reduce cramps and help with acne; quick return to fertility\n- Cons: Must be taken daily; effectiveness depends on correct use\n- Side effects: Risk of blood clots (especially for smokers or women over 35), nausea, breast tenderness, mood changes, possible increased blood pressure\n- Prescriptions: Yaz, Yasmin, Ortho TriCyclen, Alesse, Loestrin\n- OTC: OPill $20/month, Taken Daily\n\n-Transdermal Patch (e.g., Ortho Evra)\n- Prevention rate: ~91%\n- Pros: Weekly application; steady hormone delivery\n- Cons: May cause skin irritation; visible on skin; less effective if detached\n- Side effects: Similar to pills (blood clots, nausea, breast tenderness, headaches)\n\n-Vaginal Ring (e.g., NuvaRing)\n- Prevention rate: ~91%\n- Pros: Monthly insertion; lower systemic hormone levels\n- Cons: Requires comfort with insertion and removal; possible vaginal discomfort\n- Side effects: Risk of blood clots, mood changes, headaches, vaginal irritation\n\nLess Effective Methods (failure rate 10% or higher)\n-Barrier Methods\n- Male Condoms\n - Prevention rate: ~87%\n - Pros: Also protect against STIs; non-hormonal; widely available\n - Cons: Effectiveness depends on correct use; may break or slip\n - Side effects: Possible latex allergy\n- Female Condoms\n - Prevention rate: ~79%\n - Pros: Offer STI protection; female-controlled\n - Cons: More expensive; less available; may be harder to use\n - Side effects: Possible irritation or allergic reaction\n- Diaphragms and Cervical Caps\n - Prevention rate: ~83–88%\n - Pros: Reusable; non-hormonal\n - Cons: Must be used with spermicide; requires proper fitting and timing\n - Side effects: Potential irritation, allergic reactions, increased risk of UTIs\n\n-Spermicides (used alone)\n- Prevention rate: ~79%\n- Pros: Over-the-counter; can be combined with other methods\n- Cons: Lower effectiveness when used alone; requires application every time\n- Side effects: May cause irritation or allergic reactions\n\n-Fertility Awareness Methods\n- Prevention rate: ~76% (varies widely)\n- Pros: No hormones or devices; increases understanding of your cycle\n- Cons: Requires daily tracking and strict adherence; higher risk of unintended pregnancy if not followed precisely\n- Side effects: None medically\n\n-Withdrawal (Coitus Interruptus)\n- Prevention rate: ~78%\n- Pros: No cost; non-hormonal\n- Cons: Relies heavily on timing; risk from pre-ejaculate\n- Side effects: No direct side effects; may cause stress if it fails\n\nTreatments for Managing HIV:\n- **Integrase Strand Transfer Inhibitor (INSTI)–Based Regimens:**  \n  - Bictegravir/emtricitabine/tenofovir alafenamide (Biktarvy)  \n  - Dolutegravir/lamivudine/abacavir (Triumeq)\n\n- **Non-Nucleoside Reverse Transcriptase Inhibitor (NNRTI)–Based Regimen:**  \n  - Efavirenz/emtricitabine/tenofovir disoproxil fumarate (Atripla)\n\n- **Protease Inhibitor (PI)–Based Regimen:**  \n  - Darunavir (boosted with ritonavir or cobicistat) plus two NRTIs"
 
@@ -58,6 +64,37 @@ max_phrase_length = max(len(phrase) for phrase, _ in replacements)
 
 MIN_FLUSH_SIZE = max(50, max_phrase_length * 2)
 
+def think(request):
+  question = request
+
+  url = "https://openrouter.ai/api/v1/chat/completions"
+  headers = {
+      "Authorization": f"Bearer {openai_api_key}",
+      "Content-Type": "application/json"
+  }
+
+  def do_req(model, content, include_reasoning=False, reasoning=""):
+      messages = [{"role": "user", "content": content}]
+
+      if reasoning != "":
+          messages.append({"role": "assistant", "content": "<think>\n" + reasoning + "</think>\n"})
+
+      payload = {
+          "model": model,
+          "messages": messages,
+          "include_reasoning": include_reasoning,
+          "stop": "</think>"
+      }
+
+      return requests.post(url, headers=headers, data=json.dumps(payload))
+
+
+  # R1 will reliably return "done" for the content portion of the response
+  content = f"{question} Please think this through, but don't output an answer"
+  reasoning_response = do_req("deepseek/deepseek-r1", content, True)
+  reasoning = reasoning_response.json()['choices'][0]['message']['reasoning']
+  return reasoning
+
 def apply_replacements(text):
     """
     Replace all specified phrases in the text.
@@ -93,15 +130,18 @@ def chat_with_openai(message: str, history: list, temperature: float, max_new_to
         conversation.append({"role": "user", "content": user_msg})
         conversation.append({"role": "assistant", "content": assistant_msg})
     conversation.append({"role": "user", "content": message})
-    # Force the model to begin its answer with a "<think>" block.
-    conversation.append({"role": "assistant", "content": "<think> "})
 
     # Immediately yield a "thinking" status message.
     yield "HealthAssistant is Thinking! Please wait, your response will output shortly...\n\n"
 
+    think_result = think(message)
+    
+    # Force the model to begin its answer with a "<think>" block.
+    conversation.append({"role": "assistant", "content": "<think>\n"+think_result+"\n</think>"})
+    
     # Call the API with streaming enabled.
     response = client.chat.completions.create(
-        model="model-identifier",  # Replace with your actual model identifier.
+        model=ai_model,  # Replace with your actual model identifier.
         messages=conversation,
         temperature=temperature,
         max_tokens=max_new_tokens,
@@ -122,31 +162,15 @@ def chat_with_openai(message: str, history: list, temperature: float, max_new_to
         token_text = delta.content or ""
         full_response += token_text
     
-        if not think_detected:
-            # Accumulate tokens until we see the closing </think> marker.
-            buffer += token_text
-            if "</think>" in buffer:
-                think_detected = True
-                # Discard everything up to and including the "</think>" marker.
-                after_think = buffer.split("</think>", 1)[1]
-                pending_buffer += after_think
-                # Only flush if we have at least MIN_FLUSH_SIZE characters.
-                if len(pending_buffer) >= MIN_FLUSH_SIZE:
-                    # Flush all but the last max_phrase_length characters.
-                    safe_portion = pending_buffer[:-max_phrase_length] if len(pending_buffer) > max_phrase_length else ""
-                    if safe_portion:
-                        display_text += apply_replacements(safe_portion)
-                        yield display_text
-                        pending_buffer = pending_buffer[-max_phrase_length:]
-        else:
-            # After the </think> marker, add tokens to pending_buffer.
-            pending_buffer += token_text
-            if len(pending_buffer) >= MIN_FLUSH_SIZE:
-                safe_portion = pending_buffer[:-max_phrase_length] if len(pending_buffer) > max_phrase_length else ""
-                if safe_portion:
-                    display_text += apply_replacements(safe_portion)
-                    yield display_text
-                    pending_buffer = pending_buffer[-max_phrase_length:]
+        
+        # After the </think> marker, add tokens to pending_buffer.
+        pending_buffer += token_text
+        if len(pending_buffer) >= MIN_FLUSH_SIZE:
+            safe_portion = pending_buffer[:-max_phrase_length] if len(pending_buffer) > max_phrase_length else ""
+            if safe_portion:
+                display_text += apply_replacements(safe_portion)
+                yield display_text
+                pending_buffer = pending_buffer[-max_phrase_length:]
     
     # After processing all tokens, flush any remaining text.
     if pending_buffer: