Back to normal app.py v1.3

app.py

@@ -2,19 +2,21 @@
 import gradio as gr
 from core.utils import generate_first_question
 from core.mbti_analyzer import analyze_mbti
-from core.interviewer import generate_question
+from core.interviewer import generate_question, session_state
 
 def analyze_and_ask(user_text, prev_count):
-    """Пошаговый генератор — стриминг без async и без streaming=True."""
+    """Главная функция генерации (анализ → вопрос)."""
     if not user_text.strip():
         yield "⚠️ Please enter your answer.", "", prev_count
         return
 
+    user_id = "default_user"
     try:
         n = int(prev_count.split("/")[0]) + 1
     except Exception:
         n = 1
-    counter = f"{n}/30"
+
+    counter = f"{n}/8"
 
     # 1️⃣ Шаг 1 — анализ
     mbti_gen = analyze_mbti(user_text)
@@ -23,18 +25,34 @@ def analyze_and_ask(user_text, prev_count):
         mbti_text = chunk
         yield mbti_text, "💭 Interviewer is thinking...", counter
 
-    # 2️⃣ Шаг 2 — вопрос
-    interviewer_gen = generate_question("default_user", user_text)
+    # 2️⃣ Шаг 2 — генерация вопроса
+    q_gen = generate_question(user_id)
     next_q = ""
-    for chunk in interviewer_gen:
-        next_q = chunk
-        yield mbti_text, next_q, counter
+    cat = ""
+    for data in q_gen:
+        if isinstance(data, str):
+            next_q = data
+            yield mbti_text, next_q, counter
+        else:
+            cat, next_q = data
+
+    # Если все категории пройдены — показываем финальный результат
+    if next_q.startswith("✅ All"):
+        session = session_state[user_id]
+        yield (
+            f"{mbti_text}\n\nSession completed.",
+            "🎯 All MBTI axes covered.",
+            "8/8"
+        )
 
 # --------------------------------------------------------------
-# Gradio интерфейс
+# Интерфейс Gradio
 # --------------------------------------------------------------
 with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as demo:
-    gr.Markdown("## 🧠 MBTI Personality Interviewer\nОпредели личностный тип и получи следующий вопрос от интервьюера.")
+    gr.Markdown(
+        "## 🧠 MBTI Personality Interviewer\n"
+        "Определи личностный тип и получи вопросы из разных категорий MBTI."
+    )
 
     with gr.Row():
         with gr.Column(scale=1):
@@ -46,11 +64,19 @@ with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as
             btn = gr.Button("Анализировать и задать новый вопрос", variant="primary")
         with gr.Column(scale=1):
             mbti_out = gr.Textbox(label="📊 Анализ MBTI", lines=4)
-            interviewer_out = gr.Textbox(label="💬 Следующий вопрос от интервьюера", lines=3)
-            progress = gr.Textbox(label="⏳ Прогресс", value="0/30")
+            interviewer_out = gr.Textbox(label="💬 Следующий вопрос", lines=3)
+            progress = gr.Textbox(label="⏳ Прогресс", value="0/8")
 
-    btn.click(analyze_and_ask, inputs=[inp, progress], outputs=[mbti_out, interviewer_out, progress])
+    btn.click(
+        analyze_and_ask,
+        inputs=[inp, progress],
+        outputs=[mbti_out, interviewer_out, progress]
+    )
 
-    demo.load(lambda: ("", generate_first_question(), "0/30"), inputs=None, outputs=[mbti_out, interviewer_out, progress])
+    demo.load(
+        lambda: ("", generate_first_question(), "0/8"),
+        inputs=None,
+        outputs=[mbti_out, interviewer_out, progress]
+    )
 
 demo.queue(max_size=20).launch(server_name="0.0.0.0", server_port=7860)

core/interviewer.py

@@ -1,12 +1,13 @@
 # core/interviewer.py
 import random
-import uuid
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 
 INTERVIEWER_MODEL = "f3nsmart/TinyLlama-MBTI-Interviewer-LoRA"
 
 tokenizer = AutoTokenizer.from_pretrained(INTERVIEWER_MODEL)
-model = AutoModelForCausalLM.from_pretrained(INTERVIEWER_MODEL, torch_dtype="auto", device_map="auto")
+model = AutoModelForCausalLM.from_pretrained(
+    INTERVIEWER_MODEL, torch_dtype="auto", device_map="auto"
+)
 
 llm_pipe = pipeline(
     "text-generation",
@@ -17,7 +18,7 @@ llm_pipe = pipeline(
     top_p=0.9,
 )
 
-# MBTI категории
+# Основные оси MBTI
 CATEGORIES = [
     "Introversion", "Extroversion",
     "Sensing", "Intuition",
@@ -25,11 +26,11 @@ CATEGORIES = [
     "Judging", "Perceiving"
 ]
 
-# Память сессий
+# Память по сессиям
 session_state = {}
 
 def init_session(user_id: str):
-    """Инициализирует структуру данных сессии."""
+    """Создаёт новую сессию."""
     session_state[user_id] = {
         "asked": [],
         "answers": {},
@@ -38,7 +39,7 @@ def init_session(user_id: str):
     }
 
 def select_next_category(user_id: str):
-    """Выбирает следующую категорию, не повторяя."""
+    """Выбирает категорию, не повторяя."""
     s = session_state[user_id]
     remaining = [c for c in CATEGORIES if c not in s["asked"]]
     if not remaining:
@@ -48,27 +49,29 @@ def select_next_category(user_id: str):
     return next_cat
 
 def build_prompt(category: str):
-    """Формирует промпт в стиле JSON-диалога."""
+    """Формирует промпт для TinyLlama."""
     return (
-        f'{{"history": [], '
-        f'"instruction": "Ask one open-ended question about {category}. '
-        f'Avoid repeating previous phrasing. Output only the question.", '
-        f'"next_question": "?"}}'
+        f"Ask one open-ended psychological question about {category}. "
+        f"Do not repeat previous questions. "
+        f"Output only the question text."
     )
 
 def generate_question(user_id: str):
-    """Генератор вопроса для текущей категории."""
+    """Генератор вопроса по MBTI категории."""
     if user_id not in session_state:
         init_session(user_id)
 
     category = select_next_category(user_id)
     if not category:
         yield "✅ All 8 categories completed."
-        return
+        return None, None
 
+    yield f"💭 Interviewer is thinking (category: {category})..."
     prompt = build_prompt(category)
     raw = llm_pipe(prompt)[0]["generated_text"]
+
+    # Очистка
     question = raw.strip().split("\n")[0]
     if "?" not in question:
         question += "?"
-    yield f"({category}) {question}"
+    return category, question

core/mbti_analyzer.py

@@ -1,30 +1,23 @@
 # core/mbti_analyzer.py
 from transformers import pipeline
-import asyncio
 
 MBTI_MODEL = "f3nsmart/MBTIclassifier"
 mbti_pipe = pipeline("text-classification", model=MBTI_MODEL, return_all_scores=True)
 
-async def analyze_mbti_async(user_text: str):
-    """Асинхронный MBTI-анализ."""
-    loop = asyncio.get_event_loop()
-    return await loop.run_in_executor(None, lambda: mbti_pipe(user_text)[0])
-
 def analyze_mbti(user_text: str):
-    """Генератор для стриминга результата."""
+    """Пошаговый MBTI-анализ (генератор для стриминга)."""
     yield "⏳ Analyzing personality traits..."
-    res = asyncio.run(analyze_mbti_async(user_text))
-    res_sorted = sorted(res, key=lambda x: x["score"], reverse=True)
-    mbti_text = "\n".join([f"{r['label']} → {r['score']:.3f}" for r in res_sorted[:3]])
-    yield mbti_text
+    try:
+        res = mbti_pipe(user_text)[0]
+        res_sorted = sorted(res, key=lambda x: x["score"], reverse=True)
+        mbti_text = "\n".join([f"{r['label']} → {r['score']:.3f}" for r in res_sorted[:3]])
+        yield mbti_text
+    except Exception as e:
+        yield f"❌ Error: {e}"
 
 def compute_dominant_axis(results):
     """
-    results = [
-      {"label": "Introversion", "score": 0.73},
-      {"label": "Extroversion", "score": 0.27},
-      ...
-    ]
+    Преобразует оценки классификатора в MBTI-код.
     """
     axes = {
         "IE": ("Introversion", "Extroversion"),

core/utils.py

@@ -1,3 +1,3 @@
 # core/utils.py
-def generate_first_question():
-    return "Let's begin our MBTI interview. What do you enjoy doing in your free time?"
+def generate_first_question() -> str:
+    return "What do you usually enjoy doing in your free time?"