Updated structure v1.1

app.py

@@ -1,16 +1,13 @@
 # app.py
 import gradio as gr
-import asyncio
 from core.utils import generate_first_question
 from core.mbti_analyzer import analyze_mbti
 from core.interviewer import generate_question
 
-# --------------------------------------------------------------
-# Асинхронная функция для стриминга ответов
-# --------------------------------------------------------------
-async def analyze_and_ask_stream(user_text, prev_count, user_id="default_user"):
+def analyze_and_ask(user_text, prev_count):
+    """Пошаговый генератор — стриминг без async и без streaming=True."""
     if not user_text.strip():
-        yield gr.update(value="⚠️ Введите ответ."), gr.update(), gr.update()
+        yield "⚠️ Please enter your answer.", "", prev_count
         return
 
     try:
@@ -19,20 +16,19 @@ async def analyze_and_ask_stream(user_text, prev_count, user_id="default_user"):
         n = 1
     counter = f"{n}/30"
 
-    yield gr.update(), gr.update(value="⏳ Анализируем..."), gr.update(value=counter)
-
-    # Параллельные задачи
-    mbti_task = asyncio.create_task(analyze_mbti(user_text))
-    interviewer_task = asyncio.create_task(generate_question(user_id, user_text))
-
-    # Сначала MBTI
-    mbti_text = await mbti_task
-    yield gr.update(value=mbti_text), gr.update(value="💭 Интервьюер думает..."), gr.update(value=counter)
-
-    # Затем вопрос
-    next_question = await interviewer_task
-    yield gr.update(value=mbti_text), gr.update(value=next_question), gr.update(value=counter)
+    # 1️⃣ Шаг 1 — анализ
+    mbti_gen = analyze_mbti(user_text)
+    mbti_text = ""
+    for chunk in mbti_gen:
+        mbti_text = chunk
+        yield mbti_text, "💭 Interviewer is thinking...", counter
 
+    # 2️⃣ Шаг 2 — вопрос
+    interviewer_gen = generate_question("default_user", user_text)
+    next_q = ""
+    for chunk in interviewer_gen:
+        next_q = chunk
+        yield mbti_text, next_q, counter
 
 # --------------------------------------------------------------
 # Gradio интерфейс
@@ -53,9 +49,8 @@ with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as
             interviewer_out = gr.Textbox(label="💬 Следующий вопрос от интервьюера", lines=3)
             progress = gr.Textbox(label="⏳ Прогресс", value="0/30")
 
-    btn.click(analyze_and_ask_stream, inputs=[inp, progress], outputs=[mbti_out, interviewer_out, progress])
+    btn.click(analyze_and_ask, inputs=[inp, progress], outputs=[mbti_out, interviewer_out, progress])
 
-    # Первый вопрос при загрузке
     demo.load(lambda: ("", generate_first_question(), "0/30"), inputs=None, outputs=[mbti_out, interviewer_out, progress])
 
-demo.queue(streaming=True, max_size=20).launch(server_name="0.0.0.0", server_port=7860)
+demo.queue(max_size=20).launch(server_name="0.0.0.0", server_port=7860)

core/interviewer.py

@@ -4,13 +4,9 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 
 INTERVIEWER_MODEL = "f3nsmart/TinyLlama-MBTI-Interviewer-LoRA"
 
-# Инициализация
 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",
     model=model,
@@ -20,27 +16,20 @@ llm_pipe = pipeline(
     top_p=0.9,
 )
 
-# Память для пользователей
 user_memory = {}
 
 def clean_question(text: str) -> str:
-    """Удаляет инструкции, оставляя только вопрос"""
-    text = text.strip().split("\n")[0]
-    text = text.strip('"').strip("'")
+    text = text.strip().split("\n")[0].strip('"').strip("'")
     bad_tokens = ["user:", "assistant:", "instruction", "interviewer", "system:"]
     for bad in bad_tokens:
         if bad.lower() in text.lower():
             text = text.split(bad)[-1].strip()
     if not text.endswith("?"):
         text += "?"
-    if len(text.split()) < 3:
-        return "What do you usually enjoy doing in your free time?"
-    return text.strip()
+    return text if len(text.split()) > 3 else "What do you usually enjoy doing in your free time?"
 
-async def generate_question(user_id: str, user_text: str) -> str:
-    """
-    Генерирует новый вопрос с учётом предыдущих.
-    """
+def generate_question(user_id: str, user_text: str):
+    """Генератор вопроса (стриминг)."""
     prev_qs = user_memory.get(user_id, [])
     prev_joined = "; ".join(prev_qs) if prev_qs else "None"
 
@@ -53,8 +42,8 @@ async def generate_question(user_id: str, user_text: str) -> str:
         f"Interviewer:"
     )
 
-    loop = asyncio.get_event_loop()
-    raw = await loop.run_in_executor(None, lambda: llm_pipe(prompt)[0]["generated_text"])
+    yield "💭 Interviewer is thinking..."
+    raw = llm_pipe(prompt)[0]["generated_text"]
     question = clean_question(raw)
 
     valid_starts = ("What", "Why", "How", "When")
@@ -62,5 +51,5 @@ async def generate_question(user_id: str, user_text: str) -> str:
         question = "What motivates you to do the things you enjoy most?"
 
     prev_qs.append(question)
-    user_memory[user_id] = prev_qs[-10:]  # храним последние 10
-    return question
+    user_memory[user_id] = prev_qs[-10:]
+    yield question

core/mbti_analyzer.py

@@ -5,13 +5,15 @@ import asyncio
 MBTI_MODEL = "f3nsmart/MBTIclassifier"
 mbti_pipe = pipeline("text-classification", model=MBTI_MODEL, return_all_scores=True)
 
-async def analyze_mbti(user_text: str) -> str:
-    """
-    Асинхронный анализ текста на MBTI типы.
-    Возвращает 3 лучших типа с их вероятностями.
-    """
+async def analyze_mbti_async(user_text: str):
+    """Асинхронный MBTI-анализ."""
     loop = asyncio.get_event_loop()
-    res = await loop.run_in_executor(None, lambda: mbti_pipe(user_text)[0])
+    return await loop.run_in_executor(None, lambda: mbti_pipe(user_text)[0])
+
+def analyze_mbti(user_text: str):
+    """Генератор для стриминга результата."""
+    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]])
-    return mbti_text
+    yield mbti_text

core/utils.py

@@ -1,5 +1,4 @@
 # core/utils.py
 
 def generate_first_question() -> str:
-    """Первый вопрос фиксированный"""
     return "What do you usually enjoy doing in your free time?"