New model TinyLlama/TinyLlama-1.1B-Chat-v1.0. v1.0

app.py

@@ -1,111 +1,69 @@
 import gradio as gr
-import concurrent.futures
-import time
-from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoModelForCausalLM, pipeline
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, AutoModelForSequenceClassification
 
 # =========================================================
-# 1. Настройка моделей
+# 1️⃣  Настройка моделей
 # =========================================================
-MBTI_MODEL_ID = "f3nsmart/MBTIclassifier"
-LLM_MODEL_ID = "google/gemma-1.1-2b-it"  # Быстрая, контекстная и лаконичная
 
-# Классификатор (твой fine-tuned)
-mbti_tokenizer = AutoTokenizer.from_pretrained(MBTI_MODEL_ID)
-mbti_model = AutoModelForSequenceClassification.from_pretrained(MBTI_MODEL_ID)
-analyzer = pipeline("text-classification", model=mbti_model, tokenizer=mbti_tokenizer, return_all_scores=True)
+# Основная MBTI-модель (твоя fine-tuned)
+MBTI_MODEL = "f3nsmart/MBTIclassifier"
 
-# Интервьюер (Gemma)
-q_tokenizer = AutoTokenizer.from_pretrained(LLM_MODEL_ID)
-q_model = AutoModelForCausalLM.from_pretrained(LLM_MODEL_ID)
-interviewer = pipeline("text-generation", model=q_model, tokenizer=q_tokenizer)
+# Лёгкая LLM-модель для генерации вопросов
+INTERVIEWER_MODEL = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
+
+# Загружаем MBTI-классификатор
+mbti_tok = AutoTokenizer.from_pretrained(MBTI_MODEL)
+mbti_model = AutoModelForSequenceClassification.from_pretrained(MBTI_MODEL)
+mbti_pipe = pipeline("text-classification", model=mbti_model, tokenizer=mbti_tok, return_all_scores=True)
+
+# Загружаем интервьюера
+interviewer_tok = AutoTokenizer.from_pretrained(INTERVIEWER_MODEL)
+interviewer_model = AutoModelForCausalLM.from_pretrained(INTERVIEWER_MODEL)
+interviewer_pipe = pipeline("text-generation", model=interviewer_model, tokenizer=interviewer_tok, max_new_tokens=60)
 
 # =========================================================
-# 2. Основная логика
+# 2️⃣  Логика
 # =========================================================
-def classify_and_ask(user_input, question_count):
-    """
-    Классифицирует MBTI и генерирует следующий вопрос от интервьюера.
-    """
-    if not user_input.strip():
-        return "⚠️ Введите текст.", "⚠️ Вопрос не сформирован.", question_count
-
-    start_time = time.perf_counter()
-
-    # Асинхронное выполнение двух задач (аналитика + вопрос)
-    with concurrent.futures.ThreadPoolExecutor() as executor:
-        future_analysis = executor.submit(lambda: analyzer(user_input))
-        future_question = executor.submit(lambda: interviewer(
-            f"""You are an MBTI interviewer. 
-            The user answered: "{user_input}".
-            Ask one short, open-ended question that helps reveal their personality type.
-            Avoid repeating previous topics.
-            Start directly with 'How', 'Why', 'What', or 'When'. 
-            Output only the question itself, nothing else.""",
-            max_new_tokens=40,
-            temperature=0.8,
-            top_p=0.9,
-            do_sample=True,
-        ))
-
-        analysis_result = future_analysis.result()
-        question_result = future_question.result()
-
-    elapsed = time.perf_counter() - start_time
-    print(f"⏱ Время обработки запроса : {elapsed:.2f} сек")
-
-    # Парсим результаты
-    results = sorted(analysis_result[0], key=lambda x: x["score"], reverse=True)
-    top = "\n".join([f"{r['label']} → {r['score']:.3f}" for r in results[:3]])
-
-    raw_q = question_result[0]["generated_text"].strip()
-    # Убираем лишние фразы и добавляем "?" если отсутствует
-    question = raw_q.split("\n")[-1].split(":")[-1].strip()
-    if not question.endswith("?"):
-        question += "?"
-
-    question_count += 1
-    progress = f"{question_count}/30"
-
-    return top, question, progress
+
+def analyze_and_ask(user_text, question_count=1):
+    if not user_text.strip():
+        return "⚠️ Введите ответ.", "", "0/30"
+
+    # Анализ MBTI
+    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]])
+
+    # Генерация нового вопроса
+    prompt = f"You are a professional interviewer for MBTI testing. Generate one short, open-ended, natural question based on this answer: '{user_text}'. Avoid yes/no questions."
+    gen = interviewer_pipe(prompt)[0]["generated_text"]
+    gen = gen.split("question:")[-1].strip() if "question:" in gen.lower() else gen.strip()
+
+    # Счётчик
+    counter = f"{question_count}/30"
+    return mbti_text, gen, counter
 
 
 # =========================================================
-# 3. Интерфейс Gradio
+# 3️⃣  Интерфейс Gradio
 # =========================================================
-with gr.Blocks(title="MBTI Interactive Interview (Gemma)") as demo:
-    gr.Markdown(
-        "## 🧠 MBTI Personality Interviewer\n"
-        "Определи личностный тип и получи следующий вопрос от интервьюера."
-    )
-
-    question_state = gr.State(1)
+with gr.Blocks(theme=gr.themes.Soft(), title="Adaptive MBTI Interviewer") as demo:
+    gr.Markdown("## 🧠 Adaptive MBTI Interviewer\nОпредели личностный тип и получи следующий вопрос от интервьюера.")
 
     with gr.Row():
-        with gr.Column(scale=1):
-            inp = gr.Textbox(
-                label="Введите свой ответ",
-                placeholder="Например: I enjoy working with people and organizing events.",
-                lines=4,
-            )
+        with gr.Column(scale=2):
+            inp = gr.Textbox(label="Ваш ответ", placeholder="Например: I enjoy working with people and organizing events.", lines=4)
             btn = gr.Button("Анализировать и задать новый вопрос", variant="primary")
-            progress = gr.Markdown("**1/30**", elem_id="progress")
-
-        with gr.Column(scale=1):
-            out_analysis = gr.Textbox(label="📊 Анализ MBTI", lines=6)
-            out_question = gr.Textbox(
-                label="💬 Вопрос от интервьюера",
-                value="How do you usually spend your free time?",
-                lines=3,
-            )
-
-    btn.click(
-        fn=classify_and_ask,
-        inputs=[inp, question_state],
-        outputs=[out_analysis, out_question, progress],
-    )
+
+        with gr.Column(scale=2):
+            mbti_out = gr.Textbox(label="📊 Анализ MBTI", lines=5)
+            question_out = gr.Textbox(label="💬 Следующий вопрос от интервьюера", lines=3)
+            counter = gr.Textbox(label="Прогресс", value="0/30")
+
+    btn.click(fn=analyze_and_ask, inputs=[inp, counter], outputs=[mbti_out, question_out, counter])
 
 # =========================================================
-# 4. Запуск
+# 4️⃣  Запуск
 # =========================================================
 if __name__ == "__main__":
     demo.launch()