Token + T5 models switcher. app.py v1.8

app.py

@@ -1,29 +1,34 @@
 # app.py
-import gradio as gr
-import asyncio
+import os, gradio as gr, asyncio
 from itertools import cycle
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 from core.utils import generate_first_question
 from core.mbti_analyzer import analyze_mbti
 from core.interviewer import generate_question, session_state
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
+
+
+# --------------------------------------------------------------
+# 🔐 Авторизация Hugging Face
+# --------------------------------------------------------------
+HF_TOKEN = os.environ.get("HF_MBTI") or os.environ.get("HUGGINGFACEHUB_API_TOKEN")
+if HF_TOKEN:
+    print("✅ Hugging Face token detected.")
+else:
+    print("⚠️ No HF token found – only public models will work.")
 
 # --------------------------------------------------------------
-# ⚙️ Проверка и автоматический fallback T5 модели
+# ⚙️ Автоматическая загрузка доступной T5 модели с токеном
 # --------------------------------------------------------------
 def load_qg_model():
-    """
-    Пытаемся загрузить лёгкую T5-модель для генерации вопросов.
-    Если первая не доступна, переключаемся на запасную.
-    """
-    MODEL_CANDIDATES = [
-        "mrm8488/t5-small-finetuned-question-generation-ap",  # быстрая, но часто недоступна
-        "iarfmoose/t5-base-question-generator",               # качественная, чуть медленнее
-        "google/flan-t5-small"                                # fallback — всегда доступна
+    candidates = [
+        "mrm8488/t5-small-finetuned-question-generation-ap",  # fast, но часто приватная
+        "iarfmoose/t5-base-question-generator",               # стабильная
+        "google/flan-t5-small"                                # публичный fallback
     ]
-    for name in MODEL_CANDIDATES:
+    for name in candidates:
         try:
-            tok = AutoTokenizer.from_pretrained(name)
-            mdl = AutoModelForSeq2SeqLM.from_pretrained(name)
+            tok = AutoTokenizer.from_pretrained(name, token=HF_TOKEN, use_auth_token=HF_TOKEN)
+            mdl = AutoModelForSeq2SeqLM.from_pretrained(name, token=HF_TOKEN, use_auth_token=HF_TOKEN)
             print(f"✅ Loaded interviewer model: {name}")
             return pipeline(
                 "text2text-generation",
@@ -34,23 +39,22 @@ def load_qg_model():
                 no_repeat_ngram_size=4,
             )
         except Exception as e:
-            print(f"⚠️ Failed to load {name}: {e}")
-    raise RuntimeError("❌ No available T5 model for question generation.")
+            print(f"⚠️ Can't load {name}: {e}")
+    raise RuntimeError("❌ No available T5 model could be loaded even publicly.")
 
-# глобальный экземпляр пайплайна
 QG_PIPE = load_qg_model()
 
 # --------------------------------------------------------------
 # 🌀 Асинхронная анимация "Thinking..."
 # --------------------------------------------------------------
 async def async_loader(update_fn, delay=0.15):
-    frames = cycle(["⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"])
+    frames = cycle(["⠋","⠙","⠹","⠸","⠼","⠴","⠦","⠧","⠇","⠏"])
     for frame in frames:
         update_fn(f"💭 Interviewer is thinking... {frame}")
         await asyncio.sleep(delay)
 
 # --------------------------------------------------------------
-# ⚙️ Основная логика
+# 🧠 Основная логика
 # --------------------------------------------------------------
 def analyze_and_ask(user_text, prev_count):
     if not user_text.strip():
@@ -64,17 +68,14 @@ def analyze_and_ask(user_text, prev_count):
         n = 1
     counter = f"{n}/8"
 
-    # мгновенный отклик
     yield "⏳ Analyzing personality...", "💭 Interviewer is thinking... ⠋", counter
 
-    # анализ MBTI (стриминг)
     mbti_gen = analyze_mbti(user_text)
     mbti_text = ""
     for chunk in mbti_gen:
         mbti_text = chunk
         yield mbti_text, "💭 Interviewer is thinking... ⠙", counter
 
-    # генерация вопроса
     try:
         question = generate_question(user_id=user_id, user_answer=user_text, qg_pipe=QG_PIPE)
     except Exception as e:
@@ -86,8 +87,9 @@ def analyze_and_ask(user_text, prev_count):
 
     yield mbti_text, question, counter
 
+
 # --------------------------------------------------------------
-# 🧱 Интерфейс Gradio
+# 🧱 Gradio UI
 # --------------------------------------------------------------
 with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as demo:
     gr.Markdown(
@@ -100,7 +102,7 @@ with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as
             inp = gr.Textbox(
                 label="Ваш ответ",
                 placeholder="Например: I enjoy working with people and organizing events.",
-                lines=4,
+                lines=4
             )
             btn = gr.Button("Анализировать и задать новый вопрос", variant="primary")
         with gr.Column(scale=1):
@@ -112,13 +114,13 @@ with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as
         analyze_and_ask,
         inputs=[inp, progress],
         outputs=[mbti_out, interviewer_out, progress],
-        show_progress=True,
+        show_progress=True
     )
 
     demo.load(
         lambda: ("", generate_first_question(), "0/8"),
         inputs=None,
-        outputs=[mbti_out, interviewer_out, progress],
+        outputs=[mbti_out, interviewer_out, progress]
     )
 
 demo.queue(max_size=32).launch(server_name="0.0.0.0", server_port=7860)