google/flan-t5-small . app.py v1.8

app.py

@@ -1,5 +1,6 @@
 # app.py
-import os, gradio as gr, asyncio
+import gradio as gr
+import asyncio
 from itertools import cycle
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 from core.utils import generate_first_question
@@ -8,41 +9,25 @@ from core.interviewer import generate_question, session_state
 
 
 # --------------------------------------------------------------
-# 🔐 Авторизация Hugging Face
+# ✅ Всегда используем публичную модель Flan-T5-Small
 # --------------------------------------------------------------
-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.")
-
-# --------------------------------------------------------------
-# ⚙️ Автоматическая загрузка доступной T5 модели с токеном
-# --------------------------------------------------------------
-def load_qg_model():
-    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 candidates:
-        try:
-            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",
-                model=mdl,
-                tokenizer=tok,
-                max_new_tokens=40,
-                num_beams=4,
-                no_repeat_ngram_size=4,
-            )
-        except Exception as e:
-            print(f"⚠️ Can't load {name}: {e}")
-    raise RuntimeError("❌ No available T5 model could be loaded even publicly.")
+QG_MODEL = "google/flan-t5-small"
+
+try:
+    tokenizer = AutoTokenizer.from_pretrained(QG_MODEL)
+    model = AutoModelForSeq2SeqLM.from_pretrained(QG_MODEL)
+    QG_PIPE = pipeline(
+        "text2text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        max_new_tokens=40,
+        num_beams=4,
+        no_repeat_ngram_size=4,
+    )
+    print(f"✅ Loaded public interviewer model: {QG_MODEL}")
+except Exception as e:
+    raise RuntimeError(f"❌ Failed to load {QG_MODEL}: {e}")
 
-QG_PIPE = load_qg_model()
 
 # --------------------------------------------------------------
 # 🌀 Асинхронная анимация "Thinking..."
@@ -53,8 +38,9 @@ async def async_loader(update_fn, delay=0.15):
         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():
@@ -68,14 +54,17 @@ 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
     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:
@@ -89,7 +78,7 @@ def analyze_and_ask(user_text, prev_count):
 
 
 # --------------------------------------------------------------
-# 🧱 Gradio UI
+# 🧱 Интерфейс Gradio
 # --------------------------------------------------------------
 with gr.Blocks(theme=gr.themes.Soft(), title="MBTI Personality Interviewer") as demo:
     gr.Markdown(