QAway-to
commited on
Commit
·
58842ae
1
Parent(s):
09ad1b8
Change tokenizer v1.2
Browse files- core/interviewer.py +6 -6
core/interviewer.py
CHANGED
|
@@ -6,15 +6,15 @@ Generates random MBTI-style questions using a fine-tuned model.
|
|
| 6 |
🇷🇺 Модуль интервьюера.
|
| 7 |
Использует fine-tuned модель для генерации вопросов без промптов и инструкций.
|
| 8 |
"""
|
|
|
|
| 9 |
|
| 10 |
-
import random, torch, re
|
| 11 |
-
from transformers import AutoModelForSeq2SeqLM, T5Tokenizer
|
| 12 |
-
|
| 13 |
-
# --------------------------------------------------------------
|
| 14 |
-
# 1️⃣ Настройки модели
|
| 15 |
-
# --------------------------------------------------------------
|
| 16 |
QG_MODEL = "f3nsmart/ft-flan-t5-base-qgen"
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
# ❗ Используем "slow" SentencePiece токенайзер
|
| 19 |
tokenizer = T5Tokenizer.from_pretrained(QG_MODEL, use_fast=False)
|
| 20 |
model = AutoModelForSeq2SeqLM.from_pretrained(QG_MODEL)
|
|
|
|
| 6 |
🇷🇺 Модуль интервьюера.
|
| 7 |
Использует fine-tuned модель для генерации вопросов без промптов и инструкций.
|
| 8 |
"""
|
| 9 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
QG_MODEL = "f3nsmart/ft-flan-t5-base-qgen"
|
| 12 |
|
| 13 |
+
# ✅ Используем fast-tokenizer — он уже есть в модели
|
| 14 |
+
tokenizer = AutoTokenizer.from_pretrained(QG_MODEL, use_fast=True)
|
| 15 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(QG_MODEL)
|
| 16 |
+
|
| 17 |
+
|
| 18 |
# ❗ Используем "slow" SentencePiece токенайзер
|
| 19 |
tokenizer = T5Tokenizer.from_pretrained(QG_MODEL, use_fast=False)
|
| 20 |
model = AutoModelForSeq2SeqLM.from_pretrained(QG_MODEL)
|