Spaces:
Runtime error
Runtime error
Commit
·
b233041
1
Parent(s):
e590d0d
ソースコードの追加
Browse files
app.py
CHANGED
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# りんなGPT-2-medium ファインチューニングやってみた
|
2 |
+
|
3 |
+
# パッケージのインストール
|
4 |
+
!git clone https://github.com/huggingface/transformers -b v4.23.1
|
5 |
+
!pip install transformers==4.23.1
|
6 |
+
!pip install evaluate==0.3.0
|
7 |
+
!pip install sentencepiece==0.1.97
|
8 |
+
|
9 |
+
%%time
|
10 |
+
|
11 |
+
# ファインチューニングの実行
|
12 |
+
!python ./transformers/examples/pytorch/language-modeling/run_clm.py \
|
13 |
+
--model_name_or_path=rinna/japanese-gpt2-medium \
|
14 |
+
--train_file=natsumesouseki.txt \
|
15 |
+
--validation_file=natsumesouseki.txt \
|
16 |
+
--do_train \
|
17 |
+
--do_eval \
|
18 |
+
--num_train_epochs=3 \
|
19 |
+
--save_steps=5000 \
|
20 |
+
--save_total_limit=3 \
|
21 |
+
--per_device_train_batch_size=1 \
|
22 |
+
--per_device_eval_batch_size=1 \
|
23 |
+
--output_dir=output/
|
24 |
+
|
25 |
+
from transformers import T5Tokenizer, AutoModelForCausalLM
|
26 |
+
|
27 |
+
# トークナイザーとモデルの準備
|
28 |
+
tokenizer = T5Tokenizer.from_pretrained("rinna/japanese-gpt2-medium")
|
29 |
+
model = AutoModelForCausalLM.from_pretrained("rinna/japanese-gpt2-medium")
|
30 |
+
|
31 |
+
# 推論の実行
|
32 |
+
def Chat(prompt, ):
|
33 |
+
input = tokenizer.encode(prompt, return_tensors="pt")
|
34 |
+
output = model.generate(input, do_sample=True, max_length=300, num_return_sequences=5)
|
35 |
+
return print(tokenizer.batch_decode(output))
|