Spaces:
Runtime error
Runtime error
Rename inference.py to train.py
Browse files- inference.py +0 -13
- train.py +31 -0
inference.py
DELETED
@@ -1,13 +0,0 @@
|
|
1 |
-
from transformers import MT5ForConditionalGeneration, MT5Tokenizer
|
2 |
-
|
3 |
-
model = MT5ForConditionalGeneration.from_pretrained("./model")
|
4 |
-
tokenizer = MT5Tokenizer.from_pretrained("google/mt5-small")
|
5 |
-
|
6 |
-
def ask(question, context):
|
7 |
-
input_text = f"Сұрақ: {question} Контекст: {context}"
|
8 |
-
input_ids = tokenizer(input_text, return_tensors="pt").input_ids
|
9 |
-
output = model.generate(input_ids, max_length=100)
|
10 |
-
return tokenizer.decode(output[0], skip_special_tokens=True)
|
11 |
-
|
12 |
-
context = """Мәліметтер қоры дегеніміз – белгілі бір сипаттамасы бар, өзара байланыса сақталатын ақпараттар жиынтығы."""
|
13 |
-
print(ask("Мәліметтер қоры дегеніміз не?", context))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
train.py
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datasets import load_dataset
|
2 |
+
from transformers import AutoTokenizer, AutoModelForQuestionAnswering, Trainer, TrainingArguments
|
3 |
+
|
4 |
+
model_name = "ai4bharat/indic-bert"
|
5 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
6 |
+
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
|
7 |
+
|
8 |
+
dataset = load_dataset("json", data_files="qa_dataset.json")
|
9 |
+
|
10 |
+
def preprocess(examples):
|
11 |
+
inputs = tokenizer(examples['question'], examples['context'], truncation=True, padding='max_length')
|
12 |
+
return inputs
|
13 |
+
|
14 |
+
dataset = dataset.map(preprocess, batched=True)
|
15 |
+
|
16 |
+
training_args = TrainingArguments(
|
17 |
+
output_dir="./model",
|
18 |
+
evaluation_strategy="no",
|
19 |
+
per_device_train_batch_size=4,
|
20 |
+
num_train_epochs=3
|
21 |
+
)
|
22 |
+
|
23 |
+
trainer = Trainer(
|
24 |
+
model=model,
|
25 |
+
args=training_args,
|
26 |
+
train_dataset=dataset['train']
|
27 |
+
)
|
28 |
+
|
29 |
+
trainer.train()
|
30 |
+
model.save_pretrained("./model")
|
31 |
+
tokenizer.save_pretrained("./model")
|