Create obyzala.py

obyzala.py

@@ -0,0 +1,56 @@
+from transformers import AutoTokenizer, AutoModelForQuestionAnswering, Trainer, TrainingArguments
+from datasets import Dataset
+import requests
+
+pisyn = requests.get("https://raw.githubusercontent.com/Fixyres/FHeta/refs/heads/main/modules.json")
+data = [
+    {"question": "Какая твоя база данных модулей? И по какой базе ты ищешь все модули?", "answer": pisyn.text}
+]
+
+tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
+model = AutoModelForQuestionAnswering.from_pretrained("distilbert-base-uncased")
+
+dataset = Dataset.from_dict(data)
+
+def preprocess_function(examples):
+    questions = examples["question"]
+    answers = examples["answer"]
+    inputs = tokenizer(questions, padding=True, truncation=True, return_tensors="pt")
+    with tokenizer.as_target_tokenizer():
+        labels = tokenizer(answers, padding=True, truncation=True, return_tensors="pt")
+    inputs["labels"] = labels["input_ids"]
+    return inputs
+
+tokenized_datasets = dataset.map(preprocess_function, batched=True)
+
+training_args = TrainingArguments(
+    output_dir="./results",
+    num_train_epochs=3,
+    per_device_train_batch_size=8,
+    logging_dir="./logs",
+    logging_steps=10,
+)
+
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_datasets,
+)
+
+trainer.train()
+
+model.save_pretrained("./FHeta")
+tokenizer.save_pretrained("./FHeta")
+
+tokenizer = AutoTokenizer.from_pretrained("./FHeta")
+model = AutoModelForQuestionAnswering.from_pretrained("./FHeta")
+
+def get_answer(query):
+    inputs = tokenizer(query, return_tensors="pt")
+    outputs = model(**inputs)
+    answer = tokenizer.decode(outputs["logits"][0], skip_special_tokens=True)
+    return answer
+
+query = "Модуль FHeta"
+answer = get_answer(query)
+print(answer)