Update trainingscript.py

trainingscript.py

@@ -1,7 +1,8 @@
 from transformers import T5ForConditionalGeneration, T5Tokenizer, Trainer, TrainingArguments
 from datasets import load_dataset
 
-model_name = "t5-base"
+
+model_name = "google/mt5-base"
 tokenizer = T5Tokenizer.from_pretrained(model_name)
 model = T5ForConditionalGeneration.from_pretrained(model_name)
 
@@ -11,37 +12,38 @@ def preprocess_function(examples):
     for ex in examples["before after"]:
         if ex is not None:
             splits = ex.split(" before after ")
-            if len(splits) == 2:
-                before_list.append(splits[0])
-                after_list.append(splits[1])
-            else:
-                before_list.append(ex)
-                after_list.append('')
+            before_list.append(splits[0] if len(splits) == 2 else ex)
+            after_list.append(splits[1] if len(splits) == 2 else '')
         else:
             before_list.append('')
             after_list.append('')
 
-    model_inputs = tokenizer(before_list, padding="max_length", truncation=True)
-    labels = tokenizer(after_list, padding="max_length", truncation=True)
+    # Токенизация с ограничением по длине
+    model_inputs = tokenizer(before_list, padding="max_length", truncation=True, max_length=512)
+    labels = tokenizer(after_list, padding="max_length", truncation=True, max_length=512)
     model_inputs["labels"] = labels["input_ids"]
     return model_inputs
 
-
 dataset = load_dataset("csv", data_files={"train": "converted.csv"}, delimiter=" ", column_names=["before after"])
 
+
 tokenized_datasets = dataset.map(preprocess_function, batched=True)
 
+
 training_args = TrainingArguments(
-    output_dir="./results1",
+    output_dir="./results2",
     evaluation_strategy="epoch",
     save_strategy="epoch",
     learning_rate=2e-5,
-    per_device_train_batch_size=64,
-    per_device_eval_batch_size=64,
+    per_device_train_batch_size=1,
+    per_device_eval_batch_size=1,
     num_train_epochs=1,
     weight_decay=0.01,
+    gradient_accumulation_steps=64,
+    fp16=True,
 )
 
+
 trainer = Trainer(
     model=model,
     args=training_args,
@@ -49,6 +51,13 @@ trainer = Trainer(
     tokenizer=tokenizer,
 )
 
+
 trainer.train()
-model.save_pretrained("T5Autocorrection")
+
+
+for param in model.parameters():
+    param.data = param.data.contiguous()
+
+
+model.save_pretrained("T5Autocorrection", safe_serialization=False)  # Отключаем safetensors для простого сохранения
 tokenizer.save_pretrained("T5TokenizerAutocorrection")