mod allegro

allegro.py

@@ -1,16 +1,16 @@
 import os
-os.environ["TOKENIZERS_PARALLELISM"] = "false"
-
 import torch
 import weaviate
-from weaviate.classes.config import Property, DataType, Configure
+import numpy as np
-from weaviate.classes.query import Query
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, TrainingArguments, Trainer, DataCollatorForSeq2Seq
 from datasets import Dataset
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, TrainingArguments, Trainer, DataCollatorForSeq2Seq
+from weaviate.classes.config import Property, DataType, Configure
+from weaviate.classes.client import WeaviateClient
+from weaviate.classes.config import ConnectionParams
 
-# 1️⃣ Połączenie z bazą Weaviate
+# 1️⃣ Połączenie z Weaviate
-client = weaviate.WeaviateClient(
+client = WeaviateClient(
-    connection_params=weaviate.ConnectionParams.from_params(
+    connection_params=ConnectionParams.from_params(
         http_host="weaviate",
         http_port=8080,
         http_secure=False,
@@ -20,44 +20,42 @@ client = weaviate.WeaviateClient(
     )
 )
 
-# 2️⃣ Pobranie dokumentów z bazy Weaviate
+# 2️⃣ Pobranie dokumentów z Weaviate
-collection_name = "Documents"
+def fetch_documents():
-query = Query(collection_name).limit(1000)
+    query = client.query.get("Document", ["content", "fileName"]).do()
-result = client.query.run(query)
-
     documents = []
-file_names = []
+    for item in query["data"]["Get"]["Document"]:
+        file_name = item.get("fileName", "unknown_file")
+        content = item.get("content", "")
+        if content:
+            documents.append(f"fileName: {file_name}, content: {content}")
+    return documents
 
-for item in result[collection_name]['objects']:
+documents = fetch_documents()
-    documents.append(item['properties']['content'])
-    file_names.append(item['properties']['fileName'])
 
-# 3️⃣ Tworzenie datasetu
+# 3️⃣ Inicjalizacja modelu
-training_data = {
+model_name = "allegro/multislav-5lang"
-    "text": documents,
+device = "cuda" if torch.cuda.is_available() else "cpu"
-    "file_name": file_names
-}
-dataset = Dataset.from_dict(training_data)
 
-# Podział na zestaw treningowy i ewaluacyjny
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+# 4️⃣ Przygotowanie danych treningowych
+def create_training_data():
+    return Dataset.from_dict({"text": documents})
+
+dataset = create_training_data()
 split_dataset = dataset.train_test_split(test_size=0.25)
 train_dataset = split_dataset["train"]
 eval_dataset = split_dataset["test"]
 
-# 4️⃣ Ładowanie modelu Multislav
-model_name = "allegro/multislav-5lang"
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-
 # 5️⃣ Tokenizacja
-max_length = 512
 def tokenize_function(examples):
     return tokenizer(
         examples["text"],
         padding="max_length",
         truncation=True,
-        max_length=max_length
+        max_length=512
     )
 
 tokenized_train = train_dataset.map(tokenize_function, batched=True)
@@ -68,13 +66,13 @@ training_args = TrainingArguments(
     output_dir="./results",
     evaluation_strategy="steps",
     eval_steps=500,
-    save_strategy="steps",
     save_steps=500,
     learning_rate=2e-5,
-    per_device_train_batch_size=4,
+    per_device_train_batch_size=2,
-    per_device_eval_batch_size=4,
+    per_device_eval_batch_size=2,
-    num_train_epochs=10,
+    num_train_epochs=16,
     weight_decay=0.01,
+    save_total_limit=2,
     load_best_model_at_end=True,
     metric_for_best_model="loss",
     greater_is_better=False,
@@ -95,7 +93,7 @@ trainer = Trainer(
 trainer.train()
 
 # 9️⃣ Zapis modelu
-model.save_pretrained("./trained_model/multislav")
+model.save_pretrained("./models/allegro")
-tokenizer.save_pretrained("./trained_model/multislav")
+tokenizer.save_pretrained("./models/allegro")
 
 print("✅ Model został wytrenowany i zapisany!")