mod allegro

Dockerfile

@@ -0,0 +1,31 @@
+# Użyj oficjalnego obrazu Python jako bazowego
+FROM --platform=linux/amd64 python:3.9-slim
+
+# Ustaw katalog roboczy w kontenerze
+WORKDIR /app
+
+# Zainstaluj git
+RUN apt-get update && apt-get install -y git nano wget curl iputils-ping 
+
+# Skopiuj pliki wymagań (jeśli istnieją) i zainstaluj zależności
+COPY requirements.txt .
+RUN pip install --upgrade pip
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Skopiuj plik requirements.txt do kontenera
+COPY requirements.txt .
+
+# Zainstaluj zależności z pliku requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Zainstaluj Tesseract OCR
+RUN apt-get install -y tesseract-ocr
+
+# Skopiuj kod źródłowy do kontenera
+COPY . .
+COPY entrypoint.sh /entrypoint.sh
+
+RUN chmod +x /entrypoint.sh
+
+# Uruchom aplikację
+ENTRYPOINT ["/entrypoint.sh"]

allegro.py

@@ -1,119 +1,9 @@
-import os
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-import re
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer, DataCollatorForLanguageModeling
-from datasets import Dataset
 
-# Konfiguracja
+model = AutoModelForSeq2SeqLM.from_pretrained("allegro/multislav-5lang")
-os.environ["TOKENIZERS_PARALLELISM"] = "false"
+tokenizer = AutoTokenizer.from_pretrained("allegro/multislav-5lang")
-MODEL_NAME = "allegro/herbert-base-cased"
-SPECIAL_TOKENS = ["[CITATION_START]", "[CITATION_END]"]
-TEXT_FILE_PATH = "./docs/kodekspracy.txt"  # Zmień na właściwą ścieżkę
 
-def prepare_dataset_from_file(file_path):
+model.save_pretrained("./models/ably")
-    with open(file_path, 'r', encoding='utf-8') as f:
+tokenizer.save_pretrained("./models/ably")
-        text = f.read()
 
-    articles = re.findall(r'Art\.\s*\d+[a-z]*\..*?(?=\s*Art\.\s*\d+[a-z]*\.|\Z)', text, flags=re.DOTALL)
+print("✅ Model został wytrenowany i zapisany!")
-    
-    formatted_articles = []
-    for article in articles:
-        article = ' '.join(article.strip().split())
-        
-        art_match = re.match(r'Art\.\s*(\d+[a-z]*)\.?\s*(.*)', article, re.DOTALL)
-        if art_match:
-            art_number = art_match.group(1)
-            art_text = art_match.group(2)
-            
-            paragraphs = re.split(r'(§\s*\d+\.)', art_text)
-            if len(paragraphs) > 1:
-                formatted_paragraphs = []
-                for i in range(1, len(paragraphs), 2):
-                    para_num = paragraphs[i].strip()
-                    para_text = paragraphs[i+1].strip()
-                    formatted_paragraphs.append(f"{para_num} {para_text}")
-                formatted = f"[CITATION_START] Kodeks Pracy, Art. {art_number} [CITATION_END]\n" + "\n".join(formatted_paragraphs)
-            else:
-                formatted = f"[CITATION_START] Kodeks Pracy, Art. {art_number} [CITATION_END] {art_text}"
-            
-            formatted_articles.append({"text": formatted})
-            
-            questions = [
-                f"Zacytuj artykuł {art_number} Kodeksu pracy.",
-                f"Co mówi artykuł {art_number} Kodeksu pracy?",
-                f"Podaj treść artykułu {art_number} Kodeksu pracy."
-            ]
-            for question in questions:
-                formatted_articles.append({"text": f"{question}\n{formatted}"})
-    
-    return formatted_articles
-
-def main():
-    # Inicjalizacja tokenizera
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-    tokenizer.pad_token = tokenizer.eos_token
-    tokenizer.add_special_tokens({"additional_special_tokens": SPECIAL_TOKENS})
-
-    print(f"Pad token: {tokenizer.pad_token}")
-    print(f"Pad token ID: {tokenizer.pad_token_id}")
-
-    # Przygotowanie danych
-    data = prepare_dataset_from_file(TEXT_FILE_PATH)
-    dataset = Dataset.from_dict({"text": [d["text"] for d in data]})
-
-    # Tokenizacja
-    def tokenize_function(examples):
-        tokenized = tokenizer(
-            examples["text"],
-            truncation=True,
-            padding="max_length",
-            max_length=512,
-            return_tensors="pt"
-        )
-        tokenized["labels"] = tokenized["input_ids"].clone()
-        return tokenized
-
-    tokenized_dataset = dataset.map(tokenize_function, batched=True, remove_columns=dataset.column_names)
-
-    # Model i data collator
-    model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
-    model.resize_token_embeddings(len(tokenizer))
-    model.config.pad_token_id = tokenizer.pad_token_id
-    
-    data_collator = DataCollatorForLanguageModeling(
-        tokenizer=tokenizer,
-        mlm=False
-    )
-
-    # Konfiguracja treningu
-    training_args = TrainingArguments(
-        output_dir="./results",
-        num_train_epochs=32,
-        per_device_train_batch_size=2,
-        learning_rate=1e-5,
-        logging_steps=10,
-        weight_decay=0.01,
-        report_to="none",
-        save_strategy="steps",
-        save_steps=500,
-        evaluation_strategy="steps",
-        eval_steps=500,
-        load_best_model_at_end=True,
-    )
-
-    # Trainer
-    trainer = Trainer(
-        model=model,
-        args=training_args,
-        train_dataset=tokenized_dataset,
-        eval_dataset=tokenized_dataset,
-        data_collator=data_collator
-    )
-
-    print("Rozpoczęcie treningu...")
-    trainer.train()
-    trainer.save_model("./trained_model/allegro")
-    tokenizer.save_pretrained("./trained_model/allegro")
-
-if __name__ == "__main__":
-    main()

entrypoint.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+git config --global credential.helper store
+git config --global user.name ${GIT_USERNAME}
+git config --global user.email ${GIT_EMAIL}
+echo "https://${GIT_USERNAME}:${GIT_TOKEN}@${GIT_HOST}" > ~/.git-credentials
+cd /home
+git clone --single-branch --branch main/finetuning https://repo.pokash.pl/POKASH.PL/ably.do.git
+python /app/${MODELNAME}.py
+
+# Po zakończeniu głównego procesu, przejdź w tryb czuwania
+echo "Główny proces zakończony. Przechodzę w tryb czuwania..."
+tail -f /dev/null

requirements.txt

@@ -4,5 +4,13 @@ datasets>=2.13.1
 Pillow>=9.4.0
 pytesseract>=0.3.10
 python-docx>=0.8.11
-PyPDF2>=3.0.1
+pypdf
+PyPDF2
 huggingface-hub>=0.16.4
+numpy
+peft
+weaviate-client
+sentence_transformers
+faiss-gpu
+sentencepiece
+sacremoses