fine tunning

Finetunning/finetunning.py

@@ -145,7 +145,7 @@ training_args = TrainingArguments(
     per_device_train_batch_size=1,
     gradient_accumulation_steps=16,
     learning_rate=1e-4,
-    num_train_epochs=3,
+    num_train_epochs=2,
     max_steps=1000,
 
     fp16=False,      # ⚠ disable AMP

Finetunning/validation.py

@@ -8,7 +8,7 @@ from nltk.translate.bleu_score import corpus_bleu
 # Configuration
 # ----------------------------
 BASE_MODEL = "Qwen/Qwen2.5-7B-Instruct"  # base model
-LORA_DIR = "./qwen2.5-7b-uk-fr-lora"    # fine-tuned LoRA
+LORA_DIR = "./qwen2.5-7b-uk-fr-lora-2epoch"    # fine-tuned LoRA
 VALIDATION_FILE = "validation.jsonl"    # small validation subset
 MAX_INPUT_LENGTH = 1024
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

Traduction/main.py

@@ -11,7 +11,6 @@ from reportlab.pdfbase.ttfonts import TTFont
 import os, time
 
 # Configuration
-DEBUG = True
 PDF_PATH = "Traduction/TaniaBorecMemoir(Ukr).pdf"
 OLLAMA_MODEL = "traductionUkrainienVersFrancais:latest"
 OLLAMA_URL = "http://localhost:11434/api/generate"
@@ -174,7 +173,6 @@ def load_checkpoint():
             return json.load(f)
     return {"last_processed_index": -1, "results": {}}
 
-# Sauvegarde le checkpoint
 # Sauvegarde le checkpoint
 def save_checkpoint(last_index, results):
     # Trier les clés du dictionnaire results