Update djezzy.py

djezzy.py

@@ -29,6 +29,9 @@ model_name = "sentence-transformers/all-MiniLM-L6-v2"
 embedding_llm = SentenceTransformerEmbeddings(model_name=model_name)
 
 from transformers import T5Tokenizer, T5ForConditionalGeneration
+bnb_8bit_compute_dtype="uint8"
+# Activer la quantification imbriquée pour les modèles de base 8 bits
+use_nested_quant = True
 compute_dtype = getattr(torch, bnb_8bit_compute_dtype)
 
 bnb_config =BitsAndBytesConfig(