Update djezzy.py
Browse files
djezzy.py
CHANGED
@@ -29,17 +29,7 @@ model_name = "sentence-transformers/all-MiniLM-L6-v2"
|
|
29 |
embedding_llm = SentenceTransformerEmbeddings(model_name=model_name)
|
30 |
|
31 |
from transformers import T5Tokenizer, T5ForConditionalGeneration
|
32 |
-
|
33 |
-
# Activer la quantification imbriquée pour les modèles de base 8 bits
|
34 |
-
use_nested_quant = True
|
35 |
-
compute_dtype = getattr(torch, bnb_8bit_compute_dtype)
|
36 |
-
|
37 |
-
bnb_config =BitsAndBytesConfig(
|
38 |
-
load_in_8bit=True, # Activer la quantification 8 bits pour le chargement du modèle
|
39 |
-
bnb_8bit_quant_type="...",
|
40 |
-
bnb_8bit_compute_dtype="uint8", # Définissez le type de données de calcul sur un entier non signé de 8 bits
|
41 |
-
bnb_8bit_use_double_quant=False, # Désactiver la quantification imbriquée (si non applicable)
|
42 |
-
)
|
43 |
tokenizer1 = T5Tokenizer.from_pretrained("google/flan-t5-base")
|
44 |
model1 = T5ForConditionalGeneration.from_pretrained("google/flan-t5-base")
|
45 |
with tempfile.TemporaryDirectory() as temp_dir:
|
|
|
29 |
embedding_llm = SentenceTransformerEmbeddings(model_name=model_name)
|
30 |
|
31 |
from transformers import T5Tokenizer, T5ForConditionalGeneration
|
32 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
tokenizer1 = T5Tokenizer.from_pretrained("google/flan-t5-base")
|
34 |
model1 = T5ForConditionalGeneration.from_pretrained("google/flan-t5-base")
|
35 |
with tempfile.TemporaryDirectory() as temp_dir:
|