Upload model_translation.py

model_translation.py

@@ -58,6 +58,21 @@ def get_tokenizer_model_for_src_lang(src_lang: str) -> (AutoTokenizer, AutoModel
 # - e.g. 200 words
 max_words_per_chunk = 200
 
+#
+# Multilingual language pairs
+#
+from transformers import M2M100Tokenizer, M2M100ForConditionalGeneration
+
+model_name_m2m100 = "facebook/m2m100_418M"
+tokenizer_m2m100 = M2M100Tokenizer.from_pretrained(model_name_m2m100)
+model_m2m100 = M2M100ForConditionalGeneration.from_pretrained(
+    model_name_m2m100,
+    device_map="auto",
+    torch_dtype=torch.float16,
+    low_cpu_mem_usage=True,
+    load_in_8_bit=True
+)
+
 #
 # Multilingual translation model
 #
@@ -68,4 +83,6 @@ model_multilingual = AutoModelForSeq2SeqLM.from_pretrained(
     model_MADLAD_name,
     device_map="auto",
     torch_dtype=torch.float16,
-    low_cpu_mem_usage=True)
+    low_cpu_mem_usage=True,
+    load_in_8bit=True
+)