max_new_tokens

app.py

@@ -1,16 +1,17 @@
-# Based on example code of https://huggingface.co/facebook/m2m100_1.2B
+# Based on example code of https://huggingface.co/facebook/m2m100-12B-avg-5-ckpt
 
 
 import gradio as gr
 from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
 
-model = M2M100ForConditionalGeneration.from_pretrained("facebook/m2m100_1.2B")
+model = M2M100ForConditionalGeneration.from_pretrained(
+    "facebook/m2m100-12B-avg-5-ckpt")
 
-tokenizer = M2M100Tokenizer.from_pretrained("facebook/m2m100_1.2B")
+tokenizer = M2M100Tokenizer.from_pretrained("facebook/m2m100-12B-avg-5-ckpt")
 
 
 this_description = '''
-Using facebook/m2m100-12B-avg-5-ckpt pre-trained model. Some language code:
+Using facebook/m2m100-12B-avg-5-ckpt pre-trained model.
 
 Chinese(zh)
 English(en)
@@ -20,6 +21,7 @@ Sinhalese(si)
 Thai(th)
 Vietnamese(vi)
 ...
+
 '''
 
 # From facebook/m2m100-12B-avg-5-ckpt
@@ -129,7 +131,6 @@ lang_codes = {
 
 def m2m_translate(Input_Text, from_lang, to_lang):
     tokenizer.src_lang = lang_codes[from_lang]
-
     encoded_from_lang = tokenizer(Input_Text, return_tensors="pt")
 
     generated_tokens = model.generate(
@@ -143,21 +144,6 @@ def m2m_translate(Input_Text, from_lang, to_lang):
     return res[0]
 
 
-def m2m_translate2(Input_Text, from_lang, to_lang):
-    tokenizer.src_lang = lang_codes[from_lang]
-
-    encoded_from_lang = tokenizer(Input_Text, return_tensors="pt")
-
-    generated_tokens = model.generate(
-        **encoded_from_lang, forced_bos_token_id=tokenizer.get_lang_id(lang_codes[to_lang]))
-
-    res = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
-
-    return res[0]
-    # if res:
-    # return '\n'.join(res)
-
-
 iface = gr.Interface(
     fn=m2m_translate,