Update goai_helpers/goai_traduction.py

goai_helpers/goai_traduction.py

@@ -1,6 +1,6 @@
 import torch
 import spaces
-from transformers import pipeline, AutoModelForSeq2SeqLM, AutoTokenizer
+from transformers import pipeline, AutoModelForSeq2SeqLM, AutoTokenizer, GenerationConfig
 from peft import PeftModel, PeftConfig
 import os
 import unicodedata
@@ -27,12 +27,13 @@ def goai_traduction(text, src_lang, tgt_lang):
 
     tgt_lang_id = tokenizer.convert_tokens_to_ids(tgt_lang)
 
-    model.config.forced_bos_token_id = tgt_lang_id
-    model.config.decoder_start_token_id = tgt_lang_id
-
-    print("decoder_start_token_id:", model.config.decoder_start_token_id)
-    print("forced_bos_token_id:", model.config.forced_bos_token_id)
-    print("eos_token_id:", model.config.eos_token_id)
+    generation_config = GenerationConfig(
+    max_new_tokens=1024,
+    early_stopping=False,
+    decoder_start_token_id=tokenizer.convert_tokens_to_ids(tgt_lang),
+    forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_lang),
+    eos_token_id=tokenizer.eos_token_id,
+    )
 
     
     # Ajout du code de langue source
@@ -47,12 +48,11 @@ def goai_traduction(text, src_lang, tgt_lang):
     
     # Génération avec paramètres améliorés
     outputs = model.generate(
-        **inputs,
-        forced_bos_token_id=tgt_lang_id,
-        eos_token_id=tokenizer.eos_token_id,
-        max_length=512
+    **inputs,
+    generation_config=generation_config
     )
 
+
     print("Token IDs:", outputs)
     print("Tokens:", [tokenizer.decode([tok]) for tok in outputs[0]])