Update goai_helpers/goai_traduction.py

goai_helpers/goai_traduction.py

@@ -50,32 +50,16 @@ def goai_traduction(text, src_lang, tgt_lang):
     tgt_lang_id = tokenizer.convert_tokens_to_ids(tgt_lang)
     print(f"Token ID de la langue cible ({tgt_lang}): {tgt_lang_id}")
     bad_words_ids = [[tokenizer.eos_token_id]]
-    
-    for length_penalty in [1.0, 1.5, 2.0, 2.5, 3]:
-        for num_beams in [5, 10]:
-            print(f"\nTest avec length_penalty={length_penalty}, num_beams={num_beams}")
-            
 
-            outputs = model.generate(
-                **inputs,
-                forced_bos_token_id=tgt_lang_id,
-                max_new_tokens=2048,
-                early_stopping=False,
-                num_beams=num_beams,
-                no_repeat_ngram_size=0,
-                bad_words_ids=bad_words_ids,
-                length_penalty=length_penalty
-            )
-            
-        
-            translation = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
-            
-            print(f"Traduction ({len(translation)} caractères / {len(translation.split())} mots):")
-            print(translation)
-            output_ids = outputs[0]
-            print(f"- Nombre de tokens générés: {output_ids.shape[0]}")
-            print(f"- Premiers tokens générés: {output_ids[:10].tolist()}")
-            print(f"- Derniers tokens générés: {output_ids[-10:].tolist()}")
+    outputs = model.generate(
+        **inputs,
+        forced_bos_token_id=tgt_lang_id,
+        max_length=max_length,
+        min_length=max_length,
+        num_beams=5,
+        no_repeat_ngram_size=0,
+        length_penalty=2.0
+    )
             
     return translation