Remove max_seq_length from SFTTrainer (not supported in current TRL)

train.py

@@ -20,7 +20,6 @@ MODEL_ID = "google/gemma-2-2b-it"
 DATASET_ID = "talkmap/telecom-conversation-corpus"
 OUTPUT_DIR = "./gemma-2b-it-telecom"
 HUB_MODEL_ID = "ligaments-dev/gemma-2b-it-telecom"
-MAX_SEQ_LENGTH = 2048
 
 # ------------------------------------------------------------------
 # Trackio monitoring
@@ -111,7 +110,6 @@ trainer = SFTTrainer(
     args=args,
     train_dataset=train_dataset,
     processing_class=tokenizer,
-    max_seq_length=MAX_SEQ_LENGTH,
 )
 
 # ------------------------------------------------------------------