Upload scripts/train_alizee_v2_stage1_sft.py with huggingface_hub

scripts/train_alizee_v2_stage1_sft.py

@@ -71,10 +71,12 @@ print("\n📝 Loading tokenizer...")
 tokenizer = AutoTokenizer.from_pretrained(
     BASE_MODEL,
     trust_remote_code=True,
+    model_max_length=MAX_SEQ_LENGTH,  # Set max length to prevent overflow warnings
 )
 if tokenizer.pad_token is None:
     tokenizer.pad_token = tokenizer.eos_token
 tokenizer.padding_side = "right"
+tokenizer.truncation_side = "right"  # Truncate from the right for long sequences
 
 # QLoRA quantization config
 print("\n⚙️ Configuring 4-bit quantization...")