From dd86a4df3781959c4b9c6778095f64beadd6ec70 Mon Sep 17 00:00:00 2001
From: Christian Medina <37550954+cmedinasoriano@users.noreply.github.com>
Date: Thu, 2 Jul 2026 16:39:21 -0400
Subject: [PATCH] fix: rename dataset column to 'text' for SFTTrainer
 compatibility

---
 train.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/train.py b/train.py
index dd127ae..54a57fa 100644
--- a/train.py
+++ b/train.py
@@ -140,13 +140,19 @@ def train(config_path):
     text_column = config["dataset"][0].get("text_column", "text")
     print(f"Using text column: {text_column}")
     
+    # Rename column to 'text' if needed (SFTTrainer expects 'text')
+    if text_column != "text":
+        print(f"   Renaming '{text_column}' column to 'text'...")
+        dataset["train"] = dataset["train"].rename_column(text_column, "text")
+        if "test" in dataset and dataset["test"] is not None:
+            dataset["test"] = dataset["test"].rename_column(text_column, "text")
+    
     trainer = SFTTrainer(
         model=model,
         processing_class=tokenizer,
         train_dataset=dataset["train"],
         eval_dataset=dataset.get("test"),
         args=training_args,
-        dataset_text_field=text_column,
     )
     
     # Train