From ed0b1eddf8660542f85f574b255a198d9d7d0016 Mon Sep 17 00:00:00 2001 From: Christian Medina <37550954+cmedinasoriano@users.noreply.github.com> Date: Tue, 30 Jun 2026 21:05:48 -0400 Subject: [PATCH] fix: use messages format for SFTTrainer (system/user/assistant) --- training/scripts/prepare_dataset.py | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/training/scripts/prepare_dataset.py b/training/scripts/prepare_dataset.py index b10230e..c81fc4d 100755 --- a/training/scripts/prepare_dataset.py +++ b/training/scripts/prepare_dataset.py @@ -36,16 +36,20 @@ def prepare_dataset(input_file, output_file, test_size=0.05): output_text = ex['output'] - # Create conversation format + # Create chat format for SFTTrainer conversation = { - "conversations": [ + "messages": [ { - "from": "human", - "value": f"Generate a Cyron summary for this task:\n\n{instruction}\n\n{input_text}" + "role": "system", + "content": "You are a helpful coding assistant that generates project summaries." }, { - "from": "gpt", - "value": output_text + "role": "user", + "content": f"Generate a summary for this task:\n\n{instruction}\n\n{input_text}" + }, + { + "role": "assistant", + "content": output_text } ] }