From 9b615b8ffffbe8cec35460bb9a4eae2f9de007b8 Mon Sep 17 00:00:00 2001
From: HuiyingLi <willwin.lee@gmail.com>
Date: Thu, 15 Feb 2024 23:21:41 -0800
Subject: [PATCH] add prompt string to nvgpt

Signed-off-by: HuiyingLi <willwin.lee@gmail.com>
---
 nemo/collections/multimodal/data/neva/neva_dataset.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/nemo/collections/multimodal/data/neva/neva_dataset.py b/nemo/collections/multimodal/data/neva/neva_dataset.py
index ea778cd591452..38617460a5ad2 100644
--- a/nemo/collections/multimodal/data/neva/neva_dataset.py
+++ b/nemo/collections/multimodal/data/neva/neva_dataset.py
@@ -381,6 +381,8 @@ def preprocess_nvgpt(sources: dict, tokenizer, cfg,) -> Dict:
     - The function asserts that each message in a conversation alternates between the defined roles and skips messages not starting with the 'human' role.
     """
 
+    """<extra_id_0>System\nA chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.\n\n<extra_id_1>User\n{user input}\n<extra_id_1>Assistant\n<extra_id_2>quality:4,toxicity:0,humor:0,creativity:0,helpfulness:4,correctness:4,coherence:4,complexity:4,verbosity:4\n"""
+
     conv = conversation_lib.conv_nvgpt.copy()
 
     # Apply prompt templates