From 6b9dfa022e97133483f2a8d18705e1823c6ca0b3 Mon Sep 17 00:00:00 2001
From: "Teo (Timothy) Wu Haoning" <38696372+teowu@users.noreply.github.com>
Date: Tue, 7 Nov 2023 13:01:11 +0800
Subject: [PATCH] Update README.md

---
 scripts/llava_v1.5/README.md | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/scripts/llava_v1.5/README.md b/scripts/llava_v1.5/README.md
index b6006fa..a50b67e 100644
--- a/scripts/llava_v1.5/README.md
+++ b/scripts/llava_v1.5/README.md
@@ -1,6 +1,8 @@
 ## Training with Q-Instruct
 
-This document provides instruction on how to train with **Q-Instruct** dataset on LLaVA-v1.5 (7B/13B).
+This document provides instruction on how to train with **Q-Instruct** dataset on LLaVA-v1.5 (7B/13B), under the proposed two strategies (***mix*** and ***after***), shown as follows.
+
+![](strategies.png)
 
 
 ### Step 0: Pre-requisites
@@ -89,6 +91,12 @@ Please make sure you have enough computational resources before training.
 
 #### Strategy (a): Mix with High-level Datasets
 
+- *(Pre-requisite)* Get Pre-trained Multi-modal Projectors from original LLaVA-v1.5
+
+```shell
+sh scripts/llava_v1.5/get_mm_projectors.sh
+```
+
 - 13B *(requires 8x A100 80G), 21h*
 
 ```shell