From 4584208df9234d81766725ac5674b7bacbbeb029 Mon Sep 17 00:00:00 2001 From: Jun Chen Date: Mon, 17 Apr 2023 01:46:10 +0300 Subject: [PATCH] Update README.md --- README.md | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 581153e..8851234 100644 --- a/README.md +++ b/README.md @@ -24,15 +24,13 @@ More examples can be found in the [project page](https://minigpt-4.github.io). ## Introduction - MiniGPT-4 aligns a frozen visual encoder from BLIP-2 with a frozen LLM, Vicuna, using just one projection layer. -- The training of MiniGPT-4 consists of a first pretrain stage using roughly 5 million aligned image-text pairs for 10 hours on 4 A100s and a second finetuning stage using additional 3,500 carefully curated high-quality pairs for 7 minutes on 1 A100. -- MiniGPT-4 processes many emerging vision-language capabilities similar to those exhibited by GPT-4. +- We train MiniGPT-4 with two stages. The first pretraining stage is trained using roughly 5 million aligned image-text pairs with around 40 A100 hours. The second finetuning stage is trained using additional 3,500 carefully curated high-quality pairs with around 7 A100 minutes. +- MiniGPT-4 yields many emerging vision-language capabilities similar to those demonstrated in GPT-4. ![overview](figs/overview.png) - - ## Getting Started ### Installation