Add OtterHD and MagnifierBench, improve pipeline

for Pretrain, SFT, and RLHF.

README.md

@@ -53,7 +53,7 @@ For who in the mainland China: [![Open in OpenXLab](https://cdn-static.openxlab.
 <img src="https://i.postimg.cc/dtxQQzt6/demo0.png"  width="100%" height="100%">
 </div>
 
-1. 🦦 Added [OtterHD](./docs/OtterHD.md), a multimodal fine-tuned from [Fuyu-8B](https://huggingface.co/adept/fuyu-8b) to facilitate a more fine-grained interpretation of high-resolution visual input without a vision encoder. We've opensourced the finetune script for Fuyu-8B and improve training throughput by 4-5 times faster with [Flash-Attention-2](https://github.com/Dao-AILab/flash-attention). Try our finetune script at [OtterHD](./docs/OtterHD.md).
+1. 🦦 Added [OtterHD](./docs/OtterHD.md), a multimodal fine-tuned from [Fuyu-8B](https://huggingface.co/adept/fuyu-8b) to facilitate a more fine-grained interpretation of high-resolution visual input *without a explicit vision encoder module*. All image patches are linear transformed and processed together with text tokens. This is a very innovative and elegant exploration. We are fascinated and paved in this way, we opensourced the finetune script for Fuyu-8B and improve training throughput by 4-5 times faster with [Flash-Attention-2](https://github.com/Dao-AILab/flash-attention). Try our finetune script at [OtterHD](./docs/OtterHD.md).
 2. 🔍 Added [MagnifierBench](./docs/OtterHD.md), an evaluation benchmark tailored to assess whether the model can identify the tiny objects' information (1% image size) and spatial relationships.
 3. Improved pipeline for [Pretrain](pipeline/train/pretraining.py) | [SFT](pipeline/train/instruction_following.py) | [RLHF]() with (part of) current leading LMMs.
    1. **Models**: [Otter](https://arxiv.org/abs/2305.03726) | [OpenFlamingo](https://arxiv.org/abs/2308.01390) | [Idefics](https://huggingface.co/HuggingFaceM4/idefics-80b-instruct) | [Fuyu](https://huggingface.co/adept/fuyu-8b)