kydxh

IndexError: piece id is out of range.

2

I want to inference llava v1.5, but encountered error "IndexError: piece id is out of range". The environment I used is "llava 1.1.3, transformers 4.31.0, sentencepiece 0.1.99, torch 2.0.1" The...

Why resize images in MMMU for DeepSeek-VL2?

1

I notice that in evaluation of DeepSeek-VL2 on MMMU, the code resize the first image. But I don't know why. Could you please tell me the reasons? (codes from "generate_inner"...

1. I'm confused as to why for llava1.5, the model used is "liuhaotian/llava-v1.5-7b", but for llava 1.6, it's "llava-hf/llava-v1.6-vicuna-7b-hf" instead of "liuhaotian/llava-v1.6-vicuna-7b"? 2. I run the code on "llava_v1.5_7b" but...

Inputs without JPEG conversion

2

Hello, I've notice that the input images used in VLMEvalKit are converted to JPEG format. I wonder if there is version that use the original images instead of JPEG format...

您好，之前在issue中有提到说关于TextVQA等VQA的数据集的评测仍然在完善中，请问现在关于TextVQA数据集的评测是否完善了呢？

MMMU和MathVista在gpt-4o的结果

3

想请问下MMMU和MathVista上是怎么设置的prompt得到的69.9 | 62.7的结果（GPT-4o-20240806）？因为我在让模型直接回答问题时获得的结果远低于这两个数值，在用上COT之后才能接近，但是看起来在[VLMEvalKit]中并没有使用COT？或者除了prompt设置有其他单独的设置？

llava-next的代码可以被用于llava或llava1.5的推理吗？

1

请问llava-next是否是兼容之前的版本的呢？即我是否可以使用llava-next加载llava或llava1.5模型并进行推理？

### ❔ Any questions What does the parameter “eot_iter” mean in the UPGD attack class? I saw in the comments that it is explained as "number of models to estimate...

### ❔ Any questions In the demo "demo/White-box Attack on ImageNet.ipynb", atk.set_normalization_used(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) is used. So after attack, the image will be normalized. In the code,...

paper or reference for R-GAT

1

## ❓ Questions and Help I would like to know which paper or references are related to the R-GAT code (https://github.com/dmlc/dgl/tree/master/examples/pytorch/rgat). If I want to use the R-GAT model, how...