Ucas-HaoranWei / Vary-toy

Official code implementation of Vary-toy (Small Language Model Meets with Reinforced Vision Vocabulary)
565 stars 41 forks source link

可以更一下 requirements.txt 么 #1

Open tpoisonooo opened 5 months ago

tpoisonooo commented 5 months ago
  File "/root/miniconda3/envs/torch2/lib/python3.10/site-packages/transformers/configuration_utils.py", line 265, in __getattribute__
    return super().__getattribute__(key)
AttributeError: 'varyConfig' object has no attribute 'seq_length'. Did you mean: 'max_length'?

看起来版本没有对齐

tpoisonooo commented 5 months ago

https://github.com/Ucas-HaoranWei/Vary/issues/41 找到了 hard code fix.

tpoisonooo commented 5 months ago

简单跑了一下,输出:

root@b51dfc54d6c4:/workspace/GitProjects/Vary-toy/Vary-master# python3 vary/demo/run_qwen_vary.py  --model-name /models/Vary-toy/   --image-file /workspace/test-vary.png                                              
2024-01-23 19:32:24.344309: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in perfo
rmance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
You are using a model of type mmgpt to instantiate a model of type vary. This is not supported for all configurations of models and can yield errors.
[4, 23] Database licensees: The data, code, and checkpoint are intended and licensed for research use only. They are also restricted to use that follow the lic
ense agreementement of LLaMA, Vicia vina, GPT-4, Qwewn, and LLaA_Vicma.
[25] Usage and License Notices: The data, code, and checkpoint are available under the CC BY NC 4.0 International license ADL.

[wa, 6] Usage and License Notices: The data, code, and checkpoint are not made available for any other license (e.g., e.]许可证) licensed to others, except di
stributing the data to others with their approval must be sought or申 optimized for other license. The license grants the right of all rights related to all of
 these datasets belong to Loma.

输入图片: test-vary

Ucas-HaoranWei commented 5 months ago

截图长一点试一下呢?我们没训练过crop的数据,而且Vary-toy弱化了OCR

tpoisonooo commented 5 months ago

截图长一点试一下呢?我们没训练过crop的数据,而且Vary-toy弱化了OCR

那可以放出这些么?

  1. vary 的 weight
  2. vary 训练数据的 example

使用者的视角: 目前 vary 相关的 repo 有多个,包括 vary-toy 在内,没有复现论文所展示效果的直接途径。 既无训练数据,又无 pretrain。

yefangok commented 5 months ago

截图长一点试一下呢?我们没训练过crop的数据,而且Vary-toy弱化了OCR

vary-toy跟原版vary的区别是什么?看来不是单纯把LLM模型换了。

sixgod-666 commented 5 months ago

同求复现论文展示效果的方法

Ucas-HaoranWei commented 5 months ago

@tpoisonooo Vary weight的获取方式Vary repo里有详细写,训练example 论文里有详细写,请仔细阅读