unilm Assert错误:无法从命名空间推断模型类型,

svdrlsy4  于 5个月前  发布在  其他
关注(0)|答案(2)|浏览(71)

问题描述:在使用TrOCR预训练模型(如trocr-base-stage1.pt,trocr-base-printed.pt等)时,遇到了一个错误。当运行以下脚本时,会出现以下错误:AssertionError: Could not infer model type from Namespace

错误发生在以下代码段中:

if __name__ == '__main__':
    model_path = '/content/trocr-base-printed.pt'
    jpg_path = "/content/pan4.jpeg"
    beam = 5
    

model, cfg, task, generator, bpe, img_transform, device = init(model_path, beam)

sample = preprocess(jpg_path, img_transform)

text = get_text(cfg, generator, model, sample, bpe)

print(text)
print('done')


=======================================================================

### | [label] load dictionary: 50265 types

错误原因:在初始化模型时,无法从Namespace中推断出模型类型。

t1rydlwq

t1rydlwq1#

"trocr-base-printed"在我这边运行正常,"trocr-base-stage1"的错误已经修复,请再次尝试。
如果"trocr-base-printed"仍然出现错误,能否向我展示完整的错误信息?谢谢~:)

ldioqlga

ldioqlga2#

你好,
我正在面临与"trocr-large-printed"相同的问题。有任何解决方案吗?
谢谢!

相关问题