Use pre-trained model without Internet connection

[zhh210]

My network admin blocked any connection to aws so downloading pre-trained embeddings/models will fail due to network error. Is it possible to let flair point to the local directory where the embeddings/models were downloaded? Thanks.

[alanakbik]

Hello @zhh210 - yes, you can pass the full model path to initialize most embeddings and taggers. For instance, if you have a tagging model downloaded to 'home/taggers/tagger.pt', you can initialize a SequenceTagger like this:

tagger = SequenceTagger.load_from_file('home/taggers/tagger.pt')

If you want to initialize embeddings with downloaded model, you can do this by passing the full path to the default constructor, e.g.:

FlairEmbeddings('path/to/downloaded/model.pt')

[zhh210]

The example I was trying to use is glove. The download from is failing to connect aws (alan-nlp/resources/embeddings/glove.gensim) so I created copy in my local folder and in python when I was loading:
WordEmbeddings('/mydir/glove.gensim.vectors.npy')

It got an error saying:

UnpicklingError: STACK_GLOBAL requires str.

Similary when I point BytePairEmbeddings to a local folder containing en.wiki.bpe.vs100000.model, it gives error of:

Unknown language identifier

[alanakbik]

Hi @zhh210 can you try loading with WordEmbeddings('/mydir/glove.gensim') instead?

[TorusHHH]

@zhh210 老哥，需要下载两个文件，一个是glove.gensim.vectors.npy，一个是glove.gensim，然后放在同意目录下，然后用WordEmbeddings('/mydir/glove.gensim.vectors.npy')就行。

[Nicozwy]

你好，@TorusHHH 这两个 glove.gensim.vectors.npy，和glove.gensim在哪里可以下载？可否给个链接

[builtfordeve]

@TorusHHH 同问，哪里才能下载到glove.gensim和glove.gensim.vectors.npy这两个文件啊？

[Carolyn95]

你好，@TorusHHH 这两个 glove.gensim.vectors.npy，和glove.gensim在哪里可以下载？可否给个链接

可以用一台可以access internet的电脑，用flair自动下载好了然后从$HOME/.flair/embeddings中复制过去。

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[B0rner]

Is this example still valid:
tagger = SequenceTagger.load_from_file('home/taggers/tagger.pt')
??

I'm trying to run flair SequenceTagger oflline.
First I run the sctipt with internet connection to automatic download the files, unsing the following environment parameter: FLAIR_CACHE_ROOT=/models.
I created an tagger instance using: tagger = Classifier.load('flair/ner-english-ontonotes-large')

This works fine.

On the nex run, I disabled internet connection and rerun the script but using tagger = SequenceTagger.load_from_file('/models/ner-english..../pytorch_model.bin). The scrips runs into that error:

AttributeError: type object 'SequenceTagger' has no attribute 'load_from_file'

I know that site: https://flairnlp.github.io/docs but an class/methode - API documentation is missing. So I allways have to read endless lines of code to (sometimes) find which methods or parameters are valid.