command line error while pretraining ner_food_ingredients - "no such option: --init-tok2vec"

,
1

Hi,
I tried to follow the workflow adopted in prodigy's NER tutorial -- "ner_food_ingredients".
However, I'm unable to run the command in Step 4 (projects/tutorials/ner_food_ingredients at v3 · explosion/projects · GitHub) -- i.e. the command to pretrain a model and use the pretrained tok2vec weights to create tmp_model.

I get the following error on my command line -- "no such option: --init-tok2vec" (I use Windows - Anaconda Prompt)

Screenshot (385)
Screenshot (385)1462×350 19.5 KB

Has the terminal's command been updated since the release of the tutorial video for the ner_food_ingredients project on YouTube?

2

Hi @nanyasrivastav , welcome to Prodigy!

Yes, the terminal command has been updated since the tutorial. Some of the optional parameters should now be specified in a config when training your spaCy model: https://spacy.io/api/data-formats#config

But you can also override these options in the command-line by passing:

... ---paths.init_tok2vec /some/path/to/tok2vec.bin
3

Thank you for the reply! I will try them out.

4

Hi, I've been having the same issue watching the same tutorial with --init-tok2vec and tried the suggested solution of ---paths.init_tok2vec. However, ended up with the following.

Error parsing config overrides
-paths -> init_tok2vec not a section value that can be overridden

Has the terminal commands been updated? or any other potential config that I'm missing?

5

Hi @chiaseeds and welcome to the forum :slight_smile:

It looks like there is a typo in Lj's answer. Could you try instead:

--paths.init_tok2vec=path_to_your_tok2vec

If that doesn't work please send me the entire command you're running as well as your Prodigy version.
Thank you!