Training does not stop

, , ,
1

I am trying to train a multilabel classifier with the command

prodigy train --textcat-multilabel my_data --label-stats True --base-model en_core_web_lg --lang "en" --eval-split 0.2 >> ./train_logs/mylog.log

as I understand , this should create a default config file, after which training starts.

The problem I am having is that training does not seem to stop, as shown in the logs:

E    #       LOSS TOK2VEC  LOSS TEXTC...  CATS_SCORE  SPEED   SCORE 
---  ------  ------------  -------------  ----------  ------  ------
  0       0          0.00           0.61       37.34  3284.26    0.37
  3    1000          0.00         702.23       44.16  3350.89    0.44
...
1543   67000          0.00           0.64       53.53  3238.98    0.54
1567   68000          0.00           0.66       53.52  3310.02    0.54

Can someone please point me to what parameter I should be using that would allow me to stop training after a certain number of iterations?

I have played around with the below as well

prodigy train --textcat-multilabel match_eq_3_label_prodigy_input --label-stats True  --base-model en_core_web_lg --lang "en" --config test_config.cfg 
ℹ Using CPU

========================= Generating Prodigy config =========================
ℹ Using config from base model
✔ Generated training config

where i edit the training parameters as such:

[training]
max_epochs = 1
max_steps = 1000

but these dont seem to give the desired effect.

Any suggestions would be most welcome.
Thank you!!

2

Hi @JBunr,

I think you're on the right track in terms of setting the training parameters, but the reason why that config isn't being read is due to Prodigy trying to match the parameters for the base model. The solution is pretty much similar to your other post.

3

Thanks - sorry for the overlap in the two posts!