Same task shown to separate named sessions

Currently running v1.11.7 and using image.manual with feed_overlap = False. I have two different named sessions, however they are both shown the same image. I only want each image to be shown once. Is there another setting that I need to change in the config?

There are some edge cases where you may have exhausted your batches that could cause this, but I'll assume that's not the case.

Could you share your entire prodigy.json file so I can see the other settings as well? Do you have any environment variables like PRODIGY_ALLOWED_SESSIONS that might influence the sessions or the feed?

Could you also check if the base behaviour works if you resort to a NER task? I just ran an example with an examples.jsonl file that contains:

{"text": "hi my name is vincent1"}
{"text": "hi my name is vincent2"}
{"text": "hi my name is vincent3"}
{"text": "hi my name is vincent4"}
{"text": "hi my name is vincent5"}
{"text": "hi my name is vincent6"}
{"text": "hi my name is vincent7"}
{"text": "hi my name is vincent8"}
{"text": "hi my name is vincent9"}
{"text": "hi my name is vincent10"}
{"text": "hi my name is vincent11"}
{"text": "hi my name is vincent12"}

Given this and a prodigy.json that contains:

{
    "feed_overlap": false,
    "batch_size": 5
}

I was able to run:

python -m prodigy ner.manual namedemo blank:en examples.jsonl --label name

And see that indeed two different sessions have different data.

If we can confirm that you're able to run this demo locally, then we can rule out any issues with environment variables.

It was actually due to the exhausted batches. Thanks.

1 Like