I want to use rebel to extract all relationships tagged with 'medical condition treated'. Following the huggingface tutorial (https://huggingface.co/Babelscape/rebel-large), I have partly achieved this and can then filter the outputs afterwards. However, I notice that the number of outputs appears to be limited by num_beams/num_return_sequences. If I increase these above the expected number of relations (say, 1000 for a short text) I get an explosion in RAM usage (>50GB). Is there a better way to do this?
I want to use rebel to extract all relationships tagged with 'medical condition treated'. Following the huggingface tutorial (https://huggingface.co/Babelscape/rebel-large), I have partly achieved this and can then filter the outputs afterwards. However, I notice that the number of outputs appears to be limited by
num_beams/num_return_sequences. If I increase these above the expected number of relations (say, 1000 for a short text) I get an explosion in RAM usage (>50GB). Is there a better way to do this?