Justin Zhao comments

Results 42 comments of


                                            Justin Zhao

Black pre-commit and black command line disagree

Wow! Fast reply -- thanks :) Glad to hear that it's not reproducing on your end, which means it's likely just an environment issue. I'm still seeing the same disagreement...

KeyError: 'class_weights'

Hi @jiangweiatgithub, could you share the `mt_en_sch.yaml` file so that I can try to reproduce the error?

KeyError: 'class_weights'

@ShreyaR this doesn't seem like an issue that would be dataset-specific to me, though @jiangweiatgithub if we could get a small sample (i.e. head 10 lines), I'd be happy to...

combined.loss is not equal to loss of output feature?

I'll create a ticket to add this detail to our public modeling docs.

Output sequence tagger has wrong probabilities output

Hi @carlogrisetti and @w4nderlust , @geoffreyangus may have run into a related (perhaps identical) issue on Ludwig 0.5.dev, and Geoffrey has landed a fix for computing sequence probabilities on master...

AttributeError: module 'tensorflow_core.keras.layers' has no attribute 'Normalization'

Hi Christine, which notebook are you referencing and which version of Ludwig are you using? Ludwig has migrated to PyTorch, so there shouldn't be any Tensorflow errors. Perhaps some of...

Ludwig serve reserves gpu memory but doesn't utilize gpu (tf-inference pip install Ludwig 0.4)

Hi @smiraldr, I was able to get a working setup with a similar BERT-based text classification model on Ray using master (torch), after fixing a few issues, namely https://github.com/ludwig-ai/ludwig/pull/1691. We've...

Ludwig serve reserves gpu memory but doesn't utilize gpu (tf-inference pip install Ludwig 0.4)

I trained and deployed a couple of different huggingface-bert text classification models on GPUs with `ray` and `ludwig serve` and measured inference times. I've observed that the very first call...

Ludwig serve reserves gpu memory but doesn't utilize gpu (tf-inference pip install Ludwig 0.4)

@smiraldr and I met to discuss this issue in a bit more detail. We didn't reach any hard conclusions, but here were our observations. 1. Our dataset sequence lengths were...

Ludwig serve reserves gpu memory but doesn't utilize gpu (tf-inference pip install Ludwig 0.4)

Hi @smiraldr, this should be fully addressed with #2131, which was not in the 0.5.2 release but will be included in the 0.5.3 release, which should land shortly. I'll ping...