genienlp

genienlp copied to clipboard

GenieNLP has some optional dependencies like `matplotlib` and `seaborn` that are only used for certain features. These should be handled via [optional dependencies](https://setuptools.pypa.io/en/latest/userguide/dependency_management.html#optional-dependencies) so that we can run e.g. `pip...

s-jse

In https://github.com/stanford-oval/genienlp/blob/master/genienlp/ned/ned_utils.py#L38, if the download of nltk stopwords fails, it does so silently. The quiet flag suppresses errors. Error seen: ``` >>> nltk.download('stopwords') [nltk_data] Error loading stopwords: False ``` Solution...

nsmarks

Calibration fails if almond_type_mapping_path is specified

1

``` Traceback (most recent call last): File "/usr/local/bin/genienlp", line 33, in sys.exit(load_entry_point('genienlp', 'console_scripts', 'genienlp')()) File "/opt/genienlp/genienlp/__main__.py", line 108, in main subcommands[argv.subcommand][2](argv) File "/opt/genienlp/genienlp/predict.py", line 570, in main run(args, devices[0]) File...

gcampax

Pytorch 1.7 added support for mixed precision training and evaluation ([https://pytorch.org/docs/stable/amp.html?highlight=amp#module-torch.cuda.amp](https://pytorch.org/docs/stable/amp.html?highlight=amp#module-torch.cuda.amp)). This will be especially useful for speeding up inference for both semantic parsing and paraphrase generation. Also, training with...

s-jse

Change the output format of predict.py

2

We support multiple predictions per input, and soon we will add a confidence score for each prediction. Keeping the `tsv` format will make things difficult down the line. I propose...

s-jse

Huggingface recently started support for [DeepSpeed](https://github.com/microsoft/deepspeed) and [FairScale](https://github.com/facebookresearch/fairscale/) which supposedly give major speedups in both multi-gpu and single-gpu settings. The feature is still experimental but worthwhile to look into and...

Mehrad0711

Bootleg takes way too much space and memory

3

I had to bump up the ephemeral storage of BART models with Bootleg to 55G. It takes ~20 minutes for a model to start for inference, as it downloads and...

gcampax

Crash in Bootleg during request

6

The following command: "find me a movie with chris pratt" seems to reliably trigger a crash in Booleg with the currently deployed model on staging. ``` [E 210714 17:56:44 web:1789]...

gcampax