[bug] Datasets with CJK names & Evaluation comparisons with those datasets not displaying in UI

[TeoZosa]

This was a weird one I ran into. At first I thought it was from datasets being too large¹, but the problem stuck around even at smaller dataset sizes that work fine with English names. Hopefully the patch for this one isn't too much work.

Steps to reproduce

>>> import weave
>>> weave.init(project_name="my-project")
>>> for name in ["Deep Learning", "深度學習", "深層学習", "딥 러닝"]:
...     dataset = weave.Dataset(name=name, rows=[{"key": "value"}])
...     weave.publish(dataset)
📦 Published to https://wandb.ai/...
ObjectRef(entity='...', project='...', name='Deep-Learning', digest='...', extra=())
📦 Published to https://wandb.ai/...
ObjectRef(entity='...', project='...', name='深度學習', digest='...', extra=())
📦 Published to https://wandb.ai/...
ObjectRef(entity='...', project='...', name='深層学習', digest='...', extra=())
📦 Published to https://wandb.ai/...
ObjectRef(entity='...', project='...', name='딥-러닝', digest='...', extra=())

Behavior

Dataset

Datasets are confirmed loadable via the API

>>> import weave
>>> import weave.trace.weave_client
>>> weave.init(project_name="my-project")
>>> def fetch_dataset(dataset_ref: str) -> weave.Dataset:
...     dataset_ref_sanitized = weave.trace.weave_client.sanitize_object_name(dataset_ref)
...     dataset = weave.ref(dataset_ref_sanitized).get()
...     return dataset
>>> for name in ["Deep Learning", "深度學習", "深層学習", "딥 러닝"]:
...    print(f"{name}: {len(fetch_dataset(name).rows)}")
Deep Learning: 1
深度學習: 1
深層学習: 1
딥 러닝: 1  

But not viewable in the UI

Click here for screenshots

Evaluation

The Evaluation comparison view tries to load and eventually errors-out

Click here for screenshots

Footnotes

1. [bug] "Large" Datasets can't be queried  #2566 ↩

[jamie-rasmussen]

Thank you very much for this and your other recent submissions, we are investigating.

[TeoZosa]

Sounds good; thanks for jumping on this so quickly, @jamie-rasmussen!

[jamie-rasmussen]

Tracking internally as https://wandb.atlassian.net/browse/WB-21343

[tssweeney]

Hello - we have identified a fix and will deploy this week: 483ae68. This is actually a read-only issue and previously logged data should be correct.