Nikhil Thorat issues

Results 53 issues of


                                            Nikhil Thorat

Allow datasets with spaces in the name.

Markdown code block aware chunking.

We should probably have markdown-block aware chunking. When we see a code block of markdown, we should probably make it one big chunk.

Add ability to jump to a random integer offset in single-item view.

We should make the counter an input box so it can be jumped

Allow user-provided embeddings when ingesting from a source.

Default to combine_cols=True

We should fix the VALUE and SPAN key problem so we can turn combine_cold to true by default. This has come up several times now, let's prioritize.

Datasets w/ map bug: DataType has no attribute 'key_type'

``` File "/usr/local/lib/python3.10/site-packages/lilac/router_dataset.py", line 218, in select_rows_schema dataset = get_dataset(namespace, dataset_name) File "/usr/local/lib/python3.10/site-packages/lilac/db_manager.py", line 43, in get_dataset _CACHED_DATASETS[cache_key] = _DEFAULT_DATASET_CLS( File "/usr/local/lib/python3.10/site-packages/lilac/data/dataset_duckdb.py", line 337, in __init__ manifest = self.manifest() File...

Improve concept labeler

- Sometimes we have the same data point duplicated in 3 slots - Fix bouncing when labeling - Make labeling faster with keyboard shortcuts.

Alow cancelling a task

Taskas are not cancellable, we should have an X button

Allow deletion of columns created from map from the UI.

The UI currently doesnt let us delete map columns.

Break up the TaskManager.

The TaskManager should still be a global singleton (for the sake of a single place to check progress), but we should break the internals down into threads and process subclasses...