Update readme fixes

[ca16]

snippet from the current README:

...
  features:
  - name: suite_config
    struct:
    - name: name
      dtype: string
    - name: version
      dtype: string
    - name: splits
      list:
      - name: name
        dtype: string
      - name: tasks
        list:
        - name: name
          dtype: string
        - name: path
          dtype: string
        - name: primary_metric
          dtype: string
        - name: tags
          sequence: string
      - name: macro_average_weight_adjustments
        list:
        - name: tag
          dtype: string
        - name: task
          dtype: string
        - name: weight
          dtype: float64
  - name: split
    dtype: string
...

if you do dry-run to update the readme without this change

  features:
  - name: suite_config
    struct:
    - dtype: string
      name: name
    - dtype: string
      name: version
    - list:
      - dtype: string
        name: name
      - list:
        - dtype: string
          name: name
        - dtype: string
          name: path
        - dtype: string
          name: primary_metric
        - list: string
          name: tags
        name: tasks
      - list:
        - dtype: string
          name: tag
        - dtype: string
          name: task
        - dtype: float64
          name: weight
        name: macro_average_weight_adjustments
      name: splits
  - dtype: string
    name: split

if you do dry-run to update the readme with this change

  features:
  - name: suite_config
    struct:
    - name: name
      dtype: string
    - name: version
      dtype: string
    - name: splits
      list:
      - name: name
        dtype: string
      - name: tasks
        list:
        - name: name
          dtype: string
        - name: path
          dtype: string
        - name: primary_metric
          dtype: string
        - name: tags
          list: string
      - name: macro_average_weight_adjustments
        list:
        - name: tag
          dtype: string
        - name: task
          dtype: string
        - name: weight
          dtype: float64
  - name: split
    dtype: string

Now that I know what's happening, it's less confusing, but might be nice to keep the old sorting for minimal diffs?

(note I do see a list vs sequence thing too but I think you said that's just packages being out of date)

[ca16]

For reference, this is where it used to be:

agent-eval/src/agenteval/leaderboard/upload.py

Line 110 in c74cf0e

updated_yaml = yaml.dump(parsed_yaml, sort_keys=False).strip()

[ca16]

Note: I rebuilt my docker env (using --no-cache) and still see stuff like

        - name: tags
          list: string

vs

        - name: tags
          sequence: string

Maybe the first one is the latest thing?

[ca16]

I think I'm on the latest version of HF's datasets:

pip show datasets
Name: datasets
Version: 4.0.0

[ca16]

Got a 404 without this.

[rodneykinney]

Hmmm. I think for the maximum consistency we should have sorted keys, and run update_readme with the latest version of datasets. It might change the current README, but it should be functionally the same

[ca16]

Sounds good I'll take out the sorting=False! I think there's another small bug with a new line but I think I'm close to getting it...

[ca16]

Without this I got

No YAML front matter found in README.md

the next time I ran something that tried to do the readme check, after making my initial change to add a new config.

[ca16]

The commit after I uploaded stuff without sorted_keys=False

https://huggingface.co/datasets/allenai/asta-bench-internal-results/commit/ada319c806dcce3718042abf34a07bf0891b35b8

[ca16]

Published new version.