W&B EvalTable previews
Use the bundled helper to preview existing wandb.Table data as
wandb.EvalTable data. A preview creates new runs and keys; it never overwrites
the source runs or tables.
Read references/EVAL_TABLES.md before converting. It is the canonical source
for EvalTable semantics, source selection, limits, column roles, verification,
and removal.
Environment
Run the helper from an environment that provides a W&B SDK with
wandb.EvalTable and Weave. Use uv to supply missing dependencies:
uv run --with 'wandb[workspaces]>=0.28.1' --with weave \
python skills/wandb-eval-tables/scripts/table_artifact_to_eval_table.py --help
Credentials and default scope may come from WANDB_API_KEY, WANDB_ENTITY,
and WANDB_PROJECT; otherwise pass explicit entity/project arguments.
Downloads use the operating system's temporary directory, not a fixed current
working directory.
Required workflow
- Choose the narrowest source that matches the request: exact artifact, one
run, one sweep, or a project/workspace. Do not broaden an explicitly named
source.
- Run
scan. This is read-only preflight and is not proof of conversion.
- Review
eligible_summary_by_table_key, table shapes, truncations, warnings,
and existing-preview metadata. Select one table key unless the user asks for
a broader batch.
- Classify columns only when confident. Input tuples must uniquely identify
rows; score columns must be numeric or boolean; free-text labels and
rationales are outputs. Leave ambiguous tables untyped.
- Run
preview for exactly one --table-key per invocation. Keep the target
in the source project unless the user requests another writable project.
- If runs were created, run
verify-preview on the final created run and the
exact logged key. Report success only when it returns verified: true.
- Report created and skipped sources plus every truncation or failed check.
T=skills/wandb-eval-tables/scripts/table_artifact_to_eval_table.py
uv run python "$T" scan \
--source-workspace ENTITY/PROJECT --max-runs 4
uv run python "$T" preview \
--source-workspace ENTITY/PROJECT \
--target-project ENTITY/PROJECT \
--max-runs 4 \
--table-key predictions
uv run python "$T" verify-preview \
--run ENTITY/PROJECT/PREVIEW_RUN_ID \
--table-key predictions_preview
Safety
scan and preview --dry-run are read-only. preview creates W&B runs,
EvalTables, and Weave evaluations; run it only when the user requested a
conversion.
- The helper caps tables at 10,000 rows and 100 columns, warns above 1,000
rows, and refuses more than 50 eligible table artifacts for one key. Surface
every cap; never imply a truncated preview is complete.
- Do not bypass helper failures with ad-hoc logging. Narrow or correct helper
arguments, then retry only when the change is safe.
- Removal deletes both Weave evaluations and preview runs. Run
delete-preview --dry-run first, verify every target is helper-created, and
obtain confirmation before the irreversible invocation. Never delete source
runs.
1---
2name: wandb-eval-tables
3description: "Convert W&B Table artifacts into non-destructive EvalTable previews with scan-first planning, typed input/output/score columns, bounded batches, verification, and safe removal. Use when a coding agent needs to create, inspect, compare, verify, or remove W&B EvalTable previews."
4metadata:
5 source: https://skillmd.com/skills/wandb/wandb-eval-tables
6---
7
8<!--
9SPDX-FileCopyrightText: 2026 CoreWeave, Inc.
10SPDX-License-Identifier: Apache-2.0
11SPDX-PackageName: skills
12-->
13
14# W&B EvalTable previews
15
16Use the bundled helper to preview existing `wandb.Table` data as
17`wandb.EvalTable` data. A preview creates new runs and keys; it never overwrites
18the source runs or tables.
19
20Read `references/EVAL_TABLES.md` before converting. It is the canonical source
21for EvalTable semantics, source selection, limits, column roles, verification,
22and removal.
23
24## Environment
25
26Run the helper from an environment that provides a W&B SDK with
27`wandb.EvalTable` and Weave. Use `uv` to supply missing dependencies:
28
29```bash
30uv run --with 'wandb[workspaces]>=0.28.1' --with weave \
31 python skills/wandb-eval-tables/scripts/table_artifact_to_eval_table.py --help
32```
33
34Credentials and default scope may come from `WANDB_API_KEY`, `WANDB_ENTITY`,
35and `WANDB_PROJECT`; otherwise pass explicit entity/project arguments.
36Downloads use the operating system's temporary directory, not a fixed current
37working directory.
38
39## Required workflow
40
411. Choose the narrowest source that matches the request: exact artifact, one
42 run, one sweep, or a project/workspace. Do not broaden an explicitly named
43 source.
442. Run `scan`. This is read-only preflight and is not proof of conversion.
453. Review `eligible_summary_by_table_key`, table shapes, truncations, warnings,
46 and existing-preview metadata. Select one table key unless the user asks for
47 a broader batch.
484. Classify columns only when confident. Input tuples must uniquely identify
49 rows; score columns must be numeric or boolean; free-text labels and
50 rationales are outputs. Leave ambiguous tables untyped.
515. Run `preview` for exactly one `--table-key` per invocation. Keep the target
52 in the source project unless the user requests another writable project.
536. If runs were created, run `verify-preview` on the final created run and the
54 exact logged key. Report success only when it returns `verified: true`.
557. Report created and skipped sources plus every truncation or failed check.
56
57```bash
58T=skills/wandb-eval-tables/scripts/table_artifact_to_eval_table.py
59
60uv run python "$T" scan \
61 --source-workspace ENTITY/PROJECT --max-runs 4
62
63uv run python "$T" preview \
64 --source-workspace ENTITY/PROJECT \
65 --target-project ENTITY/PROJECT \
66 --max-runs 4 \
67 --table-key predictions
68
69uv run python "$T" verify-preview \
70 --run ENTITY/PROJECT/PREVIEW_RUN_ID \
71 --table-key predictions_preview
72```
73
74## Safety
75
76- `scan` and `preview --dry-run` are read-only. `preview` creates W&B runs,
77 EvalTables, and Weave evaluations; run it only when the user requested a
78 conversion.
79- The helper caps tables at 10,000 rows and 100 columns, warns above 1,000
80 rows, and refuses more than 50 eligible table artifacts for one key. Surface
81 every cap; never imply a truncated preview is complete.
82- Do not bypass helper failures with ad-hoc logging. Narrow or correct helper
83 arguments, then retry only when the change is safe.
84- Removal deletes both Weave evaluations and preview runs. Run
85 `delete-preview --dry-run` first, verify every target is helper-created, and
86 obtain confirmation before the irreversible invocation. Never delete source
87 runs.