Datasets:
Release [run release-20261111T192938Z]
#25
by Battam - opened
This view is limited to 50 files because it contains too many changes. See the raw diff here.
- CHANGELOG.md +18 -0
- README.md +67 -62
- README.zh.md +67 -62
- campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-algorithmic/LICENSE +3 -0
- campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-algorithmic/README.md +26 -0
- campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-logic/LICENSE +4 -0
- campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-logic/README.md +27 -0
- campaign_files/2026-10-hw-eda-tests-2/open-cmsis-pack-devtools-tools-svdconv-tests/README.md +0 -8
- campaign_files/2026-10-hw-eda-tests-2/open-logic/README.md +0 -10
- campaign_files/2026-10-m-issue-trackers-2/openafs-openafs-info/README.md +0 -4
- campaign_files/2026-10-m-issue-trackers-2/opendaylight-gerrit-changes/README.md +0 -4
- campaign_files/2026-10-md-media-and-documents/opendetex/README.md +0 -4
- campaign_files/2026-10-md-media-and-documents/openimageio-testsuite/README.md +0 -10
- campaign_files/2026-10-ml-serving-configs-2/ollama-chat-template-expected-prompts/README.md +0 -42
- campaign_files/2026-10-ml-serving-configs-2/open-mmlab-mmdeploy/README.md +0 -4
- campaign_files/2026-10-ml-serving-configs-2/open-mmlab-mmpretrain/README.md +0 -34
- campaign_files/2026-10-ml-serving-configs-2/open-mmlab-mmrazor/README.md +0 -46
- campaign_files/2026-10-ml-serving-configs-2/open-model-zoo-accuracy/README.md +0 -42
- campaign_files/2026-10-op-rule-calculations-2/openfisca-france-tests/README.md +0 -4
- campaign_files/2026-10-op-rule-calculations-2/openintro-loans-full-schema/README.md +0 -4
- campaign_files/2026-10-op-rule-calculations-2/openintro-mariokart/README.md +0 -4
- campaign_files/2026-10-s-security-detection-2/opa-gatekeeper-library-with-gator-suite-tests/README.md +0 -4
- campaign_files/2026-10-s-security-detection-2/opa-gatekeeper-policy-library/README.md +0 -6
- campaign_files/2026-10-sc-science-reference-data-2/openfermion-testing-data/README.md +0 -4
- campaign_files/2026-10-sc-science-reference-data/openbabel-test-files/README.md +0 -8
- campaign_files/2026-10-sw-tests-and-exercises-2/opencypher-tck/README.md +0 -4
- cards/0xdea-semgrep-rules.md +2 -2
- cards/1brc-test-samples.md +2 -2
- cards/2dpacklib.md +2 -2
- cards/abacus-integrate-tests.md +2 -2
- cards/abhishekumarr00-supply-chain-mastery.md +2 -2
- cards/academy-bus-nj-gtfs.md +2 -2
- cards/accordproject-cicero-template-library.md +2 -2
- cards/activity-schema-aggregate-in-between-1.md +2 -2
- cards/activity-schema-first-in-between-2.md +2 -2
- cards/acwf-verification.md +2 -2
- cards/adflib-adf-deleted.md +2 -2
- cards/adflib-adf-malformed.md +2 -2
- cards/adventure-works-oltp-sample.md +2 -2
- cards/agat-tests.md +2 -2
- cards/agl-gerrit-changes.md +2 -2
- cards/aiops2018-kpi.md +2 -2
- cards/akretion-factur-x-libreoffice-extension.md +2 -2
- cards/allenporter-ical-recurring-single-change.md +2 -2
- cards/altlinux-devel.md +2 -2
- cards/alumik-nkthesis-testfiles.md +2 -2
- cards/aml-dbt-rule-unit-tests.md +2 -2
- cards/ancuber-cubst-tests.md +2 -2
- cards/annabel-lee-wikisource-reading.md +2 -2
- cards/ansible-galaxy-collections.md +2 -2
CHANGELOG.md
CHANGED
|
@@ -3249,3 +3249,21 @@ Changes to the content of this index, oldest first. There is one entry for each
|
|
| 3249 |
- hf-training-records: refused
|
| 3250 |
- npm: refused
|
| 3251 |
- stackexchange-api-posts: refused
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3249 |
- hf-training-records: refused
|
| 3250 |
- npm: refused
|
| 3251 |
- stackexchange-api-posts: refused
|
| 3252 |
+
|
| 3253 |
+
## 2026-11-11: sources held back
|
| 3254 |
+
|
| 3255 |
+
- Each source below keeps its newest version that passed every check; its newer batch waits until it passes them:
|
| 3256 |
+
- borealis-dataverse-workbooks-cc-by: changed pointer group file data/borealis-dataverse-workbooks-cc-by/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 2 key(s) of the old version are in no data file of source borealis-dataverse-workbooks-cc-by after the commit (row 4 from batch 2026-10-a-spreadsheets-with-answers, row 5 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3257 |
+
- borealis-dataverse-workbooks-cc0: changed pointer group file data/borealis-dataverse-workbooks-cc0/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source borealis-dataverse-workbooks-cc0 after the commit (row 7 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3258 |
+
- cfpb-complaints: refused
|
| 3259 |
+
- data-gov-uk-workbooks: changed pointer group file data/data-gov-uk-workbooks/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 4 key(s) of the old version are in no data file of source data-gov-uk-workbooks after the commit (row 5 from batch 2026-10-a-spreadsheets-with-answers, row 6 from batch 2026-10-a-spreadsheets-with-answers, row 12 from batch 2026-10-a-spreadsheets-with-answers, 1 more); the removal record holds none of them, and none fails a publish check now
|
| 3260 |
+
- data-gov-uk-workbooks-noassertion: changed pointer group file data/data-gov-uk-workbooks-noassertion/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source data-gov-uk-workbooks-noassertion after the commit (row 6 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3261 |
+
- dataversenl-workbooks-cc-by: changed pointer group file data/dataversenl-workbooks-cc-by/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source dataversenl-workbooks-cc-by after the commit (row 17 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3262 |
+
- dataversenl-workbooks-cc0: changed pointer group file data/dataversenl-workbooks-cc0/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 3 key(s) of the old version are in no data file of source dataversenl-workbooks-cc0 after the commit (row 7 from batch 2026-10-a-spreadsheets-with-answers, row 13 from batch 2026-10-a-spreadsheets-with-answers, row 16 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3263 |
+
- dvn-fvhlgg-repro: changed pointer group file data/dvn-fvhlgg-repro/2026-10-c-repros/part-00000.parquet: 3 key(s) of the old version are in no data file of source dvn-fvhlgg-repro after the commit (row 5 from batch 2026-10-c-repros, row 7 from batch 2026-10-c-repros, row 16 from batch 2026-10-c-repros); the removal record holds none of them, and none fails a publish check now
|
| 3264 |
+
- eu-institution-workbooks-cc-by: changed pointer group file data/eu-institution-workbooks-cc-by/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 3 key(s) of the old version are in no data file of source eu-institution-workbooks-cc-by after the commit (row 2 from batch 2026-10-a-spreadsheets-with-answers, row 4 from batch 2026-10-a-spreadsheets-with-answers, row 5 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3265 |
+
- hdx-workbooks-cc0: changed pointer group file data/hdx-workbooks-cc0/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 2 key(s) of the old version are in no data file of source hdx-workbooks-cc0 after the commit (row 4 from batch 2026-10-a-spreadsheets-with-answers, row 5 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3266 |
+
- hdx-workbooks-odbl: changed pointer group file data/hdx-workbooks-odbl/2026-10-a-spreadsheets-with-answers/part-00000.parquet: 1 key(s) of the old version are in no data file of source hdx-workbooks-odbl after the commit (row 4 from batch 2026-10-a-spreadsheets-with-answers); the removal record holds none of them, and none fails a publish check now
|
| 3267 |
+
- hf-training-records: refused
|
| 3268 |
+
- npm: refused
|
| 3269 |
+
- stackexchange-api-posts: refused
|
README.md
CHANGED
|
@@ -24,25 +24,25 @@ configs:
|
|
| 24 |
<!-- artifact-pointer-layout: v2 -->
|
| 25 |
<!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
|
| 26 |
|
| 27 |
-
<img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 28 |
|
| 29 |
This dataset is the public index of the **Training Data Foundation**: find public files to build agent tasks and environments from, and fetch them with one command.
|
| 30 |
|
| 31 |
English | [简体中文](README.zh.md) · [Data page](https://huggingface.co/spaces/Battam/training-data-foundation) · [Usage video](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
|
| 32 |
|
| 33 |
<picture class="not-prose">
|
| 34 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 35 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 36 |
</picture>
|
| 37 |
<picture class="not-prose">
|
| 38 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 39 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 40 |
</picture>
|
| 41 |
|
| 42 |
## Get data in one command
|
| 43 |
|
| 44 |
```sh
|
| 45 |
-
uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 46 |
```
|
| 47 |
|
| 48 |
This fetches 3 units of every spreadsheet material set into `./data`, one folder per unit, and checks each file against the size and hashes its row records.
|
|
@@ -52,7 +52,7 @@ the same command takes the same units of this release.
|
|
| 52 |
<details>
|
| 53 |
<summary>What the command takes, and in which order</summary>
|
| 54 |
|
| 55 |
-
The command and the queries below are pinned to the release tag `release-2026-11-11.
|
| 56 |
[uv](https://docs.astral.sh/uv/) reads the script's dependencies, at the versions published before the day of this release.
|
| 57 |
|
| 58 |
The command returns units: 3 units from each material set whose form is `spreadsheets`, each in a folder of its own named after its `unit_id`.
|
|
@@ -66,8 +66,9 @@ The script reads the data files in that order (with `--limit`, the first file of
|
|
| 66 |
<summary>What the default profile leaves out</summary>
|
| 67 |
|
| 68 |
The default profile is the rules `tools/fetch.py` applies unless an option relaxes them.
|
| 69 |
-
It leaves out duplicate copies, rows that matched a benchmark or were not compared with the benchmark list yet,
|
| 70 |
Each rule has its own option, and every run says how many units each rule removed.
|
|
|
|
| 71 |
Every rule and its option, and what to keep out of an agent's environment: [`docs/for-agents.md`](docs/for-agents.md).
|
| 72 |
|
| 73 |
</details>
|
|
@@ -79,7 +80,7 @@ Choose material sets from the catalog with [DuckDB](https://duckdb.org/):
|
|
| 79 |
|
| 80 |
```sql
|
| 81 |
SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
|
| 82 |
-
FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 83 |
WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
|
| 84 |
ORDER BY collection, units DESC;
|
| 85 |
```
|
|
@@ -90,7 +91,7 @@ with `kind = 'corpus'` it returns corpora instead, counted in rows (`files`).
|
|
| 90 |
Choose files in one corpus:
|
| 91 |
|
| 92 |
```sql
|
| 93 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 94 |
SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
|
| 95 |
FROM read_parquet(getvariable('files'))
|
| 96 |
WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
|
|
@@ -112,15 +113,15 @@ A data file may hold the rows of several sources, so the query keeps this one's
|
|
| 112 |
### Task material: units by collection
|
| 113 |
|
| 114 |
<picture class="not-prose">
|
| 115 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 116 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 117 |
</picture>
|
| 118 |
<picture class="not-prose">
|
| 119 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 120 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 121 |
</picture>
|
| 122 |
|
| 123 |
-
1,
|
| 124 |
|
| 125 |
<details>
|
| 126 |
<summary>Task material by collection, as a table</summary>
|
|
@@ -131,12 +132,12 @@ A data file may hold the rows of several sources, so the query keeps this one's
|
|
| 131 |
| Disk images | 52 | 179 | 324 | 40 | 272 MB | 272 MB |
|
| 132 |
| Reproduction packages and runs | 88 | 106 | 870 | 88 | 120 MB | 120 MB |
|
| 133 |
| SEC filing reviews | 0 | 0 | 0 | 0 | 0 B | 0 B |
|
| 134 |
-
| Solver instances |
|
| 135 |
| Browser tests | 123 | 439 | 852 | 88 | 31.9 MB | 31.9 MB |
|
| 136 |
-
| Hardware design (EDA) tests | 81 | 258 | 651 | 75 | 80.8 MB | 80.
|
| 137 |
| Issue trackers | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
|
| 138 |
| Media and documents | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
|
| 139 |
-
| Model serving configurations | 103 | 398 | 1,168 | 83 | 250 MB |
|
| 140 |
| Rule calculations | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
|
| 141 |
| Security detection | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
|
| 142 |
| Science reference data | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
|
|
@@ -150,12 +151,12 @@ Upstreams are the upstream works the material sets of a collection come from.
|
|
| 150 |
### Corpora: rows by main domain
|
| 151 |
|
| 152 |
<picture class="not-prose">
|
| 153 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 154 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 155 |
</picture>
|
| 156 |
<picture class="not-prose">
|
| 157 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 158 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 159 |
</picture>
|
| 160 |
|
| 161 |
139 corpora with 1,216,362,128 rows.
|
|
@@ -188,12 +189,12 @@ Every material set and every corpus, with its counts, size, license tier and log
|
|
| 188 |
## How it is made
|
| 189 |
|
| 190 |
<picture class="not-prose">
|
| 191 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 192 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 193 |
</picture>
|
| 194 |
<picture class="not-prose">
|
| 195 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 196 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 197 |
</picture>
|
| 198 |
|
| 199 |
The terms, each step and how every number is made are folded below.
|
|
@@ -215,7 +216,7 @@ The terms, each step and how every number is made are folded below.
|
|
| 215 |
- **Collection:** the material sets of one task family, such as spreadsheets or disk images;
|
| 216 |
a corpus that holds a collection's kind of material is listed under it and stays a corpus.
|
| 217 |
- **Form:** the shape of a source's material, one id such as `spreadsheets` or `test-suites` (the catalog column `form`, the option `--form` of `tools/fetch.py`).
|
| 218 |
-
- **Default profile:** the rules `tools/fetch.py` applies unless an option relaxes them: it leaves out duplicate copies, rows that matched a benchmark or were not compared with the benchmark list yet,
|
| 219 |
|
| 220 |
Every row carries a `unit_id`: in a material set, the key of its unit;
|
| 221 |
in a corpus, the row's own key (its `uri`, then `@<range_offset>` for a byte range, then `!/<member_path>` for an archive member), and `tools/fetch.py` puts the file of a corpus row in a folder named after it.
|
|
@@ -247,9 +248,9 @@ only their sizes add up to the size of the whole index.
|
|
| 247 |
|
| 248 |
| Level | Recorded size | Deduplicated size | Rows read | Rows compared | Rows that match | Rows read, not compared | Rows without a recorded size | Their estimated size |
|
| 249 |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 250 |
-
| Whole index | 4.45 PB | 4.11 PB | 34,
|
| 251 |
| Corpora | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610 (99.6%) | 871: address unreachable 449, file gone 80, login needed 322, no size from the server 19, other error 1 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
|
| 252 |
-
| Task material | 2.88 GB | 2.
|
| 253 |
|
| 254 |
Rows read are the sample of rows the check reads in `meta/sizes.parquet`;
|
| 255 |
a row compared matched its recorded size or did not, and a row read but not compared never reached its file.
|
|
@@ -258,56 +259,60 @@ above one source its range is the sum of its sources' ranges, wider than a joint
|
|
| 258 |
|
| 259 |
The recorded size is the sum of `size_bytes` over the rows that have one (catalog column `bytes`). The check reads a sample of rows of every data file (`meta/sizes.parquet`): 20 of its rows with a recorded size, and of its rows without one its share of its source's sample, sized for their mean size within 25% at 95% confidence by the spread of the sizes read so far (at most 400 rows per source per round, 20 for a source with none read yet); its share is the rows that match their recorded size out of the rows it compared (status `match`, `size_mismatch` or `hash_mismatch`), and the rows it could not compare, which never reached their file, are counted beside it by status (`unreachable`, `gone`, `needs_auth`, `no_length`, `error`; catalog column `size_check`). The estimate for a source's rows without a recorded size (catalog columns `unsized_rows` and `unsized_estimate`) is their number times the mean measured size of such rows read, with a 95% BCa bootstrap range (`scipy.stats.bootstrap`, its default resample count); a source with fewer than 10 such rows read has none. Above one source, the estimate is the sum of its sources' estimates and its range the sum of their ranges, which is wider than a joint interval. The figures of every level are in `meta/size_figures.json`.
|
| 260 |
|
| 261 |
-
Distinct content is the sum of `size_bytes` over the rows minus the sizes of the rows that repeat another row's content (the same sha256, else sha1, else md5 and size, within the same hash scope; one copy of each content is kept, in the first source by registry order), counted after each release over the data files it published, per source in the catalog column `distinct_bytes`; archive members whose archive the same source also lists whole have other hashes, so they are not subtracted. `tools/fetch.py` removes duplicates over your own selection when it reads it. The deduplicated sizes here were counted over the data files of release `release-2026-11-11.
|
| 262 |
|
| 263 |
</details>
|
| 264 |
|
| 265 |
<details>
|
| 266 |
<summary>How the training-stage tables are counted</summary>
|
| 267 |
|
| 268 |
-
The stage table `catalog/stages.yaml` of the code repository gives each training stage a condition on fields the index already has: the kind, form and domain of a row's source, the file types of a unit
|
| 269 |
-
The release records those fields for every unit of a material set and every row of a corpus, with how many of the rules every stage shares it passes, in this order: the benchmark rule (a row that matches a public benchmark task is left out), the rule on rows not compared with the benchmark list yet
|
| 270 |
-
each stage counts what meets its condition before those rules and after each of them, and its last
|
| 271 |
-
|
| 272 |
-
|
|
|
|
| 273 |
|
| 274 |
-
|
| 275 |
|
| 276 |
#### Post-training (in use)
|
| 277 |
|
| 278 |
-
Material set units for building tasks: every file of the unit
|
| 279 |
|
| 280 |
| Step | Material sets | Units | Upstream sources |
|
| 281 |
|---|---:|---:|---:|
|
| 282 |
-
| In the stage's condition, before the rules |
|
| 283 |
-
| After the benchmark rule |
|
| 284 |
-
| After the rule on rows not compared with the benchmark list |
|
| 285 |
-
| After the
|
| 286 |
-
|
|
|
|
|
| 287 |
|
| 288 |
#### Mid-training (provisional: conditions and counts only)
|
| 289 |
|
| 290 |
-
Corpus rows of code, documentation and the terminal side
|
| 291 |
|
| 292 |
| Step | Rows | Size |
|
| 293 |
|---|---:|---:|
|
| 294 |
-
| In the stage's condition, before the rules |
|
| 295 |
-
| After the benchmark rule |
|
| 296 |
-
| After the rule on rows not compared with the benchmark list |
|
| 297 |
-
| After the
|
| 298 |
-
|
|
|
|
|
| 299 |
|
| 300 |
#### Pre-training (provisional: conditions and counts only)
|
| 301 |
|
| 302 |
-
Corpus rows of plain text
|
| 303 |
|
| 304 |
| Step | Rows | Size |
|
| 305 |
|---|---:|---:|
|
| 306 |
-
| In the stage's condition, before the rules |
|
| 307 |
-
| After the benchmark rule |
|
| 308 |
-
| After the rule on rows not compared with the benchmark list |
|
| 309 |
-
| After the
|
| 310 |
-
|
|
|
|
|
| 311 |
|
| 312 |
</details>
|
| 313 |
|
|
@@ -404,7 +409,7 @@ Every row field and every column of the catalog: [`docs/fields.md`](docs/fields.
|
|
| 404 |
| `none_known` | 3 | 0 |
|
| 405 |
| `present` | 14 | 0 |
|
| 406 |
| `suspected` | 1 | 0 |
|
| 407 |
-
| Not judged yet | 121 | 1,
|
| 408 |
| Sample files read per judged source | 9 to 30 | none |
|
| 409 |
| Days the sample files were read | 2026-10-04 and 2026-10-05 | none |
|
| 410 |
|
|
@@ -415,10 +420,10 @@ Every row field and every column of the catalog: [`docs/fields.md`](docs/fields.
|
|
| 415 |
|
| 416 |
| Sources | Corpora | Task material |
|
| 417 |
|---|---:|---:|
|
| 418 |
-
| With a recorded size for every row | 92 of 139 | 1,
|
| 419 |
-
| With a summary read from sample files | 0 of 139 | 0 of 1,
|
| 420 |
-
| With personal information read from sample files | 18 of 139 | 0 of 1,
|
| 421 |
-
| Still being collected | 25 of 139 | 0 of 1,
|
| 422 |
|
| 423 |
- Rows whose size the source does not publish are shown as unknown and left out of the size totals, never counted as 0.
|
| 424 |
- Fields not judged yet say so;
|
|
@@ -444,9 +449,9 @@ Every row field and every column of the catalog: [`docs/fields.md`](docs/fields.
|
|
| 444 |
<details>
|
| 445 |
<summary>Releases</summary>
|
| 446 |
|
| 447 |
-
- Latest release: `release-2026-11-11.
|
| 448 |
- Releases are listed in [CHANGELOG.md](CHANGELOG.md);
|
| 449 |
-
each one is a commit with its own release tag (this one is [`release-2026-11-11.
|
| 450 |
so any earlier state stays readable at its tag.
|
| 451 |
- `meta/projection.json` records the release tag and the version of the program and templates that generated this card.
|
| 452 |
- Croissant metadata: [generated by the Hub](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
|
|
@@ -479,7 +484,7 @@ A removal covers the current version of this repository (the main branch) and th
|
|
| 479 |
@misc{training_data_foundation,
|
| 480 |
title = {training-data-foundation: an index of public materials for building agent environments},
|
| 481 |
howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
|
| 482 |
-
note = {Release release-2026-11-11.
|
| 483 |
year = {2026}
|
| 484 |
}
|
| 485 |
```
|
|
|
|
| 24 |
<!-- artifact-pointer-layout: v2 -->
|
| 25 |
<!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
|
| 26 |
|
| 27 |
+
<img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/banner.svg" alt="Public web, in two layers (task material in collections; corpora), one row per file (address, size, SHA-256, license), your selection, fetched and verified">
|
| 28 |
|
| 29 |
This dataset is the public index of the **Training Data Foundation**: find public files to build agent tasks and environments from, and fetch them with one command.
|
| 30 |
|
| 31 |
English | [简体中文](README.zh.md) · [Data page](https://huggingface.co/spaces/Battam/training-data-foundation) · [Usage video](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
|
| 32 |
|
| 33 |
<picture class="not-prose">
|
| 34 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-narrow-light.svg">
|
| 35 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-wide-light.svg" alt="1,808 material sets with 4,195 units in 14 collections; 139 corpora with 1,216,362,128 rows; 4.45 PB pointed to, as the index holds addresses, not copies; 99.6% of the 33,832 rows compared match their recorded size.">
|
| 36 |
</picture>
|
| 37 |
<picture class="not-prose">
|
| 38 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-narrow-dark.svg">
|
| 39 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-wide-dark.svg" alt="1,808 material sets with 4,195 units in 14 collections; 139 corpora with 1,216,362,128 rows; 4.45 PB pointed to, as the index holds addresses, not copies; 99.6% of the 33,832 rows compared match their recorded size.">
|
| 40 |
</picture>
|
| 41 |
|
| 42 |
## Get data in one command
|
| 43 |
|
| 44 |
```sh
|
| 45 |
+
uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/tools/fetch.py --form spreadsheets --sample 3 --out ./data
|
| 46 |
```
|
| 47 |
|
| 48 |
This fetches 3 units of every spreadsheet material set into `./data`, one folder per unit, and checks each file against the size and hashes its row records.
|
|
|
|
| 52 |
<details>
|
| 53 |
<summary>What the command takes, and in which order</summary>
|
| 54 |
|
| 55 |
+
The command and the queries below are pinned to the release tag `release-2026-11-11.18`.
|
| 56 |
[uv](https://docs.astral.sh/uv/) reads the script's dependencies, at the versions published before the day of this release.
|
| 57 |
|
| 58 |
The command returns units: 3 units from each material set whose form is `spreadsheets`, each in a folder of its own named after its `unit_id`.
|
|
|
|
| 66 |
<summary>What the default profile leaves out</summary>
|
| 67 |
|
| 68 |
The default profile is the rules `tools/fetch.py` applies unless an option relaxes them.
|
| 69 |
+
It leaves out duplicate copies, rows that matched a benchmark or were not compared with the benchmark list yet, rows in a language other than English and files above 20 MB.
|
| 70 |
Each rule has its own option, and every run says how many units each rule removed.
|
| 71 |
+
The license and the personal-data mark are labels of every row, not rules of the default profile: every row comes whatever they say, and `--training-only` leaves out files whose license does not let them be shipped, `--no-personal-data` personal data.
|
| 72 |
Every rule and its option, and what to keep out of an agent's environment: [`docs/for-agents.md`](docs/for-agents.md).
|
| 73 |
|
| 74 |
</details>
|
|
|
|
| 80 |
|
| 81 |
```sql
|
| 82 |
SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
|
| 83 |
+
FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet'
|
| 84 |
WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
|
| 85 |
ORDER BY collection, units DESC;
|
| 86 |
```
|
|
|
|
| 91 |
Choose files in one corpus:
|
| 92 |
|
| 93 |
```sql
|
| 94 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'if-archive'));
|
| 95 |
SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
|
| 96 |
FROM read_parquet(getvariable('files'))
|
| 97 |
WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
|
|
|
|
| 113 |
### Task material: units by collection
|
| 114 |
|
| 115 |
<picture class="not-prose">
|
| 116 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-narrow-light.svg">
|
| 117 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-wide-light.svg" alt="Units by collection, longest bar first: Spreadsheets 536; Browser tests 439; Tests and exercises 425; Model serving configurations 398; Media and documents 391; Science reference data 342; Rule calculations 329; Security detection 328; Hardware design (EDA) tests 258; Issue trackers 258; Solver instances 197; Disk images 179; Reproduction packages and runs 106; In no collection 9; SEC filing reviews 0.">
|
| 118 |
</picture>
|
| 119 |
<picture class="not-prose">
|
| 120 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-narrow-dark.svg">
|
| 121 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-wide-dark.svg" alt="Units by collection, longest bar first: Spreadsheets 536; Browser tests 439; Tests and exercises 425; Model serving configurations 398; Media and documents 391; Science reference data 342; Rule calculations 329; Security detection 328; Hardware design (EDA) tests 258; Issue trackers 258; Solver instances 197; Disk images 179; Reproduction packages and runs 106; In no collection 9; SEC filing reviews 0.">
|
| 122 |
</picture>
|
| 123 |
|
| 124 |
+
1,808 material sets with 4,195 units, in 14 collections (1 of them with no material set yet); 1 of the material sets are in no collection.
|
| 125 |
|
| 126 |
<details>
|
| 127 |
<summary>Task material by collection, as a table</summary>
|
|
|
|
| 132 |
| Disk images | 52 | 179 | 324 | 40 | 272 MB | 272 MB |
|
| 133 |
| Reproduction packages and runs | 88 | 106 | 870 | 88 | 120 MB | 120 MB |
|
| 134 |
| SEC filing reviews | 0 | 0 | 0 | 0 | 0 B | 0 B |
|
| 135 |
+
| Solver instances | 71 | 197 | 381 | 67 | 127 MB | 127 MB |
|
| 136 |
| Browser tests | 123 | 439 | 852 | 88 | 31.9 MB | 31.9 MB |
|
| 137 |
+
| Hardware design (EDA) tests | 81 | 258 | 651 | 75 | 80.8 MB | 80.7 MB |
|
| 138 |
| Issue trackers | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
|
| 139 |
| Media and documents | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
|
| 140 |
+
| Model serving configurations | 103 | 398 | 1,168 | 83 | 250 MB | 229 MB |
|
| 141 |
| Rule calculations | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
|
| 142 |
| Security detection | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
|
| 143 |
| Science reference data | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
|
|
|
|
| 151 |
### Corpora: rows by main domain
|
| 152 |
|
| 153 |
<picture class="not-prose">
|
| 154 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-narrow-light.svg">
|
| 155 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-wide-light.svg" alt="Corpus rows by main domain, longest bar first: Data science 750,815,212; System administration 185,826,217; Many domains 109,262,742; Software engineering 45,267,701; Science 44,633,675; Operations 39,404,642; Media 12,751,612; Machine learning 9,636,889; Hardware 8,716,520; Unclassified 5,284,162; Security 4,521,631; Misc 241,125.">
|
| 156 |
</picture>
|
| 157 |
<picture class="not-prose">
|
| 158 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-narrow-dark.svg">
|
| 159 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-wide-dark.svg" alt="Corpus rows by main domain, longest bar first: Data science 750,815,212; System administration 185,826,217; Many domains 109,262,742; Software engineering 45,267,701; Science 44,633,675; Operations 39,404,642; Media 12,751,612; Machine learning 9,636,889; Hardware 8,716,520; Unclassified 5,284,162; Security 4,521,631; Misc 241,125.">
|
| 160 |
</picture>
|
| 161 |
|
| 162 |
139 corpora with 1,216,362,128 rows.
|
|
|
|
| 189 |
## How it is made
|
| 190 |
|
| 191 |
<picture class="not-prose">
|
| 192 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-narrow-light.svg">
|
| 193 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-wide-light.svg" alt="How the index is made, in four steps. 1, collect: collectors read whole upstreams into corpora, one row per file or record. 2, admit: works the search found upstream and groups of corpus rows become material sets. 3, check: licenses, personal data, overlap with benchmarks and sizes are checked before release. 4, release: one commit with its own tag, release-2026-11-11.18.">
|
| 194 |
</picture>
|
| 195 |
<picture class="not-prose">
|
| 196 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-narrow-dark.svg">
|
| 197 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-wide-dark.svg" alt="How the index is made, in four steps. 1, collect: collectors read whole upstreams into corpora, one row per file or record. 2, admit: works the search found upstream and groups of corpus rows become material sets. 3, check: licenses, personal data, overlap with benchmarks and sizes are checked before release. 4, release: one commit with its own tag, release-2026-11-11.18.">
|
| 198 |
</picture>
|
| 199 |
|
| 200 |
The terms, each step and how every number is made are folded below.
|
|
|
|
| 216 |
- **Collection:** the material sets of one task family, such as spreadsheets or disk images;
|
| 217 |
a corpus that holds a collection's kind of material is listed under it and stays a corpus.
|
| 218 |
- **Form:** the shape of a source's material, one id such as `spreadsheets` or `test-suites` (the catalog column `form`, the option `--form` of `tools/fetch.py`).
|
| 219 |
+
- **Default profile:** the rules `tools/fetch.py` applies unless an option relaxes them: it leaves out duplicate copies, rows that matched a benchmark or were not compared with the benchmark list yet, rows in a language other than English and files above 20 MB; rows of any license and rows marked as carrying personal data come unless `--training-only` or `--no-personal-data` leaves them out ([`docs/for-agents.md`](docs/for-agents.md)).
|
| 220 |
|
| 221 |
Every row carries a `unit_id`: in a material set, the key of its unit;
|
| 222 |
in a corpus, the row's own key (its `uri`, then `@<range_offset>` for a byte range, then `!/<member_path>` for an archive member), and `tools/fetch.py` puts the file of a corpus row in a folder named after it.
|
|
|
|
| 248 |
|
| 249 |
| Level | Recorded size | Deduplicated size | Rows read | Rows compared | Rows that match | Rows read, not compared | Rows without a recorded size | Their estimated size |
|
| 250 |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 251 |
+
| Whole index | 4.45 PB | 4.11 PB | 34,710 | 33,832 | 33,703 (99.6%) | 878: address unreachable 449, file gone 80, login needed 329, no size from the server 19, other error 1 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
|
| 252 |
| Corpora | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610 (99.6%) | 871: address unreachable 449, file gone 80, login needed 322, no size from the server 19, other error 1 | 312,300,338 | 202 TB, an estimate (95% range 171 TB to 245 TB), for 215,326,513 of them, from 14,161 rows read |
|
| 253 |
+
| Task material | 2.88 GB | 2.82 GB | 3,116 | 3,109 | 3,093 (99.5%) | 7: login needed 7 | 0 | none |
|
| 254 |
|
| 255 |
Rows read are the sample of rows the check reads in `meta/sizes.parquet`;
|
| 256 |
a row compared matched its recorded size or did not, and a row read but not compared never reached its file.
|
|
|
|
| 259 |
|
| 260 |
The recorded size is the sum of `size_bytes` over the rows that have one (catalog column `bytes`). The check reads a sample of rows of every data file (`meta/sizes.parquet`): 20 of its rows with a recorded size, and of its rows without one its share of its source's sample, sized for their mean size within 25% at 95% confidence by the spread of the sizes read so far (at most 400 rows per source per round, 20 for a source with none read yet); its share is the rows that match their recorded size out of the rows it compared (status `match`, `size_mismatch` or `hash_mismatch`), and the rows it could not compare, which never reached their file, are counted beside it by status (`unreachable`, `gone`, `needs_auth`, `no_length`, `error`; catalog column `size_check`). The estimate for a source's rows without a recorded size (catalog columns `unsized_rows` and `unsized_estimate`) is their number times the mean measured size of such rows read, with a 95% BCa bootstrap range (`scipy.stats.bootstrap`, its default resample count); a source with fewer than 10 such rows read has none. Above one source, the estimate is the sum of its sources' estimates and its range the sum of their ranges, which is wider than a joint interval. The figures of every level are in `meta/size_figures.json`.
|
| 261 |
|
| 262 |
+
Distinct content is the sum of `size_bytes` over the rows minus the sizes of the rows that repeat another row's content (the same sha256, else sha1, else md5 and size, within the same hash scope; one copy of each content is kept, in the first source by registry order), counted after each release over the data files it published, per source in the catalog column `distinct_bytes`; archive members whose archive the same source also lists whole have other hashes, so they are not subtracted. `tools/fetch.py` removes duplicates over your own selection when it reads it. The deduplicated sizes here were counted over the data files of release `release-2026-11-11.15`.
|
| 263 |
|
| 264 |
</details>
|
| 265 |
|
| 266 |
<details>
|
| 267 |
<summary>How the training-stage tables are counted</summary>
|
| 268 |
|
| 269 |
+
The stage table `catalog/stages.yaml` of the code repository gives each training stage a condition on fields the index already has: the kind, form and domain of a row's source, the file types of a unit and, for some stages, rules of their own such as a size limit.
|
| 270 |
+
The release records those fields for every unit of a material set and every row of a corpus, with how many of the rules every stage shares it passes, in this order: the benchmark rule (a row that matches a public benchmark task is left out), the rule on rows not compared with the benchmark list yet and deduplication (`meta/stage_fields.parquet`);
|
| 271 |
+
each stage counts what meets its condition before those rules and after each of them, and its count after the last one is what it can take.
|
| 272 |
+
The training permit and the personal-data mark are labels of every row, neither a condition nor a rule of any stage.
|
| 273 |
+
Each table below gives what a training stage can take, counted before the rules and after each of them, in their order, down to the stage's count;
|
| 274 |
+
its last two rows count how much of the stage's count is allowed for training (every file's training permit yes or conditional) and how much is marked as carrying personal data.
|
| 275 |
|
| 276 |
+
This release has not counted its stage fields (`meta/stage_fields.parquet`) yet, so every count below says not counted yet.
|
| 277 |
|
| 278 |
#### Post-training (in use)
|
| 279 |
|
| 280 |
+
Material set units for building tasks: every file of the unit in English or no natural language, and at most 20 MB.
|
| 281 |
|
| 282 |
| Step | Material sets | Units | Upstream sources |
|
| 283 |
|---|---:|---:|---:|
|
| 284 |
+
| In the stage's condition, before the rules | not counted yet | not counted yet | not counted yet |
|
| 285 |
+
| After the benchmark rule | not counted yet | not counted yet | not counted yet |
|
| 286 |
+
| After the rule on rows not compared with the benchmark list | not counted yet | not counted yet | not counted yet |
|
| 287 |
+
| After deduplication: the stage's count | not counted yet | not counted yet | not counted yet |
|
| 288 |
+
| Of the stage's count, allowed for training (yes or conditional) | not counted yet | not counted yet | not counted yet |
|
| 289 |
+
| Of the stage's count, marked as carrying personal data | not counted yet | not counted yet | not counted yet |
|
| 290 |
|
| 291 |
#### Mid-training (provisional: conditions and counts only)
|
| 292 |
|
| 293 |
+
Corpus rows of code, documentation and the terminal side: conditions and counts only.
|
| 294 |
|
| 295 |
| Step | Rows | Size |
|
| 296 |
|---|---:|---:|
|
| 297 |
+
| In the stage's condition, before the rules | not counted yet | not counted yet |
|
| 298 |
+
| After the benchmark rule | not counted yet | not counted yet |
|
| 299 |
+
| After the rule on rows not compared with the benchmark list | not counted yet | not counted yet |
|
| 300 |
+
| After deduplication: the stage's count | not counted yet | not counted yet |
|
| 301 |
+
| Of the stage's count, allowed for training (yes or conditional) | not counted yet | not counted yet |
|
| 302 |
+
| Of the stage's count, marked as carrying personal data | not counted yet | not counted yet |
|
| 303 |
|
| 304 |
#### Pre-training (provisional: conditions and counts only)
|
| 305 |
|
| 306 |
+
Corpus rows of plain text, the mid-training types not counted here: conditions and counts only.
|
| 307 |
|
| 308 |
| Step | Rows | Size |
|
| 309 |
|---|---:|---:|
|
| 310 |
+
| In the stage's condition, before the rules | not counted yet | not counted yet |
|
| 311 |
+
| After the benchmark rule | not counted yet | not counted yet |
|
| 312 |
+
| After the rule on rows not compared with the benchmark list | not counted yet | not counted yet |
|
| 313 |
+
| After deduplication: the stage's count | not counted yet | not counted yet |
|
| 314 |
+
| Of the stage's count, allowed for training (yes or conditional) | not counted yet | not counted yet |
|
| 315 |
+
| Of the stage's count, marked as carrying personal data | not counted yet | not counted yet |
|
| 316 |
|
| 317 |
</details>
|
| 318 |
|
|
|
|
| 409 |
| `none_known` | 3 | 0 |
|
| 410 |
| `present` | 14 | 0 |
|
| 411 |
| `suspected` | 1 | 0 |
|
| 412 |
+
| Not judged yet | 121 | 1,808 |
|
| 413 |
| Sample files read per judged source | 9 to 30 | none |
|
| 414 |
| Days the sample files were read | 2026-10-04 and 2026-10-05 | none |
|
| 415 |
|
|
|
|
| 420 |
|
| 421 |
| Sources | Corpora | Task material |
|
| 422 |
|---|---:|---:|
|
| 423 |
+
| With a recorded size for every row | 92 of 139 | 1,808 of 1,808 |
|
| 424 |
+
| With a summary read from sample files | 0 of 139 | 0 of 1,808 |
|
| 425 |
+
| With personal information read from sample files | 18 of 139 | 0 of 1,808 |
|
| 426 |
+
| Still being collected | 25 of 139 | 0 of 1,808 |
|
| 427 |
|
| 428 |
- Rows whose size the source does not publish are shown as unknown and left out of the size totals, never counted as 0.
|
| 429 |
- Fields not judged yet say so;
|
|
|
|
| 449 |
<details>
|
| 450 |
<summary>Releases</summary>
|
| 451 |
|
| 452 |
+
- Latest release: `release-2026-11-11.18` (its newest CHANGELOG entry is dated 2026-11-11).
|
| 453 |
- Releases are listed in [CHANGELOG.md](CHANGELOG.md);
|
| 454 |
+
each one is a commit with its own release tag (this one is [`release-2026-11-11.18`](https://huggingface.co/datasets/Battam/training-data-foundation/tree/release-2026-11-11.18)),
|
| 455 |
so any earlier state stays readable at its tag.
|
| 456 |
- `meta/projection.json` records the release tag and the version of the program and templates that generated this card.
|
| 457 |
- Croissant metadata: [generated by the Hub](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
|
|
|
|
| 484 |
@misc{training_data_foundation,
|
| 485 |
title = {training-data-foundation: an index of public materials for building agent environments},
|
| 486 |
howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
|
| 487 |
+
note = {Release release-2026-11-11.18},
|
| 488 |
year = {2026}
|
| 489 |
}
|
| 490 |
```
|
README.zh.md
CHANGED
|
@@ -11,25 +11,25 @@ tags:
|
|
| 11 |
---
|
| 12 |
<!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
|
| 13 |
|
| 14 |
-
<img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 15 |
|
| 16 |
本数据集是**数据基座**的公开索引:在这里找到可以用来构建智能体任务和环境的公开文件,并用一条命令把它们取下来。
|
| 17 |
|
| 18 |
[English](README.md) | 简体中文 · [数据页](https://huggingface.co/spaces/Battam/training-data-foundation) · [使用示例视频](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
|
| 19 |
|
| 20 |
<picture class="not-prose">
|
| 21 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 22 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 23 |
</picture>
|
| 24 |
<picture class="not-prose">
|
| 25 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 26 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 27 |
</picture>
|
| 28 |
|
| 29 |
## 一条命令取数据
|
| 30 |
|
| 31 |
```sh
|
| 32 |
-
uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 33 |
```
|
| 34 |
|
| 35 |
这条命令从每个电子表格材料集各取 3 个单元放进 `./data`,每个单元一个文件夹,并按每个文件所在行记录的大小和哈希值校验它。
|
|
@@ -39,7 +39,7 @@ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/r
|
|
| 39 |
<details>
|
| 40 |
<summary>命令取什么、按什么顺序取</summary>
|
| 41 |
|
| 42 |
-
这条命令和下面的查询都固定在发布标签 `release-2026-11-11.
|
| 43 |
[uv](https://docs.astral.sh/uv/) 会读取脚本声明的依赖,取这个发布当天之前发布的版本。
|
| 44 |
|
| 45 |
命令返回单元:形式为 `spreadsheets` 的每个材料集各 3 个单元,每个单元放在以它的 `unit_id` 命名的单独文件夹里。
|
|
@@ -53,8 +53,9 @@ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/r
|
|
| 53 |
<summary>默认配置去掉什么</summary>
|
| 54 |
|
| 55 |
默认配置是 `tools/fetch.py` 在没有选项放宽时应用的规则。
|
| 56 |
-
它去掉重复的副本、与基准测试相符或还没有与基准测试清单比对的行、
|
| 57 |
每条规则都有自己的选项,每次运行都会说明每条规则去掉了多少个单元。
|
|
|
|
| 58 |
每条规则及其选项,以及哪些内容不能放进智能体的环境:[`docs/for-agents.zh.md`](docs/for-agents.zh.md)。
|
| 59 |
|
| 60 |
</details>
|
|
@@ -66,7 +67,7 @@ uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/r
|
|
| 66 |
|
| 67 |
```sql
|
| 68 |
SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
|
| 69 |
-
FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 70 |
WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
|
| 71 |
ORDER BY collection, units DESC;
|
| 72 |
```
|
|
@@ -77,7 +78,7 @@ ORDER BY collection, units DESC;
|
|
| 77 |
在一个语料库里挑选文件:
|
| 78 |
|
| 79 |
```sql
|
| 80 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 81 |
SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
|
| 82 |
FROM read_parquet(getvariable('files'))
|
| 83 |
WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
|
|
@@ -99,15 +100,15 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 99 |
### 任务材料:按合集的单元数
|
| 100 |
|
| 101 |
<picture class="not-prose">
|
| 102 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 103 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 104 |
</picture>
|
| 105 |
<picture class="not-prose">
|
| 106 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 107 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 108 |
</picture>
|
| 109 |
|
| 110 |
-
1,
|
| 111 |
|
| 112 |
<details>
|
| 113 |
<summary>按合集的任务材料(表格)</summary>
|
|
@@ -118,12 +119,12 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 118 |
| 磁盘镜像 | 52 | 179 | 324 | 40 | 272 MB | 272 MB |
|
| 119 |
| 复现包和运行 | 88 | 106 | 870 | 88 | 120 MB | 120 MB |
|
| 120 |
| SEC 申报审查 | 0 | 0 | 0 | 0 | 0 B | 0 B |
|
| 121 |
-
| 求解器实例 |
|
| 122 |
| 浏览器测试 | 123 | 439 | 852 | 88 | 31.9 MB | 31.9 MB |
|
| 123 |
-
| 硬件设计(EDA)测试 | 81 | 258 | 651 | 75 | 80.8 MB | 80.
|
| 124 |
| 问题跟踪系统 | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
|
| 125 |
| 媒体和文档 | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
|
| 126 |
-
| 模型服务配置 | 103 | 398 | 1,168 | 83 | 250 MB |
|
| 127 |
| 规则计算 | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
|
| 128 |
| 安全检测 | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
|
| 129 |
| 科学参考数据 | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
|
|
@@ -137,12 +138,12 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 137 |
### 语料库:按主要领域的行数
|
| 138 |
|
| 139 |
<picture class="not-prose">
|
| 140 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 141 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 142 |
</picture>
|
| 143 |
<picture class="not-prose">
|
| 144 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 145 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 146 |
</picture>
|
| 147 |
|
| 148 |
139 个语料库,共 1,216,362,128 行。
|
|
@@ -175,12 +176,12 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 175 |
## 怎样做成
|
| 176 |
|
| 177 |
<picture class="not-prose">
|
| 178 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 179 |
-
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 180 |
</picture>
|
| 181 |
<picture class="not-prose">
|
| 182 |
-
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 183 |
-
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.
|
| 184 |
</picture>
|
| 185 |
|
| 186 |
术语、每一步,以及每个数字怎样得出,都折叠在下面。
|
|
@@ -202,7 +203,7 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 202 |
- **合集:** 同一个任务类别的材料集,例如电子表格或磁盘镜像;
|
| 203 |
装着某个合集那一类材料的语料库列在该合集下,仍然是语料库。
|
| 204 |
- **形式:** 一个来源的材料的形态,一个标识,例如 `spreadsheets` 或 `test-suites`(目录列 `form`,`tools/fetch.py` 的选项 `--form`)。
|
| 205 |
-
- **默认配置:** `tools/fetch.py` 在没有选项放宽时应用的规则:去掉重复的副本、与基准测试相符或还没有与基准测试清单比对的行、
|
| 206 |
|
| 207 |
每一行都带有 `unit_id`:在材料集中,是它所属单元的键;
|
| 208 |
在语料库中,是该行自己的键(它的 `uri`,字节范围再接 `@<range_offset>`,存档成员再接 `!/<member_path>`),`tools/fetch.py` 把语料库一行的文件放进以它命名的文件夹。
|
|
@@ -234,9 +235,9 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 234 |
|
| 235 |
| 层级 | 记录的大小 | 去重后大小 | 读过的行 | 比对过的行 | 一致的行 | 读过但未比对的行 | 没有记录大小的行 | 这些行的估计大小 |
|
| 236 |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 237 |
-
| 整个索引 | 4.45 PB | 4.11 PB | 34,
|
| 238 |
| 语料库 | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610(99.6%) | 871:地址无法访问 449 行、文件已不存在 80 行、需要登录 322 行、服务器未给出大小 19 行、其他错误 1 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
|
| 239 |
-
| 任务材料 | 2.88 GB | 2.
|
| 240 |
|
| 241 |
读过的行是核对在 `meta/sizes.parquet` 中读取的抽样行;
|
| 242 |
比对过的行与记录的大小一致或不一致,读过但未比对的行没有取到它的文件。
|
|
@@ -245,56 +246,60 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 245 |
|
| 246 |
记录的大小是有大小的各行 `size_bytes` 之和(目录列 `bytes`)。核对读取每个数据文件中抽取的一组行(`meta/sizes.parquet`):有记录大小的行抽 20 行;没有记录大小的行按其所在来源的样本分摊,样本量按目前读到的大小的离散程度定,使其平均大小在 95% 置信度下误差不超过 25%(每个来源每轮最多 400 行,还没有读数的来源取 20 行);其比例是比对过的行(状态 `match`、`size_mismatch` 或 `hash_mismatch`)中与记录的大小一致的行所占的比例,没能比对的行(没有取到文件)按状态另行计数(`unreachable`、`gone`、`needs_auth`、`no_length`、`error`;目录列 `size_check`)。一个来源没有记录大小的行的估计(目录列 `unsized_rows` 和 `unsized_estimate`)是这些行的行数乘以读过的这类行的平均实测大小,并给出 95% BCa 自助法范围(`scipy.stats.bootstrap`,用其默认重抽样次数);读过的这类行少于 10 行的来源没有估计。在来源之上的层级,估计是其各来源估计之和,范围是各来源范围之和,比联合区间更宽。每个层级的数字都在 `meta/size_figures.json` 中。
|
| 247 |
|
| 248 |
-
去重后内容是各行 `size_bytes` 之和,减去与另一行内容重复的那些行的大小(sha256 相同;没有 sha256 时 sha1 相同;再没有时 md5 和大小都相同;并且哈希范围相同;每种内容只保留一份,留在按注册顺序排在最前的来源中),在每次发布之后按它发布的数据文件重新计算,每个来源的值在目录列 `distinct_bytes` 中;同一来源也整个列出的存档,其成员的哈希与存档不同,所以不扣除。`tools/fetch.py` 读取你自己的选择时会去掉其中的重复。 这里的去重后大小是按发布 `release-2026-11-11.
|
| 249 |
|
| 250 |
</details>
|
| 251 |
|
| 252 |
<details>
|
| 253 |
<summary>训练阶段表怎样计数</summary>
|
| 254 |
|
| 255 |
-
代码仓库的阶段表 `catalog/stages.yaml` 给每个训练阶段一个条件,条件只用索引已有的字段:一行所属来源的种类、形式和领域,一个单元的文件类型
|
| 256 |
-
发布为材料集的每个单元和语料库的每一行记录这些字段,并记录它按以下顺序通过了各阶段共用规则中的几条:基准测试规则(与公开基准测试任务相符的行被去掉)、还没有与基准测试清单比对的行的规则
|
| 257 |
-
每个阶段在这些规则之前和每条规则之后统计符合它的条件的材料,最后一
|
| 258 |
-
|
| 259 |
-
每张表的
|
|
|
|
| 260 |
|
| 261 |
-
阶段字段(`meta/stage_fields.parquet`)
|
| 262 |
|
| 263 |
#### 后训练(在用)
|
| 264 |
|
| 265 |
-
用于构建任务的材料集单元:该单元的每个文件
|
| 266 |
|
| 267 |
| 步骤 | 材料集数 | 单元数 | 上游来源数 |
|
| 268 |
|---|---:|---:|---:|
|
| 269 |
-
| 符合该阶段的条件,在各条规则之前 |
|
| 270 |
-
| 基准测试规则之后 |
|
| 271 |
-
| 未与基准测试清单比对的行的规则之后 |
|
| 272 |
-
|
|
| 273 |
-
|
|
|
|
|
| 274 |
|
| 275 |
#### 中训练(暂定:只给条件和计数)
|
| 276 |
|
| 277 |
-
代码、文档和终端侧的语料库行
|
| 278 |
|
| 279 |
| 步骤 | 行数 | 大小 |
|
| 280 |
|---|---:|---:|
|
| 281 |
-
| 符合该阶段的条件,在各条规则之前 |
|
| 282 |
-
| 基准测试规则之后 |
|
| 283 |
-
| 未与基准测试清单比对的行的规则之后 |
|
| 284 |
-
|
|
| 285 |
-
|
|
|
|
|
| 286 |
|
| 287 |
#### 预训练(暂定:只给条件和计数)
|
| 288 |
|
| 289 |
-
纯文本的语料库行,
|
| 290 |
|
| 291 |
| 步骤 | 行数 | 大小 |
|
| 292 |
|---|---:|---:|
|
| 293 |
-
| 符合该阶段的条件,在各条规则之前 |
|
| 294 |
-
| 基准测试规则之后 |
|
| 295 |
-
| 未与基准测试清单比对的行的规则之后 |
|
| 296 |
-
|
|
| 297 |
-
|
|
|
|
|
| 298 |
|
| 299 |
</details>
|
| 300 |
|
|
@@ -393,7 +398,7 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 393 |
| `none_known` | 3 | 0 |
|
| 394 |
| `present` | 14 | 0 |
|
| 395 |
| `suspected` | 1 | 0 |
|
| 396 |
-
| 尚未判读 | 121 | 1,
|
| 397 |
| 每个已判读的来源读过的样本文件数 | 9 到 30 | 无 |
|
| 398 |
| 读取样本文件的日期 | 2026-10-04 和 2026-10-05 | 无 |
|
| 399 |
|
|
@@ -404,10 +409,10 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 404 |
|
| 405 |
| 来源 | 语料库 | 任务材料 |
|
| 406 |
|---|---:|---:|
|
| 407 |
-
| 每一行都有记录的大小 | 139 个中的 92 个 | 1,
|
| 408 |
-
| 已从样本文件读出摘要 | 139 个中的 0 个 | 1,
|
| 409 |
-
| 已从样本文件读出个人信息 | 139 个中的 18 个 | 1,
|
| 410 |
-
| 仍在采集中 | 139 个中的 25 个 | 1,
|
| 411 |
|
| 412 |
- 来源没有公布大小的行显示为未知,不计入大小合计,也从不按 0 计。
|
| 413 |
- 尚未判读的字段会注明;
|
|
@@ -433,9 +438,9 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 433 |
<details>
|
| 434 |
<summary>发布</summary>
|
| 435 |
|
| 436 |
-
- 最新发布:`release-2026-11-11.
|
| 437 |
- 各次发布列在 [CHANGELOG.md](CHANGELOG.md) 中;
|
| 438 |
-
每次发布是一个带有自己发布标签的提交(本次为 [`release-2026-11-11.
|
| 439 |
因此任何较早的状态都可以在它的标签处读到。
|
| 440 |
- `meta/projection.json` 记录发布标签,以及生成本卡片的程序和模板的版本。
|
| 441 |
- Croissant 元数据:[由 Hub 生成](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
|
|
@@ -476,7 +481,7 @@ ORDER BY size_bytes LIMIT 20;
|
|
| 476 |
@misc{training_data_foundation,
|
| 477 |
title = {training-data-foundation: an index of public materials for building agent environments},
|
| 478 |
howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
|
| 479 |
-
note = {Release release-2026-11-11.
|
| 480 |
year = {2026}
|
| 481 |
}
|
| 482 |
```
|
|
|
|
| 11 |
---
|
| 12 |
<!-- Generated by artpool.index_projection from meta/, cards/*.yaml, the field registry and its templates. Every release regenerates it, so edits to this file do not last. -->
|
| 13 |
|
| 14 |
+
<img class="not-prose" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/banner.zh.svg" alt="公开网络,分两层(按合集的任务材料;语料库),每个文件一行(地址、大小、SHA-256、许可证),你的选择,下载并校验">
|
| 15 |
|
| 16 |
本数据集是**数据基座**的公开索引:在这里找到可以用来构建智能体任务和环境的公开文件,并用一条命令把它们取下来。
|
| 17 |
|
| 18 |
[English](README.md) | 简体中文 · [数据页](https://huggingface.co/spaces/Battam/training-data-foundation) · [使用示例视频](https://huggingface.co/spaces/Battam/training-data-foundation#usage-video)
|
| 19 |
|
| 20 |
<picture class="not-prose">
|
| 21 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-narrow-light.zh.svg">
|
| 22 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-wide-light.zh.svg" alt="1,808 个材料集,共 4,195 个单元,分在 14 个合集;139 个语料库,共 1,216,362,128 行;指向的大小共 4.45 PB,索引存的是地址,不是文件副本;比对过的 33,832 行中,99.6% 与记录的大小一致。">
|
| 23 |
</picture>
|
| 24 |
<picture class="not-prose">
|
| 25 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-narrow-dark.zh.svg">
|
| 26 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/numbers-wide-dark.zh.svg" alt="1,808 个材料集,共 4,195 个单元,分在 14 个合集;139 个语料库,共 1,216,362,128 行;指向的大小共 4.45 PB,索引存的是地址,不是文件副本;比对过的 33,832 行中,99.6% 与记录的大小一致。">
|
| 27 |
</picture>
|
| 28 |
|
| 29 |
## 一条命令取数据
|
| 30 |
|
| 31 |
```sh
|
| 32 |
+
uv run https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/tools/fetch.py --form spreadsheets --sample 3 --out ./data
|
| 33 |
```
|
| 34 |
|
| 35 |
这条命令从每个电子表格材料集各取 3 个单元放进 `./data`,每个单元一个文件夹,并按每个文件所在行记录的大小和哈希值校验它。
|
|
|
|
| 39 |
<details>
|
| 40 |
<summary>命令取什么、按什么顺序取</summary>
|
| 41 |
|
| 42 |
+
这条命令和下面的查询都固定在发布标签 `release-2026-11-11.18` 上。
|
| 43 |
[uv](https://docs.astral.sh/uv/) 会读取脚本声明的依赖,取这个发布当天之前发布的版本。
|
| 44 |
|
| 45 |
命令返回单元:形式为 `spreadsheets` 的每个材料集各 3 个单元,每个单元放在以它的 `unit_id` 命名的单独文件夹里。
|
|
|
|
| 53 |
<summary>默认配置去掉什么</summary>
|
| 54 |
|
| 55 |
默认配置是 `tools/fetch.py` 在没有选项放宽时应用的规则。
|
| 56 |
+
它去掉重复的副本、与基准测试相符或还没有与基准测试清单比对的行、记录为英语以外语言的行,以及大于 20 MB 的文件。
|
| 57 |
每条规则都有自己的选项,每次运行都会说明每条规则去掉了多少个单元。
|
| 58 |
+
许可证和个人数据标记是每一行的标签,不是默认配置的规则:不论它们是什么,每一行都会进来;`--training-only` 去掉许可证不允许随附分发的文件,`--no-personal-data` 去掉个人数据。
|
| 59 |
每条规则及其选项,以及哪些内容不能放进智能体的环境:[`docs/for-agents.zh.md`](docs/for-agents.zh.md)。
|
| 60 |
|
| 61 |
</details>
|
|
|
|
| 67 |
|
| 68 |
```sql
|
| 69 |
SELECT source, collection, title, units, files, top_types, license.tier AS license_tier, needs_auth
|
| 70 |
+
FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet'
|
| 71 |
WHERE kind = 'material_set' AND needs_auth = 'no' AND license.tier IN ('permissive', 'share-alike')
|
| 72 |
ORDER BY collection, units DESC;
|
| 73 |
```
|
|
|
|
| 78 |
在一个语料库里挑选文件:
|
| 79 |
|
| 80 |
```sql
|
| 81 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'if-archive'));
|
| 82 |
SELECT uri, size_bytes, lower(hex(sha256)) AS sha256
|
| 83 |
FROM read_parquet(getvariable('files'))
|
| 84 |
WHERE source = 'if-archive' AND ext = 'pdf' AND size_bytes BETWEEN 100000 AND 5000000
|
|
|
|
| 100 |
### 任务材料:按合集的单元数
|
| 101 |
|
| 102 |
<picture class="not-prose">
|
| 103 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-narrow-light.zh.svg">
|
| 104 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-wide-light.zh.svg" alt="按合集的单元数,条形由长到短:电子表格 536;浏览器测试 439;测试和练习 425;模型服务配置 398;媒体和文档 391;科学参考数据 342;规则计算 329;安全检测 328;硬件设计(EDA)测试 258;问题跟踪系统 258;求解器实例 197;磁盘镜像 179;复现包和运行 106;不属于任何合集 9;SEC 申报审查 0。">
|
| 105 |
</picture>
|
| 106 |
<picture class="not-prose">
|
| 107 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-narrow-dark.zh.svg">
|
| 108 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/material-wide-dark.zh.svg" alt="按合集的单元数,条形由长到短:电子表�� 536;浏览器测试 439;测试和练习 425;模型服务配置 398;媒体和文档 391;科学参考数据 342;规则计算 329;安全检测 328;硬件设计(EDA)测试 258;问题跟踪系统 258;求解器实例 197;磁盘镜像 179;复现包和运行 106;不属于任何合集 9;SEC 申报审查 0。">
|
| 109 |
</picture>
|
| 110 |
|
| 111 |
+
1,808 个材料集,共 4,195 个单元,分在 14 个合集中(其中 1 个合集还没有材料集);其中 1 个材料集不属于任何合集。
|
| 112 |
|
| 113 |
<details>
|
| 114 |
<summary>按合集的任务材料(表格)</summary>
|
|
|
|
| 119 |
| 磁盘镜像 | 52 | 179 | 324 | 40 | 272 MB | 272 MB |
|
| 120 |
| 复现包和运行 | 88 | 106 | 870 | 88 | 120 MB | 120 MB |
|
| 121 |
| SEC 申报审查 | 0 | 0 | 0 | 0 | 0 B | 0 B |
|
| 122 |
+
| 求解器实例 | 71 | 197 | 381 | 67 | 127 MB | 127 MB |
|
| 123 |
| 浏览器测试 | 123 | 439 | 852 | 88 | 31.9 MB | 31.9 MB |
|
| 124 |
+
| 硬件设计(EDA)测试 | 81 | 258 | 651 | 75 | 80.8 MB | 80.7 MB |
|
| 125 |
| 问题跟踪系统 | 212 | 258 | 682 | 212 | 196 MB | 196 MB |
|
| 126 |
| 媒体和文档 | 276 | 391 | 1,088 | 269 | 569 MB | 569 MB |
|
| 127 |
+
| 模型服务配置 | 103 | 398 | 1,168 | 83 | 250 MB | 229 MB |
|
| 128 |
| 规则计算 | 177 | 329 | 727 | 154 | 196 MB | 186 MB |
|
| 129 |
| 安全检测 | 110 | 328 | 983 | 77 | 684 MB | 652 MB |
|
| 130 |
| 科学参考数据 | 161 | 342 | 671 | 157 | 63.8 MB | 63.8 MB |
|
|
|
|
| 138 |
### 语料库:按主要领域的行数
|
| 139 |
|
| 140 |
<picture class="not-prose">
|
| 141 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-narrow-light.zh.svg">
|
| 142 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-wide-light.zh.svg" alt="按主要领域的语料库行数,条形由长到短:数据科学 750,815,212;系统管理 185,826,217;多领域 109,262,742;软件工程 45,267,701;科学 44,633,675;运营 39,404,642;媒体 12,751,612;机器学习 9,636,889;硬件 8,716,520;未分类 5,284,162;安全 4,521,631;杂项 241,125。">
|
| 143 |
</picture>
|
| 144 |
<picture class="not-prose">
|
| 145 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-narrow-dark.zh.svg">
|
| 146 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/corpora-wide-dark.zh.svg" alt="按主要领域的语料库行数,条形由长到短:数据科学 750,815,212;系统管理 185,826,217;多领域 109,262,742;软件工程 45,267,701;科学 44,633,675;运营 39,404,642;媒体 12,751,612;机器学习 9,636,889;硬件 8,716,520;未分类 5,284,162;安全 4,521,631;杂项 241,125。">
|
| 147 |
</picture>
|
| 148 |
|
| 149 |
139 个语料库,共 1,216,362,128 行。
|
|
|
|
| 176 |
## 怎样做成
|
| 177 |
|
| 178 |
<picture class="not-prose">
|
| 179 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-narrow-light.zh.svg">
|
| 180 |
+
<img class="not-prose dark:hidden" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-wide-light.zh.svg" alt="索引分四步做成。1,采集:采集器整批读取完整上游,每个文件或每条记录记一行,成为语料库。2,收录:检索在上游找到的作品和成组的语料库行,收录为材料集。3,检查:发布前核对许可证、个人信息、与基准测试的重叠和大小。4,发布:一个带有自己标签的提交,release-2026-11-11.18。">
|
| 181 |
</picture>
|
| 182 |
<picture class="not-prose">
|
| 183 |
+
<source media="(max-width: 1023px)" srcset="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-narrow-dark.zh.svg">
|
| 184 |
+
<img class="not-prose hidden dark:block" src="https://huggingface.co/datasets/Battam/training-data-foundation/resolve/release-2026-11-11.18/docs/figures/steps-wide-dark.zh.svg" alt="索引分四步做成。1,采集:采集器整批读取完整上游,每个文件或每条记录记一行,成为语料库。2,收录:检索在上游找到的作品和成组的语料库行,收录为材料集。3,检查:发布前核对许可证、个人信息、与基准测试的重叠和大小。4,发布:一个带有自己标签的提交,release-2026-11-11.18。">
|
| 185 |
</picture>
|
| 186 |
|
| 187 |
术语、每一步,以及每个数字怎样得出,都折叠在下面。
|
|
|
|
| 203 |
- **合集:** 同一个任务类别的材料集,例如电子表格或磁盘镜像;
|
| 204 |
装着某个合集那一类材料的语料库列在该合集下,仍然是语料库。
|
| 205 |
- **形式:** 一个来源的材料的形态,一个标识,例如 `spreadsheets` 或 `test-suites`(目录列 `form`,`tools/fetch.py` 的选项 `--form`)。
|
| 206 |
+
- **默认配置:** `tools/fetch.py` 在没有选项放宽时应用的规则:去掉重复的副本、与基准测试相符或还没有与基准测试清单比对的行、记录为英语以外语言的行,以及大于 20 MB 的文件;任何许可证的行和标为带有个人数据的行都会进来,除非 `--training-only` 或 `--no-personal-data` 把它们去掉([`docs/for-agents.zh.md`](docs/for-agents.zh.md))。
|
| 207 |
|
| 208 |
每一行都带有 `unit_id`:在材料集中,是它所属单元的键;
|
| 209 |
在语料库中,是该行自己的键(它的 `uri`,字节范围再接 `@<range_offset>`,存档成员再接 `!/<member_path>`),`tools/fetch.py` 把语料库一行的文件放进以它命名的文件夹。
|
|
|
|
| 235 |
|
| 236 |
| 层级 | 记录的大小 | 去重后大小 | 读过的行 | 比对过的行 | 一致的行 | 读过但未比对的行 | 没有记录大小的行 | 这些行的估计大小 |
|
| 237 |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 238 |
+
| 整个索引 | 4.45 PB | 4.11 PB | 34,710 | 33,832 | 33,703(99.6%) | 878:地址无法访问 449 行、文件已不存在 80 行、需要登录 329 行、服务器���给出大小 19 行、其他错误 1 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
|
| 239 |
| 语料库 | 4.45 PB | 4.11 PB | 31,594 | 30,723 | 30,610(99.6%) | 871:地址无法访问 449 行、文件已不存在 80 行、需要登录 322 行、服务器未给出大小 19 行、其他错误 1 行 | 312,300,338 | 202 TB,这是估计值(95% 范围 171 TB 到 245 TB),针对其中 215,326,513 行,依据读过的 14,161 行 |
|
| 240 |
+
| 任务材料 | 2.88 GB | 2.82 GB | 3,116 | 3,109 | 3,093(99.5%) | 7:需要登录 7 行 | 0 | 无 |
|
| 241 |
|
| 242 |
读过的行是核对在 `meta/sizes.parquet` 中读取的抽样行;
|
| 243 |
比对过的行与记录的大小一致或不一致,读过但未比对的行没有取到它的文件。
|
|
|
|
| 246 |
|
| 247 |
记录的大小是有大小的各行 `size_bytes` 之和(目录列 `bytes`)。核对读取每个数据文件中抽取的一组行(`meta/sizes.parquet`):有记录大小的行抽 20 行;没有记录大小的行按其所在来源的样本分摊,样本量按目前读到的大小的离散程度定,使其平均大小在 95% 置信度下误差不超过 25%(每个来源每轮最多 400 行,还没有读数的来源取 20 行);其比例是比对过的行(状态 `match`、`size_mismatch` 或 `hash_mismatch`)中与记录的大小一致的行所占的比例,没能比对的行(没有取到文件)按状态另行计数(`unreachable`、`gone`、`needs_auth`、`no_length`、`error`;目录列 `size_check`)。一个来源没有记录大小的行的估计(目录列 `unsized_rows` 和 `unsized_estimate`)是这些行的行数乘以读过的这类行的平均实测大小,并给出 95% BCa 自助法范围(`scipy.stats.bootstrap`,用其默认重抽样次数);读过的这类行少于 10 行的来源没有估计。在来源之上的层级,估计是其各来源估计之和,范围是各来源范围之和,比联合区间更宽。每个层级的数字都在 `meta/size_figures.json` 中。
|
| 248 |
|
| 249 |
+
去重后内容是各行 `size_bytes` 之和,减去与另一行内容重复的那些行的大小(sha256 相同;没有 sha256 时 sha1 相同;再没有时 md5 和大小都相同;并且哈希范围相同;每种内容只保留一份,留在按注册顺序排在最前的来源中),在每次发布之后按它发布的数据文件重新计算,每个来源的值在目录列 `distinct_bytes` 中;同一来源也整个列出的存档,其成员的哈希与存档不同,所以不扣除。`tools/fetch.py` 读取你自己的选择时会去掉其中的重复。 这里的去重后大小是按发布 `release-2026-11-11.15` 的数据文件算出的。
|
| 250 |
|
| 251 |
</details>
|
| 252 |
|
| 253 |
<details>
|
| 254 |
<summary>训练阶段表怎样计数</summary>
|
| 255 |
|
| 256 |
+
代码仓库的阶段表 `catalog/stages.yaml` 给每个训练阶段一个条件,条件只用索引已有的字段:一行所属来源的种类、形式和领域,一个单元的文件类型,以及有些阶段自己的规则,例如大小上限。
|
| 257 |
+
发布为材料集的每个单元和语料库的每一行记录这些字段,并记录它按以下顺序通过了各阶段共用规则中的几条:基准测试规则(与公开基准测试任务相符的行被去掉)、还没有与基准测试清单比对的行的规则和去重(`meta/stage_fields.parquet`);
|
| 258 |
+
每个阶段在这些规则之前和每条规则之后统计符合它的条件的材料,最后一条规则之后的计数就是它能用的材料。
|
| 259 |
+
训练许可和个人数据标记是每一行的标签,既不是任何阶段的条件,也不是规则。
|
| 260 |
+
下面每张表给出一个训练阶段能用的材料,在各条规则之前和按顺序每条规则之后计数,直到该阶段的计数;
|
| 261 |
+
每张表的最后两行统计该阶段的计数中有多少允许训练(每个文件的训练许可为 yes 或 conditional),有多少标为带有个人数据。
|
| 262 |
|
| 263 |
+
本次发布还没有计算它的阶段字段(`meta/stage_fields.parquet`),所以下面每个计数都写作尚未计数。
|
| 264 |
|
| 265 |
#### 后训练(在用)
|
| 266 |
|
| 267 |
+
用于构建任务的材料集单元:该单元的每个文件为英语或不含自然语言,且最多 20 MB。
|
| 268 |
|
| 269 |
| 步骤 | 材料集数 | 单元数 | 上游来源数 |
|
| 270 |
|---|---:|---:|---:|
|
| 271 |
+
| 符合该阶段的条件,在各条规则之前 | 尚未计数 | 尚未计数 | 尚未计数 |
|
| 272 |
+
| 基准测试规则之后 | 尚未计数 | 尚未计数 | 尚未计数 |
|
| 273 |
+
| 未与基准测试清单比对的行的规则之后 | 尚未计数 | 尚未计数 | 尚未计数 |
|
| 274 |
+
| 去重之后:该阶段的计数 | 尚未计数 | 尚未计数 | 尚未计数 |
|
| 275 |
+
| 该阶段的计数中,允许训练的(yes 或 conditional) | 尚未计数 | 尚未计数 | 尚未计数 |
|
| 276 |
+
| 该阶段的计数中,标为带有个人数据的 | 尚未计数 | 尚未计数 | 尚未计数 |
|
| 277 |
|
| 278 |
#### 中训练(暂定:只给条件和计数)
|
| 279 |
|
| 280 |
+
代码、文档和终端侧的语料库行:仅列条件和数量。
|
| 281 |
|
| 282 |
| 步骤 | 行数 | 大小 |
|
| 283 |
|---|---:|---:|
|
| 284 |
+
| 符合该阶段的条件,在各条规则之前 | 尚未计数 | 尚未计数 |
|
| 285 |
+
| 基准测试规则之后 | 尚未计数 | 尚未计数 |
|
| 286 |
+
| 未与基准测试清单比对的行的规则之后 | 尚未计数 | 尚未计数 |
|
| 287 |
+
| 去重之后:该阶段的计数 | 尚未计数 | 尚未计数 |
|
| 288 |
+
| 该阶段的计数中,允许训练的(yes 或 conditional) | 尚未计数 | 尚未计数 |
|
| 289 |
+
| 该阶段的计数中,标为带有个人数据的 | 尚未计数 | 尚未计数 |
|
| 290 |
|
| 291 |
#### 预训练(暂定:只给条件和计数)
|
| 292 |
|
| 293 |
+
纯文本的语料库行,不在此统计中期训练类型:仅列条件和数量。
|
| 294 |
|
| 295 |
| 步骤 | 行数 | 大小 |
|
| 296 |
|---|---:|---:|
|
| 297 |
+
| 符合该阶段的条件,在各条规则之前 | 尚未计数 | 尚未计数 |
|
| 298 |
+
| 基准测试规则之后 | 尚未计数 | 尚未计数 |
|
| 299 |
+
| 未与基准测试清单比对的行的规则之后 | 尚未计数 | 尚未计数 |
|
| 300 |
+
| 去重之后:该阶段的计数 | 尚未计数 | 尚未计数 |
|
| 301 |
+
| 该阶段的计数中,允许训练的(yes 或 conditional) | 尚未计数 | 尚未计数 |
|
| 302 |
+
| 该阶段的计数中,标为带有个人数据的 | 尚未计数 | 尚未计数 |
|
| 303 |
|
| 304 |
</details>
|
| 305 |
|
|
|
|
| 398 |
| `none_known` | 3 | 0 |
|
| 399 |
| `present` | 14 | 0 |
|
| 400 |
| `suspected` | 1 | 0 |
|
| 401 |
+
| 尚未判读 | 121 | 1,808 |
|
| 402 |
| 每个已判读的来源读过的样本文件数 | 9 到 30 | 无 |
|
| 403 |
| 读取样本文件的日期 | 2026-10-04 和 2026-10-05 | 无 |
|
| 404 |
|
|
|
|
| 409 |
|
| 410 |
| 来源 | 语料库 | 任务材料 |
|
| 411 |
|---|---:|---:|
|
| 412 |
+
| 每一行都有记录的大小 | 139 个中的 92 个 | 1,808 个中的 1,808 个 |
|
| 413 |
+
| 已从样本文件读出摘要 | 139 个中的 0 个 | 1,808 个中的 0 个 |
|
| 414 |
+
| 已从样本文件读出个人信息 | 139 个中的 18 个 | 1,808 个中的 0 个 |
|
| 415 |
+
| 仍在采集中 | 139 个中的 25 个 | 1,808 个中的 0 个 |
|
| 416 |
|
| 417 |
- 来源没有公布大小的行显示为未知,不计入大小合计,也从不按 0 计。
|
| 418 |
- 尚未判读的字段会注明;
|
|
|
|
| 438 |
<details>
|
| 439 |
<summary>发布</summary>
|
| 440 |
|
| 441 |
+
- 最新发布:`release-2026-11-11.18`(其最新的 CHANGELOG 条目日期为 2026-11-11)。
|
| 442 |
- 各次发布列在 [CHANGELOG.md](CHANGELOG.md) 中;
|
| 443 |
+
每次发布是一个带有自己发布标签的提交(本次为 [`release-2026-11-11.18`](https://huggingface.co/datasets/Battam/training-data-foundation/tree/release-2026-11-11.18)),
|
| 444 |
因此任何较早的状态都可以在它的标签处读到。
|
| 445 |
- `meta/projection.json` 记录发布标签,以及生成本卡片的程序和模板的版本。
|
| 446 |
- Croissant 元数据:[由 Hub 生成](https://huggingface.co/api/datasets/Battam/training-data-foundation/croissant);
|
|
|
|
| 481 |
@misc{training_data_foundation,
|
| 482 |
title = {training-data-foundation: an index of public materials for building agent environments},
|
| 483 |
howpublished = {https://huggingface.co/datasets/Battam/training-data-foundation},
|
| 484 |
+
note = {Release release-2026-11-11.18},
|
| 485 |
year = {2026}
|
| 486 |
}
|
| 487 |
```
|
campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-algorithmic/LICENSE
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Licence: Apache-2.0
|
| 2 |
+
|
| 3 |
+
- Apache-2.0: https://github.com/open-thought/reasoning-gym/blob/49b07130b3fcd12f2d064bba7c43869543a0e7e7/LICENSE (publisher: open-thought)
|
campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-algorithmic/README.md
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# open-thought-reasoning-gym-algorithmic
|
| 2 |
+
|
| 3 |
+
**What this is:** 2 units (4 files) published by open-thought. Collection theme: Seeded puzzle generators (Simon Tatham puzzle collection class, with generation command, seed and solution).
|
| 4 |
+
|
| 5 |
+
**What one unit is:** Each unit is a few named upstream files, delivered together, kept at its upstream path. Format: JSONL of generated problems with answers, plus a generation record.
|
| 6 |
+
|
| 7 |
+
**Where it comes from:** Upstream: open-thought/reasoning-gym. Kind: generated data.
|
| 8 |
+
|
| 9 |
+
**Self-contained:** Each unit is self-contained. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
+
|
| 11 |
+
**Tools to open and check:** Check: Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (cryptarithm.py from line 287) accepts any mapping that verify_cryptarithm_solution (line 24) finds valid: 1.0 (line 329); a parseable but invalid mapping gets 0.01 (line 331); no or unparseable answer 0.0 (lines 306 and 316). Only 1.0 counts as correct. Unique solution: not required, since every valid mapping is accepted. Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (jugs.py from line 297) replays the agent's moves with verify_solution (line 118): 1.0 if the target is reached (line 319), 0.01 for a parseable move list that fails (line 317), 0.0 otherwise (lines 311 and 321). Only 1.0 counts as correct. Unique solution: not required, since any valid move list is accepted.
|
| 12 |
+
|
| 13 |
+
**Licence (SPDX):** Apache-2.0
|
| 14 |
+
|
| 15 |
+
**Provenance:** https://github.com/open-thought/reasoning-gym
|
| 16 |
+
|
| 17 |
+
Units:
|
| 18 |
+
|
| 19 |
+
- `generated/` (3 files): Reasoning-gym cryptarithm generator (reasoning_gym/algorithmic/cryptarithm.py at commit 49b07130), generated letter-addition puzzles with a verifying scorer.
|
| 20 |
+
Format: JSONL of generated problems with answers, plus a generation record
|
| 21 |
+
Check: Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (cryptarithm.py from line 287) accepts any mapping that verify_cryptarithm_solution (line 24) finds valid: 1.0 (line 329); a parseable but invalid mapping gets 0.01 (line 331); no or unparseable answer 0.0 (lines 306 and 316). Only 1.0 counts as correct. Unique solution: not required, since every valid mapping is accepted.
|
| 22 |
+
Dependencies: reasoning-gym at git commit 49b07130b3fcd12f2d064bba7c43869543a0e7e7 (pyproject version 0.1.26.dev0, installed from the commit, not from PyPI 0.1.25); generator only, not packed; its pip requirements resolved inside the pinned image; the full pip freeze goes into the generation record; not packed; Python 3.12; not packed; build-time tools (decision 8 of 2026-10-06): pip as shipped in the pinned base image, setuptools 80.9.0 and wheel 0.45.1 installed before the generator; their versions, with javac/gcc/cmake where present, go into the build_tools field of every generation record; not packed
|
| 23 |
+
- `generated/jugs.py/` (1 file): Reasoning-gym jugs generator (reasoning_gym/algorithmic/jugs.py at commit 49b07130), generated water jug pouring puzzles with a replaying scorer.
|
| 24 |
+
Format: JSONL of generated problems with answers, plus a generation record
|
| 25 |
+
Check: Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (jugs.py from line 297) replays the agent's moves with verify_solution (line 118): 1.0 if the target is reached (line 319), 0.01 for a parseable move list that fails (line 317), 0.0 otherwise (lines 311 and 321). Only 1.0 counts as correct. Unique solution: not required, since any valid move list is accepted.
|
| 26 |
+
Dependencies: reasoning-gym at git commit 49b07130b3fcd12f2d064bba7c43869543a0e7e7 (pyproject version 0.1.26.dev0, installed from the commit, not from PyPI 0.1.25); generator only, not packed; its pip requirements resolved inside the pinned image; the full pip freeze goes into the generation record; not packed; Python 3.12; not packed; build-time tools (decision 8 of 2026-10-06): pip as shipped in the pinned base image, setuptools 80.9.0 and wheel 0.45.1 installed before the generator; their versions, with javac/gcc/cmake where present, go into the build_tools field of every generation record; not packed
|
campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-logic/LICENSE
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Licence: Apache-2.0 AND MIT AND Apache-2.0
|
| 2 |
+
|
| 3 |
+
- Apache-2.0 AND MIT: https://github.com/open-thought/reasoning-gym/blob/49b07130b3fcd12f2d064bba7c43869543a0e7e7/LICENSE (publisher: open-thought)
|
| 4 |
+
- Apache-2.0: https://github.com/open-thought/reasoning-gym/blob/49b07130b3fcd12f2d064bba7c43869543a0e7e7/LICENSE (publisher: open-thought)
|
campaign_files/2026-10-g-solver-instances-2/open-thought-reasoning-gym-logic/README.md
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# open-thought-reasoning-gym-logic
|
| 2 |
+
|
| 3 |
+
**What this is:** 2 units (6 files) published by open-thought. Collection theme: Seeded puzzle generators (Simon Tatham puzzle collection class, with generation command, seed and solution).
|
| 4 |
+
|
| 5 |
+
**What one unit is:** Each unit is a few named upstream files, delivered together, kept at its upstream path. Format: JSONL of generated problems with answers, plus a generation record.
|
| 6 |
+
|
| 7 |
+
**Where it comes from:** Upstream: open-thought/reasoning-gym. Kind: generated data.
|
| 8 |
+
|
| 9 |
+
**Self-contained:** Each unit is self-contained. 1 of 2 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
+
|
| 11 |
+
**Tools to open and check:** Check: Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (zebra_puzzles.py from line 62) lowercases both sides and removes newlines before an exact comparison (line 76), returning 1.0 (line 77) or 0.0 (line 78). Unique solution: yes, by the has_unique_solution check of the generator. Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (self_reference.py from line 358) compares the agent's answer with str(entry['answer']) exactly, returning 1.0 (line 373) or 0.0 (line 374). Unique solution: the answer is a single integer computed by exhaustive count.
|
| 12 |
+
|
| 13 |
+
**Licence (SPDX):** Apache-2.0 AND MIT AND Apache-2.0
|
| 14 |
+
|
| 15 |
+
**Provenance:** https://github.com/open-thought/reasoning-gym
|
| 16 |
+
|
| 17 |
+
Units:
|
| 18 |
+
|
| 19 |
+
- `generated/` (5 files): Reasoning-gym zebra_puzzles generator (reasoning_gym/logic/zebra_puzzles.py at commit 49b07130), generated Einstein-style logic grid puzzles with answers.
|
| 20 |
+
Format: JSONL of generated problems with answers, plus a generation record
|
| 21 |
+
Check: Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (zebra_puzzles.py from line 62) lowercases both sides and removes newlines before an exact comparison (line 76), returning 1.0 (line 77) or 0.0 (line 78). Unique solution: yes, by the has_unique_solution check of the generator.
|
| 22 |
+
Companions (not included): reasoning_gym/logic/contrib/logic_puzzle/README.md; reasoning_gym/logic/contrib/logic_puzzle/generate.py; LICENSE
|
| 23 |
+
Dependencies: reasoning-gym at git commit 49b07130b3fcd12f2d064bba7c43869543a0e7e7 (pyproject version 0.1.26.dev0, installed from the commit, not from PyPI 0.1.25); generator only, not packed; its pip requirements resolved inside the pinned image; the full pip freeze goes into the generation record; not packed; Python 3.12; not packed; build-time tools (decision 8 of 2026-10-06): pip as shipped in the pinned base image, setuptools 80.9.0 and wheel 0.45.1 installed before the generator; their versions, with javac/gcc/cmake where present, go into the build_tools field of every generation record; not packed
|
| 24 |
+
- `generated/self_reference.py/` (1 file): Reasoning-gym self_reference generator (reasoning_gym/logic/self_reference.py at commit 49b07130), generated self-referential truth puzzles (count the consistent truth assignments) with answers.
|
| 25 |
+
Format: JSONL of generated problems with answers, plus a generation record
|
| 26 |
+
Check: Judged by reasoning_gym's own score_answer for this dataset (decision 5 of 2026-10-06), run in the pinned reasoning-gym image (commit 49b07130, recorded in dependencies) with no network, on the stored entry (question, answer, metadata) and the agent's answer; correct iff score_answer returns 1.0. No hand-written judge. score_answer (self_reference.py from line 358) compares the agent's answer with str(entry['answer']) exactly, returning 1.0 (line 373) or 0.0 (line 374). Unique solution: the answer is a single integer computed by exhaustive count.
|
| 27 |
+
Dependencies: reasoning-gym at git commit 49b07130b3fcd12f2d064bba7c43869543a0e7e7 (pyproject version 0.1.26.dev0, installed from the commit, not from PyPI 0.1.25); generator only, not packed; its pip requirements resolved inside the pinned image; the full pip freeze goes into the generation record; not packed; Python 3.12; not packed; build-time tools (decision 8 of 2026-10-06): pip as shipped in the pinned base image, setuptools 80.9.0 and wheel 0.45.1 installed before the generator; their versions, with javac/gcc/cmake where present, go into the build_tools field of every generation record; not packed
|
campaign_files/2026-10-hw-eda-tests-2/open-cmsis-pack-devtools-tools-svdconv-tests/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: generated_header_strings. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: Task: generate the CMSIS header from PosMaskDim.svd. Judge: every expected define string pair in CheckPosMaskDimFields must appear in the generated header byte for byte; or build SVDConv and run the two tests, passing when no EXPECT fails. Task: generate the header and the peripheral, register and field memory maps from option_n.svd. Judge: each expected address string and regex of the tests must match the generated listing, and the enum defines must equal the expected values; or build SVDConv and run the named tests. Task: generate the header from EnumComboWidth.svd. Judge: the generator must not raise diagnostic M227 and the enum defines must equal the expected values exactly; or build SVDConv and run the named tests.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,24 +19,18 @@ Files:
|
|
| 21 |
- `tools/svdconv/test/data/posMaskDim/PosMaskDim.svd`: SVDConv fixture PosMaskDim: header _Pos and _Msk defines for dim-expanded fields must match the test.
|
| 22 |
Format: CMSIS-SVD XML; C++ GoogleTest source as answer carrier
|
| 23 |
Origin: hand-written SVD fixture of the SVDConv test suite; expected strings set by hand in the test source.
|
| 24 |
-
Reference result: generated_header_strings
|
| 25 |
-
Reference source: tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp lines 333-380 CheckPosMaskDimFields: line 352 `{ "#define TIM_DATA_PIN0_Pos", "(0UL)" }` through line 367 TIM_DATA_PIN7_Msk "(0x80UL)"; lines 186-214 ExpandsRegisterAndFieldDimensionsInMemoryMap
|
| 26 |
Check: Task: generate the CMSIS header from PosMaskDim.svd. Judge: every expected define string pair in CheckPosMaskDimFields must appear in the generated header byte for byte; or build SVDConv and run the two tests, passing when no EXPECT fails.
|
| 27 |
Companions (not included): LICENSE; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp
|
| 28 |
Dependencies: Arm SVDConv from devtools at the pinned commit (Apache-2.0, not packed) for the upstream assertion; GoogleTest 1.12 or later for building the upstream tests (not packed)
|
| 29 |
- `tools/svdconv/test/data/option_n/option_n.svd`: SVDConv fixture option_n: memory map listings and enum header defines for derived peripherals and nested clusters.
|
| 30 |
Format: CMSIS-SVD XML; C++ GoogleTest source as answer carrier
|
| 31 |
Origin: hand-written SVD fixture of the SVDConv test suite; expected strings set by hand in the test source.
|
| 32 |
-
Reference result: generated_header_strings
|
| 33 |
-
Reference source: tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp lines 134-184 GeneratesMemoryMapsAtRequestedDetailLevel (line 160 base addresses 0x40003000, 0x40004000, 0xe000ee08; lines 170-175 register address patterns); tools/svdconv/test/integtests/src/SvdConvEnumHeaderTests.cpp lines 73-80 DCB_DSCSR_Clust_DSCSR_CDS_Disable 0 and Enable 1; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp lines 98-116 CheckOption_n and line 118 CheckSfdGeneration
|
| 34 |
Check: Task: generate the header and the peripheral, register and field memory maps from option_n.svd. Judge: each expected address string and regex of the tests must match the generated listing, and the enum defines must equal the expected values; or build SVDConv and run the named tests.
|
| 35 |
Companions (not included): LICENSE; tools/svdconv/test/integtests/src/SvdConvEnumHeaderTests.cpp; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp
|
| 36 |
Dependencies: Arm SVDConv from devtools at the pinned commit (Apache-2.0, not packed) for the upstream assertion; GoogleTest 1.12 or later for building the upstream tests (not packed)
|
| 37 |
- `tools/svdconv/test/data/enumComboWidth/EnumComboWidth.svd`: SVDConv fixture EnumComboWidth: enumerated values on narrow fields must be emitted without the M227 width warning.
|
| 38 |
Format: CMSIS-SVD XML; C++ GoogleTest source as answer carrier
|
| 39 |
Origin: hand-written SVD fixture of the SVDConv test suite; expected strings set by hand in the test source.
|
| 40 |
-
Reference result: generated_header_strings
|
| 41 |
-
Reference source: tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp lines 382-430 CheckEnumComboWidthLimit (M227 at line 398); tools/svdconv/test/integtests/src/SvdConvEnumHeaderTests.cpp line 42 expects no M227 and lines 64-70 TEST_CTRL_SIX_BIT_ONE=1U, TEST_CTRL_SEVEN_BIT_NINE=9U
|
| 42 |
Check: Task: generate the header from EnumComboWidth.svd. Judge: the generator must not raise diagnostic M227 and the enum defines must equal the expected values exactly; or build SVDConv and run the named tests.
|
| 43 |
Companions (not included): LICENSE; tools/svdconv/test/integtests/src/SvdConvEnumHeaderTests.cpp; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp
|
| 44 |
Dependencies: Arm SVDConv from devtools at the pinned commit (Apache-2.0, not packed) for the upstream assertion; GoogleTest 1.12 or later for building the upstream tests (not packed)
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: Task: generate the CMSIS header from PosMaskDim.svd. Judge: every expected define string pair in CheckPosMaskDimFields must appear in the generated header byte for byte; or build SVDConv and run the two tests, passing when no EXPECT fails. Task: generate the header and the peripheral, register and field memory maps from option_n.svd. Judge: each expected address string and regex of the tests must match the generated listing, and the enum defines must equal the expected values; or build SVDConv and run the named tests. Task: generate the header from EnumComboWidth.svd. Judge: the generator must not raise diagnostic M227 and the enum defines must equal the expected values exactly; or build SVDConv and run the named tests.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `tools/svdconv/test/data/posMaskDim/PosMaskDim.svd`: SVDConv fixture PosMaskDim: header _Pos and _Msk defines for dim-expanded fields must match the test.
|
| 20 |
Format: CMSIS-SVD XML; C++ GoogleTest source as answer carrier
|
| 21 |
Origin: hand-written SVD fixture of the SVDConv test suite; expected strings set by hand in the test source.
|
|
|
|
|
|
|
| 22 |
Check: Task: generate the CMSIS header from PosMaskDim.svd. Judge: every expected define string pair in CheckPosMaskDimFields must appear in the generated header byte for byte; or build SVDConv and run the two tests, passing when no EXPECT fails.
|
| 23 |
Companions (not included): LICENSE; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp
|
| 24 |
Dependencies: Arm SVDConv from devtools at the pinned commit (Apache-2.0, not packed) for the upstream assertion; GoogleTest 1.12 or later for building the upstream tests (not packed)
|
| 25 |
- `tools/svdconv/test/data/option_n/option_n.svd`: SVDConv fixture option_n: memory map listings and enum header defines for derived peripherals and nested clusters.
|
| 26 |
Format: CMSIS-SVD XML; C++ GoogleTest source as answer carrier
|
| 27 |
Origin: hand-written SVD fixture of the SVDConv test suite; expected strings set by hand in the test source.
|
|
|
|
|
|
|
| 28 |
Check: Task: generate the header and the peripheral, register and field memory maps from option_n.svd. Judge: each expected address string and regex of the tests must match the generated listing, and the enum defines must equal the expected values; or build SVDConv and run the named tests.
|
| 29 |
Companions (not included): LICENSE; tools/svdconv/test/integtests/src/SvdConvEnumHeaderTests.cpp; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp
|
| 30 |
Dependencies: Arm SVDConv from devtools at the pinned commit (Apache-2.0, not packed) for the upstream assertion; GoogleTest 1.12 or later for building the upstream tests (not packed)
|
| 31 |
- `tools/svdconv/test/data/enumComboWidth/EnumComboWidth.svd`: SVDConv fixture EnumComboWidth: enumerated values on narrow fields must be emitted without the M227 width warning.
|
| 32 |
Format: CMSIS-SVD XML; C++ GoogleTest source as answer carrier
|
| 33 |
Origin: hand-written SVD fixture of the SVDConv test suite; expected strings set by hand in the test source.
|
|
|
|
|
|
|
| 34 |
Check: Task: generate the header from EnumComboWidth.svd. Judge: the generator must not raise diagnostic M227 and the enum defines must equal the expected values exactly; or build SVDConv and run the named tests.
|
| 35 |
Companions (not included): LICENSE; tools/svdconv/test/integtests/src/SvdConvEnumHeaderTests.cpp; tools/svdconv/test/integtests/src/SvdConvIntegTests.cpp
|
| 36 |
Dependencies: Arm SVDConv from devtools at the pinned commit (Apache-2.0, not packed) for the upstream assertion; GoogleTest 1.12 or later for building the upstream tests (not packed)
|
campaign_files/2026-10-hw-eda-tests-2/open-logic/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 4 of 4 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: self_checking_testbench_pass_fail. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (`All passed`, exit code 0); any check_axi_stream or check_equal error fails it. Comparison is exact bit equality, no tolerance. Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance; OSVVM random values come from the seed set by InitSeed on the instance name, so a run is repeatable. Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance. Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test passed (exit code 0). Comparison is exact bit equality, no tolerance.
|
| 14 |
|
| 15 |
**Licence (SPDX):** LicenseRef-PSI-HDL-Library-License-1.0
|
|
@@ -21,32 +19,24 @@ Files:
|
|
| 21 |
- `test/base/olo_base_crc/olo_base_crc_tb.vhd`: Open-logic olo_base_crc: configurable CRC generator over AXI-Stream with a VUnit testbench of hard-coded expected CRCs.
|
| 22 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 23 |
Origin: self-checking VUnit testbench written by the library author for his own CRC core (tool_fixture): the expected CRC values are hard-coded per configuration
|
| 24 |
-
Reference result: self_checking_testbench_pass_fail
|
| 25 |
-
Reference source: the testbench holds the expected CRC for each test message per (CrcWidth_g, DataWidth_g) pair in its function getResponse and checks every output beat with check_axi_stream (data and last flag) or check_equal; directed tests Reset, Identity, single beat, multi beat, multi packet with first and last flags, backpressure and input delay
|
| 26 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (`All passed`, exit code 0); any check_axi_stream or check_equal error fails it. Comparison is exact bit equality, no tolerance.
|
| 27 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_crc.vhd; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd
|
| 28 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
| 29 |
- `test/base/olo_base_dyn_sft/olo_base_dyn_sft_tb.vhd`: Open-logic olo_base_dyn_sft: pipelined dynamic barrel shifter with a VUnit testbench checking against an integer model.
|
| 30 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 31 |
Origin: self-checking VUnit testbench written by the library author for his own shifter (tool_fixture): an integer reference model in the testbench gives the expected output
|
| 32 |
-
Reference result: self_checking_testbench_pass_fail
|
| 33 |
-
Reference source: the testbench procedure testShift computes the expected output with an integer model (left shift: value*2**shift mod 2**16; right shift with sign extension: floor division; right shift without: unsigned division) and checks the DUT output with check_axi_stream; test AllShifts covers every shift 0 to MaxShift_g with random values, test Random_vData sends 101 random values
|
| 34 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance; OSVVM random values come from the seed set by InitSeed on the instance name, so a run is repeatable.
|
| 35 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_dyn_sft.vhd; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd
|
| 36 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
| 37 |
- `test/base/olo_base_prbs/olo_base_prbs4_tb.vhd`: Open-logic olo_base_prbs (PRBS4): pseudo-random bit sequence generator with a VUnit testbench of the hard-coded sequence.
|
| 38 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 39 |
Origin: self-checking VUnit testbench written by the library author for his own PRBS generator (tool_fixture): the expected sequence and states are hard-coded
|
| 40 |
-
Reference result: self_checking_testbench_pass_fail
|
| 41 |
-
Reference source: the testbench holds the 15-bit PRBS4 sequence `010110010001111` and the table of the 15 LFSR states; test CheckSequence checks every output symbol against the sequence, test SetState loads each state and checks the continuation, both with check_axi_stream
|
| 42 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance.
|
| 43 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd; src/base/vhdl/olo_base_prbs.vhd
|
| 44 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
| 45 |
- `test/base/olo_base_arb_rr/olo_base_arb_rr_tb.vhd`: Open-logic olo_base_arb_rr: round-robin arbiter with a VUnit testbench of hard-coded expected grants.
|
| 46 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 47 |
Origin: self-checking VUnit testbench written by the library author for his own arbiter (tool_fixture): the expected grants are hard-coded
|
| 48 |
-
Reference result: self_checking_testbench_pass_fail
|
| 49 |
-
Reference source: the testbench drives request patterns and checks the expected grant vector after each step with 71 hard-coded check_equal calls; tests SingleBit, MultiBit, ReadyLow and ExampleFromDoc (the waveform of the documentation)
|
| 50 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test passed (exit code 0). Comparison is exact bit equality, no tolerance.
|
| 51 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_arb_prio.vhd; src/base/vhdl/olo_base_arb_rr.vhd; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd
|
| 52 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 4 of 4 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (`All passed`, exit code 0); any check_axi_stream or check_equal error fails it. Comparison is exact bit equality, no tolerance. Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance; OSVVM random values come from the seed set by InitSeed on the instance name, so a run is repeatable. Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance. Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test passed (exit code 0). Comparison is exact bit equality, no tolerance.
|
| 12 |
|
| 13 |
**Licence (SPDX):** LicenseRef-PSI-HDL-Library-License-1.0
|
|
|
|
| 19 |
- `test/base/olo_base_crc/olo_base_crc_tb.vhd`: Open-logic olo_base_crc: configurable CRC generator over AXI-Stream with a VUnit testbench of hard-coded expected CRCs.
|
| 20 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 21 |
Origin: self-checking VUnit testbench written by the library author for his own CRC core (tool_fixture): the expected CRC values are hard-coded per configuration
|
|
|
|
|
|
|
| 22 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (`All passed`, exit code 0); any check_axi_stream or check_equal error fails it. Comparison is exact bit equality, no tolerance.
|
| 23 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_crc.vhd; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd
|
| 24 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
| 25 |
- `test/base/olo_base_dyn_sft/olo_base_dyn_sft_tb.vhd`: Open-logic olo_base_dyn_sft: pipelined dynamic barrel shifter with a VUnit testbench checking against an integer model.
|
| 26 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 27 |
Origin: self-checking VUnit testbench written by the library author for his own shifter (tool_fixture): an integer reference model in the testbench gives the expected output
|
|
|
|
|
|
|
| 28 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance; OSVVM random values come from the seed set by InitSeed on the instance name, so a run is repeatable.
|
| 29 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_dyn_sft.vhd; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd
|
| 30 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
| 31 |
- `test/base/olo_base_prbs/olo_base_prbs4_tb.vhd`: Open-logic olo_base_prbs (PRBS4): pseudo-random bit sequence generator with a VUnit testbench of the hard-coded sequence.
|
| 32 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 33 |
Origin: self-checking VUnit testbench written by the library author for his own PRBS generator (tool_fixture): the expected sequence and states are hard-coded
|
|
|
|
|
|
|
| 34 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test of every configuration passed (exit code 0). Comparison is exact bit equality, no tolerance.
|
| 35 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd; src/base/vhdl/olo_base_prbs.vhd
|
| 36 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
| 37 |
- `test/base/olo_base_arb_rr/olo_base_arb_rr_tb.vhd`: Open-logic olo_base_arb_rr: round-robin arbiter with a VUnit testbench of hard-coded expected grants.
|
| 38 |
Format: VHDL-2008 DUT with a VUnit (OSVVM-based) self-checking testbench and a Python run script
|
| 39 |
Origin: self-checking VUnit testbench written by the library author for his own arbiter (tool_fixture): the expected grants are hard-coded
|
|
|
|
|
|
|
| 40 |
Check: Build in a sandbox with the versions listed in dependencies. Upstream runs `python3 sim/run.py --ghdl` from sim/, which adds every file under src/ to library olo and the testbenches to olo_tb; The unit passes when VUnit reports every test passed (exit code 0). Comparison is exact bit equality, no tolerance.
|
| 41 |
Companions (not included): LGPL2_1.txt; License.txt; sim/run.py; sim/test_configs/olo_base.py; src/base/vhdl/olo_base_arb_prio.vhd; src/base/vhdl/olo_base_arb_rr.vhd; src/base/vhdl/olo_base_pkg_array.vhd; src/base/vhdl/olo_base_pkg_logic.vhd; src/base/vhdl/olo_base_pkg_math.vhd; src/base/vhdl/olo_base_pkg_string.vhd
|
| 42 |
Dependencies: VUnit 5.0.0.dev7 (vunit_hdl==5.0.0.dev7 pinned in .github/workflows/requirements.txt line 1 at the unit commit; brings the OSVVM library and the VUnit verification components used by the testbench); open source, not packed; GHDL v6.0.0 (sim/run.py line 86 makes GHDL the default simulator; version not pinned upstream; latest release on or before the unit commit date 2026-10-01, release v6.0.0 dated 2026-03-07 on https://github.com/ghdl/ghdl/releases/tag/v6.0.0); open source, not packed; Python 3
|
campaign_files/2026-10-m-issue-trackers-2/openafs-openafs-info/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** The file is self-contained.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: Facts computed from the thread headers. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
|
| 14 |
|
| 15 |
**Licence (SPDX):** NOASSERTION
|
|
@@ -21,7 +19,5 @@ Files:
|
|
| 21 |
- `m141-16/2019-March.txt.gz`: [OpenAFS] new RW root.cell (10 messages, 2019-03-06 to 2019-03-08).
|
| 22 |
Format: Pipermail month archive, mbox text gzip-compressed: full headers (From with obfuscated address and name, Date, Subject, Message-ID, In-Reply-To, References) and plain-text bodies. Cut the thread by root Message-ID through the In-Reply-To closure (last References entry when In-Reply-To is missing).
|
| 23 |
Origin: Real list traffic in a closed past month of the list's Pipermail archive. The unit is one thread, cut out of the month file by its root Message-ID; the month file is the pinned pointer and is delivered whole.
|
| 24 |
-
Reference result: facts computed from the thread headers
|
| 25 |
-
Reference source: Message-ID, In-Reply-To, References, From and Date headers of the messages in the thread, read from the pinned month file.
|
| 26 |
Check: Exact match against values computed from the pinned month file: parse it as mbox, drop repeated copies of a Message-ID already seen (first copy kept), take the root Message-ID, collect every message whose In-Reply-To (or last References entry when In-Reply-To is missing) chains back to it, then count, take dates in UTC and compare sender names. Message-IDs are compared after removing angle brackets and double quotes. People are counted by person: From entries with the same address, or the same name after removing accents and letter case, are one person (the names in scale are the first spelling of each). The first-reply interval is the exact difference in whole seconds between the Date headers of the root and the earliest reply, converted to UTC; Date headers have one-second resolution, so no rounding is involved. Deterministic. The program was checked on 2 threads of this source by a second, independent parser (splits the raw text on mbox From lines and reads headers with regular expressions, no mailbox or email module): the chosen thread (also read message by message) and '[OpenAFS] About `dafileserver` vs `fileserver` differences (for small cells)' (11 messages, root <CA+Tk8fyFniNdMxULx6f=_Khm9mGJUUb9xjACho=FMyS4vN6G_w@mail.gmail.com>); message count, people, last sender and time, first-reply seconds and direct replies agreed on both.
|
| 27 |
Tools: Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** The file is self-contained.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
|
| 12 |
|
| 13 |
**Licence (SPDX):** NOASSERTION
|
|
|
|
| 19 |
- `m141-16/2019-March.txt.gz`: [OpenAFS] new RW root.cell (10 messages, 2019-03-06 to 2019-03-08).
|
| 20 |
Format: Pipermail month archive, mbox text gzip-compressed: full headers (From with obfuscated address and name, Date, Subject, Message-ID, In-Reply-To, References) and plain-text bodies. Cut the thread by root Message-ID through the In-Reply-To closure (last References entry when In-Reply-To is missing).
|
| 21 |
Origin: Real list traffic in a closed past month of the list's Pipermail archive. The unit is one thread, cut out of the month file by its root Message-ID; the month file is the pinned pointer and is delivered whole.
|
|
|
|
|
|
|
| 22 |
Check: Exact match against values computed from the pinned month file: parse it as mbox, drop repeated copies of a Message-ID already seen (first copy kept), take the root Message-ID, collect every message whose In-Reply-To (or last References entry when In-Reply-To is missing) chains back to it, then count, take dates in UTC and compare sender names. Message-IDs are compared after removing angle brackets and double quotes. People are counted by person: From entries with the same address, or the same name after removing accents and letter case, are one person (the names in scale are the first spelling of each). The first-reply interval is the exact difference in whole seconds between the Date headers of the root and the earliest reply, converted to UTC; Date headers have one-second resolution, so no rounding is involved. Deterministic. The program was checked on 2 threads of this source by a second, independent parser (splits the raw text on mbox From lines and reads headers with regular expressions, no mailbox or email module): the chosen thread (also read message by message) and '[OpenAFS] About `dafileserver` vs `fileserver` differences (for small cells)' (11 messages, root <CA+Tk8fyFniNdMxULx6f=_Khm9mGJUUb9xjACho=FMyS4vN6G_w@mail.gmail.com>); message count, people, last sender and time, first-reply seconds and direct replies agreed on both.
|
| 23 |
Tools: Read the archive as mbox with Python's mailbox and email modules (or mutt -f / notmuch on the decompressed file); rebuild the thread from Message-ID, In-Reply-To and References headers.
|
campaign_files/2026-10-m-issue-trackers-2/opendaylight-gerrit-changes/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** The file is self-contained.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: Facts computed from the change's own review timeline (status, submit time and submitter, patch sets, votes, reviewers added or removed, review comments). Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** curl the Gerrit REST API (GET https://git.opendaylight.org/gerrit/changes/{number}/detail), strip the first line ")]}'" and read the JSON with jq or Python json; the git-review CLI and Gerrit's ssh 'gerrit query --format=JSON --all-approvals --comments' reach the same data where ssh access is open
|
| 14 |
|
| 15 |
**Licence (SPDX):** NOASSERTION
|
|
@@ -21,7 +19,5 @@ Files:
|
|
| 21 |
- `changes/115743/detail.json`: Transportpce change 115743: Support for OpenConfig renderer.
|
| 22 |
Format: Gerrit REST API ChangeInfo JSON from the /detail endpoint, preceded by the anti-XSSI line ")]}'": status, created, updated, submitted, submitter, owner, labels with every vote (DETAILED_LABELS), messages (author, real_author, date, tag, _revision_number, message text) and reviewer_updates (updated, updated_by, reviewer, state REVIEWER/CC/REMOVED). Patch-set uploads are messages tagged autogenerated:gerrit:newPatchSet; votes appear in message text as 'Patch Set N: Label+V' and in labels.*.all. The endpoint returns all messages and reviewer updates of the change, so the record is complete.
|
| 23 |
Origin: Live Gerrit REST interface of a closed (merged) code review change, collected as a NOASSERTION pointer row under spec general rule 4: the detail JSON was fetched twice apart and was byte-identical. Unit file kept only outside the repository for checking and then deleted.
|
| 24 |
-
Reference result: facts computed from the change's own review timeline (status, submit time and submitter, patch sets, votes, reviewers added or removed, review comments)
|
| 25 |
-
Reference source: The pinned response: changes/115743/detail.json.
|
| 26 |
Check: Exact match against values read from the pinned JSON: status, submitted and submitter; reviewer_updates entries (reviewer, state, updated_by, updated); messages tagged autogenerated:gerrit:newPatchSet (patch-set count, _revision_number); votes from labels.*.all (value, _account_id) and from message first lines 'Patch Set N: Label+V'. A human review comment is a message whose author is not a bot or service account, whose tag does not start with 'autogenerated:', and whose text has content beyond the 'Patch Set N: votes' first line or names inline comments '(N comments)'. Bot and service accounts excluded from the comment count: jenkins-releng, odl.gha.bot, odl.required.gha, accounts tagged SERVICE_USER, and any account whose name, username or email matches the pattern (bot|zuul|jenkins|ci|luci|gerrit code review|service user|automation|buildbot|lint|sonar|tryjob|cq|autosubmit|copybara|tricium|gerritforge|verifier|checker|scheduler). Deterministic, no model judgement. Before answers are issued, the extraction program must pass on 2 hand-checked units of this source (spec general rule 5); hand-checking is pending for the delivery step.
|
| 27 |
Tools: curl the Gerrit REST API (GET https://git.opendaylight.org/gerrit/changes/{number}/detail), strip the first line ")]}'" and read the JSON with jq or Python json; the git-review CLI and Gerrit's ssh 'gerrit query --format=JSON --all-approvals --comments' reach the same data where ssh access is open
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** The file is self-contained.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** curl the Gerrit REST API (GET https://git.opendaylight.org/gerrit/changes/{number}/detail), strip the first line ")]}'" and read the JSON with jq or Python json; the git-review CLI and Gerrit's ssh 'gerrit query --format=JSON --all-approvals --comments' reach the same data where ssh access is open
|
| 12 |
|
| 13 |
**Licence (SPDX):** NOASSERTION
|
|
|
|
| 19 |
- `changes/115743/detail.json`: Transportpce change 115743: Support for OpenConfig renderer.
|
| 20 |
Format: Gerrit REST API ChangeInfo JSON from the /detail endpoint, preceded by the anti-XSSI line ")]}'": status, created, updated, submitted, submitter, owner, labels with every vote (DETAILED_LABELS), messages (author, real_author, date, tag, _revision_number, message text) and reviewer_updates (updated, updated_by, reviewer, state REVIEWER/CC/REMOVED). Patch-set uploads are messages tagged autogenerated:gerrit:newPatchSet; votes appear in message text as 'Patch Set N: Label+V' and in labels.*.all. The endpoint returns all messages and reviewer updates of the change, so the record is complete.
|
| 21 |
Origin: Live Gerrit REST interface of a closed (merged) code review change, collected as a NOASSERTION pointer row under spec general rule 4: the detail JSON was fetched twice apart and was byte-identical. Unit file kept only outside the repository for checking and then deleted.
|
|
|
|
|
|
|
| 22 |
Check: Exact match against values read from the pinned JSON: status, submitted and submitter; reviewer_updates entries (reviewer, state, updated_by, updated); messages tagged autogenerated:gerrit:newPatchSet (patch-set count, _revision_number); votes from labels.*.all (value, _account_id) and from message first lines 'Patch Set N: Label+V'. A human review comment is a message whose author is not a bot or service account, whose tag does not start with 'autogenerated:', and whose text has content beyond the 'Patch Set N: votes' first line or names inline comments '(N comments)'. Bot and service accounts excluded from the comment count: jenkins-releng, odl.gha.bot, odl.required.gha, accounts tagged SERVICE_USER, and any account whose name, username or email matches the pattern (bot|zuul|jenkins|ci|luci|gerrit code review|service user|automation|buildbot|lint|sonar|tryjob|cq|autosubmit|copybara|tricium|gerritforge|verifier|checker|scheduler). Deterministic, no model judgement. Before answers are issued, the extraction program must pass on 2 hand-checked units of this source (spec general rule 5); hand-checking is pending for the delivery step.
|
| 23 |
Tools: curl the Gerrit REST API (GET https://git.opendaylight.org/gerrit/changes/{number}/detail), strip the first line ")]}'" and read the JSON with jq or Python json; the git-review CLI and Gerrit's ssh 'gerrit query --format=JSON --all-approvals --comments' reach the same data where ssh access is open
|
campaign_files/2026-10-md-media-and-documents/opendetex/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: Expected plain text of a LaTeX document. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** detex in.tex (command line; the test calls the built binary as ../delatex); stdout compared with correct.txt by diff
|
| 14 |
|
| 15 |
**Licence (SPDX):** NCSA
|
|
@@ -21,8 +19,6 @@ Files:
|
|
| 21 |
- `test/in.tex`: OpenDetex test: LaTeX document with included part stripped to plain text.
|
| 22 |
Format: LaTeX input (in.tex plus the included part.tex) and one expected text file
|
| 23 |
Origin: a LaTeX test document (with \input{part}) and its expected plain text written by the OpenDetex maintainers; test.pl is the judge
|
| 24 |
-
Reference result: expected plain text of a LaTeX document
|
| 25 |
-
Reference source: test/correct.txt
|
| 26 |
Check: give the agent in.tex and part.tex and ask for the plain text detex prints with default options; require diff to report no difference (exact, line by line); the \includegraphics of img/overview at in.tex line 52 is not needed for text output
|
| 27 |
Tools: detex in.tex (command line; the test calls the built binary as ../delatex); stdout compared with correct.txt by diff
|
| 28 |
Companions (not included): test/part.tex; test/correct.txt; test.pl; COPYRIGHT
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** detex in.tex (command line; the test calls the built binary as ../delatex); stdout compared with correct.txt by diff
|
| 12 |
|
| 13 |
**Licence (SPDX):** NCSA
|
|
|
|
| 19 |
- `test/in.tex`: OpenDetex test: LaTeX document with included part stripped to plain text.
|
| 20 |
Format: LaTeX input (in.tex plus the included part.tex) and one expected text file
|
| 21 |
Origin: a LaTeX test document (with \input{part}) and its expected plain text written by the OpenDetex maintainers; test.pl is the judge
|
|
|
|
|
|
|
| 22 |
Check: give the agent in.tex and part.tex and ask for the plain text detex prints with default options; require diff to report no difference (exact, line by line); the \includegraphics of img/overview at in.tex line 52 is not needed for text output
|
| 23 |
Tools: detex in.tex (command line; the test calls the built binary as ../delatex); stdout compared with correct.txt by diff
|
| 24 |
Companions (not included): test/part.tex; test/correct.txt; test.pl; COPYRIGHT
|
campaign_files/2026-10-md-media-and-documents/openimageio-testsuite/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 4 of 4 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: Reference output: a over b, 256 by 256 RGBA half-float OpenEXR; reference conversion: the YCCK JPEG converted to an RGB TIFF (52 by 52); reference conversion: an 8-bit CMYK Photoshop file converted to an RGB TIFF (200 by 150); two reference outputs: non-finite pixels replaced by 0 (black.exr) and by the mean of the finite 3 by 3 neighbours (box3.exr), 64 by 64 float OpenEXR. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Tools are named in the check written per file in the notes.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,32 +19,24 @@ Files:
|
|
| 21 |
- `testsuite/oiiotool-composite/src/a.exr`: OpenImageIO oiiotool-composite test: a.exr over b.exr with the reference a_over_b.exr.
|
| 22 |
Format: two OpenEXR inputs (256 by 256 RGBA, 51303 and 55412 bytes) plus the reference OpenEXR (91905 bytes)
|
| 23 |
Origin: oiiotool regression test of the over operation; testsuite/oiiotool-composite/run.py line 13: oiiotool ("src/a.exr --over src/b.exr -o a_over_b.exr")
|
| 24 |
-
Reference result: reference output: a over b, 256 by 256 RGBA half-float OpenEXR
|
| 25 |
-
Reference source: testsuite/oiiotool-composite/ref/a_over_b.exr: the reference output listed in run.py line 20 (outputs) and compared by checkref in testsuite/runtest.py lines 394 to 425
|
| 26 |
Check: hide testsuite/oiiotool-composite/ref/a_over_b.exr; the agent composites testsuite/oiiotool-composite/src/a.exr over testsuite/oiiotool-composite/src/b.exr (premultiplied over: out = A + B times (1 minus alpha of A)) and writes an OpenEXR; compare with the reference under the idiff thresholds below. The round computed over itself in python from the two inputs and matched the reference within 1.19e-7
|
| 27 |
Companions (not included): testsuite/oiiotool-composite/src/b.exr; testsuite/oiiotool-composite/ref/a_over_b.exr; testsuite/oiiotool-composite/run.py; testsuite/runtest.py; LICENSE.md
|
| 28 |
Dependencies: an OpenEXR reader and writer (OpenImageIO oiiotool, ffmpeg for reading, OpenEXR library).
|
| 29 |
- `testsuite/jpeg/src/YCbCrK.jpg`: OpenImageIO jpeg test: YCCK JPEG converted to the reference RGB TIFF.
|
| 30 |
Format: JPEG (YCCK, 52 by 52, 2707 bytes) plus the reference TIFF (RGB 8-bit, 52 by 52, 4230 bytes)
|
| 31 |
Origin: OpenImageIO JPEG reader test for a YCCK (Adobe CMYK) JPEG; testsuite/jpeg/run.py line 10: oiiotool ("src/YCbCrK.jpg -o rgb-from-YCbCrK.tif")
|
| 32 |
-
Reference result: reference conversion: the YCCK JPEG converted to an RGB TIFF (52 by 52)
|
| 33 |
-
Reference source: testsuite/jpeg/ref/rgb-from-YCbCrK.tif: the reference output listed in run.py line 12 (outputs) and compared by checkref in testsuite/runtest.py lines 394 to 425
|
| 34 |
Check: hide testsuite/jpeg/ref/rgb-from-YCbCrK.tif; the agent decodes testsuite/jpeg/src/YCbCrK.jpg (4-channel YCCK JPEG) to RGB and writes a 52 by 52 RGB TIFF; compare with the reference under the idiff thresholds below. Not checked independently this round (ffmpeg reads the JPEG as yuva444p and has no CMYK inversion)
|
| 35 |
Companions (not included): testsuite/jpeg/ref/rgb-from-YCbCrK.tif; testsuite/jpeg/run.py; testsuite/runtest.py; LICENSE.md
|
| 36 |
Dependencies: a JPEG decoder that handles Adobe YCCK (OpenImageIO, libjpeg-turbo with CMYK conversion, ImageMagick).
|
| 37 |
- `testsuite/psd-colormodes/src/pattern2-8-cmyk.psd`: OpenImageIO psd-colormodes test: 8-bit CMYK PSD with the reference RGB TIFF.
|
| 38 |
Format: PSD (8-bit CMYK, 200 by 150, 622872 bytes) plus the reference TIFF (RGB 8-bit, 200 by 150, 10967 bytes)
|
| 39 |
Origin: OpenImageIO PSD reader test over colour modes; testsuite/psd-colormodes/run.py lines 21 to 25 convert each src file with rw_command to f+".tif"
|
| 40 |
-
Reference result: reference conversion: an 8-bit CMYK Photoshop file converted to an RGB TIFF (200 by 150)
|
| 41 |
-
Reference source: testsuite/psd-colormodes/ref/pattern2-8-cmyk.psd.tif: the reference written by the rw_command of run.py (testsuite/runtest.py lines 354 to 369) and compared by checkref with anymatch (run.py line 9)
|
| 42 |
Check: hide testsuite/psd-colormodes/ref/pattern2-8-cmyk.psd.tif; the agent reads testsuite/psd-colormodes/src/pattern2-8-cmyk.psd (8-bit CMYK PSD), converts to RGB and writes a 200 by 150 TIFF; compare with the reference under the idiff thresholds below (failthresh raised to 0.008 by this test). The round decoded the PSD with ffmpeg 7.1 and matched the reference within 1/255
|
| 43 |
Companions (not included): testsuite/psd-colormodes/ref/pattern2-8-cmyk.psd.tif; testsuite/psd-colormodes/run.py; testsuite/runtest.py; LICENSE.md
|
| 44 |
Dependencies: a PSD reader with CMYK to RGB conversion (OpenImageIO, ffmpeg, ImageMagick, GIMP).
|
| 45 |
- `testsuite/oiiotool-fixnan/src/bad.exr`: OpenImageIO oiiotool-fixnan test: bad.exr repaired by black and box3 with both references.
|
| 46 |
Format: OpenEXR input (64 by 64 float, 625 bytes) plus two reference OpenEXR files (618 and 611 bytes)
|
| 47 |
Origin: oiiotool regression test of NaN and infinity repair; testsuite/oiiotool-fixnan/run.py line 11: --fixnan black -o black.exr; line 12: --fixnan box3 -o box3.exr
|
| 48 |
-
Reference result: two reference outputs: non-finite pixels replaced by 0 (black.exr) and by the mean of the finite 3 by 3 neighbours (box3.exr), 64 by 64 float OpenEXR
|
| 49 |
-
Reference source: testsuite/oiiotool-fixnan/ref/black.exr and testsuite/oiiotool-fixnan/ref/box3.exr: the reference outputs listed in run.py line 28 and compared by checkref in testsuite/runtest.py lines 394 to 425
|
| 50 |
Check: hide both references; the agent repairs testsuite/oiiotool-fixnan/src/bad.exr (the R plane has an infinity at pixel (4,4) and a NaN at (12,4)) twice: replacing non-finite values with 0, and with the mean of the finite values in the 3 by 3 neighbourhood; compare each result with its reference under the idiff thresholds below. The round recomputed both repairs in python from bad.exr and matched both references exactly (max difference 0)
|
| 51 |
Companions (not included): testsuite/oiiotool-fixnan/ref/black.exr; testsuite/oiiotool-fixnan/ref/box3.exr; testsuite/oiiotool-fixnan/run.py; testsuite/runtest.py; LICENSE.md
|
| 52 |
Dependencies: an OpenEXR reader and writer (OpenImageIO oiiotool, ffmpeg for reading, OpenEXR library).
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 4 of 4 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Tools are named in the check written per file in the notes.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `testsuite/oiiotool-composite/src/a.exr`: OpenImageIO oiiotool-composite test: a.exr over b.exr with the reference a_over_b.exr.
|
| 20 |
Format: two OpenEXR inputs (256 by 256 RGBA, 51303 and 55412 bytes) plus the reference OpenEXR (91905 bytes)
|
| 21 |
Origin: oiiotool regression test of the over operation; testsuite/oiiotool-composite/run.py line 13: oiiotool ("src/a.exr --over src/b.exr -o a_over_b.exr")
|
|
|
|
|
|
|
| 22 |
Check: hide testsuite/oiiotool-composite/ref/a_over_b.exr; the agent composites testsuite/oiiotool-composite/src/a.exr over testsuite/oiiotool-composite/src/b.exr (premultiplied over: out = A + B times (1 minus alpha of A)) and writes an OpenEXR; compare with the reference under the idiff thresholds below. The round computed over itself in python from the two inputs and matched the reference within 1.19e-7
|
| 23 |
Companions (not included): testsuite/oiiotool-composite/src/b.exr; testsuite/oiiotool-composite/ref/a_over_b.exr; testsuite/oiiotool-composite/run.py; testsuite/runtest.py; LICENSE.md
|
| 24 |
Dependencies: an OpenEXR reader and writer (OpenImageIO oiiotool, ffmpeg for reading, OpenEXR library).
|
| 25 |
- `testsuite/jpeg/src/YCbCrK.jpg`: OpenImageIO jpeg test: YCCK JPEG converted to the reference RGB TIFF.
|
| 26 |
Format: JPEG (YCCK, 52 by 52, 2707 bytes) plus the reference TIFF (RGB 8-bit, 52 by 52, 4230 bytes)
|
| 27 |
Origin: OpenImageIO JPEG reader test for a YCCK (Adobe CMYK) JPEG; testsuite/jpeg/run.py line 10: oiiotool ("src/YCbCrK.jpg -o rgb-from-YCbCrK.tif")
|
|
|
|
|
|
|
| 28 |
Check: hide testsuite/jpeg/ref/rgb-from-YCbCrK.tif; the agent decodes testsuite/jpeg/src/YCbCrK.jpg (4-channel YCCK JPEG) to RGB and writes a 52 by 52 RGB TIFF; compare with the reference under the idiff thresholds below. Not checked independently this round (ffmpeg reads the JPEG as yuva444p and has no CMYK inversion)
|
| 29 |
Companions (not included): testsuite/jpeg/ref/rgb-from-YCbCrK.tif; testsuite/jpeg/run.py; testsuite/runtest.py; LICENSE.md
|
| 30 |
Dependencies: a JPEG decoder that handles Adobe YCCK (OpenImageIO, libjpeg-turbo with CMYK conversion, ImageMagick).
|
| 31 |
- `testsuite/psd-colormodes/src/pattern2-8-cmyk.psd`: OpenImageIO psd-colormodes test: 8-bit CMYK PSD with the reference RGB TIFF.
|
| 32 |
Format: PSD (8-bit CMYK, 200 by 150, 622872 bytes) plus the reference TIFF (RGB 8-bit, 200 by 150, 10967 bytes)
|
| 33 |
Origin: OpenImageIO PSD reader test over colour modes; testsuite/psd-colormodes/run.py lines 21 to 25 convert each src file with rw_command to f+".tif"
|
|
|
|
|
|
|
| 34 |
Check: hide testsuite/psd-colormodes/ref/pattern2-8-cmyk.psd.tif; the agent reads testsuite/psd-colormodes/src/pattern2-8-cmyk.psd (8-bit CMYK PSD), converts to RGB and writes a 200 by 150 TIFF; compare with the reference under the idiff thresholds below (failthresh raised to 0.008 by this test). The round decoded the PSD with ffmpeg 7.1 and matched the reference within 1/255
|
| 35 |
Companions (not included): testsuite/psd-colormodes/ref/pattern2-8-cmyk.psd.tif; testsuite/psd-colormodes/run.py; testsuite/runtest.py; LICENSE.md
|
| 36 |
Dependencies: a PSD reader with CMYK to RGB conversion (OpenImageIO, ffmpeg, ImageMagick, GIMP).
|
| 37 |
- `testsuite/oiiotool-fixnan/src/bad.exr`: OpenImageIO oiiotool-fixnan test: bad.exr repaired by black and box3 with both references.
|
| 38 |
Format: OpenEXR input (64 by 64 float, 625 bytes) plus two reference OpenEXR files (618 and 611 bytes)
|
| 39 |
Origin: oiiotool regression test of NaN and infinity repair; testsuite/oiiotool-fixnan/run.py line 11: --fixnan black -o black.exr; line 12: --fixnan box3 -o box3.exr
|
|
|
|
|
|
|
| 40 |
Check: hide both references; the agent repairs testsuite/oiiotool-fixnan/src/bad.exr (the R plane has an infinity at pixel (4,4) and a NaN at (12,4)) twice: replacing non-finite values with 0, and with the mean of the finite values in the 3 by 3 neighbourhood; compare each result with its reference under the idiff thresholds below. The round recomputed both repairs in python from bad.exr and matched both references exactly (max difference 0)
|
| 41 |
Companions (not included): testsuite/oiiotool-fixnan/ref/black.exr; testsuite/oiiotool-fixnan/ref/box3.exr; testsuite/oiiotool-fixnan/run.py; testsuite/runtest.py; LICENSE.md
|
| 42 |
Dependencies: an OpenEXR reader and writer (OpenImageIO oiiotool, ffmpeg for reading, OpenEXR library).
|
campaign_files/2026-10-ml-serving-configs-2/ollama-chat-template-expected-prompts/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether each unit needs other upstream files is not stated. 20 of 20 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: expected_model_outputs. Where the reference result comes from and how a solution is checked is written per unit in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 14 |
|
| 15 |
**Licence (SPDX):** MIT
|
|
@@ -21,160 +19,120 @@ Units:
|
|
| 21 |
- `template/testdata/alfred.gotmpl/` (3 files): Ollama chat template alfred.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 22 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 23 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 24 |
-
Reference result: expected_model_outputs
|
| 25 |
-
Reference source: the files under template/testdata/alfred.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 26 |
Check: Parse template/alfred.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/alfred.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/alfred.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 27 |
Companions (not included): template/alfred.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 28 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 29 |
- `template/testdata/alpaca.gotmpl/` (3 files): Ollama chat template alpaca.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 30 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 31 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 32 |
-
Reference result: expected_model_outputs
|
| 33 |
-
Reference source: the files under template/testdata/alpaca.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 34 |
Check: Parse template/alpaca.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/alpaca.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/alpaca.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 35 |
Companions (not included): template/alpaca.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 36 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 37 |
- `template/testdata/chatml.gotmpl/` (3 files): Ollama chat template chatml.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 38 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 39 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 40 |
-
Reference result: expected_model_outputs
|
| 41 |
-
Reference source: the files under template/testdata/chatml.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 42 |
Check: Parse template/chatml.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/chatml.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/chatml.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 43 |
Companions (not included): template/chatml.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 44 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 45 |
- `template/testdata/chatqa.gotmpl/` (3 files): Ollama chat template chatqa.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 46 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 47 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 48 |
-
Reference result: expected_model_outputs
|
| 49 |
-
Reference source: the files under template/testdata/chatqa.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 50 |
Check: Parse template/chatqa.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/chatqa.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/chatqa.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 51 |
Companions (not included): template/chatqa.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 52 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 53 |
- `template/testdata/codellama-70b-instruct.gotmpl/` (3 files): Ollama chat template codellama-70b-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 54 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 55 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 56 |
-
Reference result: expected_model_outputs
|
| 57 |
-
Reference source: the files under template/testdata/codellama-70b-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 58 |
Check: Parse template/codellama-70b-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/codellama-70b-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/codellama-70b-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 59 |
Companions (not included): template/codellama-70b-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 60 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 61 |
- `template/testdata/command-r.gotmpl/` (3 files): Ollama chat template command-r.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 62 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 63 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 64 |
-
Reference result: expected_model_outputs
|
| 65 |
-
Reference source: the files under template/testdata/command-r.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 66 |
Check: Parse template/command-r.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/command-r.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/command-r.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 67 |
Companions (not included): template/command-r.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 68 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 69 |
- `template/testdata/falcon-instruct.gotmpl/` (3 files): Ollama chat template falcon-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 70 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 71 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 72 |
-
Reference result: expected_model_outputs
|
| 73 |
-
Reference source: the files under template/testdata/falcon-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 74 |
Check: Parse template/falcon-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/falcon-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/falcon-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 75 |
Companions (not included): template/falcon-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 76 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 77 |
- `template/testdata/gemma-instruct.gotmpl/` (3 files): Ollama chat template gemma-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 78 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 79 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 80 |
-
Reference result: expected_model_outputs
|
| 81 |
-
Reference source: the files under template/testdata/gemma-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 82 |
Check: Parse template/gemma-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/gemma-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/gemma-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 83 |
Companions (not included): template/gemma-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 84 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 85 |
- `template/testdata/gemma3-instruct.gotmpl/` (3 files): Ollama chat template gemma3-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 86 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 87 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 88 |
-
Reference result: expected_model_outputs
|
| 89 |
-
Reference source: the files under template/testdata/gemma3-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 90 |
Check: Parse template/gemma3-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/gemma3-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/gemma3-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 91 |
Companions (not included): template/gemma3-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 92 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 93 |
- `template/testdata/granite-instruct.gotmpl/` (3 files): Ollama chat template granite-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 94 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 95 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 96 |
-
Reference result: expected_model_outputs
|
| 97 |
-
Reference source: the files under template/testdata/granite-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 98 |
Check: Parse template/granite-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/granite-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/granite-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 99 |
Companions (not included): template/granite-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 100 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 101 |
- `template/testdata/llama2-chat.gotmpl/` (3 files): Ollama chat template llama2-chat.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 102 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 103 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 104 |
-
Reference result: expected_model_outputs
|
| 105 |
-
Reference source: the files under template/testdata/llama2-chat.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 106 |
Check: Parse template/llama2-chat.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/llama2-chat.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/llama2-chat.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 107 |
Companions (not included): template/llama2-chat.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 108 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 109 |
- `template/testdata/llama3-instruct.gotmpl/` (3 files): Ollama chat template llama3-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 110 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 111 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 112 |
-
Reference result: expected_model_outputs
|
| 113 |
-
Reference source: the files under template/testdata/llama3-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 114 |
Check: Parse template/llama3-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/llama3-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/llama3-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 115 |
Companions (not included): template/llama3-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 116 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 117 |
- `template/testdata/magicoder.gotmpl/` (3 files): Ollama chat template magicoder.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 118 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 119 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 120 |
-
Reference result: expected_model_outputs
|
| 121 |
-
Reference source: the files under template/testdata/magicoder.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 122 |
Check: Parse template/magicoder.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/magicoder.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/magicoder.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 123 |
Companions (not included): template/magicoder.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 124 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 125 |
- `template/testdata/mistral-instruct.gotmpl/` (3 files): Ollama chat template mistral-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 126 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 127 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 128 |
-
Reference result: expected_model_outputs
|
| 129 |
-
Reference source: the files under template/testdata/mistral-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 130 |
Check: Parse template/mistral-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/mistral-instruct.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/mistral-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 131 |
Companions (not included): template/mistral-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 132 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 133 |
- `template/testdata/openchat.gotmpl/` (3 files): Ollama chat template openchat.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 134 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 135 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 136 |
-
Reference result: expected_model_outputs
|
| 137 |
-
Reference source: the files under template/testdata/openchat.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 138 |
Check: Parse template/openchat.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/openchat.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/openchat.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 139 |
Companions (not included): template/openchat.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 140 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 141 |
- `template/testdata/phi-3.gotmpl/` (3 files): Ollama chat template phi-3.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 142 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 143 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 144 |
-
Reference result: expected_model_outputs
|
| 145 |
-
Reference source: the files under template/testdata/phi-3.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 146 |
Check: Parse template/phi-3.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/phi-3.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/phi-3.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 147 |
Companions (not included): template/phi-3.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 148 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 149 |
- `template/testdata/solar-instruct.gotmpl/` (3 files): Ollama chat template solar-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 150 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 151 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 152 |
-
Reference result: expected_model_outputs
|
| 153 |
-
Reference source: the files under template/testdata/solar-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 154 |
Check: Parse template/solar-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/solar-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/solar-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 155 |
Companions (not included): template/solar-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 156 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 157 |
- `template/testdata/starcoder2-instruct.gotmpl/` (3 files): Ollama chat template starcoder2-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 158 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 159 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 160 |
-
Reference result: expected_model_outputs
|
| 161 |
-
Reference source: the files under template/testdata/starcoder2-instruct.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 162 |
Check: Parse template/starcoder2-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/starcoder2-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/starcoder2-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 163 |
Companions (not included): template/starcoder2-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 164 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 165 |
- `template/testdata/vicuna.gotmpl/` (3 files): Ollama chat template vicuna.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 166 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 167 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 168 |
-
Reference result: expected_model_outputs
|
| 169 |
-
Reference source: the files under template/testdata/vicuna.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 170 |
Check: Parse template/vicuna.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/vicuna.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/vicuna.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 171 |
Companions (not included): template/vicuna.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 172 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 173 |
- `template/testdata/zephyr.gotmpl/` (3 files): Ollama chat template zephyr.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 174 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 175 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
| 176 |
-
Reference result: expected_model_outputs
|
| 177 |
-
Reference source: the files under template/testdata/zephyr.gotmpl/, each named by the roles of its conversation; the conversations are fixed in TestTemplate, template/template_test.go lines 64 to 87
|
| 178 |
Check: Parse template/zephyr.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/zephyr.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/zephyr.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 179 |
Companions (not included): template/zephyr.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 180 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether each unit needs other upstream files is not stated. 20 of 20 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 12 |
|
| 13 |
**Licence (SPDX):** MIT
|
|
|
|
| 19 |
- `template/testdata/alfred.gotmpl/` (3 files): Ollama chat template alfred.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 20 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 21 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 22 |
Check: Parse template/alfred.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/alfred.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/alfred.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 23 |
Companions (not included): template/alfred.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 24 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 25 |
- `template/testdata/alpaca.gotmpl/` (3 files): Ollama chat template alpaca.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 26 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 27 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 28 |
Check: Parse template/alpaca.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/alpaca.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/alpaca.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 29 |
Companions (not included): template/alpaca.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 30 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 31 |
- `template/testdata/chatml.gotmpl/` (3 files): Ollama chat template chatml.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 32 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 33 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 34 |
Check: Parse template/chatml.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/chatml.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/chatml.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 35 |
Companions (not included): template/chatml.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 36 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 37 |
- `template/testdata/chatqa.gotmpl/` (3 files): Ollama chat template chatqa.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 38 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 39 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 40 |
Check: Parse template/chatqa.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/chatqa.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/chatqa.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 41 |
Companions (not included): template/chatqa.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 42 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 43 |
- `template/testdata/codellama-70b-instruct.gotmpl/` (3 files): Ollama chat template codellama-70b-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 44 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 45 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 46 |
Check: Parse template/codellama-70b-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/codellama-70b-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/codellama-70b-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 47 |
Companions (not included): template/codellama-70b-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 48 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 49 |
- `template/testdata/command-r.gotmpl/` (3 files): Ollama chat template command-r.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 50 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 51 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 52 |
Check: Parse template/command-r.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/command-r.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/command-r.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 53 |
Companions (not included): template/command-r.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 54 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 55 |
- `template/testdata/falcon-instruct.gotmpl/` (3 files): Ollama chat template falcon-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 56 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 57 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 58 |
Check: Parse template/falcon-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/falcon-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/falcon-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 59 |
Companions (not included): template/falcon-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 60 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 61 |
- `template/testdata/gemma-instruct.gotmpl/` (3 files): Ollama chat template gemma-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 62 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 63 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 64 |
Check: Parse template/gemma-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/gemma-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/gemma-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 65 |
Companions (not included): template/gemma-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 66 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 67 |
- `template/testdata/gemma3-instruct.gotmpl/` (3 files): Ollama chat template gemma3-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 68 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 69 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 70 |
Check: Parse template/gemma3-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/gemma3-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/gemma3-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 71 |
Companions (not included): template/gemma3-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 72 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 73 |
- `template/testdata/granite-instruct.gotmpl/` (3 files): Ollama chat template granite-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 74 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 75 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 76 |
Check: Parse template/granite-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/granite-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/granite-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 77 |
Companions (not included): template/granite-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 78 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 79 |
- `template/testdata/llama2-chat.gotmpl/` (3 files): Ollama chat template llama2-chat.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 80 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 81 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 82 |
Check: Parse template/llama2-chat.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/llama2-chat.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/llama2-chat.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 83 |
Companions (not included): template/llama2-chat.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 84 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 85 |
- `template/testdata/llama3-instruct.gotmpl/` (3 files): Ollama chat template llama3-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 86 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 87 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 88 |
Check: Parse template/llama3-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/llama3-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/llama3-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 89 |
Companions (not included): template/llama3-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 90 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 91 |
- `template/testdata/magicoder.gotmpl/` (3 files): Ollama chat template magicoder.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 92 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 93 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 94 |
Check: Parse template/magicoder.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/magicoder.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/magicoder.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 95 |
Companions (not included): template/magicoder.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 96 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 97 |
- `template/testdata/mistral-instruct.gotmpl/` (3 files): Ollama chat template mistral-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 98 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 99 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 100 |
Check: Parse template/mistral-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/mistral-instruct.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/mistral-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 101 |
Companions (not included): template/mistral-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 102 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 103 |
- `template/testdata/openchat.gotmpl/` (3 files): Ollama chat template openchat.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 104 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 105 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 106 |
Check: Parse template/openchat.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/openchat.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/openchat.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 107 |
Companions (not included): template/openchat.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 108 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 109 |
- `template/testdata/phi-3.gotmpl/` (3 files): Ollama chat template phi-3.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 110 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 111 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 112 |
Check: Parse template/phi-3.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/phi-3.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/phi-3.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 113 |
Companions (not included): template/phi-3.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 114 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 115 |
- `template/testdata/solar-instruct.gotmpl/` (3 files): Ollama chat template solar-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 116 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 117 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 118 |
Check: Parse template/solar-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/solar-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/solar-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 119 |
Companions (not included): template/solar-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 120 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 121 |
- `template/testdata/starcoder2-instruct.gotmpl/` (3 files): Ollama chat template starcoder2-instruct.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 122 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 123 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 124 |
Check: Parse template/starcoder2-instruct.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/starcoder2-instruct.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/starcoder2-instruct.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 125 |
Companions (not included): template/starcoder2-instruct.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 126 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 127 |
- `template/testdata/vicuna.gotmpl/` (3 files): Ollama chat template vicuna.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 128 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 129 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 130 |
Check: Parse template/vicuna.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/vicuna.gotmpl/<roles> byte for byte after removing one trailing space (lines 119 to 122). Run: go test ./template -run TestTemplate/vicuna.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 131 |
Companions (not included): template/vicuna.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 132 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
| 133 |
- `template/testdata/zephyr.gotmpl/` (3 files): Ollama chat template zephyr.gotmpl: rendering three fixed conversations must give the published prompt bytes.
|
| 134 |
Format: Go text/template prompt template with one expected rendered prompt per conversation shape
|
| 135 |
Origin: expected prompts committed by the Ollama maintainers as fixtures of the chat template tests
|
|
|
|
|
|
|
| 136 |
Check: Parse template/zephyr.gotmpl with ollama template.Parse and Execute it with Values{Messages: the conversation} for each of the three role sequences of TestTemplate (template_test.go lines 63 to 128); the output must equal template/testdata/zephyr.gotmpl/<roles> byte for byte. Run: go test ./template -run TestTemplate/zephyr.gotmpl at the pinned commit. For a task, give the template and the conversations and hide the expected files; the agent must render the exact prompt string the server would send to the model.
|
| 137 |
Companions (not included): template/zephyr.gotmpl; template/template_test.go; template/template.go; LICENSE
|
| 138 |
Dependencies: Go toolchain of the go.mod at commit c2b7368d4156656ddb9a23b43f721a841b9c23e0 and the ollama module at that commit (template.go imports github.com/ollama/ollama/api and text/template); not bundled
|
campaign_files/2026-10-ml-serving-configs-2/open-mmlab-mmdeploy/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: aggregate_metric_reference. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: For a model and backend pipeline listed in mmpretrain.yml, convert the PyTorch checkpoint with the named deploy config, evaluate the backend model on the ImageNet-1k validation set, and pass when the backend accuracy is at least the metafile value minus 1 percentage point (tolerance 1 at mmpretrain.yml lines 12 and 15; the check backend_value >= metric_value - tolerance at tools/regression_test.py line 439). The check is one-sided as the source writes it. Status stays pending until recomputed.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,8 +19,6 @@ Files:
|
|
| 21 |
- `tests/regression/mmpretrain.yml`: Mmdeploy regression-test task config for mmpretrain: 14 model families converted to ONNX Runtime, TensorRT, OpenVINO, ncnn, PPLNN, TorchScript and other backends, each checked against its mmpretrain metafile accuracy.
|
| 22 |
Format: YAML task config (anchors and aliases) read by tools/regression_test.py
|
| 23 |
Origin: the regression-test task config of a model deployment framework, kept in its tests tree
|
| 24 |
-
Reference result: aggregate_metric_reference
|
| 25 |
-
Reference source: the Top 1 and Top 5 Accuracy of each model entry in the mmpretrain metafile named on the models list (mmpretrain.yml lines 114 to 264), as tools/regression_test.py reads them (metric_info at mmpretrain.yml lines 9 to 15 maps the metafile metric names to the result keys accuracy/top1 and accuracy/top5)
|
| 26 |
Check: For a model and backend pipeline listed in mmpretrain.yml, convert the PyTorch checkpoint with the named deploy config, evaluate the backend model on the ImageNet-1k validation set, and pass when the backend accuracy is at least the metafile value minus 1 percentage point (tolerance 1 at mmpretrain.yml lines 12 and 15; the check backend_value >= metric_value - tolerance at tools/regression_test.py line 439). The check is one-sided as the source writes it. Status stays pending until recomputed
|
| 27 |
Companions (not included): tools/regression_test.py; docs/en/07-developer-guide/regression_test.md; LICENSE
|
| 28 |
Dependencies: checkpoints from the Weights URLs of the mmpretrain metafiles (download.openmmlab.com), published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: For a model and backend pipeline listed in mmpretrain.yml, convert the PyTorch checkpoint with the named deploy config, evaluate the backend model on the ImageNet-1k validation set, and pass when the backend accuracy is at least the metafile value minus 1 percentage point (tolerance 1 at mmpretrain.yml lines 12 and 15; the check backend_value >= metric_value - tolerance at tools/regression_test.py line 439). The check is one-sided as the source writes it. Status stays pending until recomputed.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `tests/regression/mmpretrain.yml`: Mmdeploy regression-test task config for mmpretrain: 14 model families converted to ONNX Runtime, TensorRT, OpenVINO, ncnn, PPLNN, TorchScript and other backends, each checked against its mmpretrain metafile accuracy.
|
| 20 |
Format: YAML task config (anchors and aliases) read by tools/regression_test.py
|
| 21 |
Origin: the regression-test task config of a model deployment framework, kept in its tests tree
|
|
|
|
|
|
|
| 22 |
Check: For a model and backend pipeline listed in mmpretrain.yml, convert the PyTorch checkpoint with the named deploy config, evaluate the backend model on the ImageNet-1k validation set, and pass when the backend accuracy is at least the metafile value minus 1 percentage point (tolerance 1 at mmpretrain.yml lines 12 and 15; the check backend_value >= metric_value - tolerance at tools/regression_test.py line 439). The check is one-sided as the source writes it. Status stays pending until recomputed
|
| 23 |
Companions (not included): tools/regression_test.py; docs/en/07-developer-guide/regression_test.md; LICENSE
|
| 24 |
Dependencies: checkpoints from the Weights URLs of the mmpretrain metafiles (download.openmmlab.com), published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
campaign_files/2026-10-ml-serving-configs-2/open-mmlab-mmpretrain/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether each unit needs other upstream files is not stated. 16 of 16 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: aggregate_metric_reference. Where the reference result comes from and how a solution is checked is written per unit in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,128 +19,96 @@ Units:
|
|
| 21 |
- `configs/mobilenet_v2/` (3 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MobileNet V2 (1 model (mobilenet-v2_8xb32_in1k: Top 1 71.86, Top 5 90.42)).
|
| 22 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 23 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
| 24 |
-
Reference result: aggregate_metric_reference
|
| 25 |
-
Reference source: configs/mobilenet_v2/metafile.yml: Top 1 and Top 5 Accuracy (percent, 2 decimals) per model on ImageNet-1k, measured by OpenMMLab with the listed config and checkpoint; the README results table repeats the same values
|
| 26 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/mobilenet_v2/README.md line 71) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 27 |
Companions (not included): configs/_base_/models/mobilenet_v2_1x.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/schedules/imagenet_bs256_epochstep.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 28 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab (no third-party weights).
|
| 29 |
- `configs/shufflenet_v1/` (3 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for ShuffleNet V1 (1 model (shufflenet-v1-1x_16xb64_in1k: Top 1 68.13, Top 5 87.81)).
|
| 30 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 31 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
| 32 |
-
Reference result: aggregate_metric_reference
|
| 33 |
-
Reference source: configs/shufflenet_v1/metafile.yml: Top 1 and Top 5 Accuracy (percent, 2 decimals) per model on ImageNet-1k, measured by OpenMMLab with the listed config and checkpoint; the README results table repeats the same values
|
| 34 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/shufflenet_v1/README.md line 57) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 35 |
Companions (not included): configs/_base_/models/shufflenet_v1_1x.py; configs/_base_/datasets/imagenet_bs64_pil_resize.py; configs/_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 36 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab (no third-party weights).
|
| 37 |
- `configs/shufflenet_v2/` (3 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for ShuffleNet V2 (1 model (shufflenet-v2-1x_16xb64_in1k: Top 1 69.55, Top 5 88.92)).
|
| 38 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 39 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
| 40 |
-
Reference result: aggregate_metric_reference
|
| 41 |
-
Reference source: configs/shufflenet_v2/metafile.yml: Top 1 and Top 5 Accuracy (percent, 2 decimals) per model on ImageNet-1k, measured by OpenMMLab with the listed config and checkpoint; the README results table repeats the same values
|
| 42 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/shufflenet_v2/README.md line 57) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 43 |
Companions (not included): configs/_base_/models/shufflenet_v2_1x.py; configs/_base_/datasets/imagenet_bs64_pil_resize.py; configs/_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 44 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab (no third-party weights).
|
| 45 |
- `configs/mobilenet_v3/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MobileNet V3 (6 models over 4 configs (small-050 57.91/80.19, small-075 65.23/85.44, small 66.68/86.74, small 3rdparty 67.66/87.41, large 73.49/91.31, large 3rdparty 74.04/91.34; Top 1/Top 5)).
|
| 46 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 47 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
| 48 |
-
Reference result: aggregate_metric_reference
|
| 49 |
-
Reference source: configs/mobilenet_v3/metafile.yml: Top 1 and Top 5 Accuracy (percent, 2 decimals) per model on ImageNet-1k, measured by OpenMMLab with the listed config and checkpoint; the README results table repeats the same values
|
| 50 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/mobilenet_v3/README.md line 69) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 51 |
Companions (not included): configs/_base_/models/mobilenet_v3/mobilenet_v3_large_imagenet.py; configs/_base_/models/mobilenet_v3/mobilenet_v3_small_imagenet.py; configs/_base_/models/mobilenet_v3/mobilenet_v3_small_050_imagenet.py; configs/_base_/models/mobilenet_v3/mobilenet_v3_small_075_imagenet.py; configs/_base_/datasets/imagenet_bs128_mbv3.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 52 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); two models trained by OpenMMLab; four marked 3rdparty, converted from torchvision (configs/mobilenet_v3/README.md line 87; metafile.yml lines 81 and 110 point to download.pytorch.org) or from timm releases (metafile.yml lines 36 and 52).
|
| 53 |
- `configs/repvgg/` (15 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RepVGG A0 to D2se (structural reparameterization: multi-branch training-time blocks fused into a plain 3x3 conv stack) (12 model entries (repvgg-A0_8xb32_in1k: Top 1 72.37, Top 5 90.56; repvgg-A1_8xb32_in1k: Top 1 74.23, Top 5 91.80; repvgg-A2_8xb32_in1k: Top 1 76.49, Top 5 93.09; repvgg-B0_8xb32_in1k: Top 1 75.27, Top 5 92.21; repvgg-B1_8xb32_in1k: Top 1 78.19, Top 5 94.04; repvgg-B1g2_8xb32_in1k: Top 1 77.87, Top 5 93.99; repvgg-B1g4_8xb32_in1k: Top 1 77.81, Top 5 93.77; repvgg-B2_8xb32_in1k: Top 1 78.58, Top 5 94.23; repvgg-B2g4_8xb32_in1k: Top 1 79.44, Top 5 94.72; repvgg-B3_8xb32_in1k: Top 1 80.58, Top 5 95.33; repvgg-B3g4_8xb32_in1k: Top 1 80.26, Top 5 95.15; repvgg-D2se_3rdparty_in1k: Top 1 81.81, Top 5 95.94)).
|
| 54 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 55 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
| 56 |
-
Reference result: aggregate_metric_reference
|
| 57 |
-
Reference source: configs/repvgg/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and training-time checkpoint; the README results table repeats the values; the deploy configs under configs/repvgg/deploy/ (or the *deploy* configs of the directory) give the inference-time model the converted checkpoint must reproduce
|
| 58 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/repvgg/metafile.yml (the command given in configs/repvgg/README.md line 75); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/repvgg/README.md line 89; README line 89 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 59 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/repvgg-A0_in1k.py; configs/_base_/models/repvgg-B3_lbs-mixup_in1k.py; configs/_base_/schedules/imagenet_bs256_coslr.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 60 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 1 of 12 entries (repvgg-D2se_3rdparty_in1k) is weights converted by OpenMMLab from the authors' official release (README note at configs/repvgg/README.md line 130: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 61 |
- `configs/mobileone/` (12 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MobileOne S0 to S4 (reparameterizable mobile backbone) (5 model entries (mobileone-s0_8xb32_in1k: Top 1 71.34, Top 5 89.87; mobileone-s1_8xb32_in1k: Top 1 75.72, Top 5 92.54; mobileone-s2_8xb32_in1k: Top 1 77.37, Top 5 93.34; mobileone-s3_8xb32_in1k: Top 1 78.06, Top 5 93.83; mobileone-s4_8xb32_in1k: Top 1 79.69, Top 5 94.46)).
|
| 62 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 63 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
| 64 |
-
Reference result: aggregate_metric_reference
|
| 65 |
-
Reference source: configs/mobileone/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and training-time checkpoint; the README results table repeats the values; the deploy configs under configs/mobileone/deploy/ (or the *deploy* configs of the directory) give the inference-time model the converted checkpoint must reproduce
|
| 66 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/mobileone/metafile.yml (the command given in configs/mobileone/README.md line 72); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the README of this directory does not name the tool; it calls model.backbone.switch_to_deploy() (tools/model_converters/reparameterize_model.py line 18) and the deploy configs under configs/mobileone/deploy/ set backbone deploy=True) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 67 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/mobileone/mobileone_s0.py; configs/_base_/models/mobileone/mobileone_s1.py; configs/_base_/models/mobileone/mobileone_s2.py; configs/_base_/models/mobileone/mobileone_s3.py; configs/_base_/models/mobileone/mobileone_s4.py; configs/_base_/schedules/imagenet_bs256_coslr_coswd_300e.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 68 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 69 |
- `configs/replknet/` (10 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RepLKNet-31B, 31L and XL (very large reparameterizable kernels) (6 model entries (replknet-31B_3rdparty_in1k: Top 1 83.48, Top 5 96.57; replknet-31B_3rdparty_in1k-384px: Top 1 84.84, Top 5 97.34; replknet-31B_in21k-pre_3rdparty_in1k: Top 1 85.20, Top 5 97.56; replknet-31B_in21k-pre_3rdparty_in1k-384px: Top 1 85.99, Top 5 97.75; replknet-31L_in21k-pre_3rdparty_in1k-384px: Top 1 86.63, Top 5 98.00; replknet-XL_meg73m-pre_3rdparty_in1k-320px: Top 1 87.57, Top 5 98.39)).
|
| 70 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 71 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
| 72 |
-
Reference result: aggregate_metric_reference
|
| 73 |
-
Reference source: configs/replknet/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and training-time checkpoint; the README results table repeats the values; the deploy configs under configs/replknet/deploy/ (or the *deploy* configs of the directory) give the inference-time model the converted checkpoint must reproduce
|
| 74 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/replknet/metafile.yml (the command given in configs/replknet/README.md line 53); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/replknet/README.md line 61; README line 61 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 75 |
Companions (not included): configs/_base_/datasets/imagenet_bs16_pil_bicubic_384.py; configs/_base_/datasets/imagenet_bs32_pil_bicubic.py; configs/_base_/datasets/imagenet_bs8_pil_bicubic_320.py; configs/_base_/default_runtime.py; configs/_base_/models/replknet-31B_in1k.py; configs/_base_/models/replknet-31L_in1k.py; configs/_base_/models/replknet-XL_in1k.py; configs/_base_/schedules/imagenet_bs256_coslr.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 76 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 6 of 6 entries (replknet-31B_3rdparty_in1k, replknet-31B_3rdparty_in1k-384px, replknet-31B_in21k-pre_3rdparty_in1k, replknet-31B_in21k-pre_3rdparty_in1k-384px, replknet-31L_in21k-pre_3rdparty_in1k-384px, replknet-XL_meg73m-pre_3rdparty_in1k-320px) are weights converted by OpenMMLab from the authors' official release (README note at configs/replknet/README.md line 96: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 77 |
- `configs/repmlp/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RepMLPNet-Base (reparameterized MLP blocks fused from conv branches) (2 model entries (repmlp-base_3rdparty_8xb64_in1k: Top 1 80.41, Top 5 95.14; repmlp-base_3rdparty_8xb64_in1k-256px: Top 1 81.11, Top 5 95.50)).
|
| 78 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 79 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
| 80 |
-
Reference result: aggregate_metric_reference
|
| 81 |
-
Reference source: configs/repmlp/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and training-time checkpoint; the README results table repeats the values; the deploy configs under configs/repmlp/deploy/ (or the *deploy* configs of the directory) give the inference-time model the converted checkpoint must reproduce
|
| 82 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/repmlp/metafile.yml (the command given in configs/repmlp/README.md line 53); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/repmlp/README.md line 61; README line 61 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 83 |
Companions (not included): configs/_base_/datasets/imagenet_bs64_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/repmlp-base_224.py; configs/_base_/schedules/imagenet_bs1024_adamw_swin.py; configs/_base_/schedules/imagenet_bs4096_AdamW.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 84 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 2 of 2 entries (repmlp-base_3rdparty_8xb64_in1k, repmlp-base_3rdparty_8xb64_in1k-256px) are weights converted by OpenMMLab from the authors' official release (README note at configs/repmlp/README.md line 92: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 85 |
- `configs/riformer/` (22 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RIFormer S12 to M48 at 224 and 384 px (token mixer removed by reparameterizing an affine transform) (10 model entries (riformer-s12_in1k: Top 1 76.90, Top 5 93.06; riformer-s24_in1k: Top 1 80.28, Top 5 94.80; riformer-s36_in1k: Top 1 81.29, Top 5 95.41; riformer-m36_in1k: Top 1 82.57, Top 5 95.99; riformer-m48_in1k: Top 1 82.75, Top 5 96.11; riformer-s12_in1k-384: Top 1 78.29, Top 5 93.93; riformer-s24_in1k-384: Top 1 81.36, Top 5 95.40; riformer-s36_in1k-384: Top 1 82.22, Top 5 95.95; riformer-m36_in1k-384: Top 1 83.39, Top 5 96.40; riformer-m48_in1k-384: Top 1 83.70, Top 5 96.60)).
|
| 86 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 87 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
| 88 |
-
Reference result: aggregate_metric_reference
|
| 89 |
-
Reference source: configs/riformer/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and training-time checkpoint; the README results table repeats the values; the deploy configs under configs/riformer/deploy/ (or the *deploy* configs of the directory) give the inference-time model the converted checkpoint must reproduce
|
| 90 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/riformer/metafile.yml (the command given in configs/riformer/README.md line 93); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/riformer/README.md lines 99, 121, 133; README line 121 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py; README lines 70-75 show the source's own check that the switched model gives the same logits (torch.allclose(out, out_deploy, rtol=1e-4, atol=1e-5))) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 91 |
Companions (not included): configs/_base_/datasets/imagenet_bs128_poolformer_medium_224.py; configs/_base_/datasets/imagenet_bs128_poolformer_small_224.py; configs/_base_/datasets/imagenet_bs128_riformer_medium_384.py; configs/_base_/datasets/imagenet_bs128_riformer_small_384.py; configs/_base_/default_runtime.py; configs/_base_/schedules/imagenet_bs1024_adamw_swin.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 92 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 93 |
- `configs/levit/` (12 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for LeViT 128S to 384 (hybrid conv-attention models whose deploy configs fuse batch norm into the linear and conv layers) (5 model entries (levit-128s_3rdparty_in1k: Top 1 76.51, Top 5 92.90; levit-128_3rdparty_in1k: Top 1 78.58, Top 5 93.95; levit-192_3rdparty_in1k: Top 1 79.86, Top 5 94.75; levit-256_3rdparty_in1k: Top 1 81.59, Top 5 95.46; levit-384_3rdparty_in1k: Top 1 82.59, Top 5 95.95)).
|
| 94 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 95 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
| 96 |
-
Reference result: aggregate_metric_reference
|
| 97 |
-
Reference source: configs/levit/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and training-time checkpoint; the README results table repeats the values; the deploy configs under configs/levit/deploy/ (or the *deploy* configs of the directory) give the inference-time model the converted checkpoint must reproduce
|
| 98 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/levit/metafile.yml (the command given in configs/levit/README.md line 51); Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 99 |
Companions (not included): configs/_base_/datasets/imagenet_bs64_swin_224.py; configs/_base_/default_runtime.py; configs/_base_/models/levit-256-p16.py; configs/_base_/schedules/imagenet_bs2048_adamw_levit.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 100 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 5 of 5 entries (levit-128s_3rdparty_in1k, levit-128_3rdparty_in1k, levit-192_3rdparty_in1k, levit-256_3rdparty_in1k, levit-384_3rdparty_in1k) are weights converted by OpenMMLab from the authors' official release (README note at configs/levit/README.md line 68: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 101 |
- `configs/mocov2/` (4 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MoCo v2 self-supervised ResNet-50, linear evaluation (1 model entry (resnet50_mocov2-pre_8xb32-linear-steplr-100e_in1k: Top 1 67.5)).
|
| 102 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 103 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
| 104 |
-
Reference result: aggregate_metric_reference
|
| 105 |
-
Reference source: configs/mocov2/metafile.yml: Top 1 Accuracy (percent, 1 decimal) of the linear evaluation entries on ImageNet-1k (a linear classifier trained on frozen self-supervised features), measured by OpenMMLab with the listed benchmark config and linear-head checkpoint; the README results table repeats the values
|
| 106 |
Check: For each linear-evaluation entry of configs/mocov2/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/mocov2/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/mocov2/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 107 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_mocov2.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_sgd_coslr_200e.py; configs/_base_/schedules/imagenet_sgd_steplr_100e.py; tools/test.py; LICENSE
|
| 108 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 109 |
- `configs/simclr/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for SimCLR self-supervised ResNet-50 (200 and 800 epoch pretraining), linear evaluation (2 model entries (resnet50_simclr-200e-pre_8xb512-linear-coslr-90e_in1k: Top 1 66.9; resnet50_simclr-800e-pre_8xb512-linear-coslr-90e_in1k: Top 1 69.2)).
|
| 110 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 111 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
| 112 |
-
Reference result: aggregate_metric_reference
|
| 113 |
-
Reference source: configs/simclr/metafile.yml: Top 1 Accuracy (percent, 1 decimal) of the linear evaluation entries on ImageNet-1k (a linear classifier trained on frozen self-supervised features), measured by OpenMMLab with the listed benchmark config and linear-head checkpoint; the README results table repeats the values
|
| 114 |
Check: For each linear-evaluation entry of configs/simclr/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/simclr/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/simclr/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 115 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/datasets/imagenet_bs32_simclr.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_lars_coslr_200e.py; configs/_base_/schedules/imagenet_lars_coslr_90e.py; tools/test.py; LICENSE
|
| 116 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 117 |
- `configs/simsiam/` (5 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for SimSiam self-supervised ResNet-50 (100 and 200 epoch pretraining), linear evaluation (2 model entries (resnet50_simsiam-100e-pre_8xb512-linear-coslr-90e_in1k: Top 1 68.3; resnet50_simsiam-200e-pre_8xb512-linear-coslr-90e_in1k: Top 1 69.8)).
|
| 118 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 119 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
| 120 |
-
Reference result: aggregate_metric_reference
|
| 121 |
-
Reference source: configs/simsiam/metafile.yml: Top 1 Accuracy (percent, 1 decimal) of the linear evaluation entries on ImageNet-1k (a linear classifier trained on frozen self-supervised features), measured by OpenMMLab with the listed benchmark config and linear-head checkpoint; the README results table repeats the values
|
| 122 |
Check: For each linear-evaluation entry of configs/simsiam/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/simsiam/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/simsiam/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 123 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_mocov2.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_lars_coslr_90e.py; configs/_base_/schedules/imagenet_sgd_coslr_200e.py; tools/test.py; LICENSE
|
| 124 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 125 |
- `configs/barlowtwins/` (5 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for Barlow Twins self-supervised ResNet-50, linear evaluation (1 model entry (resnet50_barlowtwins-pre_8xb32-linear-coslr-100e_in1k: Top 1 71.8)).
|
| 126 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 127 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
| 128 |
-
Reference result: aggregate_metric_reference
|
| 129 |
-
Reference source: configs/barlowtwins/metafile.yml: Top 1 Accuracy (percent, 1 decimal) of the linear evaluation entries on ImageNet-1k (a linear classifier trained on frozen self-supervised features), measured by OpenMMLab with the listed benchmark config and linear-head checkpoint; the README results table repeats the values
|
| 130 |
Check: For each linear-evaluation entry of configs/barlowtwins/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/barlowtwins/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/barlowtwins/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 131 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_byol.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_sgd_coslr_100e.py; tools/test.py; LICENSE
|
| 132 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 133 |
- `configs/densenet/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for DenseNet-121, 169, 201 and 161 (weights converted from torchvision) (4 model entries (densenet121_3rdparty_in1k: Top 1 74.96, Top 5 92.21; densenet169_3rdparty_in1k: Top 1 76.08, Top 5 93.11; densenet201_3rdparty_in1k: Top 1 77.32, Top 5 93.64; densenet161_3rdparty_in1k: Top 1 77.61, Top 5 93.83)).
|
| 134 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 135 |
Origin: published model-zoo results of a classification toolbox, kept in its configs tree
|
| 136 |
-
Reference result: aggregate_metric_reference
|
| 137 |
-
Reference source: configs/densenet/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and checkpoint; the README results table repeats the values
|
| 138 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/densenet/metafile.yml at the pinned commit (the command given in configs/densenet/README.md line 51) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). Status stays pending until recomputed
|
| 139 |
Companions (not included): configs/_base_/datasets/imagenet_bs64.py; configs/_base_/default_runtime.py; configs/_base_/models/densenet/densenet121.py; configs/_base_/models/densenet/densenet161.py; configs/_base_/models/densenet/densenet169.py; configs/_base_/models/densenet/densenet201.py; configs/_base_/schedules/imagenet_bs256.py; tools/test.py; LICENSE
|
| 140 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 4 of 4 entries (densenet121_3rdparty_in1k, densenet169_3rdparty_in1k, densenet201_3rdparty_in1k, densenet161_3rdparty_in1k) are weights converted by OpenMMLab from the authors' official release (README note at configs/densenet/README.md line 67: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 141 |
- `configs/vgg/` (11 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for VGG-11 to VGG-19 with and without batch norm (8 model entries (vgg11_8xb32_in1k: Top 1 68.75, Top 5 88.87; vgg13_8xb32_in1k: Top 1 70.02, Top 5 89.46; vgg16_8xb32_in1k: Top 1 71.62, Top 5 90.49; vgg19_8xb32_in1k: Top 1 72.41, Top 5 90.8; vgg11bn_8xb32_in1k: Top 1 70.67, Top 5 90.16; vgg13bn_8xb32_in1k: Top 1 72.12, Top 5 90.66; vgg16bn_8xb32_in1k: Top 1 73.74, Top 5 91.66; vgg19bn_8xb32_in1k: Top 1 74.68, Top 5 92.27)).
|
| 142 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 143 |
Origin: published model-zoo results of a classification toolbox, kept in its configs tree
|
| 144 |
-
Reference result: aggregate_metric_reference
|
| 145 |
-
Reference source: configs/vgg/metafile.yml: Top 1 and Top 5 Accuracy (percent) per model entry on ImageNet-1k, measured by OpenMMLab with the listed config and checkpoint; the README results table repeats the values
|
| 146 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/vgg/metafile.yml at the pinned commit (the command given in configs/vgg/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal or 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). Status stays pending until recomputed
|
| 147 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/datasets/voc_bs16.py; configs/_base_/default_runtime.py; configs/_base_/models/vgg11.py; configs/_base_/models/vgg11bn.py; configs/_base_/models/vgg13.py; configs/_base_/models/vgg13bn.py; configs/_base_/models/vgg16.py; configs/_base_/models/vgg16bn.py; configs/_base_/models/vgg19.py; configs/_base_/models/vgg19bn.py; configs/_base_/schedules/imagenet_bs256.py; tools/test.py; LICENSE
|
| 148 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether each unit needs other upstream files is not stated. 16 of 16 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `configs/mobilenet_v2/` (3 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MobileNet V2 (1 model (mobilenet-v2_8xb32_in1k: Top 1 71.86, Top 5 90.42)).
|
| 20 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 21 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
|
|
|
|
|
|
| 22 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/mobilenet_v2/README.md line 71) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 23 |
Companions (not included): configs/_base_/models/mobilenet_v2_1x.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/schedules/imagenet_bs256_epochstep.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 24 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab (no third-party weights).
|
| 25 |
- `configs/shufflenet_v1/` (3 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for ShuffleNet V1 (1 model (shufflenet-v1-1x_16xb64_in1k: Top 1 68.13, Top 5 87.81)).
|
| 26 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 27 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
|
|
|
|
|
|
| 28 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/shufflenet_v1/README.md line 57) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 29 |
Companions (not included): configs/_base_/models/shufflenet_v1_1x.py; configs/_base_/datasets/imagenet_bs64_pil_resize.py; configs/_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 30 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab (no third-party weights).
|
| 31 |
- `configs/shufflenet_v2/` (3 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for ShuffleNet V2 (1 model (shufflenet-v2-1x_16xb64_in1k: Top 1 69.55, Top 5 88.92)).
|
| 32 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 33 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
|
|
|
|
|
|
| 34 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/shufflenet_v2/README.md line 57) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 35 |
Companions (not included): configs/_base_/models/shufflenet_v2_1x.py; configs/_base_/datasets/imagenet_bs64_pil_resize.py; configs/_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 36 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab (no third-party weights).
|
| 37 |
- `configs/mobilenet_v3/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MobileNet V3 (6 models over 4 configs (small-050 57.91/80.19, small-075 65.23/85.44, small 66.68/86.74, small 3rdparty 67.66/87.41, large 73.49/91.31, large 3rdparty 74.04/91.34; Top 1/Top 5)).
|
| 38 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 39 |
Origin: published model-zoo results of an evaluation and training framework, kept in its configs tree
|
|
|
|
|
|
|
| 40 |
Check: For a model entry, run tools/test.py <Config> <Weights> at the pinned commit (the command given in configs/mobilenet_v3/README.md line 69) on the ImageNet-1k validation set, and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within 0.005 percentage points (half a unit of the second decimal). One image is 0.002 percentage points, so the tolerance allows a difference of two images; status stays pending until recomputed
|
| 41 |
Companions (not included): configs/_base_/models/mobilenet_v3/mobilenet_v3_large_imagenet.py; configs/_base_/models/mobilenet_v3/mobilenet_v3_small_imagenet.py; configs/_base_/models/mobilenet_v3/mobilenet_v3_small_050_imagenet.py; configs/_base_/models/mobilenet_v3/mobilenet_v3_small_075_imagenet.py; configs/_base_/datasets/imagenet_bs128_mbv3.py; configs/_base_/default_runtime.py; tools/test.py; LICENSE
|
| 42 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); two models trained by OpenMMLab; four marked 3rdparty, converted from torchvision (configs/mobilenet_v3/README.md line 87; metafile.yml lines 81 and 110 point to download.pytorch.org) or from timm releases (metafile.yml lines 36 and 52).
|
| 43 |
- `configs/repvgg/` (15 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RepVGG A0 to D2se (structural reparameterization: multi-branch training-time blocks fused into a plain 3x3 conv stack) (12 model entries (repvgg-A0_8xb32_in1k: Top 1 72.37, Top 5 90.56; repvgg-A1_8xb32_in1k: Top 1 74.23, Top 5 91.80; repvgg-A2_8xb32_in1k: Top 1 76.49, Top 5 93.09; repvgg-B0_8xb32_in1k: Top 1 75.27, Top 5 92.21; repvgg-B1_8xb32_in1k: Top 1 78.19, Top 5 94.04; repvgg-B1g2_8xb32_in1k: Top 1 77.87, Top 5 93.99; repvgg-B1g4_8xb32_in1k: Top 1 77.81, Top 5 93.77; repvgg-B2_8xb32_in1k: Top 1 78.58, Top 5 94.23; repvgg-B2g4_8xb32_in1k: Top 1 79.44, Top 5 94.72; repvgg-B3_8xb32_in1k: Top 1 80.58, Top 5 95.33; repvgg-B3g4_8xb32_in1k: Top 1 80.26, Top 5 95.15; repvgg-D2se_3rdparty_in1k: Top 1 81.81, Top 5 95.94)).
|
| 44 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 45 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
|
|
|
|
|
|
| 46 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/repvgg/metafile.yml (the command given in configs/repvgg/README.md line 75); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/repvgg/README.md line 89; README line 89 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 47 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/repvgg-A0_in1k.py; configs/_base_/models/repvgg-B3_lbs-mixup_in1k.py; configs/_base_/schedules/imagenet_bs256_coslr.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 48 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 1 of 12 entries (repvgg-D2se_3rdparty_in1k) is weights converted by OpenMMLab from the authors' official release (README note at configs/repvgg/README.md line 130: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 49 |
- `configs/mobileone/` (12 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MobileOne S0 to S4 (reparameterizable mobile backbone) (5 model entries (mobileone-s0_8xb32_in1k: Top 1 71.34, Top 5 89.87; mobileone-s1_8xb32_in1k: Top 1 75.72, Top 5 92.54; mobileone-s2_8xb32_in1k: Top 1 77.37, Top 5 93.34; mobileone-s3_8xb32_in1k: Top 1 78.06, Top 5 93.83; mobileone-s4_8xb32_in1k: Top 1 79.69, Top 5 94.46)).
|
| 50 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 51 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
|
|
|
|
|
|
| 52 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/mobileone/metafile.yml (the command given in configs/mobileone/README.md line 72); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the README of this directory does not name the tool; it calls model.backbone.switch_to_deploy() (tools/model_converters/reparameterize_model.py line 18) and the deploy configs under configs/mobileone/deploy/ set backbone deploy=True) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 53 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/mobileone/mobileone_s0.py; configs/_base_/models/mobileone/mobileone_s1.py; configs/_base_/models/mobileone/mobileone_s2.py; configs/_base_/models/mobileone/mobileone_s3.py; configs/_base_/models/mobileone/mobileone_s4.py; configs/_base_/schedules/imagenet_bs256_coslr_coswd_300e.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 54 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 55 |
- `configs/replknet/` (10 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RepLKNet-31B, 31L and XL (very large reparameterizable kernels) (6 model entries (replknet-31B_3rdparty_in1k: Top 1 83.48, Top 5 96.57; replknet-31B_3rdparty_in1k-384px: Top 1 84.84, Top 5 97.34; replknet-31B_in21k-pre_3rdparty_in1k: Top 1 85.20, Top 5 97.56; replknet-31B_in21k-pre_3rdparty_in1k-384px: Top 1 85.99, Top 5 97.75; replknet-31L_in21k-pre_3rdparty_in1k-384px: Top 1 86.63, Top 5 98.00; replknet-XL_meg73m-pre_3rdparty_in1k-320px: Top 1 87.57, Top 5 98.39)).
|
| 56 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 57 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
|
|
|
|
|
|
| 58 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/replknet/metafile.yml (the command given in configs/replknet/README.md line 53); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/replknet/README.md line 61; README line 61 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 59 |
Companions (not included): configs/_base_/datasets/imagenet_bs16_pil_bicubic_384.py; configs/_base_/datasets/imagenet_bs32_pil_bicubic.py; configs/_base_/datasets/imagenet_bs8_pil_bicubic_320.py; configs/_base_/default_runtime.py; configs/_base_/models/replknet-31B_in1k.py; configs/_base_/models/replknet-31L_in1k.py; configs/_base_/models/replknet-XL_in1k.py; configs/_base_/schedules/imagenet_bs256_coslr.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 60 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 6 of 6 entries (replknet-31B_3rdparty_in1k, replknet-31B_3rdparty_in1k-384px, replknet-31B_in21k-pre_3rdparty_in1k, replknet-31B_in21k-pre_3rdparty_in1k-384px, replknet-31L_in21k-pre_3rdparty_in1k-384px, replknet-XL_meg73m-pre_3rdparty_in1k-320px) are weights converted by OpenMMLab from the authors' official release (README note at configs/replknet/README.md line 96: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 61 |
- `configs/repmlp/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RepMLPNet-Base (reparameterized MLP blocks fused from conv branches) (2 model entries (repmlp-base_3rdparty_8xb64_in1k: Top 1 80.41, Top 5 95.14; repmlp-base_3rdparty_8xb64_in1k-256px: Top 1 81.11, Top 5 95.50)).
|
| 62 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 63 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
|
|
|
|
|
|
| 64 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/repmlp/metafile.yml (the command given in configs/repmlp/README.md line 53); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/repmlp/README.md line 61; README line 61 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 65 |
Companions (not included): configs/_base_/datasets/imagenet_bs64_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/repmlp-base_224.py; configs/_base_/schedules/imagenet_bs1024_adamw_swin.py; configs/_base_/schedules/imagenet_bs4096_AdamW.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 66 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 2 of 2 entries (repmlp-base_3rdparty_8xb64_in1k, repmlp-base_3rdparty_8xb64_in1k-256px) are weights converted by OpenMMLab from the authors' official release (README note at configs/repmlp/README.md line 92: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 67 |
- `configs/riformer/` (22 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for RIFormer S12 to M48 at 224 and 384 px (token mixer removed by reparameterizing an affine transform) (10 model entries (riformer-s12_in1k: Top 1 76.90, Top 5 93.06; riformer-s24_in1k: Top 1 80.28, Top 5 94.80; riformer-s36_in1k: Top 1 81.29, Top 5 95.41; riformer-m36_in1k: Top 1 82.57, Top 5 95.99; riformer-m48_in1k: Top 1 82.75, Top 5 96.11; riformer-s12_in1k-384: Top 1 78.29, Top 5 93.93; riformer-s24_in1k-384: Top 1 81.36, Top 5 95.40; riformer-s36_in1k-384: Top 1 82.22, Top 5 95.95; riformer-m36_in1k-384: Top 1 83.39, Top 5 96.40; riformer-m48_in1k-384: Top 1 83.70, Top 5 96.60)).
|
| 68 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 69 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
|
|
|
|
|
|
| 70 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/riformer/metafile.yml (the command given in configs/riformer/README.md line 93); (b) convert the training-time checkpoint to its inference-time form with python tools/model_converters/reparameterize_model.py <Config> <Weights> <deploy.pth> (the conversion command in configs/riformer/README.md lines 99, 121, 133; README line 121 writes the older path tools/convert_models/, the file at the pinned commit is tools/model_converters/reparameterize_model.py; README lines 70-75 show the source's own check that the switched model gives the same logits (torch.allclose(out, out_deploy, rtol=1e-4, atol=1e-5))) and run tools/test.py with the matching deploy config and the converted checkpoint. Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 71 |
Companions (not included): configs/_base_/datasets/imagenet_bs128_poolformer_medium_224.py; configs/_base_/datasets/imagenet_bs128_poolformer_small_224.py; configs/_base_/datasets/imagenet_bs128_riformer_medium_384.py; configs/_base_/datasets/imagenet_bs128_riformer_small_384.py; configs/_base_/default_runtime.py; configs/_base_/schedules/imagenet_bs1024_adamw_swin.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 72 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 73 |
- `configs/levit/` (12 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for LeViT 128S to 384 (hybrid conv-attention models whose deploy configs fuse batch norm into the linear and conv layers) (5 model entries (levit-128s_3rdparty_in1k: Top 1 76.51, Top 5 92.90; levit-128_3rdparty_in1k: Top 1 78.58, Top 5 93.95; levit-192_3rdparty_in1k: Top 1 79.86, Top 5 94.75; levit-256_3rdparty_in1k: Top 1 81.59, Top 5 95.46; levit-384_3rdparty_in1k: Top 1 82.59, Top 5 95.95)).
|
| 74 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 75 |
Origin: published model-zoo results of a classification toolbox for a family of reparameterizable backbones, with the conversion tool and deploy configs, kept in its configs tree
|
|
|
|
|
|
|
| 76 |
Check: Two checks per model entry, both at the pinned commit on the ImageNet-1k validation set (50,000 images): (a) run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/levit/metafile.yml (the command given in configs/levit/README.md line 51); Each check passes when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). The source publishes no separate accuracy for the converted model: the README presents the conversion as switching to the inference-time architecture with fewer parameters and less computation, so the converted model is judged against the same metafile value; a difference of more than the tolerance in (b) only is the conversion failing. Status stays pending until recomputed
|
| 77 |
Companions (not included): configs/_base_/datasets/imagenet_bs64_swin_224.py; configs/_base_/default_runtime.py; configs/_base_/models/levit-256-p16.py; configs/_base_/schedules/imagenet_bs2048_adamw_levit.py; tools/test.py; tools/model_converters/reparameterize_model.py; LICENSE
|
| 78 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 5 of 5 entries (levit-128s_3rdparty_in1k, levit-128_3rdparty_in1k, levit-192_3rdparty_in1k, levit-256_3rdparty_in1k, levit-384_3rdparty_in1k) are weights converted by OpenMMLab from the authors' official release (README note at configs/levit/README.md line 68: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 79 |
- `configs/mocov2/` (4 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for MoCo v2 self-supervised ResNet-50, linear evaluation (1 model entry (resnet50_mocov2-pre_8xb32-linear-steplr-100e_in1k: Top 1 67.5)).
|
| 80 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 81 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
|
|
|
|
|
|
| 82 |
Check: For each linear-evaluation entry of configs/mocov2/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/mocov2/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/mocov2/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 83 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_mocov2.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_sgd_coslr_200e.py; configs/_base_/schedules/imagenet_sgd_steplr_100e.py; tools/test.py; LICENSE
|
| 84 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 85 |
- `configs/simclr/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for SimCLR self-supervised ResNet-50 (200 and 800 epoch pretraining), linear evaluation (2 model entries (resnet50_simclr-200e-pre_8xb512-linear-coslr-90e_in1k: Top 1 66.9; resnet50_simclr-800e-pre_8xb512-linear-coslr-90e_in1k: Top 1 69.2)).
|
| 86 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 87 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
|
|
|
|
|
|
| 88 |
Check: For each linear-evaluation entry of configs/simclr/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/simclr/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/simclr/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 89 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/datasets/imagenet_bs32_simclr.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_lars_coslr_200e.py; configs/_base_/schedules/imagenet_lars_coslr_90e.py; tools/test.py; LICENSE
|
| 90 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 91 |
- `configs/simsiam/` (5 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for SimSiam self-supervised ResNet-50 (100 and 200 epoch pretraining), linear evaluation (2 model entries (resnet50_simsiam-100e-pre_8xb512-linear-coslr-90e_in1k: Top 1 68.3; resnet50_simsiam-200e-pre_8xb512-linear-coslr-90e_in1k: Top 1 69.8)).
|
| 92 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 93 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
|
|
|
|
|
|
| 94 |
Check: For each linear-evaluation entry of configs/simsiam/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/simsiam/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/simsiam/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 95 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_mocov2.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_lars_coslr_90e.py; configs/_base_/schedules/imagenet_sgd_coslr_200e.py; tools/test.py; LICENSE
|
| 96 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 97 |
- `configs/barlowtwins/` (5 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for Barlow Twins self-supervised ResNet-50, linear evaluation (1 model entry (resnet50_barlowtwins-pre_8xb32-linear-coslr-100e_in1k: Top 1 71.8)).
|
| 98 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 99 |
Origin: published model-zoo results of a pretraining toolbox (self-supervised methods merged from mmselfsup), kept in its configs tree
|
|
|
|
|
|
|
| 100 |
Check: For each linear-evaluation entry of configs/barlowtwins/metafile.yml (the entries with Results), run python tools/test.py <Config> <Weights> with the benchmark Config under configs/barlowtwins/benchmarks/ and the linear-head checkpoint of the entry (the command given in configs/barlowtwins/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 matches the metafile value within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal (the metafile prints one decimal, so a band of 100 images). The pretraining entries without Results give the backbone checkpoints and are not answers. Status stays pending until recomputed
|
| 101 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_byol.py; configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/default_runtime.py; configs/_base_/models/resnet50.py; configs/_base_/schedules/imagenet_sgd_coslr_100e.py; tools/test.py; LICENSE
|
| 102 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 103 |
- `configs/densenet/` (6 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for DenseNet-121, 169, 201 and 161 (weights converted from torchvision) (4 model entries (densenet121_3rdparty_in1k: Top 1 74.96, Top 5 92.21; densenet169_3rdparty_in1k: Top 1 76.08, Top 5 93.11; densenet201_3rdparty_in1k: Top 1 77.32, Top 5 93.64; densenet161_3rdparty_in1k: Top 1 77.61, Top 5 93.83)).
|
| 104 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 105 |
Origin: published model-zoo results of a classification toolbox, kept in its configs tree
|
|
|
|
|
|
|
| 106 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/densenet/metafile.yml at the pinned commit (the command given in configs/densenet/README.md line 51) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). Status stays pending until recomputed
|
| 107 |
Companions (not included): configs/_base_/datasets/imagenet_bs64.py; configs/_base_/default_runtime.py; configs/_base_/models/densenet/densenet121.py; configs/_base_/models/densenet/densenet161.py; configs/_base_/models/densenet/densenet169.py; configs/_base_/models/densenet/densenet201.py; configs/_base_/schedules/imagenet_bs256.py; tools/test.py; LICENSE
|
| 108 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); 4 of 4 entries (densenet121_3rdparty_in1k, densenet169_3rdparty_in1k, densenet201_3rdparty_in1k, densenet161_3rdparty_in1k) are weights converted by OpenMMLab from the authors' official release (README note at configs/densenet/README.md line 67: "Models with * are converted from the official repo"), whose own terms apply. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 109 |
- `configs/vgg/` (11 files): Mmpretrain model-zoo metafile, configs and published ImageNet-1k accuracies for VGG-11 to VGG-19 with and without batch norm (8 model entries (vgg11_8xb32_in1k: Top 1 68.75, Top 5 88.87; vgg13_8xb32_in1k: Top 1 70.02, Top 5 89.46; vgg16_8xb32_in1k: Top 1 71.62, Top 5 90.49; vgg19_8xb32_in1k: Top 1 72.41, Top 5 90.8; vgg11bn_8xb32_in1k: Top 1 70.67, Top 5 90.16; vgg13bn_8xb32_in1k: Top 1 72.12, Top 5 90.66; vgg16bn_8xb32_in1k: Top 1 73.74, Top 5 91.66; vgg19bn_8xb32_in1k: Top 1 74.68, Top 5 92.27)).
|
| 110 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 111 |
Origin: published model-zoo results of a classification toolbox, kept in its configs tree
|
|
|
|
|
|
|
| 112 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/vgg/metafile.yml at the pinned commit (the command given in configs/vgg/README.md line 57) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile values within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal or 0.005 percentage points for values printed with 2 decimals (one image is 0.002 points). Status stays pending until recomputed
|
| 113 |
Companions (not included): configs/_base_/datasets/imagenet_bs32_pil_resize.py; configs/_base_/datasets/voc_bs16.py; configs/_base_/default_runtime.py; configs/_base_/models/vgg11.py; configs/_base_/models/vgg11bn.py; configs/_base_/models/vgg13.py; configs/_base_/models/vgg13bn.py; configs/_base_/models/vgg16.py; configs/_base_/models/vgg16bn.py; configs/_base_/models/vgg19.py; configs/_base_/models/vgg19bn.py; configs/_base_/schedules/imagenet_bs256.py; tools/test.py; LICENSE
|
| 114 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com); trained by OpenMMLab or the method authors and published by OpenMMLab. ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
campaign_files/2026-10-ml-serving-configs-2/open-mmlab-mmrazor/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether each unit needs other upstream files is not stated. 22 of 22 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: aggregate_metric_reference. Where the reference result comes from and how a solution is checked is written per unit in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,176 +19,132 @@ Units:
|
|
| 21 |
- `configs/quantization/ptq/base/` (12 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for post-training quantization (PTQ, OpenVINO and TensorRT int8 backends) (6 model entries (ptq_openvino_mbv2_8xb32_in1k_calib32xb32: Top 1 70.224; ptq_openvino_resnet18_8xb32_in1k_calib32xb32: Top 1 69.742; ptq_openvino_resnet50_8xb32_in1k_calib32xb32: Top 1 76.374; ptq_tensorrt_mbv2_8xb32_in1k_calib32xb32: Top 1 70.324; ptq_tensorrt_resnet18_8xb32_in1k_calib32xb32: Top 1 69.762; ptq_tensorrt_resnet50_8xb32_in1k_calib32xb32: Top 1 76.372)).
|
| 22 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 23 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 24 |
-
Reference result: aggregate_metric_reference
|
| 25 |
-
Reference source: configs/quantization/ptq/base/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 26 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/quantization/ptq/base/metafile.yml at the pinned commit (README.md line 56 (python tools/test.py ${CONFIG} ${CKPT}); recomputing the calibration itself is README.md line 50 (python tools/ptq.py ${CONFIG})) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.0005 percentage points for values printed with 3 decimals (on ImageNet one image is 0.002 points, so this is an exact image count). The four COCO entries (metafile lines 66, 82, 146, 162) are excluded from judging (COCO images carry Flickr terms; general rule on data licences); the mmdet deploy configs they use are not fetched. Status stays pending until recomputed
|
| 27 |
Companions (not included): tools/ptq.py; configs/quantization/deploy_cfgs/mmcls/classification_openvino_dynamic-224x224.py; configs/quantization/deploy_cfgs/mmcls/classification_tensorrt-int8-explicit_dynamic-224x224.py; tools/test.py; LICENSE
|
| 28 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 29 |
- `configs/quantization/qat/base/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for quantization-aware training (QAT, OpenVINO backend) of ResNet-18 (1 model entry (qat_openvino_resnet18_10e_8xb32_in1k.py: Top 1 69.98)).
|
| 30 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 31 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 32 |
-
Reference result: aggregate_metric_reference
|
| 33 |
-
Reference source: configs/quantization/qat/base/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 34 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/quantization/qat/base/metafile.yml at the pinned commit (README.md line 42 (python tools/test.py ${CONFIG} ${CKPT})) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 35 |
Companions (not included): tools/test.py; LICENSE
|
| 36 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 37 |
- `configs/quantization/qat/lsq/` (4 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for LSQ quantization-aware training (OpenVINO backend) of ResNet-18, 10 and 100 epochs (2 model entries (lsq_openvino_resnet18_8xb32_10e_in1k.py: Top 1 69.418; lsq_openvino_resnet18_8xb32_100e_in1k.py: Top 1 69.992)).
|
| 38 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 39 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 40 |
-
Reference result: aggregate_metric_reference
|
| 41 |
-
Reference source: configs/quantization/qat/lsq/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 42 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/quantization/qat/lsq/metafile.yml at the pinned commit (README.md line 43 (python tools/test.py ${CONFIG} ${CKPT})) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.0005 percentage points for values printed with 3 decimals (on ImageNet one image is 0.002 points, so this is an exact image count). Status stays pending until recomputed
|
| 43 |
Companions (not included): tools/test.py; LICENSE
|
| 44 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 45 |
- `configs/pruning/mmcls/l1-norm/` (9 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for L1-norm channel pruning of ResNet-34 (three pruning ratios a, b, c) (3 model entries (l1-norm_resnet34_8xb32_in1k_a: Top 1 73.61; l1-norm_resnet34_8xb32_in1k_b: Top 1 73.20; l1-norm_resnet34_8xb32_in1k_c: Top 1 73.89)).
|
| 46 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 47 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 48 |
-
Reference result: aggregate_metric_reference
|
| 49 |
-
Reference source: configs/pruning/mmcls/l1-norm/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 50 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/pruning/mmcls/l1-norm/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 51 |
Companions (not included): tools/test.py; LICENSE
|
| 52 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 53 |
- `configs/pruning/mmcls/group_fisher/mobilenet/` (8 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for Group Fisher channel pruning of MobileNet V2 (activation and FLOPs criteria) (2 model entries (group_fisher_act_finetune_mobilenet-v2_8xb32_in1k: Top 1 70.82; group_fisher_flops_finetune_mobilenet-v2_8xb32_in1k: Top 1 70.87)).
|
| 54 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 55 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 56 |
-
Reference result: aggregate_metric_reference
|
| 57 |
-
Reference source: configs/pruning/mmcls/group_fisher/mobilenet/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 58 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/pruning/mmcls/group_fisher/mobilenet/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 59 |
Companions (not included): configs/pruning/mmcls/group_fisher/README.md; tools/test.py; LICENSE
|
| 60 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 61 |
- `configs/pruning/mmcls/group_fisher/resnet50/` (9 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for Group Fisher channel pruning of ResNet-50 (activation and FLOPs criteria) (2 model entries (group_fisher_act_finetune_resnet50_8xb32_in1k: Top 1 75.22; group_fisher_flops_finetune_resnet50_8xb32_in1k: Top 1 75.61)).
|
| 62 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 63 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 64 |
-
Reference result: aggregate_metric_reference
|
| 65 |
-
Reference source: configs/pruning/mmcls/group_fisher/resnet50/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 66 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/pruning/mmcls/group_fisher/resnet50/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). group_fisher_act_finetune_resnet50_8xb32_in1k_dist.py line 59 inherits ./resnet_group_fisher_prune.py, which is not in the tree at this commit; that distillation variant has no metafile entry and is not judged. Status stays pending until recomputed
|
| 67 |
Companions (not included): configs/pruning/mmcls/group_fisher/README.md; tools/test.py; LICENSE
|
| 68 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 69 |
- `configs/distill/mmcls/kd/` (5 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for logit knowledge distillation (KD) into ResNet-18, MobileNet V2 and ShuffleNet V2 (3 model entries (kd_logits_resnet34_resnet18_8xb32_in1k: Top 1 71.81; kd_logits_resnet50_mobilenet-v2_8xb32_in1k: Top 1 73.56; kd_logits_resnet50_shufflenet-v2-1x_16xb64_in1k: Top 1 70.87)).
|
| 70 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 71 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 72 |
-
Reference result: aggregate_metric_reference
|
| 73 |
-
Reference source: configs/distill/mmcls/kd/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 74 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/kd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 75 |
Companions (not included): tools/test.py; LICENSE
|
| 76 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 77 |
- `configs/distill/mmcls/dkd/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for decoupled knowledge distillation (DKD) ResNet-34 to ResNet-18 (1 model entry (dkd_resnet34_resnet18_8xb32_in1k: Top 1 71.368, Top 5 90.256)).
|
| 78 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 79 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 80 |
-
Reference result: aggregate_metric_reference
|
| 81 |
-
Reference source: configs/distill/mmcls/dkd/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 82 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/dkd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.0005 percentage points for values printed with 3 decimals (on ImageNet one image is 0.002 points, so this is an exact image count). Status stays pending until recomputed
|
| 83 |
Companions (not included): tools/test.py; LICENSE
|
| 84 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 85 |
- `configs/distill/mmcls/abloss/` (4 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for activation-boundary loss distillation (ABLoss) ResNet-50 to ResNet-18 (1 model entry (abloss_logits_resnet50_resnet18_8xb32_in1k: Top 1 69.94)).
|
| 86 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 87 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 88 |
-
Reference result: aggregate_metric_reference
|
| 89 |
-
Reference source: configs/distill/mmcls/abloss/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 90 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/abloss/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 91 |
Companions (not included): tools/test.py; LICENSE
|
| 92 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 93 |
- `configs/distill/mmcls/rkd/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for relational knowledge distillation (RKD) ResNet-34 to ResNet-18 (1 model entry (rkd_neck_resnet34_resnet18_8xb32_in1k: Top 1 70.23)).
|
| 94 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 95 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 96 |
-
Reference result: aggregate_metric_reference
|
| 97 |
-
Reference source: configs/distill/mmcls/rkd/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 98 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/rkd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 99 |
Companions (not included): tools/test.py; LICENSE
|
| 100 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 101 |
- `configs/distill/mmcls/wsld/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for weighted soft label distillation (WSLD) ResNet-34 to ResNet-18 (1 model entry (wsld_logits_resnet34_resnet18_8xb32_in1k: Top 1 71.54)).
|
| 102 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 103 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 104 |
-
Reference result: aggregate_metric_reference
|
| 105 |
-
Reference source: configs/distill/mmcls/wsld/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 106 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/wsld/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 107 |
Companions (not included): tools/test.py; LICENSE
|
| 108 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 109 |
- `configs/distill/mmcls/fitnets/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for FitNets backbone and logit distillation ResNet-50 to ResNet-18 (1 model entry (fitnets_backbone_logits_resnet50_resnet18_8xb32_in1k: Top 1 70.58)).
|
| 110 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 111 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 112 |
-
Reference result: aggregate_metric_reference
|
| 113 |
-
Reference source: configs/distill/mmcls/fitnets/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 114 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/fitnets/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 115 |
Companions (not included): tools/test.py; LICENSE
|
| 116 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 117 |
- `configs/distill/mmcls/ofd/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for overhaul feature distillation (OFD) WRN-28-4 to WRN-16-2 on CIFAR-10 (1 model entry (ofd_backbone_resnet50_resnet18_8xb16_cifar10: Top 1 94.21)).
|
| 118 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 119 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 120 |
-
Reference result: aggregate_metric_reference
|
| 121 |
-
Reference source: configs/distill/mmcls/ofd/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on CIFAR-10, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 122 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/ofd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the CIFAR-10 test set (10,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on CIFAR-10 test, 10,000 images, one image is 0.01 points, so this is an exact image count). Status stays pending until recomputed
|
| 123 |
Companions (not included): configs/vanilla/mmcls/wide-resnet/wrn16-w2_b16x8_cifar10.py; configs/vanilla/mmcls/wide-resnet/wrn28-w4_b16x8_cifar10.py; configs/_base_/vanilla_models/wrn16_2_cifar10.py; tools/test.py; LICENSE
|
| 124 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0).
|
| 125 |
- `configs/nas/mmcls/autoslim/` (8 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for AutoSlim searched MobileNet V2 1.5x subnets (530M, 320M, 220M FLOPs) (3 model entries (autoslim_mbv2_1.5x_subnet_8xb256_in1k_flops-530M: Top 1 74.23, Top 5 91.73; autoslim_mbv2_1.5x_subnet_8xb256_in1k_flops-320M: Top 1 72.73, Top 5 90.84; autoslim_mbv2_1.5x_subnet_8xb256_in1k_flops-220M: Top 1 71.4, Top 5 90.08)).
|
| 126 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 127 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 128 |
-
Reference result: aggregate_metric_reference
|
| 129 |
-
Reference source: configs/nas/mmcls/autoslim/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 130 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/nas/mmcls/autoslim/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal; 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 131 |
Companions (not included): configs/_base_/settings/imagenet_bs2048_autoslim_pil.py; configs/_base_/settings/imagenet_bs2048_autoslim.py; configs/_base_/settings/imagenet_bs1024_spos.py; tools/test.py; LICENSE
|
| 132 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab (the metafile marks the collection "Converted From: JiahuiYu/slimmable_networks", so the weights are converted from that third-party release); per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 133 |
- `configs/nas/mmcls/spos/` (11 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for SPOS searched ShuffleNet V2 subnet (1 model entry (spos_shufflenet_subnet_8xb128_in1k: Top 1 73.87, Top 5 91.60)).
|
| 134 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 135 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 136 |
-
Reference result: aggregate_metric_reference
|
| 137 |
-
Reference source: configs/nas/mmcls/spos/metafile.yml: Top 1 (and where given Top 5) Accuracy in percent per model entry on ImageNet-1k, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the values
|
| 138 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/nas/mmcls/spos/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). faster-rcnn_nas_backbone_fpn_1x_coco.py is a COCO detection config without a metafile entry and is not judged. Status stays pending until recomputed
|
| 139 |
Companions (not included): configs/_base_/settings/imagenet_bs1024_spos.py; configs/_base_/nas_backbones/spos_shufflenet_supernet.py; configs/_base_/nas_backbones/spos_mobilenet_supernet.py; tools/test.py; LICENSE
|
| 140 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 141 |
- `configs/distill/mmcls/byot/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-100 accuracies for Be Your Own Teacher (BYOT) self-distillation of ResNet-18 on CIFAR-100 (1 model entry (byot_resnet18_8xb16_cifar100: Top 1 80.66, Top 5 95.76)).
|
| 142 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 143 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 144 |
-
Reference result: aggregate_metric_reference
|
| 145 |
-
Reference source: configs/distill/mmcls/byot/metafile.yml: Top 1 and Top 5 Accuracy in percent per model entry on CIFAR-100 test split, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the value
|
| 146 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/byot/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-100 test split (10,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). Status stays pending until recomputed
|
| 147 |
Companions (not included): configs/_base_/datasets/mmcls/cifar100_bs16_auto_aug.py; configs/_base_/datasets/mmcls/pipelines/auto_aug_cifar.py; tools/test.py; LICENSE
|
| 148 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-100 test split (10,000 32x32 images with fine labels, Krizhevsky 2009), which the mmcls CIFAR100 dataset class downloads to data/cifar100 from www.cs.toronto.edu; the dataset page states no licence.
|
| 149 |
- `configs/distill/mmcls/dafl/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for data-free learning (DAFL) distillation from ResNet-34 into ResNet-18 on CIFAR-10 (1 model entry (dafl_logits_resnet34_resnet18_8xb256_cifar10: Top 1 93.27)).
|
| 150 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 151 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 152 |
-
Reference result: aggregate_metric_reference
|
| 153 |
-
Reference source: configs/distill/mmcls/dafl/metafile.yml: Top 1 Accuracy in percent per model entry on CIFAR-10 test split, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the value
|
| 154 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/dafl/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). The student is trained without the CIFAR-10 training images (data-free distillation with a generator); the evaluation is the ordinary test split. Status stays pending until recomputed
|
| 155 |
Companions (not included): tools/test.py; LICENSE
|
| 156 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 157 |
- `configs/distill/mmcls/deit/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for DeiT-Base distilled from a RegNetY-160 teacher on ImageNet-1k (1 model entry (deit-base_regnety160_pt-16xb64_in1k: Top 1 83.24, Top 5 96.33)).
|
| 158 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 159 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 160 |
-
Reference result: aggregate_metric_reference
|
| 161 |
-
Reference source: configs/distill/mmcls/deit/metafile.yml: Top 1 and Top 5 Accuracy in percent per model entry on ImageNet-1k validation set, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the value
|
| 162 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/deit/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.002 points here). The teacher RegNetY-160 checkpoint (dl.fbaipublicfiles.com) is only needed for training, not for the test run. Status stays pending until recomputed
|
| 163 |
Companions (not included): tools/test.py; LICENSE
|
| 164 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 165 |
- `configs/distill/mmcls/dfad/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for data-free adversarial distillation (DFAD) from ResNet-34 into ResNet-18 on CIFAR-10 (1 model entry (dfad_logits_resnet34_resnet18_8xb32_cifar10: Top 1 92.80)).
|
| 166 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 167 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 168 |
-
Reference result: aggregate_metric_reference
|
| 169 |
-
Reference source: configs/distill/mmcls/dfad/metafile.yml: Top 1 Accuracy in percent per model entry on CIFAR-10 test split, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the value
|
| 170 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/dfad/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). The student is trained without the CIFAR-10 training images (data-free adversarial distillation); the evaluation is the ordinary test split. Status stays pending until recomputed
|
| 171 |
Companions (not included): tools/test.py; LICENSE
|
| 172 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 173 |
- `configs/distill/mmcls/factor_transfer/` (4 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for factor transfer distillation from ResNet-50 into ResNet-18 on CIFAR-10 (1 model entry (factor-transfer_backbone_resnet50_resnet18_8xb16_cifar10_train: Top 1 94.8800)).
|
| 174 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 175 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 176 |
-
Reference result: aggregate_metric_reference
|
| 177 |
-
Reference source: configs/distill/mmcls/factor_transfer/metafile.yml line 36: Top 1 Accuracy 94.8800 (percent, 4 decimals) on CIFAR-10 for the train-stage model; the README results table (configs/distill/mmcls/factor_transfer/README.md line 15) prints 94.86 for the same model, a 0.02 point (two images) mismatch with the metafile.
|
| 178 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/factor_transfer/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 5e-05 percentage points (one image is 0.01 points here). Status stays pending until recomputed
|
| 179 |
Companions (not included): tools/test.py; LICENSE
|
| 180 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 181 |
- `configs/distill/mmcls/zskt/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for zero-shot knowledge transfer (ZSKT) from ResNet-34 into ResNet-18 on CIFAR-10 (1 model entry (zskt_backbone_logits_resnet34_resnet18_8xb16_cifar10: Top 1 93.05)).
|
| 182 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 183 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 184 |
-
Reference result: aggregate_metric_reference
|
| 185 |
-
Reference source: configs/distill/mmcls/zskt/metafile.yml: Top 1 Accuracy in percent per model entry on CIFAR-10 test split, measured by OpenMMLab with the listed Config and Weights; the README results table repeats the value
|
| 186 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/zskt/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). The student is trained without the CIFAR-10 training images (zero-shot transfer with an adversarial generator); the evaluation is the ordinary test split. Status stays pending until recomputed
|
| 187 |
Companions (not included): tools/test.py; LICENSE
|
| 188 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 189 |
- `configs/nas/mmcls/darts/` (7 files): Mmrazor DARTS subnet configs, searched-architecture files and published CIFAR-10 accuracies of two DARTS subnets (README subnet table: the mmrazor-searched subnet Top 1 97.32, Top 5 99.94; the paper subnet Top 1 97.27, Top 5 99.98).
|
| 190 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README; the subnet architecture files are YAML (fix_subnet)
|
| 191 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
| 192 |
-
Reference result: aggregate_metric_reference
|
| 193 |
-
Reference source: configs/nas/mmcls/darts/README.md lines 53-54 (subnet table): the subnet with config darts_subnet_1xb96_cifar10_2.0_mmrazor.py (fix_subnet DARTS_SUBNET_CIFAR_MMRAZOR_97.32.yaml) and the 97.32 checkpoint gives Top-1 97.32, Top-5 99.94; the subnet with config darts_subnet_1xb96_cifar10_2.0.py (fix_subnet DARTS_SUBNET_CIFAR_PAPER_ALIAS.yaml, init checkpoint the 97.27 one at line 40 of that config) gives Top-1 97.27, Top-5 99.98. metafile.yml (lines 16-28) pairs the 97.32/99.94 values and the 97.32 checkpoint with darts_subnet_1xb96_cifar10_2.0.py instead.
|
| 194 |
Check: For each of the two subnets, run python tools/test.py <config> <checkpoint> with the config and checkpoint the README subnet table pairs (README line 53: _2.0_mmrazor.py with the 97.32 checkpoint; line 54: _2.0.py with the 97.27 checkpoint; the README Step 3 command at line 35 uses tools/dist_test.sh on one GPU) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the table within 0.005 percentage points (one image is 0.01 points, so the values must match exactly). Status stays pending until recomputed
|
| 195 |
Companions (not included): configs/_base_/settings/cifar10_darts_subnet.py; configs/_base_/nas_backbones/darts_supernet.py; tools/test.py; LICENSE
|
| 196 |
Dependencies: checkpoints from the model links of the README subnet table (darts_subnetnet_1xb96_cifar10_acc-97.32_20211222-e5727921_latest.pth and darts_subnetnet_1xb96_cifar10_acc-97.27_20211222-17e42600_latest.pth on openmmlab-share.oss-cn-hangzhou.aliyuncs.com, the 97.32 one also on download.openmmlab.com per metafile.yml line 28), published by OpenMMLab. CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether each unit needs other upstream files is not stated. 22 of 22 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `configs/quantization/ptq/base/` (12 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for post-training quantization (PTQ, OpenVINO and TensorRT int8 backends) (6 model entries (ptq_openvino_mbv2_8xb32_in1k_calib32xb32: Top 1 70.224; ptq_openvino_resnet18_8xb32_in1k_calib32xb32: Top 1 69.742; ptq_openvino_resnet50_8xb32_in1k_calib32xb32: Top 1 76.374; ptq_tensorrt_mbv2_8xb32_in1k_calib32xb32: Top 1 70.324; ptq_tensorrt_resnet18_8xb32_in1k_calib32xb32: Top 1 69.762; ptq_tensorrt_resnet50_8xb32_in1k_calib32xb32: Top 1 76.372)).
|
| 20 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 21 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 22 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/quantization/ptq/base/metafile.yml at the pinned commit (README.md line 56 (python tools/test.py ${CONFIG} ${CKPT}); recomputing the calibration itself is README.md line 50 (python tools/ptq.py ${CONFIG})) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.0005 percentage points for values printed with 3 decimals (on ImageNet one image is 0.002 points, so this is an exact image count). The four COCO entries (metafile lines 66, 82, 146, 162) are excluded from judging (COCO images carry Flickr terms; general rule on data licences); the mmdet deploy configs they use are not fetched. Status stays pending until recomputed
|
| 23 |
Companions (not included): tools/ptq.py; configs/quantization/deploy_cfgs/mmcls/classification_openvino_dynamic-224x224.py; configs/quantization/deploy_cfgs/mmcls/classification_tensorrt-int8-explicit_dynamic-224x224.py; tools/test.py; LICENSE
|
| 24 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 25 |
- `configs/quantization/qat/base/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for quantization-aware training (QAT, OpenVINO backend) of ResNet-18 (1 model entry (qat_openvino_resnet18_10e_8xb32_in1k.py: Top 1 69.98)).
|
| 26 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 27 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 28 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/quantization/qat/base/metafile.yml at the pinned commit (README.md line 42 (python tools/test.py ${CONFIG} ${CKPT})) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 29 |
Companions (not included): tools/test.py; LICENSE
|
| 30 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 31 |
- `configs/quantization/qat/lsq/` (4 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for LSQ quantization-aware training (OpenVINO backend) of ResNet-18, 10 and 100 epochs (2 model entries (lsq_openvino_resnet18_8xb32_10e_in1k.py: Top 1 69.418; lsq_openvino_resnet18_8xb32_100e_in1k.py: Top 1 69.992)).
|
| 32 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 33 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 34 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/quantization/qat/lsq/metafile.yml at the pinned commit (README.md line 43 (python tools/test.py ${CONFIG} ${CKPT})) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.0005 percentage points for values printed with 3 decimals (on ImageNet one image is 0.002 points, so this is an exact image count). Status stays pending until recomputed
|
| 35 |
Companions (not included): tools/test.py; LICENSE
|
| 36 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 37 |
- `configs/pruning/mmcls/l1-norm/` (9 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for L1-norm channel pruning of ResNet-34 (three pruning ratios a, b, c) (3 model entries (l1-norm_resnet34_8xb32_in1k_a: Top 1 73.61; l1-norm_resnet34_8xb32_in1k_b: Top 1 73.20; l1-norm_resnet34_8xb32_in1k_c: Top 1 73.89)).
|
| 38 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 39 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 40 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/pruning/mmcls/l1-norm/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 41 |
Companions (not included): tools/test.py; LICENSE
|
| 42 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 43 |
- `configs/pruning/mmcls/group_fisher/mobilenet/` (8 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for Group Fisher channel pruning of MobileNet V2 (activation and FLOPs criteria) (2 model entries (group_fisher_act_finetune_mobilenet-v2_8xb32_in1k: Top 1 70.82; group_fisher_flops_finetune_mobilenet-v2_8xb32_in1k: Top 1 70.87)).
|
| 44 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 45 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 46 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/pruning/mmcls/group_fisher/mobilenet/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 47 |
Companions (not included): configs/pruning/mmcls/group_fisher/README.md; tools/test.py; LICENSE
|
| 48 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 49 |
- `configs/pruning/mmcls/group_fisher/resnet50/` (9 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for Group Fisher channel pruning of ResNet-50 (activation and FLOPs criteria) (2 model entries (group_fisher_act_finetune_resnet50_8xb32_in1k: Top 1 75.22; group_fisher_flops_finetune_resnet50_8xb32_in1k: Top 1 75.61)).
|
| 50 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 51 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 52 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/pruning/mmcls/group_fisher/resnet50/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). group_fisher_act_finetune_resnet50_8xb32_in1k_dist.py line 59 inherits ./resnet_group_fisher_prune.py, which is not in the tree at this commit; that distillation variant has no metafile entry and is not judged. Status stays pending until recomputed
|
| 53 |
Companions (not included): configs/pruning/mmcls/group_fisher/README.md; tools/test.py; LICENSE
|
| 54 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 55 |
- `configs/distill/mmcls/kd/` (5 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for logit knowledge distillation (KD) into ResNet-18, MobileNet V2 and ShuffleNet V2 (3 model entries (kd_logits_resnet34_resnet18_8xb32_in1k: Top 1 71.81; kd_logits_resnet50_mobilenet-v2_8xb32_in1k: Top 1 73.56; kd_logits_resnet50_shufflenet-v2-1x_16xb64_in1k: Top 1 70.87)).
|
| 56 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 57 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 58 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/kd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 59 |
Companions (not included): tools/test.py; LICENSE
|
| 60 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 61 |
- `configs/distill/mmcls/dkd/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for decoupled knowledge distillation (DKD) ResNet-34 to ResNet-18 (1 model entry (dkd_resnet34_resnet18_8xb32_in1k: Top 1 71.368, Top 5 90.256)).
|
| 62 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 63 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 64 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/dkd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.0005 percentage points for values printed with 3 decimals (on ImageNet one image is 0.002 points, so this is an exact image count). Status stays pending until recomputed
|
| 65 |
Companions (not included): tools/test.py; LICENSE
|
| 66 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 67 |
- `configs/distill/mmcls/abloss/` (4 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for activation-boundary loss distillation (ABLoss) ResNet-50 to ResNet-18 (1 model entry (abloss_logits_resnet50_resnet18_8xb32_in1k: Top 1 69.94)).
|
| 68 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 69 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 70 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/abloss/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 71 |
Companions (not included): tools/test.py; LICENSE
|
| 72 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 73 |
- `configs/distill/mmcls/rkd/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for relational knowledge distillation (RKD) ResNet-34 to ResNet-18 (1 model entry (rkd_neck_resnet34_resnet18_8xb32_in1k: Top 1 70.23)).
|
| 74 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 75 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 76 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/rkd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 77 |
Companions (not included): tools/test.py; LICENSE
|
| 78 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 79 |
- `configs/distill/mmcls/wsld/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for weighted soft label distillation (WSLD) ResNet-34 to ResNet-18 (1 model entry (wsld_logits_resnet34_resnet18_8xb32_in1k: Top 1 71.54)).
|
| 80 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 81 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 82 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/wsld/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 83 |
Companions (not included): tools/test.py; LICENSE
|
| 84 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 85 |
- `configs/distill/mmcls/fitnets/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for FitNets backbone and logit distillation ResNet-50 to ResNet-18 (1 model entry (fitnets_backbone_logits_resnet50_resnet18_8xb32_in1k: Top 1 70.58)).
|
| 86 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 87 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 88 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/fitnets/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 89 |
Companions (not included): tools/test.py; LICENSE
|
| 90 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 91 |
- `configs/distill/mmcls/ofd/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for overhaul feature distillation (OFD) WRN-28-4 to WRN-16-2 on CIFAR-10 (1 model entry (ofd_backbone_resnet50_resnet18_8xb16_cifar10: Top 1 94.21)).
|
| 92 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 93 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 94 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/distill/mmcls/ofd/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the CIFAR-10 test set (10,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on CIFAR-10 test, 10,000 images, one image is 0.01 points, so this is an exact image count). Status stays pending until recomputed
|
| 95 |
Companions (not included): configs/vanilla/mmcls/wide-resnet/wrn16-w2_b16x8_cifar10.py; configs/vanilla/mmcls/wide-resnet/wrn28-w4_b16x8_cifar10.py; configs/_base_/vanilla_models/wrn16_2_cifar10.py; tools/test.py; LICENSE
|
| 96 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0).
|
| 97 |
- `configs/nas/mmcls/autoslim/` (8 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for AutoSlim searched MobileNet V2 1.5x subnets (530M, 320M, 220M FLOPs) (3 model entries (autoslim_mbv2_1.5x_subnet_8xb256_in1k_flops-530M: Top 1 74.23, Top 5 91.73; autoslim_mbv2_1.5x_subnet_8xb256_in1k_flops-320M: Top 1 72.73, Top 5 90.84; autoslim_mbv2_1.5x_subnet_8xb256_in1k_flops-220M: Top 1 71.4, Top 5 90.08)).
|
| 98 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 99 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 100 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/nas/mmcls/autoslim/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.05 percentage points for values printed with 1 decimal; 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). Status stays pending until recomputed
|
| 101 |
Companions (not included): configs/_base_/settings/imagenet_bs2048_autoslim_pil.py; configs/_base_/settings/imagenet_bs2048_autoslim.py; configs/_base_/settings/imagenet_bs1024_spos.py; tools/test.py; LICENSE
|
| 102 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab (the metafile marks the collection "Converted From: JiahuiYu/slimmable_networks", so the weights are converted from that third-party release); per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 103 |
- `configs/nas/mmcls/spos/` (11 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for SPOS searched ShuffleNet V2 subnet (1 model entry (spos_shufflenet_subnet_8xb128_in1k: Top 1 73.87, Top 5 91.60)).
|
| 104 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 105 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 106 |
Check: For a model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of that entry in configs/nas/mmcls/spos/metafile.yml at the pinned commit (the standard OpenMMLab test entry point tools/test.py at the pinned commit (the README test commands of this directory point at older paths)) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 (and accuracy/top5 where the metafile gives Top 5) match the metafile values within half a unit of the last printed decimal: 0.005 percentage points for values printed with 2 decimals (on ImageNet one image is 0.002 points, so two images). faster-rcnn_nas_backbone_fpn_1x_coco.py is a COCO detection config without a metafile entry and is not judged. Status stays pending until recomputed
|
| 107 |
Companions (not included): configs/_base_/settings/imagenet_bs1024_spos.py; configs/_base_/nas_backbones/spos_shufflenet_supernet.py; configs/_base_/nas_backbones/spos_mobilenet_supernet.py; tools/test.py; LICENSE
|
| 108 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the mmcls dataset config expects.
|
| 109 |
- `configs/distill/mmcls/byot/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-100 accuracies for Be Your Own Teacher (BYOT) self-distillation of ResNet-18 on CIFAR-100 (1 model entry (byot_resnet18_8xb16_cifar100: Top 1 80.66, Top 5 95.76)).
|
| 110 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 111 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 112 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/byot/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-100 test split (10,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). Status stays pending until recomputed
|
| 113 |
Companions (not included): configs/_base_/datasets/mmcls/cifar100_bs16_auto_aug.py; configs/_base_/datasets/mmcls/pipelines/auto_aug_cifar.py; tools/test.py; LICENSE
|
| 114 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-100 test split (10,000 32x32 images with fine labels, Krizhevsky 2009), which the mmcls CIFAR100 dataset class downloads to data/cifar100 from www.cs.toronto.edu; the dataset page states no licence.
|
| 115 |
- `configs/distill/mmcls/dafl/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for data-free learning (DAFL) distillation from ResNet-34 into ResNet-18 on CIFAR-10 (1 model entry (dafl_logits_resnet34_resnet18_8xb256_cifar10: Top 1 93.27)).
|
| 116 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 117 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 118 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/dafl/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). The student is trained without the CIFAR-10 training images (data-free distillation with a generator); the evaluation is the ordinary test split. Status stays pending until recomputed
|
| 119 |
Companions (not included): tools/test.py; LICENSE
|
| 120 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 121 |
- `configs/distill/mmcls/deit/` (3 files): Mmrazor model-zoo metafile, configs and published ImageNet-1k accuracies for DeiT-Base distilled from a RegNetY-160 teacher on ImageNet-1k (1 model entry (deit-base_regnety160_pt-16xb64_in1k: Top 1 83.24, Top 5 96.33)).
|
| 122 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 123 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 124 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/deit/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the ImageNet-1k validation set (50,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.002 points here). The teacher RegNetY-160 checkpoint (dl.fbaipublicfiles.com) is only needed for training, not for the test run. Status stays pending until recomputed
|
| 125 |
Companions (not included): tools/test.py; LICENSE
|
| 126 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). ImageNet-1k (ILSVRC2012) validation set, 50,000 images with labels, laid out under data/imagenet as the dataset config expects.
|
| 127 |
- `configs/distill/mmcls/dfad/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for data-free adversarial distillation (DFAD) from ResNet-34 into ResNet-18 on CIFAR-10 (1 model entry (dfad_logits_resnet34_resnet18_8xb32_cifar10: Top 1 92.80)).
|
| 128 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 129 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 130 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/dfad/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). The student is trained without the CIFAR-10 training images (data-free adversarial distillation); the evaluation is the ordinary test split. Status stays pending until recomputed
|
| 131 |
Companions (not included): tools/test.py; LICENSE
|
| 132 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 133 |
- `configs/distill/mmcls/factor_transfer/` (4 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for factor transfer distillation from ResNet-50 into ResNet-18 on CIFAR-10 (1 model entry (factor-transfer_backbone_resnet50_resnet18_8xb16_cifar10_train: Top 1 94.8800)).
|
| 134 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 135 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 136 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/factor_transfer/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 5e-05 percentage points (one image is 0.01 points here). Status stays pending until recomputed
|
| 137 |
Companions (not included): tools/test.py; LICENSE
|
| 138 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 139 |
- `configs/distill/mmcls/zskt/` (3 files): Mmrazor model-zoo metafile, configs and published CIFAR-10 accuracies for zero-shot knowledge transfer (ZSKT) from ResNet-34 into ResNet-18 on CIFAR-10 (1 model entry (zskt_backbone_logits_resnet34_resnet18_8xb16_cifar10: Top 1 93.05)).
|
| 140 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README
|
| 141 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 142 |
Check: For the model entry, run python tools/test.py <Config> <Weights> with the Config and Weights of the entry in configs/distill/mmcls/zskt/metafile.yml at the pinned commit (the standard OpenMMLab test entry point; the README of this directory gives only training commands) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 match the metafile value within half a unit of the last printed decimal: 0.005 percentage points (one image is 0.01 points here). The student is trained without the CIFAR-10 training images (zero-shot transfer with an adversarial generator); the evaluation is the ordinary test split. Status stays pending until recomputed
|
| 143 |
Companions (not included): tools/test.py; LICENSE
|
| 144 |
Dependencies: checkpoints from the Weights URLs of metafile.yml (download.openmmlab.com), published by OpenMMLab; per-checkpoint licence not stated separately (the repository is Apache-2.0). CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
| 145 |
- `configs/nas/mmcls/darts/` (7 files): Mmrazor DARTS subnet configs, searched-architecture files and published CIFAR-10 accuracies of two DARTS subnets (README subnet table: the mmrazor-searched subnet Top 1 97.32, Top 5 99.94; the paper subnet Top 1 97.27, Top 5 99.98).
|
| 146 |
Format: YAML metafile (Collections and Models with Metadata, Results, Weights, Config) plus Python config files (mmengine Config) and a Markdown README; the subnet architecture files are YAML (fix_subnet)
|
| 147 |
Origin: published model-zoo results of a model-compression toolbox (quantization, pruning, distillation, NAS), kept in its configs tree
|
|
|
|
|
|
|
| 148 |
Check: For each of the two subnets, run python tools/test.py <config> <checkpoint> with the config and checkpoint the README subnet table pairs (README line 53: _2.0_mmrazor.py with the 97.32 checkpoint; line 54: _2.0.py with the 97.27 checkpoint; the README Step 3 command at line 35 uses tools/dist_test.sh on one GPU) on the CIFAR-10 test split (10,000 images), and pass when the reported accuracy/top1 and accuracy/top5 match the table within 0.005 percentage points (one image is 0.01 points, so the values must match exactly). Status stays pending until recomputed
|
| 149 |
Companions (not included): configs/_base_/settings/cifar10_darts_subnet.py; configs/_base_/nas_backbones/darts_supernet.py; tools/test.py; LICENSE
|
| 150 |
Dependencies: checkpoints from the model links of the README subnet table (darts_subnetnet_1xb96_cifar10_acc-97.32_20211222-e5727921_latest.pth and darts_subnetnet_1xb96_cifar10_acc-97.27_20211222-17e42600_latest.pth on openmmlab-share.oss-cn-hangzhou.aliyuncs.com, the 97.32 one also on download.openmmlab.com per metafile.yml line 28), published by OpenMMLab. CIFAR-10 test split (10,000 32x32 images with labels, Krizhevsky 2009), which the mmcls CIFAR10 dataset class downloads to data/cifar10 from www.cs.toronto.edu; the dataset page states no licence.
|
campaign_files/2026-10-ml-serving-configs-2/open-model-zoo-accuracy/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** 0 of 20 units are self-contained; for the others it is not stated. 20 of 20 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: aggregate_accuracy_metric. Where the reference result comes from and how a solution is checked is written per unit in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,160 +19,120 @@ Units:
|
|
| 21 |
- `models/public/gpt-2/` (3 files): Open Model Zoo gpt-2: GPT-2 (124M) language model, published accuracy with the Accuracy Checker config that reproduces it.
|
| 22 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 23 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 24 |
-
Reference result: aggregate_accuracy_metric
|
| 25 |
-
Reference source: perplexity 0.29 (models/public/gpt-2/accuracy-check.yml line 24; README line 26 "29.00%"); README section "## Accuracy" at line 20
|
| 26 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/gpt-2/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 27 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 28 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/gpt-2/model.yml (licence: huggingface.co/gpt2 (MIT per the model card); model.yml line 127: license: https://huggingface.co/gpt2); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; WikiText-2 raw test split (CC BY-SA, not shipped; data/dataset_definitions.yml line 1440)
|
| 29 |
- `models/public/wav2vec2-base/` (3 files): Open Model Zoo wav2vec2-base: wav2vec 2.0 base speech recognition, published accuracy with the Accuracy Checker config that reproduces it.
|
| 30 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 31 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 32 |
-
Reference result: aggregate_accuracy_metric
|
| 33 |
-
Reference source: wer 0.0339 (models/public/wav2vec2-base/accuracy-check.yml line 21; README line 22 3.39%); README section "## Accuracy" at line 18
|
| 34 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/wav2vec2-base/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 35 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 36 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/wav2vec2-base/model.yml (licence: fairseq (MIT); model.yml line 117: license: https://raw.githubusercontent.com/pytorch/fairseq/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; LibriSpeech test-clean (CC BY 4.0, not shipped; data/dataset_definitions.yml lines 1111 and 1373)
|
| 37 |
- `models/public/resnet-50-pytorch/` (3 files): Open Model Zoo resnet-50-pytorch: ResNet-50 image classification (torchvision weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 38 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 39 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 40 |
-
Reference result: aggregate_accuracy_metric
|
| 41 |
-
Reference source: accuracy top1 0.76128 (models/public/resnet-50-pytorch/accuracy-check.yml line 41; README line 29 76.128%); accuracy top5 0.92858 (models/public/resnet-50-pytorch/accuracy-check.yml line 45; README line 30 92.858%); README section "## Accuracy" at line 25
|
| 42 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/resnet-50-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 43 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 44 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/resnet-50-pytorch/model.yml (licence: torchvision (BSD-3-Clause); model.yml line 52: license: https://raw.githubusercontent.com/pytorch/vision/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped; named in data/dataset_definitions.yml line 183)
|
| 45 |
- `models/public/mobilenet-v2-pytorch/` (3 files): Open Model Zoo mobilenet-v2-pytorch: MobileNet V2 image classification (torchvision weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 46 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 47 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 48 |
-
Reference result: aggregate_accuracy_metric
|
| 49 |
-
Reference source: accuracy top1 0.719 (models/public/mobilenet-v2-pytorch/accuracy-check.yml line 44; README line 29 71.9%); accuracy top5 0.903 (models/public/mobilenet-v2-pytorch/accuracy-check.yml line 48; README line 30 90.3%); README section "## Accuracy" at line 25
|
| 50 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/mobilenet-v2-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 51 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 52 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/mobilenet-v2-pytorch/model.yml (licence: torchvision (BSD-3-Clause); model.yml line 52: license: https://raw.githubusercontent.com/pytorch/vision/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped; named in data/dataset_definitions.yml line 183)
|
| 53 |
- `models/public/convnext-tiny/` (3 files): Open Model Zoo convnext-tiny: ConvNeXt-Tiny image classification (timm weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 54 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 55 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 56 |
-
Reference result: aggregate_accuracy_metric
|
| 57 |
-
Reference source: accuracy top1 0.8205 (models/public/convnext-tiny/accuracy-check.yml line 28; README line 26 82.05%); accuracy top5 0.9586 (models/public/convnext-tiny/accuracy-check.yml line 32; README line 27 95.86%); README section "## Accuracy" at line 22
|
| 58 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/convnext-tiny/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 59 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 60 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/convnext-tiny/model.yml (licence: timm (Apache-2.0); model.yml line 65: license: https://raw.githubusercontent.com/rwightman/pytorch-image-models/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped; named in data/dataset_definitions.yml line 183)
|
| 61 |
- `models/public/quartznet-15x5-en/` (3 files): Open Model Zoo quartznet-15x5-en: QuartzNet 15x5 English speech recognition (NeMo), published accuracy with the Accuracy Checker config that reproduces it.
|
| 62 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 63 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 64 |
-
Reference result: aggregate_accuracy_metric
|
| 65 |
-
Reference source: wer 0.0386 (models/public/quartznet-15x5-en/accuracy-check.yml line 19; README line 24 3.86%); README section "## Accuracy" at line 20
|
| 66 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/quartznet-15x5-en/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 67 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 68 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/quartznet-15x5-en/model.yml (licence: NVIDIA NeMo (Apache-2.0); model.yml line 113: license: https://raw.githubusercontent.com/NVIDIA/NeMo/main/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; LibriSpeech test-clean (CC BY 4.0, not shipped; data/dataset_definitions.yml lines 1111 and 1373)
|
| 69 |
- `models/public/yolo-v3-onnx/` (3 files): Open Model Zoo yolo-v3-onnx: YOLO v3 object detection (ONNX model zoo), published accuracy with the Accuracy Checker config that reproduces it.
|
| 70 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 71 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 72 |
-
Reference result: aggregate_accuracy_metric
|
| 73 |
-
Reference source: map 0.483 (models/public/yolo-v3-onnx/accuracy-check.yml line 33; README line 22 48.30%); coco_precision 0.4707 (models/public/yolo-v3-onnx/accuracy-check.yml line 37; README line 23 47.07%); README section "## Accuracy" at line 16
|
| 74 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/yolo-v3-onnx/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 75 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 76 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/yolo-v3-onnx/model.yml (licence: onnx/models (Apache-2.0 per the zoo); model.yml line 41: license: https://raw.githubusercontent.com/onnx/models/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; COCO val2017 images and instances annotations (not shipped; data/dataset_definitions.yml lines 72 and 97)
|
| 77 |
- `models/public/ssd_mobilenet_v1_coco/` (3 files): Open Model Zoo ssd_mobilenet_v1_coco: SSD MobileNet V1 object detection (TensorFlow Object Detection API), published accuracy with the Accuracy Checker config that reproduces it.
|
| 78 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 79 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 80 |
-
Reference result: aggregate_accuracy_metric
|
| 81 |
-
Reference source: coco_precision 0.233212 (models/public/ssd_mobilenet_v1_coco/accuracy-check.yml line 16; README line 20 23.3212%); README section "## Accuracy" at line 16
|
| 82 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/ssd_mobilenet_v1_coco/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 83 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 84 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/ssd_mobilenet_v1_coco/model.yml (licence: tensorflow/models (Apache-2.0); model.yml line 40: license: https://raw.githubusercontent.com/tensorflow/models/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; COCO val2017 images and instances annotations (not shipped; data/dataset_definitions.yml lines 72 and 97)
|
| 85 |
- `models/public/resnet-18-pytorch/` (3 files): Open Model Zoo resnet-18-pytorch: ResNet-18 image classification (torchvision weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 86 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 87 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 88 |
-
Reference result: aggregate_accuracy_metric
|
| 89 |
-
Reference source: accuracy top1 0.69754 (models/public/resnet-18-pytorch/accuracy-check.yml line(s) 45 and 76; README line 29 69.754%); accuracy top5 0.89088 (models/public/resnet-18-pytorch/accuracy-check.yml line(s) 49 and 80; README line 30 89.088%)
|
| 90 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/resnet-18-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 91 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 92 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/resnet-18-pytorch/model.yml (licence: torchvision (BSD-3-Clause); model.yml line 52: license: https://raw.githubusercontent.com/pytorch/vision/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped); Accuracy Checker dataset definition imagenet_1000_classes at data/dataset_definitions.yml line 183
|
| 93 |
- `models/public/efficientnet-b0-pytorch/` (3 files): Open Model Zoo efficientnet-b0-pytorch: EfficientNet-B0 image classification (gen-efficientnet-pytorch weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 94 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 95 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 96 |
-
Reference result: aggregate_accuracy_metric
|
| 97 |
-
Reference source: accuracy top1 0.777 (models/public/efficientnet-b0-pytorch/accuracy-check.yml line(s) 36 and 68; README line 28 77.70%); accuracy top5 0.9352 (models/public/efficientnet-b0-pytorch/accuracy-check.yml line(s) 40 and 72; README line 29 93.52%)
|
| 98 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/efficientnet-b0-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 99 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 100 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/efficientnet-b0-pytorch/model.yml (licence: gen-efficientnet-pytorch (Apache-2.0); model.yml line 77: license: https://raw.githubusercontent.com/rwightman/gen-efficientnet-pytorch/a36e2b2cd1bd122a508a6fffeaa7606890f8c882/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped); Accuracy Checker dataset definition imagenet_1000_classes at data/dataset_definitions.yml line 183
|
| 101 |
- `models/public/regnetx-3.2gf/` (3 files): Open Model Zoo regnetx-3.2gf: RegNetX-3.2GF image classification (pycls weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 102 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 103 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 104 |
-
Reference result: aggregate_accuracy_metric
|
| 105 |
-
Reference source: accuracy top1 0.7815 (models/public/regnetx-3.2gf/accuracy-check.yml line(s) 31 and 58; README line 22 78.15%); accuracy top5 0.9409 (models/public/regnetx-3.2gf/accuracy-check.yml line(s) 35 and 62; README line 23 94.09%)
|
| 106 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/regnetx-3.2gf/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 107 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 108 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/regnetx-3.2gf/model.yml (licence: pycls (MIT); model.yml line 115: license: https://raw.githubusercontent.com/facebookresearch/pycls/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped); Accuracy Checker dataset definition imagenet_1000_classes at data/dataset_definitions.yml line 183
|
| 109 |
- `models/public/detr-resnet50/` (3 files): Open Model Zoo detr-resnet50: DETR ResNet-50 object detection on COCO, published accuracy with the Accuracy Checker config that reproduces it.
|
| 110 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 111 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 112 |
-
Reference result: aggregate_accuracy_metric
|
| 113 |
-
Reference source: coco_precision 0.4236 (models/public/detr-resnet50/accuracy-check.yml line(s) 25 and 49; README line 25 42.36%); coco_orig_precision 0.3927 (models/public/detr-resnet50/accuracy-check.yml line(s) 27 and 51; README line 24 39.27%)
|
| 114 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/detr-resnet50/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 115 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 116 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/detr-resnet50/model.yml (licence: facebookresearch/detr (Apache-2.0); model.yml line 97: license: https://raw.githubusercontent.com/facebookresearch/detr/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_91_classes at data/dataset_definitions.yml line 72
|
| 117 |
- `models/public/yolox-tiny/` (3 files): Open Model Zoo yolox-tiny: YOLOX-Tiny object detection on COCO, published accuracy with the Accuracy Checker config that reproduces it.
|
| 118 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 119 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 120 |
-
Reference result: aggregate_accuracy_metric
|
| 121 |
-
Reference source: map 0.4785 (models/public/yolox-tiny/accuracy-check.yml line(s) 33 and 71; README line 24 47.85%); coco_precision AP@0.5 0.5256 (models/public/yolox-tiny/accuracy-check.yml line(s) 38 and 76; README line 25 52.56%); coco_precision AP@0.5:0.05:95 0.3182 (models/public/yolox-tiny/accuracy-check.yml line(s) 43 and 81; README line 26 31.82%)
|
| 122 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/yolox-tiny/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 123 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 124 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/yolox-tiny/model.yml (licence: Megvii YOLOX (Apache-2.0); model.yml line 88: license: https://raw.githubusercontent.com/Megvii-BaseDetection/YOLOX/main/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_80_class_without_background at data/dataset_definitions.yml line 97
|
| 125 |
- `models/public/ssdlite_mobilenet_v2/` (3 files): Open Model Zoo ssdlite_mobilenet_v2: SSDLite MobileNetV2 object detection on COCO (TensorFlow Object Detection API weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 126 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 127 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 128 |
-
Reference result: aggregate_accuracy_metric
|
| 129 |
-
Reference source: coco_precision 0.242946 (models/public/ssdlite_mobilenet_v2/accuracy-check.yml line(s) 16; README line 20 24.2946%)
|
| 130 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/ssdlite_mobilenet_v2/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 131 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 132 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/ssdlite_mobilenet_v2/model.yml (licence: tensorflow/models (Apache-2.0); model.yml line 41: license: https://raw.githubusercontent.com/tensorflow/models/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_91_classes at data/dataset_definitions.yml line 72
|
| 133 |
- `models/public/yolo-v3-tiny-onnx/` (3 files): Open Model Zoo yolo-v3-tiny-onnx: Tiny YOLOv3 object detection on COCO (ONNX model zoo weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 134 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 135 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 136 |
-
Reference result: aggregate_accuracy_metric
|
| 137 |
-
Reference source: map 0.1707 (models/public/yolo-v3-tiny-onnx/accuracy-check.yml line(s) 33 and 68; README line 22 17.07%); coco_precision 0.1364 (models/public/yolo-v3-tiny-onnx/accuracy-check.yml line(s) 37 and 72; README line 23 13.64%)
|
| 138 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/yolo-v3-tiny-onnx/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 139 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 140 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/yolo-v3-tiny-onnx/model.yml (licence: onnx/models (Apache-2.0); model.yml line 41: license: https://raw.githubusercontent.com/onnx/models/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_80_class_without_background at data/dataset_definitions.yml line 97
|
| 141 |
- `models/public/efficientdet-d0-tf/` (3 files): Open Model Zoo efficientdet-d0-tf: EfficientDet-D0 object detection on COCO (google/automl weights, TensorFlow), published accuracy with the Accuracy Checker config that reproduces it.
|
| 142 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 143 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 144 |
-
Reference result: aggregate_accuracy_metric
|
| 145 |
-
Reference source: coco_precision 0.3195 (models/public/efficientdet-d0-tf/accuracy-check.yml line(s) 24; README line 49 31.95%)
|
| 146 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/efficientdet-d0-tf/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 147 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 148 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/efficientdet-d0-tf/model.yml (licence: google/automl (Apache-2.0); model.yml line 36: license: https://raw.githubusercontent.com/google/automl/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_90_class_without_background at data/dataset_definitions.yml line 119
|
| 149 |
- `models/public/nanodet-plus-m-1.5x-416/` (3 files): Open Model Zoo nanodet-plus-m-1.5x-416: NanoDet-Plus-m-1.5x-416 object detection on COCO, published accuracy with the Accuracy Checker config that reproduces it.
|
| 150 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 151 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 152 |
-
Reference result: aggregate_accuracy_metric
|
| 153 |
-
Reference source: coco_precision 0.3453 (models/public/nanodet-plus-m-1.5x-416/accuracy-check.yml line(s) 23; README line 25 34.53%); coco_orig_precision 0.3377 (models/public/nanodet-plus-m-1.5x-416/accuracy-check.yml line(s) 25; README line 24 33.77%)
|
| 154 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/nanodet-plus-m-1.5x-416/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 155 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 156 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/nanodet-plus-m-1.5x-416/model.yml (licence: RangiLyu/nanodet (Apache-2.0); model.yml line 251: license: https://raw.githubusercontent.com/RangiLyu/nanodet/main/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_80_class_without_background at data/dataset_definitions.yml line 97
|
| 157 |
- `models/public/deeplabv3/` (3 files): Open Model Zoo deeplabv3: DeepLabV3 MobileNetV2 semantic segmentation on PASCAL VOC 2012, published accuracy with the Accuracy Checker config that reproduces it.
|
| 158 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 159 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 160 |
-
Reference result: aggregate_accuracy_metric
|
| 161 |
-
Reference source: mean_iou 0.6841 (models/public/deeplabv3/accuracy-check.yml line(s) 30; README line 20 68.41%)
|
| 162 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/deeplabv3/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 163 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 164 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/deeplabv3/model.yml (licence: tensorflow/models (Apache-2.0); model.yml line 38: license: https://raw.githubusercontent.com/tensorflow/models/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; PASCAL VOC 2012 segmentation validation split (images under Flickr terms, annotations from the VOC project; not shipped); Accuracy Checker dataset definition VOC2012_Segmentation at data/dataset_definitions.yml line 261
|
| 165 |
- `models/public/hrnet-v2-c1-segmentation/` (3 files): Open Model Zoo hrnet-v2-c1-segmentation: HRNetV2-C1 semantic segmentation on ADE20K, published accuracy with the Accuracy Checker config that reproduces it.
|
| 166 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 167 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 168 |
-
Reference result: aggregate_accuracy_metric
|
| 169 |
-
Reference source: pix_accuracy 0.7769 (models/public/hrnet-v2-c1-segmentation/accuracy-check.yml line(s) 30 and 59; README line 25 77.69%); mean_IoU 0.3302 (models/public/hrnet-v2-c1-segmentation/accuracy-check.yml line(s) 34 and 63; README line 26 33.02%)
|
| 170 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/hrnet-v2-c1-segmentation/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 171 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 172 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/hrnet-v2-c1-segmentation/model.yml (licence: CSAILVision semantic-segmentation-pytorch (BSD-3-Clause); model.yml line 122: license: https://raw.githubusercontent.com/CSAILVision/semantic-segmentation-pytorch/9aff40de31ee4b21f18514d31e5d6e4ba056924d/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ADE20K ADEChallengeData2016 validation split (2000 images, MIT CSAIL terms; not shipped); Accuracy Checker dataset definition ADEChallengeData2016 at data/dataset_definitions.yml line 1237
|
| 173 |
- `models/public/mozilla-deepspeech-0.8.2/` (3 files): Open Model Zoo mozilla-deepspeech-0.8.2: Mozilla DeepSpeech 0.8.2 speech recognition on LibriSpeech test-clean (with language model), published accuracy with the Accuracy Checker config that reproduces it.
|
| 174 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 175 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
| 176 |
-
Reference result: aggregate_accuracy_metric
|
| 177 |
-
Reference source: wer 0.0839 (models/public/mozilla-deepspeech-0.8.2/accuracy-check.yml line(s) 60; README line 27 8.39%)
|
| 178 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/mozilla-deepspeech-0.8.2/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside. The config decodes with ctc_beam_search_decoder_with_lm, beam_size 32 (accuracy-check.yml lines 6 and 9), which matches README line 27 (8.39%, beam_width = 32, Python CTC decoder); README lines 28 and 29 (6.13% and 6.15% at beam_width 500) are other settings and are not the reference.
|
| 179 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 180 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/mozilla-deepspeech-0.8.2/model.yml (licence: mozilla/DeepSpeech (MPL-2.0); model.yml line 50: license: https://raw.githubusercontent.com/mozilla/DeepSpeech/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; LibriSpeech test-clean (CC BY 4.0, not shipped); Accuracy Checker dataset definition librispeech-test-clean at data/dataset_definitions.yml line 1111
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** 0 of 20 units are self-contained; for the others it is not stated. 20 of 20 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Tools are named in the check written per unit in the notes.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `models/public/gpt-2/` (3 files): Open Model Zoo gpt-2: GPT-2 (124M) language model, published accuracy with the Accuracy Checker config that reproduces it.
|
| 20 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 21 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 22 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/gpt-2/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 23 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 24 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/gpt-2/model.yml (licence: huggingface.co/gpt2 (MIT per the model card); model.yml line 127: license: https://huggingface.co/gpt2); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; WikiText-2 raw test split (CC BY-SA, not shipped; data/dataset_definitions.yml line 1440)
|
| 25 |
- `models/public/wav2vec2-base/` (3 files): Open Model Zoo wav2vec2-base: wav2vec 2.0 base speech recognition, published accuracy with the Accuracy Checker config that reproduces it.
|
| 26 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 27 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 28 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/wav2vec2-base/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 29 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 30 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/wav2vec2-base/model.yml (licence: fairseq (MIT); model.yml line 117: license: https://raw.githubusercontent.com/pytorch/fairseq/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; LibriSpeech test-clean (CC BY 4.0, not shipped; data/dataset_definitions.yml lines 1111 and 1373)
|
| 31 |
- `models/public/resnet-50-pytorch/` (3 files): Open Model Zoo resnet-50-pytorch: ResNet-50 image classification (torchvision weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 32 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 33 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 34 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/resnet-50-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 35 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 36 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/resnet-50-pytorch/model.yml (licence: torchvision (BSD-3-Clause); model.yml line 52: license: https://raw.githubusercontent.com/pytorch/vision/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped; named in data/dataset_definitions.yml line 183)
|
| 37 |
- `models/public/mobilenet-v2-pytorch/` (3 files): Open Model Zoo mobilenet-v2-pytorch: MobileNet V2 image classification (torchvision weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 38 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 39 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 40 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/mobilenet-v2-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 41 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 42 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/mobilenet-v2-pytorch/model.yml (licence: torchvision (BSD-3-Clause); model.yml line 52: license: https://raw.githubusercontent.com/pytorch/vision/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped; named in data/dataset_definitions.yml line 183)
|
| 43 |
- `models/public/convnext-tiny/` (3 files): Open Model Zoo convnext-tiny: ConvNeXt-Tiny image classification (timm weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 44 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 45 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 46 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/convnext-tiny/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 47 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 48 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/convnext-tiny/model.yml (licence: timm (Apache-2.0); model.yml line 65: license: https://raw.githubusercontent.com/rwightman/pytorch-image-models/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped; named in data/dataset_definitions.yml line 183)
|
| 49 |
- `models/public/quartznet-15x5-en/` (3 files): Open Model Zoo quartznet-15x5-en: QuartzNet 15x5 English speech recognition (NeMo), published accuracy with the Accuracy Checker config that reproduces it.
|
| 50 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 51 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 52 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/quartznet-15x5-en/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 53 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 54 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/quartznet-15x5-en/model.yml (licence: NVIDIA NeMo (Apache-2.0); model.yml line 113: license: https://raw.githubusercontent.com/NVIDIA/NeMo/main/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; LibriSpeech test-clean (CC BY 4.0, not shipped; data/dataset_definitions.yml lines 1111 and 1373)
|
| 55 |
- `models/public/yolo-v3-onnx/` (3 files): Open Model Zoo yolo-v3-onnx: YOLO v3 object detection (ONNX model zoo), published accuracy with the Accuracy Checker config that reproduces it.
|
| 56 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 57 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 58 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/yolo-v3-onnx/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 59 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 60 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/yolo-v3-onnx/model.yml (licence: onnx/models (Apache-2.0 per the zoo); model.yml line 41: license: https://raw.githubusercontent.com/onnx/models/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; COCO val2017 images and instances annotations (not shipped; data/dataset_definitions.yml lines 72 and 97)
|
| 61 |
- `models/public/ssd_mobilenet_v1_coco/` (3 files): Open Model Zoo ssd_mobilenet_v1_coco: SSD MobileNet V1 object detection (TensorFlow Object Detection API), published accuracy with the Accuracy Checker config that reproduces it.
|
| 62 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 63 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 64 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/ssd_mobilenet_v1_coco/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 65 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 66 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/ssd_mobilenet_v1_coco/model.yml (licence: tensorflow/models (Apache-2.0); model.yml line 40: license: https://raw.githubusercontent.com/tensorflow/models/master/LICENSE); the URLs point at upstream masters and are pinned by the sha256 fields in model.yml; COCO val2017 images and instances annotations (not shipped; data/dataset_definitions.yml lines 72 and 97)
|
| 67 |
- `models/public/resnet-18-pytorch/` (3 files): Open Model Zoo resnet-18-pytorch: ResNet-18 image classification (torchvision weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 68 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 69 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 70 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/resnet-18-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 71 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 72 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/resnet-18-pytorch/model.yml (licence: torchvision (BSD-3-Clause); model.yml line 52: license: https://raw.githubusercontent.com/pytorch/vision/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped); Accuracy Checker dataset definition imagenet_1000_classes at data/dataset_definitions.yml line 183
|
| 73 |
- `models/public/efficientnet-b0-pytorch/` (3 files): Open Model Zoo efficientnet-b0-pytorch: EfficientNet-B0 image classification (gen-efficientnet-pytorch weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 74 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 75 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 76 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/efficientnet-b0-pytorch/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 77 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 78 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/efficientnet-b0-pytorch/model.yml (licence: gen-efficientnet-pytorch (Apache-2.0); model.yml line 77: license: https://raw.githubusercontent.com/rwightman/gen-efficientnet-pytorch/a36e2b2cd1bd122a508a6fffeaa7606890f8c882/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped); Accuracy Checker dataset definition imagenet_1000_classes at data/dataset_definitions.yml line 183
|
| 79 |
- `models/public/regnetx-3.2gf/` (3 files): Open Model Zoo regnetx-3.2gf: RegNetX-3.2GF image classification (pycls weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 80 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 81 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 82 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/regnetx-3.2gf/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 83 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 84 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/regnetx-3.2gf/model.yml (licence: pycls (MIT); model.yml line 115: license: https://raw.githubusercontent.com/facebookresearch/pycls/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ImageNet ILSVRC2012 validation set (50000 images; licence-gated download from image-net.org, not shipped); Accuracy Checker dataset definition imagenet_1000_classes at data/dataset_definitions.yml line 183
|
| 85 |
- `models/public/detr-resnet50/` (3 files): Open Model Zoo detr-resnet50: DETR ResNet-50 object detection on COCO, published accuracy with the Accuracy Checker config that reproduces it.
|
| 86 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 87 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 88 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/detr-resnet50/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 89 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 90 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/detr-resnet50/model.yml (licence: facebookresearch/detr (Apache-2.0); model.yml line 97: license: https://raw.githubusercontent.com/facebookresearch/detr/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_91_classes at data/dataset_definitions.yml line 72
|
| 91 |
- `models/public/yolox-tiny/` (3 files): Open Model Zoo yolox-tiny: YOLOX-Tiny object detection on COCO, published accuracy with the Accuracy Checker config that reproduces it.
|
| 92 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 93 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 94 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/yolox-tiny/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 95 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 96 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/yolox-tiny/model.yml (licence: Megvii YOLOX (Apache-2.0); model.yml line 88: license: https://raw.githubusercontent.com/Megvii-BaseDetection/YOLOX/main/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_80_class_without_background at data/dataset_definitions.yml line 97
|
| 97 |
- `models/public/ssdlite_mobilenet_v2/` (3 files): Open Model Zoo ssdlite_mobilenet_v2: SSDLite MobileNetV2 object detection on COCO (TensorFlow Object Detection API weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 98 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 99 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 100 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/ssdlite_mobilenet_v2/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 101 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 102 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/ssdlite_mobilenet_v2/model.yml (licence: tensorflow/models (Apache-2.0); model.yml line 41: license: https://raw.githubusercontent.com/tensorflow/models/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_91_classes at data/dataset_definitions.yml line 72
|
| 103 |
- `models/public/yolo-v3-tiny-onnx/` (3 files): Open Model Zoo yolo-v3-tiny-onnx: Tiny YOLOv3 object detection on COCO (ONNX model zoo weights), published accuracy with the Accuracy Checker config that reproduces it.
|
| 104 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 105 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 106 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/yolo-v3-tiny-onnx/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 107 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 108 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/yolo-v3-tiny-onnx/model.yml (licence: onnx/models (Apache-2.0); model.yml line 41: license: https://raw.githubusercontent.com/onnx/models/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_80_class_without_background at data/dataset_definitions.yml line 97
|
| 109 |
- `models/public/efficientdet-d0-tf/` (3 files): Open Model Zoo efficientdet-d0-tf: EfficientDet-D0 object detection on COCO (google/automl weights, TensorFlow), published accuracy with the Accuracy Checker config that reproduces it.
|
| 110 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 111 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 112 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/efficientdet-d0-tf/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 113 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 114 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/efficientdet-d0-tf/model.yml (licence: google/automl (Apache-2.0); model.yml line 36: license: https://raw.githubusercontent.com/google/automl/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_90_class_without_background at data/dataset_definitions.yml line 119
|
| 115 |
- `models/public/nanodet-plus-m-1.5x-416/` (3 files): Open Model Zoo nanodet-plus-m-1.5x-416: NanoDet-Plus-m-1.5x-416 object detection on COCO, published accuracy with the Accuracy Checker config that reproduces it.
|
| 116 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 117 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 118 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/nanodet-plus-m-1.5x-416/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 119 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 120 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/nanodet-plus-m-1.5x-416/model.yml (licence: RangiLyu/nanodet (Apache-2.0); model.yml line 251: license: https://raw.githubusercontent.com/RangiLyu/nanodet/main/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; COCO val2017 images and instances annotations (annotations CC BY 4.0, images under Flickr terms; not shipped); Accuracy Checker dataset definition ms_coco_detection_80_class_without_background at data/dataset_definitions.yml line 97
|
| 121 |
- `models/public/deeplabv3/` (3 files): Open Model Zoo deeplabv3: DeepLabV3 MobileNetV2 semantic segmentation on PASCAL VOC 2012, published accuracy with the Accuracy Checker config that reproduces it.
|
| 122 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 123 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 124 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/deeplabv3/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 125 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 126 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/deeplabv3/model.yml (licence: tensorflow/models (Apache-2.0); model.yml line 38: license: https://raw.githubusercontent.com/tensorflow/models/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; PASCAL VOC 2012 segmentation validation split (images under Flickr terms, annotations from the VOC project; not shipped); Accuracy Checker dataset definition VOC2012_Segmentation at data/dataset_definitions.yml line 261
|
| 127 |
- `models/public/hrnet-v2-c1-segmentation/` (3 files): Open Model Zoo hrnet-v2-c1-segmentation: HRNetV2-C1 semantic segmentation on ADE20K, published accuracy with the Accuracy Checker config that reproduces it.
|
| 128 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 129 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 130 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/hrnet-v2-c1-segmentation/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside.
|
| 131 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 132 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/hrnet-v2-c1-segmentation/model.yml (licence: CSAILVision semantic-segmentation-pytorch (BSD-3-Clause); model.yml line 122: license: https://raw.githubusercontent.com/CSAILVision/semantic-segmentation-pytorch/9aff40de31ee4b21f18514d31e5d6e4ba056924d/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; ADE20K ADEChallengeData2016 validation split (2000 images, MIT CSAIL terms; not shipped); Accuracy Checker dataset definition ADEChallengeData2016 at data/dataset_definitions.yml line 1237
|
| 133 |
- `models/public/mozilla-deepspeech-0.8.2/` (3 files): Open Model Zoo mozilla-deepspeech-0.8.2: Mozilla DeepSpeech 0.8.2 speech recognition on LibriSpeech test-clean (with language model), published accuracy with the Accuracy Checker config that reproduces it.
|
| 134 |
Format: Markdown model card with an accuracy table, Accuracy Checker YAML config with reference values, model.yml download and conversion recipe
|
| 135 |
Origin: accuracy measured and published by the Open Model Zoo maintainers for the converted model; the reference values in accuracy-check.yml are the same numbers
|
|
|
|
|
|
|
| 136 |
Check: Download and convert the model with omz_downloader and omz_converter at the pinned commit, run accuracy_check -c models/public/mozilla-deepspeech-0.8.2/accuracy-check.yml on the CPU device with the dataset in dependencies; the metric printed for each reference must equal the reference within half a unit of the last digit printed in the README. Annotation caches are produced by the judge's own run, never taken from outside. The config decodes with ctc_beam_search_decoder_with_lm, beam_size 32 (accuracy-check.yml lines 6 and 9), which matches README line 27 (8.39%, beam_width = 32, Python CTC decoder); README lines 28 and 29 (6.13% and 6.15% at beam_width 500) are other settings and are not the reference.
|
| 137 |
Companions (not included): data/dataset_definitions.yml; tools/accuracy_checker/README.md; LICENSE
|
| 138 |
Dependencies: OpenVINO Open Model Zoo tools at commit a6946b6d6ce42cbf4278df20275fab199655fc7d: omz_downloader and omz_converter (model.yml) and accuracy_check (tools/accuracy_checker), Python; OpenVINO runtime CPU plugin, version to be pinned with the build; original model weights downloaded by omz_downloader from the URLs in models/public/mozilla-deepspeech-0.8.2/model.yml (licence: mozilla/DeepSpeech (MPL-2.0); model.yml line 50: license: https://raw.githubusercontent.com/mozilla/DeepSpeech/master/LICENSE; licence file fetched this round); the URLs are pinned by the sha384 checksum fields in model.yml; LibriSpeech test-clean (CC BY 4.0, not shipped); Accuracy Checker dataset definition librispeech-test-clean at data/dataset_definitions.yml line 1111
|
campaign_files/2026-10-op-rule-calculations-2/openfisca-france-tests/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: reference_amounts_in_test. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** a Python script applying the 2011 bracket schedule, the 10% professional expense allowance cap and the CEHR bands; or the openfisca test runner in a terminal
|
| 14 |
|
| 15 |
**Licence (SPDX):** AGPL-3.0-only
|
|
@@ -21,8 +19,6 @@ Files:
|
|
| 21 |
- `tests/formulas/irpp_cehr.yaml`: French exceptional contribution on high incomes (contribution exceptionnelle sur les hauts revenus) and income tax due for 2011, 10 cases.
|
| 22 |
Format: OpenFisca YAML test file; case names in French
|
| 23 |
Origin: expected values written by the OpenFisca developers for the 2011 French income tax and CEHR (3% of reference income between 250,000 and 500,000 euros for a single person, 4% above; thresholds doubled for couples, CGI article 223 sexies). These expected values were worked out by the developers, not published by a tax authority; they are used as given, with no separate official calibration.
|
| 24 |
-
Reference result: reference_amounts_in_test
|
| 25 |
-
Reference source: outputs contribution_exceptionnelle_hauts_revenus or impot_revenu_restant_a_payer of each case (negative impot_revenu_restant_a_payer means tax owed)
|
| 26 |
Check: give the agent each household's salary and prior-year reference income for 2011; it reports the CEHR or the income tax still due; pass when the absolute difference from the expected amount is within the margin each case sets
|
| 27 |
Tools: a Python script applying the 2011 bracket schedule, the 10% professional expense allowance cap and the CEHR bands; or the openfisca test runner in a terminal
|
| 28 |
Dependencies: the 2011 French income tax parameters (openfisca-france parameters; not packed)
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** a Python script applying the 2011 bracket schedule, the 10% professional expense allowance cap and the CEHR bands; or the openfisca test runner in a terminal
|
| 12 |
|
| 13 |
**Licence (SPDX):** AGPL-3.0-only
|
|
|
|
| 19 |
- `tests/formulas/irpp_cehr.yaml`: French exceptional contribution on high incomes (contribution exceptionnelle sur les hauts revenus) and income tax due for 2011, 10 cases.
|
| 20 |
Format: OpenFisca YAML test file; case names in French
|
| 21 |
Origin: expected values written by the OpenFisca developers for the 2011 French income tax and CEHR (3% of reference income between 250,000 and 500,000 euros for a single person, 4% above; thresholds doubled for couples, CGI article 223 sexies). These expected values were worked out by the developers, not published by a tax authority; they are used as given, with no separate official calibration.
|
|
|
|
|
|
|
| 22 |
Check: give the agent each household's salary and prior-year reference income for 2011; it reports the CEHR or the income tax still due; pass when the absolute difference from the expected amount is within the margin each case sets
|
| 23 |
Tools: a Python script applying the 2011 bracket schedule, the 10% professional expense allowance cap and the CEHR bands; or the openfisca test runner in a terminal
|
| 24 |
Dependencies: the 2011 French income tax parameters (openfisca-france parameters; not packed)
|
campaign_files/2026-10-op-rule-calculations-2/openintro-loans-full-schema/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: published_metric_values. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** read the CSV with Python csv; derive bankruptcy = public_record_bankrupt > 0 and credit_util = total_credit_utilized / total_credit_limit; fit OLS interest_rate ~ bankruptcy, ~ verified_income and the full model of IMS Table 8.5; write coefficient tables and adjusted R2 to JSON
|
| 14 |
|
| 15 |
**Licence (SPDX):** NOASSERTION
|
|
@@ -21,7 +19,5 @@ Files:
|
|
| 21 |
- `data/csv/loans_full_schema.csv`: loans_full_schema: 10,000 Lending Club loans with 55 variables, with the published interest-rate regression tables of IMS chapter 8.
|
| 22 |
Format: CSV, 10,000 rows plus header, 55 variables, 2,548,030 bytes
|
| 23 |
Origin: real Lending Club loans issued in the first quarter of 2018 (data page: "This data set represents thousands of loans made through the Lending Club platform")
|
| 24 |
-
Reference result: published_metric_values
|
| 25 |
-
Reference source: IMS model-mlr.html Table 8.3 (interest_rate ~ bankruptcy), Table 8.4 (~ verified_income) and Tables 8.5 and 8.6 (full model with debt_to_income, credit_util, bankruptcy, term, credit_checks, issue_month)
|
| 26 |
Check: give the CSV; ask for the three OLS fits with the stated derived variables (rows with missing debt_to_income or credit_util dropped, n 9,974 for the full model); compare coefficients, SE and t with the book to their printed precision
|
| 27 |
Tools: read the CSV with Python csv; derive bankruptcy = public_record_bankrupt > 0 and credit_util = total_credit_utilized / total_credit_limit; fit OLS interest_rate ~ bankruptcy, ~ verified_income and the full model of IMS Table 8.5; write coefficient tables and adjusted R2 to JSON
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** read the CSV with Python csv; derive bankruptcy = public_record_bankrupt > 0 and credit_util = total_credit_utilized / total_credit_limit; fit OLS interest_rate ~ bankruptcy, ~ verified_income and the full model of IMS Table 8.5; write coefficient tables and adjusted R2 to JSON
|
| 12 |
|
| 13 |
**Licence (SPDX):** NOASSERTION
|
|
|
|
| 19 |
- `data/csv/loans_full_schema.csv`: loans_full_schema: 10,000 Lending Club loans with 55 variables, with the published interest-rate regression tables of IMS chapter 8.
|
| 20 |
Format: CSV, 10,000 rows plus header, 55 variables, 2,548,030 bytes
|
| 21 |
Origin: real Lending Club loans issued in the first quarter of 2018 (data page: "This data set represents thousands of loans made through the Lending Club platform")
|
|
|
|
|
|
|
| 22 |
Check: give the CSV; ask for the three OLS fits with the stated derived variables (rows with missing debt_to_income or credit_util dropped, n 9,974 for the full model); compare coefficients, SE and t with the book to their printed precision
|
| 23 |
Tools: read the CSV with Python csv; derive bankruptcy = public_record_bankrupt > 0 and credit_util = total_credit_utilized / total_credit_limit; fit OLS interest_rate ~ bankruptcy, ~ verified_income and the full model of IMS Table 8.5; write coefficient tables and adjusted R2 to JSON
|
campaign_files/2026-10-op-rule-calculations-2/openintro-mariokart/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: published_metric_values. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** read the CSV with Python csv; drop the two auctions with total_pr >= 100; fit OLS total_pr ~ cond_new and total_pr ~ cond_new + stock_photo + duration + wheels; write coefficient tables to JSON
|
| 14 |
|
| 15 |
**Licence (SPDX):** NOASSERTION
|
|
@@ -21,7 +19,5 @@ Files:
|
|
| 21 |
- `data/csv/mariokart.csv`: Mariokart: 143 eBay auctions of Mario Kart for Wii (Oct 2009) with total price, condition, photo, duration and wheels, with the published regression tables of IMS chapter 27.
|
| 22 |
Format: CSV, 143 rows plus header, 13,795 bytes
|
| 23 |
Origin: real eBay auction transactions collected in October 2009
|
| 24 |
-
Reference result: published_metric_values
|
| 25 |
-
Reference source: IMS inf-model-applications.html Table 27.3 (total_pr ~ cond_new) and Table 27.4 (total_pr ~ cond_new + stock_photo + duration + wheels), fitted on the 141 auctions left after the two outliers
|
| 26 |
Check: give the CSV; ask to drop the two outlier auctions (total_pr 100 or more, they bundle other items) and fit the two OLS models; compare coefficients, SE and t with the book to their printed precision
|
| 27 |
Tools: read the CSV with Python csv; drop the two auctions with total_pr >= 100; fit OLS total_pr ~ cond_new and total_pr ~ cond_new + stock_photo + duration + wheels; write coefficient tables to JSON
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** read the CSV with Python csv; drop the two auctions with total_pr >= 100; fit OLS total_pr ~ cond_new and total_pr ~ cond_new + stock_photo + duration + wheels; write coefficient tables to JSON
|
| 12 |
|
| 13 |
**Licence (SPDX):** NOASSERTION
|
|
|
|
| 19 |
- `data/csv/mariokart.csv`: Mariokart: 143 eBay auctions of Mario Kart for Wii (Oct 2009) with total price, condition, photo, duration and wheels, with the published regression tables of IMS chapter 27.
|
| 20 |
Format: CSV, 143 rows plus header, 13,795 bytes
|
| 21 |
Origin: real eBay auction transactions collected in October 2009
|
|
|
|
|
|
|
| 22 |
Check: give the CSV; ask to drop the two outlier auctions (total_pr 100 or more, they bundle other items) and fit the two OLS models; compare coefficients, SE and t with the book to their printed precision
|
| 23 |
Tools: read the CSV with Python csv; drop the two auctions with total_pr >= 100; fit OLS total_pr ~ cond_new and total_pr ~ cond_new + stock_photo + duration + wheels; write coefficient tables to JSON
|
campaign_files/2026-10-s-security-detection-2/opa-gatekeeper-library-with-gator-suite-tests/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: per_resource_verdict. Where the reference result comes from and how a solution is checked is written per unit in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: Task: given the template, the constraint and the 6 sample objects (file names replaced by neutral ids), say for each whether the constraint reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 6.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,8 +19,6 @@ Units:
|
|
| 21 |
- `library/pod-security-policy/proc-mount/` (11 files): Gatekeeper K8sPSPProcMount: decide for 6 Pod admission requests whether the proc-mount constraint reports a violation.
|
| 22 |
Format: Gatekeeper ConstraintTemplate YAML with embedded Rego, a constraint (procMount Default, exemptImages safeimages.com/*), 6 sample objects, and suite.yaml (expected violations yes or no per case)
|
| 23 |
Origin: test fixture written by the rule maintainers to pin the expected findings of their own rule (synthetic, not recorded production code or telemetry)
|
| 24 |
-
Reference result: per_resource_verdict
|
| 25 |
-
Reference source: suite.yaml: violations yes for example-disallowed (line 13) and disallowed-ephemeral (line 25); violations no for example-allowed, example-allowed-missing, update, image-exempt-prefix-match (lines 17, 21, 29, 33)
|
| 26 |
Check: Task: given the template, the constraint and the 6 sample objects (file names replaced by neutral ids), say for each whether the constraint reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 6
|
| 27 |
Companions (not included): LICENSE
|
| 28 |
Dependencies: gator CLI (Gatekeeper) at this commit (Apache-2.0): needed only to rerun the scan; not used by the judge; version pinned when the pool is built; not packed
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** The unit is self-contained. 1 of 1 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: Task: given the template, the constraint and the 6 sample objects (file names replaced by neutral ids), say for each whether the constraint reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 6.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `library/pod-security-policy/proc-mount/` (11 files): Gatekeeper K8sPSPProcMount: decide for 6 Pod admission requests whether the proc-mount constraint reports a violation.
|
| 20 |
Format: Gatekeeper ConstraintTemplate YAML with embedded Rego, a constraint (procMount Default, exemptImages safeimages.com/*), 6 sample objects, and suite.yaml (expected violations yes or no per case)
|
| 21 |
Origin: test fixture written by the rule maintainers to pin the expected findings of their own rule (synthetic, not recorded production code or telemetry)
|
|
|
|
|
|
|
| 22 |
Check: Task: given the template, the constraint and the 6 sample objects (file names replaced by neutral ids), say for each whether the constraint reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 6
|
| 23 |
Companions (not included): LICENSE
|
| 24 |
Dependencies: gator CLI (Gatekeeper) at this commit (Apache-2.0): needed only to rerun the scan; not used by the judge; version pinned when the pool is built; not packed
|
campaign_files/2026-10-s-security-detection-2/opa-gatekeeper-policy-library/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Each unit is self-contained. 2 of 2 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: per_case_verdict. Where the reference result comes from and how a solution is checked is written per unit in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: Task: for each (constraint, object) pair of suite.yaml, with case names hidden, say whether the template reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 8; Task: for each (constraint, Ingress) pair (case names hidden) say whether a violation is reported. Score: exact match of yes or no against suite.yaml; accuracy over 4.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,16 +19,12 @@ Units:
|
|
| 21 |
- `library/pod-security-policy/read-only-root-filesystem/` (13 files): Gatekeeper read-only-root-filesystem: decide for 8 test cases (3 constraints) whether the constraint reports a violation.
|
| 22 |
Format: Rego ConstraintTemplate (template.yaml), sample constraints and objects under samples/, and suite.yaml giving the expected violations yes or no per case
|
| 23 |
Origin: test fixture written by the rule maintainers to pin the expected findings of their own rule (synthetic, not recorded production code or telemetry)
|
| 24 |
-
Reference result: per_case_verdict
|
| 25 |
-
Reference source: suite.yaml: violations yes for example-disallowed (line 13), disallowed-ephemeral (line 22), image-with-different-prefix-must-set-readOnlyRootFilesystem (line 51); violations no for example-allowed (line 18), update (line 27), exact-exemption (line 31), allow-normally-disallowed (line 39), image-with-exempt-prefix-readOnlyRootFilesystem-not-required (line 47)
|
| 26 |
Check: Task: for each (constraint, object) pair of suite.yaml, with case names hidden, say whether the template reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 8
|
| 27 |
Companions (not included): LICENSE
|
| 28 |
Dependencies: gator CLI (Gatekeeper) at this commit (Apache-2.0): needed only to rerun the scan; not used by the judge; version pinned when the pool is built; not packed
|
| 29 |
- `library/general/httpsonly/` (10 files): Gatekeeper httpsonly: decide for 4 Ingress objects under 2 constraints whether the Ingress is HTTPS only.
|
| 30 |
Format: Rego ConstraintTemplate (template.yaml), two sample constraints (tlsOptional false and true) with allowed and disallowed Ingress objects, and suite.yaml with expected violations
|
| 31 |
Origin: test fixture written by the rule maintainers to pin the expected findings of their own rule (synthetic, not recorded production code or telemetry)
|
| 32 |
-
Reference result: per_case_verdict
|
| 33 |
-
Reference source: suite.yaml: violations no for example-allowed (line 13) and example-allowed-tls-optional (line 25); violations yes for example-disallowed (line 17) and example-disallowed-tls-optional (line 29)
|
| 34 |
Check: Task: for each (constraint, Ingress) pair (case names hidden) say whether a violation is reported. Score: exact match of yes or no against suite.yaml; accuracy over 4
|
| 35 |
Companions (not included): LICENSE
|
| 36 |
Dependencies: gator CLI (Gatekeeper) at this commit (Apache-2.0): needed only to rerun the scan; not used by the judge; version pinned when the pool is built; not packed
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Each unit is self-contained. 2 of 2 units name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per unit in the notes. Dependencies named upstream are listed per unit in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: Task: for each (constraint, object) pair of suite.yaml, with case names hidden, say whether the template reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 8; Task: for each (constraint, Ingress) pair (case names hidden) say whether a violation is reported. Score: exact match of yes or no against suite.yaml; accuracy over 4.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `library/pod-security-policy/read-only-root-filesystem/` (13 files): Gatekeeper read-only-root-filesystem: decide for 8 test cases (3 constraints) whether the constraint reports a violation.
|
| 20 |
Format: Rego ConstraintTemplate (template.yaml), sample constraints and objects under samples/, and suite.yaml giving the expected violations yes or no per case
|
| 21 |
Origin: test fixture written by the rule maintainers to pin the expected findings of their own rule (synthetic, not recorded production code or telemetry)
|
|
|
|
|
|
|
| 22 |
Check: Task: for each (constraint, object) pair of suite.yaml, with case names hidden, say whether the template reports a violation. Score: exact match of yes or no against suite.yaml; accuracy over 8
|
| 23 |
Companions (not included): LICENSE
|
| 24 |
Dependencies: gator CLI (Gatekeeper) at this commit (Apache-2.0): needed only to rerun the scan; not used by the judge; version pinned when the pool is built; not packed
|
| 25 |
- `library/general/httpsonly/` (10 files): Gatekeeper httpsonly: decide for 4 Ingress objects under 2 constraints whether the Ingress is HTTPS only.
|
| 26 |
Format: Rego ConstraintTemplate (template.yaml), two sample constraints (tlsOptional false and true) with allowed and disallowed Ingress objects, and suite.yaml with expected violations
|
| 27 |
Origin: test fixture written by the rule maintainers to pin the expected findings of their own rule (synthetic, not recorded production code or telemetry)
|
|
|
|
|
|
|
| 28 |
Check: Task: for each (constraint, Ingress) pair (case names hidden) say whether a violation is reported. Score: exact match of yes or no against suite.yaml; accuracy over 4
|
| 29 |
Companions (not included): LICENSE
|
| 30 |
Dependencies: gator CLI (Gatekeeper) at this commit (Apache-2.0): needed only to rerun the scan; not used by the judge; version pinned when the pool is built; not packed
|
campaign_files/2026-10-sc-science-reference-data-2/openfermion-testing-data/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: reference_numbers_with_tolerance. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: Give the agent the H2 geometry (0.7414 A), basis STO-3G and multiplicity 1 and ask for the HF, MP2, CISD and CCSD total energies; each must agree with the reference to 4 decimal places as unittest assertAlmostEqual(places=4) checks, that is round(actual - expected, 4) == 0; the HDF5 values are held back from the agent.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,8 +19,6 @@ Files:
|
|
| 21 |
- `src/openfermion/testing/data/H2_sto-3g_singlet_0.7414.hdf5`: OpenFermion H2 STO-3G singlet at 0.7414 A: Hartree-Fock, MP2, CISD and CCSD energies with stored integrals.
|
| 22 |
Format: HDF5 (MolecularData: geometry, basis, one- and two-body integrals, energies; 45,872 bytes; no pickle) and Python test source
|
| 23 |
Origin: test data of OpenFermion (tool_fixture): MolecularData computed with an electronic structure plugin and committed as HDF5
|
| 24 |
-
Reference result: reference_numbers_with_tolerance
|
| 25 |
-
Reference source: src/openfermion/chem/molecular_data_test.py lines 274 to 277: hf_energy -1.1167, mp2_energy -1.1299, cisd_energy -1.1373, ccsd_energy -1.1373 (Hartree); geometry at line 45 [('H', (0,0,0)), ('H', (0,0,0.7414))], basis sto-3g, singlet
|
| 26 |
Check: give the agent the H2 geometry (0.7414 A), basis STO-3G and multiplicity 1 and ask for the HF, MP2, CISD and CCSD total energies; each must agree with the reference to 4 decimal places as unittest assertAlmostEqual(places=4) checks, that is round(actual - expected, 4) == 0; the HDF5 values are held back from the agent
|
| 27 |
Companions (not included): src/openfermion/chem/molecular_data_test.py
|
| 28 |
Dependencies: an open-source quantum chemistry code; h5py to read the HDF5
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 1 of 1 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: Give the agent the H2 geometry (0.7414 A), basis STO-3G and multiplicity 1 and ask for the HF, MP2, CISD and CCSD total energies; each must agree with the reference to 4 decimal places as unittest assertAlmostEqual(places=4) checks, that is round(actual - expected, 4) == 0; the HDF5 values are held back from the agent.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `src/openfermion/testing/data/H2_sto-3g_singlet_0.7414.hdf5`: OpenFermion H2 STO-3G singlet at 0.7414 A: Hartree-Fock, MP2, CISD and CCSD energies with stored integrals.
|
| 20 |
Format: HDF5 (MolecularData: geometry, basis, one- and two-body integrals, energies; 45,872 bytes; no pickle) and Python test source
|
| 21 |
Origin: test data of OpenFermion (tool_fixture): MolecularData computed with an electronic structure plugin and committed as HDF5
|
|
|
|
|
|
|
| 22 |
Check: give the agent the H2 geometry (0.7414 A), basis STO-3G and multiplicity 1 and ask for the HF, MP2, CISD and CCSD total energies; each must agree with the reference to 4 decimal places as unittest assertAlmostEqual(places=4) checks, that is round(actual - expected, 4) == 0; the HDF5 values are held back from the agent
|
| 23 |
Companions (not included): src/openfermion/chem/molecular_data_test.py
|
| 24 |
Dependencies: an open-source quantum chemistry code; h5py to read the HDF5
|
campaign_files/2026-10-sc-science-reference-data/openbabel-test-files/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: reference_values_in_file; label_per_item. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: One file (forcefield.sdf, 18 molecules), two question sets. Question A (MMFF94 energies, mmff94results.txt): agent computes the MMFF94 energy per molecule (obenergy -ff MMFF94 or RDKit MMFF) and reports in order; a molecule fails as in ffmmff94.cpp line 145 'if ( fabs(atof(buffer) - energy ) > 1.0e-3)' (kcal/mol). Question B (Gasteiger charges and dipoles, charge-gasteiger.txt and dipole-gasteiger.txt): agent computes Gasteiger-Marsili charges (obabel --partialcharge gasteiger or own code) and the charge dipole per molecule; an atom fails as in charge_gasteiger.cpp line 146 'if ( fabs(atom->GetPartialCharge() - atof(buffer)) > 1.0e-3 )' and a dipole fails as in line 131 'if ( fabs(result.length_2()) > 1.0e-4)' (result is the difference between computed and reference dipole vectors); no relaxed tolerance; agent parses each SMILES and reports formula (exact string), MW and exact mass; score formula exact and masses within 1e-3; shuffle both lists without labels; agent marks each pattern valid or invalid (may use a SMARTS parser); score accuracy, and count invalid patterns wrongly accepted (false accept rate).
|
| 14 |
|
| 15 |
**Licence (SPDX):** GPL-2.0-only
|
|
@@ -21,21 +19,15 @@ Files:
|
|
| 21 |
- `test/files/forcefield.sdf`: Open Babel forcefield.sdf (18 molecules): MMFF94 total energies, Gasteiger partial charges and dipole vectors with reference results.
|
| 22 |
Format: SDF (18 molecules) + plain text value lists
|
| 23 |
Origin: Open Babel test fixture with reference energies (upstream regression-test input with the reference values the project's own test harness checks against); the 18 molecules were built by an Open Babel developer (each title line is a local Ghemical project path, /Volumes/Home/ghutchis/Desktop/UFF/<name>.gpr, and the file header is OpenBabel-generated), so no third-party origin
|
| 24 |
-
Reference result: reference_values_in_file
|
| 25 |
-
Reference source: Question A: mmff94results.txt, one MMFF94 energy (kcal/mol) per molecule in SDF order (ffmmff94.cpp line 145, 1.0e-3). Question B: charge-gasteiger.txt, one Gasteiger charge per atom in file order (608 atoms); dipole-gasteiger.txt, one dipole vector per molecule (charge_gasteiger.cpp line 146, 1.0e-3; line 131, 1.0e-4)
|
| 26 |
Check: one file (forcefield.sdf, 18 molecules), two question sets. Question A (MMFF94 energies, mmff94results.txt): agent computes the MMFF94 energy per molecule (obenergy -ff MMFF94 or RDKit MMFF) and reports in order; a molecule fails as in ffmmff94.cpp line 145 'if ( fabs(atof(buffer) - energy ) > 1.0e-3)' (kcal/mol). Question B (Gasteiger charges and dipoles, charge-gasteiger.txt and dipole-gasteiger.txt): agent computes Gasteiger-Marsili charges (obabel --partialcharge gasteiger or own code) and the charge dipole per molecule; an atom fails as in charge_gasteiger.cpp line 146 'if ( fabs(atom->GetPartialCharge() - atof(buffer)) > 1.0e-3 )' and a dipole fails as in line 131 'if ( fabs(result.length_2()) > 1.0e-4)' (result is the difference between computed and reference dipole vectors); no relaxed tolerance
|
| 27 |
Companions (not included): test/files/mmff94results.txt (18 energies); test/ffmmff94.cpp (tolerance 1.0e-3, line 145); test/files/charge-gasteiger.txt (608 atomic charges); test/files/dipole-gasteiger.txt (18 dipole vectors); test/charge_gasteiger.cpp (tolerances 1.0e-3 charge, line 146; 1.0e-4 dipole, line 131)
|
| 28 |
- `test/files/attype.00.smi`: Open Babel attype.00.smi: molecular formula, molecular weight and exact mass for 274 SMILES with reference table.
|
| 29 |
Format: SMILES list + plain text table
|
| 30 |
Origin: Open Babel test fixture (upstream regression-test input with the reference values the project's own test harness checks against)
|
| 31 |
-
Reference result: reference_values_in_file
|
| 32 |
-
Reference source: formularesults.txt: per molecule Hill formula, molecular weight and exact mass; formula.cpp checks to 1e-3
|
| 33 |
Check: agent parses each SMILES and reports formula (exact string), MW and exact mass; score formula exact and masses within 1e-3
|
| 34 |
Companions (not included): test/files/formularesults.txt (formula, MW, exact mass per molecule); test/formula.cpp (tolerance 1e-3)
|
| 35 |
- `test/files/validsmarts.txt`: Open Babel SMARTS validity: 231 valid and 33 invalid SMARTS patterns to classify.
|
| 36 |
Format: plain text pattern lists
|
| 37 |
Origin: Open Babel parser test lists, hand-written by developers
|
| 38 |
-
Reference result: label_per_item
|
| 39 |
-
Reference source: membership in validsmarts.txt (231 patterns after dropping 16 comment lines) versus invalid-smarts.txt (33); tests require parse success or failure respectively
|
| 40 |
Check: shuffle both lists without labels; agent marks each pattern valid or invalid (may use a SMARTS parser); score accuracy, and count invalid patterns wrongly accepted (false accept rate)
|
| 41 |
Companions (not included): test/files/invalid-smarts.txt; test/smartsparse.cpp; test/invalidsmarts.cpp
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Not self-contained: each file is used together with other files from the same upstream. 3 of 3 files name companion files (other upstream files read alongside); companions are not included in this delivery and are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: One file (forcefield.sdf, 18 molecules), two question sets. Question A (MMFF94 energies, mmff94results.txt): agent computes the MMFF94 energy per molecule (obenergy -ff MMFF94 or RDKit MMFF) and reports in order; a molecule fails as in ffmmff94.cpp line 145 'if ( fabs(atof(buffer) - energy ) > 1.0e-3)' (kcal/mol). Question B (Gasteiger charges and dipoles, charge-gasteiger.txt and dipole-gasteiger.txt): agent computes Gasteiger-Marsili charges (obabel --partialcharge gasteiger or own code) and the charge dipole per molecule; an atom fails as in charge_gasteiger.cpp line 146 'if ( fabs(atom->GetPartialCharge() - atof(buffer)) > 1.0e-3 )' and a dipole fails as in line 131 'if ( fabs(result.length_2()) > 1.0e-4)' (result is the difference between computed and reference dipole vectors); no relaxed tolerance; agent parses each SMILES and reports formula (exact string), MW and exact mass; score formula exact and masses within 1e-3; shuffle both lists without labels; agent marks each pattern valid or invalid (may use a SMARTS parser); score accuracy, and count invalid patterns wrongly accepted (false accept rate).
|
| 12 |
|
| 13 |
**Licence (SPDX):** GPL-2.0-only
|
|
|
|
| 19 |
- `test/files/forcefield.sdf`: Open Babel forcefield.sdf (18 molecules): MMFF94 total energies, Gasteiger partial charges and dipole vectors with reference results.
|
| 20 |
Format: SDF (18 molecules) + plain text value lists
|
| 21 |
Origin: Open Babel test fixture with reference energies (upstream regression-test input with the reference values the project's own test harness checks against); the 18 molecules were built by an Open Babel developer (each title line is a local Ghemical project path, /Volumes/Home/ghutchis/Desktop/UFF/<name>.gpr, and the file header is OpenBabel-generated), so no third-party origin
|
|
|
|
|
|
|
| 22 |
Check: one file (forcefield.sdf, 18 molecules), two question sets. Question A (MMFF94 energies, mmff94results.txt): agent computes the MMFF94 energy per molecule (obenergy -ff MMFF94 or RDKit MMFF) and reports in order; a molecule fails as in ffmmff94.cpp line 145 'if ( fabs(atof(buffer) - energy ) > 1.0e-3)' (kcal/mol). Question B (Gasteiger charges and dipoles, charge-gasteiger.txt and dipole-gasteiger.txt): agent computes Gasteiger-Marsili charges (obabel --partialcharge gasteiger or own code) and the charge dipole per molecule; an atom fails as in charge_gasteiger.cpp line 146 'if ( fabs(atom->GetPartialCharge() - atof(buffer)) > 1.0e-3 )' and a dipole fails as in line 131 'if ( fabs(result.length_2()) > 1.0e-4)' (result is the difference between computed and reference dipole vectors); no relaxed tolerance
|
| 23 |
Companions (not included): test/files/mmff94results.txt (18 energies); test/ffmmff94.cpp (tolerance 1.0e-3, line 145); test/files/charge-gasteiger.txt (608 atomic charges); test/files/dipole-gasteiger.txt (18 dipole vectors); test/charge_gasteiger.cpp (tolerances 1.0e-3 charge, line 146; 1.0e-4 dipole, line 131)
|
| 24 |
- `test/files/attype.00.smi`: Open Babel attype.00.smi: molecular formula, molecular weight and exact mass for 274 SMILES with reference table.
|
| 25 |
Format: SMILES list + plain text table
|
| 26 |
Origin: Open Babel test fixture (upstream regression-test input with the reference values the project's own test harness checks against)
|
|
|
|
|
|
|
| 27 |
Check: agent parses each SMILES and reports formula (exact string), MW and exact mass; score formula exact and masses within 1e-3
|
| 28 |
Companions (not included): test/files/formularesults.txt (formula, MW, exact mass per molecule); test/formula.cpp (tolerance 1e-3)
|
| 29 |
- `test/files/validsmarts.txt`: Open Babel SMARTS validity: 231 valid and 33 invalid SMARTS patterns to classify.
|
| 30 |
Format: plain text pattern lists
|
| 31 |
Origin: Open Babel parser test lists, hand-written by developers
|
|
|
|
|
|
|
| 32 |
Check: shuffle both lists without labels; agent marks each pattern valid or invalid (may use a SMARTS parser); score accuracy, and count invalid patterns wrongly accepted (false accept rate)
|
| 33 |
Companions (not included): test/files/invalid-smarts.txt; test/smartsparse.cpp; test/invalidsmarts.cpp
|
campaign_files/2026-10-sw-tests-and-exercises-2/opencypher-tck/README.md
CHANGED
|
@@ -8,8 +8,6 @@
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
| 11 |
-
**Answers:** No answer files are included. Reference result: expected_output. Where the reference result comes from and how a solution is checked is written per file in the notes.
|
| 12 |
-
|
| 13 |
**Tools to open and check:** Check: Run each scenario on the graph database under test (set up the given graph, execute the query) and compare the returned rows with the expected table as a multiset when the step says in any order, or check the raised error type; exact values, no tolerance.
|
| 14 |
|
| 15 |
**Licence (SPDX):** Apache-2.0
|
|
@@ -21,7 +19,5 @@ Files:
|
|
| 21 |
- `tck/features/clauses/match/Match1.feature`: openCypher Technology Compatibility Kit, MATCH clause scenarios (Match1.feature): 11 Gherkin scenarios of graph setup, Cypher query and expected result table.
|
| 22 |
Format: Gherkin feature file; each scenario gives the graph (Given), the Cypher query (When executing query) and the expected result table or error (Then)
|
| 23 |
Origin: tool_fixture: an official test case of the project's own conformance or regression suite, with its expected result
|
| 24 |
-
Reference result: expected_output
|
| 25 |
-
Reference source: the Then step of each scenario: the expected result table (in any order) or the expected error type and phase
|
| 26 |
Check: run each scenario on the graph database under test (set up the given graph, execute the query) and compare the returned rows with the expected table as a multiset when the step says in any order, or check the raised error type; exact values, no tolerance
|
| 27 |
Dependencies: an openCypher implementation with a TCK runner
|
|
|
|
| 8 |
|
| 9 |
**Self-contained:** Whether the file needs other upstream files is not stated. Dependencies named upstream are listed per file in the notes.
|
| 10 |
|
|
|
|
|
|
|
| 11 |
**Tools to open and check:** Check: Run each scenario on the graph database under test (set up the given graph, execute the query) and compare the returned rows with the expected table as a multiset when the step says in any order, or check the raised error type; exact values, no tolerance.
|
| 12 |
|
| 13 |
**Licence (SPDX):** Apache-2.0
|
|
|
|
| 19 |
- `tck/features/clauses/match/Match1.feature`: openCypher Technology Compatibility Kit, MATCH clause scenarios (Match1.feature): 11 Gherkin scenarios of graph setup, Cypher query and expected result table.
|
| 20 |
Format: Gherkin feature file; each scenario gives the graph (Given), the Cypher query (When executing query) and the expected result table or error (Then)
|
| 21 |
Origin: tool_fixture: an official test case of the project's own conformance or regression suite, with its expected result
|
|
|
|
|
|
|
| 22 |
Check: run each scenario on the graph database under test (set up the given graph, execute the query) and compare the returned rows with the expected table as a multiset when the step says in any order, or check the raised error type; exact values, no tolerance
|
| 23 |
Dependencies: an openCypher implementation with a TCK runner
|
cards/0xdea-semgrep-rules.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = '0xdea-semgrep-rules'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = '0xdea-semgrep-rules'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = '0xdea-semgrep-rules'
|
cards/1brc-test-samples.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = '1brc-test-samples'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = '1brc-test-samples'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = '1brc-test-samples'
|
cards/2dpacklib.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = '2dpacklib'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = '2dpacklib'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = '2dpacklib'
|
cards/abacus-integrate-tests.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'abacus-integrate-tests'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'abacus-integrate-tests'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'abacus-integrate-tests'
|
cards/abhishekumarr00-supply-chain-mastery.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'abhishekumarr00-supply-chain-mastery'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'abhishekumarr00-supply-chain-mastery'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'abhishekumarr00-supply-chain-mastery'
|
cards/academy-bus-nj-gtfs.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'academy-bus-nj-gtfs'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'academy-bus-nj-gtfs'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'academy-bus-nj-gtfs'
|
cards/accordproject-cicero-template-library.md
CHANGED
|
@@ -52,10 +52,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 52 |
|
| 53 |
## Query this source at a fixed release
|
| 54 |
|
| 55 |
-
Pinned to the release tag `release-2026-11-11.
|
| 56 |
|
| 57 |
```sql
|
| 58 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 59 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 60 |
FROM read_parquet(getvariable('files'))
|
| 61 |
WHERE source = 'accordproject-cicero-template-library'
|
|
|
|
| 52 |
|
| 53 |
## Query this source at a fixed release
|
| 54 |
|
| 55 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 56 |
|
| 57 |
```sql
|
| 58 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'accordproject-cicero-template-library'));
|
| 59 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 60 |
FROM read_parquet(getvariable('files'))
|
| 61 |
WHERE source = 'accordproject-cicero-template-library'
|
cards/activity-schema-aggregate-in-between-1.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'activity-schema-aggregate-in-between-1'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'activity-schema-aggregate-in-between-1'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'activity-schema-aggregate-in-between-1'
|
cards/activity-schema-first-in-between-2.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'activity-schema-first-in-between-2'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'activity-schema-first-in-between-2'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'activity-schema-first-in-between-2'
|
cards/acwf-verification.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'acwf-verification'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'acwf-verification'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'acwf-verification'
|
cards/adflib-adf-deleted.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'adflib-adf-deleted'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'adflib-adf-deleted'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'adflib-adf-deleted'
|
cards/adflib-adf-malformed.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'adflib-adf-malformed'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'adflib-adf-malformed'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'adflib-adf-malformed'
|
cards/adventure-works-oltp-sample.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'adventure-works-oltp-sample'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'adventure-works-oltp-sample'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'adventure-works-oltp-sample'
|
cards/agat-tests.md
CHANGED
|
@@ -52,10 +52,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 52 |
|
| 53 |
## Query this source at a fixed release
|
| 54 |
|
| 55 |
-
Pinned to the release tag `release-2026-11-11.
|
| 56 |
|
| 57 |
```sql
|
| 58 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 59 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 60 |
FROM read_parquet(getvariable('files'))
|
| 61 |
WHERE source = 'agat-tests'
|
|
|
|
| 52 |
|
| 53 |
## Query this source at a fixed release
|
| 54 |
|
| 55 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 56 |
|
| 57 |
```sql
|
| 58 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'agat-tests'));
|
| 59 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 60 |
FROM read_parquet(getvariable('files'))
|
| 61 |
WHERE source = 'agat-tests'
|
cards/agl-gerrit-changes.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'agl-gerrit-changes'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'agl-gerrit-changes'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'agl-gerrit-changes'
|
cards/aiops2018-kpi.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'aiops2018-kpi'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'aiops2018-kpi'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'aiops2018-kpi'
|
cards/akretion-factur-x-libreoffice-extension.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'akretion-factur-x-libreoffice-extension'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'akretion-factur-x-libreoffice-extension'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'akretion-factur-x-libreoffice-extension'
|
cards/allenporter-ical-recurring-single-change.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'allenporter-ical-recurring-single-change'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'allenporter-ical-recurring-single-change'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'allenporter-ical-recurring-single-change'
|
cards/altlinux-devel.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'altlinux-devel'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'altlinux-devel'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'altlinux-devel'
|
cards/alumik-nkthesis-testfiles.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'alumik-nkthesis-testfiles'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'alumik-nkthesis-testfiles'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'alumik-nkthesis-testfiles'
|
cards/aml-dbt-rule-unit-tests.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'aml-dbt-rule-unit-tests'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'aml-dbt-rule-unit-tests'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'aml-dbt-rule-unit-tests'
|
cards/ancuber-cubst-tests.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'ancuber-cubst-tests'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'ancuber-cubst-tests'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'ancuber-cubst-tests'
|
cards/annabel-lee-wikisource-reading.md
CHANGED
|
@@ -50,10 +50,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
-
Pinned to the release tag `release-2026-11-11.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'annabel-lee-wikisource-reading'
|
|
|
|
| 50 |
|
| 51 |
## Query this source at a fixed release
|
| 52 |
|
| 53 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 54 |
|
| 55 |
```sql
|
| 56 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'annabel-lee-wikisource-reading'));
|
| 57 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 58 |
FROM read_parquet(getvariable('files'))
|
| 59 |
WHERE source = 'annabel-lee-wikisource-reading'
|
cards/ansible-galaxy-collections.md
CHANGED
|
@@ -48,10 +48,10 @@ Size quantiles: unknown (no size distribution recorded for this source).
|
|
| 48 |
|
| 49 |
## Query this source at a fixed release
|
| 50 |
|
| 51 |
-
Pinned to the release tag `release-2026-11-11.
|
| 52 |
|
| 53 |
```sql
|
| 54 |
-
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.
|
| 55 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 56 |
FROM read_parquet(getvariable('files'))
|
| 57 |
WHERE source = 'ansible-galaxy-collections'
|
|
|
|
| 48 |
|
| 49 |
## Query this source at a fixed release
|
| 50 |
|
| 51 |
+
Pinned to the release tag `release-2026-11-11.18`.
|
| 52 |
|
| 53 |
```sql
|
| 54 |
+
SET VARIABLE files = (SELECT list('hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/' || f) FROM (SELECT unnest(data_files) AS f FROM 'hf://datasets/Battam/training-data-foundation@release-2026-11-11.18/meta/catalog.parquet' WHERE source = 'ansible-galaxy-collections'));
|
| 55 |
SELECT uri, ext, size_bytes, lower(hex(sha256)) AS sha256
|
| 56 |
FROM read_parquet(getvariable('files'))
|
| 57 |
WHERE source = 'ansible-galaxy-collections'
|