Datasets:
Download cards/ctranslate2-engine-test-expected-values.yaml from Battam/training-data-foundation: direct link, hf CLI and curl.
- Browser
- Download file 11.2 kB
-
https://huggingface.co/datasets/Battam/training-data-foundation/resolve/main/cards/ctranslate2-engine-test-expected-values.yaml
- Command line
-
hf download hf://datasets/Battam/training-data-foundation/cards/ctranslate2-engine-test-expected-values.yaml
-
curl -L -o ctranslate2-engine-test-expected-values.yaml https://huggingface.co/datasets/Battam/training-data-foundation/resolve/main/cards/ctranslate2-engine-test-expected-values.yaml
11.2 kB
| # Source card (card_format agentic-env-index). Readers: agents that choose sources and fetch files to build | |
| # terminal tasks. Field names follow the corpus card format where the meaning is the same; provenance is where | |
| # the data comes from and its licence, labeling is who judged the card. | |
| # evidence.<field path>.fill: auto (computed from the index), judgment (from sample files downloaded and read), | |
| # derived (fixed rule over other fields), unfilled (no sample evidence yet, left empty on purpose; basis says | |
| # why and the plan), not_applicable (this class of card never has a value for the field; basis says why). | |
| # Sample numbers #NN refer to sample.files. Ratios are counts in the sample, not estimates for the source. | |
| card_version: 5 | |
| card_format: agentic-env-index | |
| source: ctranslate2-engine-test-expected-values | |
| title: "CTranslate2 engine tests: expected values for operators, layers and Whisper alignment" | |
| description: "Files published by SYSTRAN and The OpenNMT Authors (CTranslate2). Collection theme: Inference serving configs and benchmark results (serving-framework and inference-engine repositories with published results in configs, checkable on CPU)." | |
| audit: {status: draft, reviewed_by: null} | |
| provenance: | |
| url: "https://github.com/OpenNMT/CTranslate2" | |
| licence: MIT | |
| training_ok: "yes" | |
| redistribution_ok: "yes" | |
| licence_tier_counts: null | |
| license: | |
| note: "The units are under MIT, as recorded from their publishers; the licence file is campaign_files/2026-10-ml-serving-configs-2/ctranslate2-engine-test-expected-values/LICENSE." | |
| tier: permissive | |
| ship: "yes" | |
| unit: | |
| units: {unit: 3} | |
| size: {bytes: 91159, files: 5, truncated_listing: false} | |
| campaign: | |
| tag: "campaign:2026-10-ml-serving-configs-2" | |
| domain_folder: D4_machine-learning | |
| leaf: D4.2 | |
| unit: "One upstream file per unit, kept at its path in the upstream repository or site. Format: JSON test case (num_frames, text_tokens, expected_text_token_probs_sum, expected_alignments) with the NumPy audio array it belongs to; GoogleTest C++ source holding the inputs and expected StorageView values inline, with the comparison helpers header; GoogleTest C++ source holding inputs and expected values inline, with the comparison helpers header." | |
| origin: "Upstream: OpenNMT/CTranslate2. Kind: test or example files shipped with a tool. Expected values written by the CTranslate2 maintainers into the test sources." | |
| standalone: "Whether each file needs other upstream files is not stated. Each file names 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." | |
| tools: "Check: Convert openai/whisper-tiny.en with ctranslate2.converters.TransformersConverter, compute features of jfk.npy with transformers.WhisperProcessor at 16000 Hz, run Whisper.align on CPU with start sequence [50257], the text_tokens and num_frames of the JSON (lines 893 to 903); the alignment list must equal expected_alignments exactly (lines 911 to 913) and the sum of text_token_probs must be within 1e-3 of expected_text_token_probs_sum (lines 907 to 909). For a task, give the audio, the model and the tokens and hide the two expected fields. Load jfk.npy with numpy.load(allow_pickle=False). Build CTranslate2 at the pinned commit with -DBUILD_TESTS=ON and run ctranslate2_test --gtest_filter=*CPU*:OpTest.*; the CPU suites are instantiated at lines 1500 to 1503. Floating-point tests compare with expect_storage_eq at the suite error 1e-5, integer and exact tests with abs_diff 0 (test_utils.h lines 26 to 47 and 63 to 70; at abs_diff 0 floats use EXPECT_FLOAT_EQ); the sampling test Multinomial compares frequencies within 0.05 (line 1078). For a task, give the operator name and inputs of a test and hide its expected literal; the agent computes the outputs. Build CTranslate2 at the pinned commit with -DBUILD_TESTS=ON and run the LayerTest and CPU LayerDeviceFPTest suites (instantiated at lines 261 to 263). Integer position and padder tests compare exactly (expect_storage_eq default abs_diff 0); Alibi uses max(1e-5, 1e-4) (line 37); RotaryEmbedding uses the suite error 1e-5; the position encoder uses 1e-5 (lines 247 and 256). For a task, give the layer and inputs and hide the expected literal." | |
| files_in_this_repository: | |
| paths: [README.md, LICENSE] | |
| where: campaign_files/2026-10-ml-serving-configs-2/ctranslate2-engine-test-expected-values/ | |
| licence: "These files (README.md, the LICENSE summary when it is generated, and any file written for a package of it, such as a package README.md or run script) are released under CC0 1.0, like the catalog metadata, unless their row records another licence. The unit files themselves stay with their publishers under the licences listed above." | |
| rebuild: "See docs/selecting.md, section \"Rebuilding a batch from a collection tag\"." | |
| status: ok | |
| status_reasons: [] | |
| duplicate_of: null | |
| prior_taxon: null | |
| checks: | |
| errors: [] | |
| warnings: [] | |
| history: [] | |
| evidence: | |
| card_format: | |
| fill: derived | |
| basis: "fixed name of this card format; the corpus cards are another format and keep their own card_version numbers" | |
| checks.errors: {fill: derived, basis: "a card with a failing check is not written"} | |
| checks.history: | |
| fill: not_applicable | |
| basis: "the card generator has no repair rounds: a failing check stops the build instead of being repaired, so there is never a round to record; this holds only while the card generators have no repair step: if one is added, this entry leaves the list" | |
| checks.warnings: | |
| fill: derived | |
| basis: "licence warnings of a collection are listed in its ingest report, not on the card" | |
| derived.coverage: | |
| fill: unfilled | |
| basis: "scheduled: computed from label.unit_kinds and the pointer rows, and label.unit_kinds is not judged yet for this source; plan: waits for a later step that computes it from the judged fields" | |
| derived.unit_kinds: | |
| fill: unfilled | |
| basis: "scheduled: computed from label.unit_kinds and the pointer rows, and label.unit_kinds is not judged yet for this source; plan: waits for a later step that computes it from the judged fields" | |
| duplicate_of: {fill: derived, basis: "one card per source name in a collection"} | |
| label.contamination.benchmark: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.contamination.evidence: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.contamination.scope: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.domain_span: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.excluded_parts: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.notes: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.source_kind: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.summary: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| label.unit_kinds: | |
| fill: unfilled | |
| basis: "scheduled: the field applies to this source, but its sample files have not been read and judged yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| labeling.brief_sha: | |
| fill: unfilled | |
| basis: "scheduled: written with the judgment of this source, which has not been made yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| labeling.labeler: | |
| fill: unfilled | |
| basis: "scheduled: written with the judgment of this source, which has not been made yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| labeling.model: | |
| fill: unfilled | |
| basis: "scheduled: written with the judgment of this source, which has not been made yet; plan: judged from a stratified sample by one model, with key fields re-judged by a second model" | |
| license.ship: | |
| fill: derived | |
| basis: "derived from provenance.training_ok and redistribution_ok by the fixed pair-to-tier rule when the card was read; training_ok keeps the value written on the card" | |
| license.tier: | |
| fill: derived | |
| basis: "derived from provenance.training_ok and redistribution_ok by the fixed pair-to-tier rule when the card was read; training_ok keeps the value written on the card" | |
| prior_taxon: | |
| fill: not_applicable | |
| basis: "the prior taxon is a code from the seed list the corpus cards were started from; sources of a collection are not drawn from that list and never are: no step that adds sources draws from it; that list is private material, so even if the same source is also in it one day, its code does not go on a public card" | |
| probe: | |
| fill: unfilled | |
| basis: "scheduled: no review result is recorded for this source yet; plan: waits until these sources are judged, the two-model review is done and we spot-check them, then filled from that review (our own review, not a second model re-reading the whole card: two_model counts how many judged fields an independent second model agreed with; spot_check is the conclusion and date of our own spot check)" | |
| provenance.licence_tier_counts: | |
| fill: derived | |
| basis: "derived from provenance.training_ok and redistribution_ok by the fixed pair-to-tier rule when the card was read; training_ok keeps the value written on the card" | |
| size.truncated_listing: {fill: derived, basis: "unit counts come from every row of the collection"} | |
| status: {fill: derived, basis: "a selection card that fails the public text check is not written"} | |
| status_reasons: {fill: derived, basis: "empty when status is ok"} | |
| probe: null | |
| label: | |
| summary: null | |
| source_kind: null | |
| notes: null | |
| excluded_parts: null | |
| domain_span: null | |
| contamination: {scope: null, benchmark: null, evidence: null} | |
| unit_kinds: null | |
| derived: {coverage: null, unit_kinds: null} | |
| labeling: {labeler: null, model: null, brief_sha: null} | |