Dataset · Speech recognition
trn-indcnfr-hi-pilot
by Sulabh Katiyar sulabhkatiyar/trn-indcnfr-hi-pilot
Append-only teacher-output cache used to distill a ternary Hindi ASR student. It stores, per audio clip, ONLY: - rowid (a deterministic source-shard/position pointer), - top-k (k=64) CTC logits (vocab id + log-prob per kept entry, blank always kept), …
Dataset Card
By Sulabh Katiyar, published under cc-by-4.0, revision 94ef66d7bcd6.
trn-indcnfr-hi-pilot — Teacher-Output Cache for Ternary-ASR Distillation
Append-only teacher-output cache used to distill a ternary Hindi ASR student. It stores, per audio clip, ONLY:
row_id(a deterministic source-shard/position pointer),- top-k (k=64) CTC logits (vocab id + log-prob per kept entry, blank always kept),
- selected encoder hidden states (last-3 blocks: layers 14 / 15 / 16), fp16.
No audio and no ground-truth transcripts are stored or redistributed.
Provenance
- Teacher:
ai4bharat/indicconformer_stt_hi_hybrid_ctc_rnnt_large— license MIT. - Source corpora (audio & transcripts NOT redistributed here; used only to compute teacher outputs):
ai4bharat/Kathbath(hindi) — CC-BY-4.0.ai4bharat/Shrutilipi(hindi) — CC-BY-4.0.- Attribution: AI4Bharat.
License
CC-BY-4.0. Redistributing the derived teacher outputs (logits + hidden states + ids) is not redistribution of the source works; attribution to AI4Bharat is required.
Structure
default 29 rows
| Split | Rows | Size |
|---|---|---|
| train | 29 | 14.8 KB |
Details
- Repository
- sulabhkatiyar/trn-indcnfr-hi-pilot
- Publisher
- Sulabh Katiyar
- Task category
- Speech recognition
- Tags
- knowledge-distillation, teacher-cache, ctc
- Size category
- Not stated by the source
- Languages
- hi
- Revision
- 94ef66d7bcd69188b5a02e4dbeaa20805aa2f272
- Last updated
- 2026-10-03
Files
60 files, 17.4 GB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| manifest/kathbath/hindi/train/src-00000-block-00000.json | Data | 1.0 KB | — |
| manifest/kathbath/hindi/train/src-00000-block-00001.json | Data | 1.0 KB | — |
| manifest/kathbath/hindi/train/src-00001-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00001-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00002-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00002-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00003-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00003-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00004-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00004-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00005-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00005-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00006-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00006-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00007-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00007-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00008-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00008-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00009-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00009-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00010-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00010-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00011-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00011-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00012-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00012-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00013-block-00000.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00013-block-00001.json | Data | 1.1 KB | — |
| manifest/kathbath/hindi/train/src-00014-block-00000.json | Data | 1.1 KB | — |
| README.md | Documentation | 1.4 KB | — |
| shards/kathbath/hindi/train/src-00000-block-00000.npz | Other | 838.1 MB | ffe25fce52bd |
| shards/kathbath/hindi/train/src-00000-block-00001.npz | Other | 356.6 MB | 9b300cd72df3 |
| shards/kathbath/hindi/train/src-00001-block-00000.npz | Other | 826.1 MB | bde78f9a7fab |
| shards/kathbath/hindi/train/src-00001-block-00001.npz | Other | 362.0 MB | eded8c2c4339 |
| shards/kathbath/hindi/train/src-00002-block-00000.npz | Other | 831.0 MB | 1e230fd31180 |
| shards/kathbath/hindi/train/src-00002-block-00001.npz | Other | 359.1 MB | b436a4b97895 |
| shards/kathbath/hindi/train/src-00003-block-00000.npz | Other | 823.3 MB | 4ca1db9fa70e |
| shards/kathbath/hindi/train/src-00003-block-00001.npz | Other | 359.0 MB | 9969ac89f142 |
| shards/kathbath/hindi/train/src-00004-block-00000.npz | Other | 839.3 MB | 0eb05cce16a0 |
| shards/kathbath/hindi/train/src-00004-block-00001.npz | Other | 366.9 MB | 72747c7fb625 |
| shards/kathbath/hindi/train/src-00005-block-00000.npz | Other | 836.7 MB | 28584324d7ea |
| shards/kathbath/hindi/train/src-00005-block-00001.npz | Other | 365.1 MB | 543fb4531730 |
| shards/kathbath/hindi/train/src-00006-block-00000.npz | Other | 844.5 MB | 7cf63492ccf5 |
| shards/kathbath/hindi/train/src-00006-block-00001.npz | Other | 366.7 MB | b7253e88839e |
| shards/kathbath/hindi/train/src-00007-block-00000.npz | Other | 830.7 MB | bc427ba3b4c0 |
| shards/kathbath/hindi/train/src-00007-block-00001.npz | Other | 362.7 MB | bd5fc372bf31 |
| shards/kathbath/hindi/train/src-00008-block-00000.npz | Other | 833.4 MB | fe40589f764a |
| shards/kathbath/hindi/train/src-00008-block-00001.npz | Other | 367.5 MB | b4b1d7f89445 |
| shards/kathbath/hindi/train/src-00009-block-00000.npz | Other | 830.0 MB | 1220d9a5bae7 |
| shards/kathbath/hindi/train/src-00009-block-00001.npz | Other | 361.8 MB | 4ee8e0bfcd02 |
| shards/kathbath/hindi/train/src-00010-block-00000.npz | Other | 837.3 MB | 4cf900f62c60 |
| shards/kathbath/hindi/train/src-00010-block-00001.npz | Other | 363.3 MB | 0a0d19d2df4a |
| shards/kathbath/hindi/train/src-00011-block-00000.npz | Other | 838.7 MB | 874cb1c2c230 |
| shards/kathbath/hindi/train/src-00011-block-00001.npz | Other | 362.1 MB | 71d8c201084b |
| shards/kathbath/hindi/train/src-00012-block-00000.npz | Other | 833.6 MB | 66adae702147 |
| shards/kathbath/hindi/train/src-00012-block-00001.npz | Other | 360.6 MB | b0c69f935bde |
| shards/kathbath/hindi/train/src-00013-block-00000.npz | Other | 831.2 MB | 54eb253c5689 |
| shards/kathbath/hindi/train/src-00013-block-00001.npz | Other | 362.9 MB | 2d8cd06a6897 |
| shards/kathbath/hindi/train/src-00014-block-00000.npz | Other | 626.7 MB | 0be2bbcaa82e |
| .gitattributes | Repository | 2.5 KB | — |
License and Download
- License
- cc-by-4.0
- Access
- No access gate
Released by Sulabh Katiyar through its official repository on Hugging Face. Read the license.