File size: 811 Bytes
354cfa6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 | Parseh ONNX aligner — Hindi
Derived from ai4bharat/indicwav2vec-hindi at immutable revision e746fa3e8de3f6629b9ded29d9c1c565c566f3db.
Base model: AI4Bharat IndicWav2Vec (exact upstream base revision not stated in public Hub metadata).
Changes made: converted the CTC checkpoint to ONNX (opset 17) and dynamically quantized MatMul weights to int8. No retraining was performed.
Training material named by the source card:
- None declared
The original source model licence is apache-2.0. This derivative carries the applicable licence text and required attribution; no endorsement is implied.
AI4Bharat’s IndicWav2Vec repository separately states that MIT applies to its pretrained, fine-tuned and language models; this is retained as upstream provenance alongside this checkpoint’s Apache-2.0 Hub tag.
|