You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

sl-cyan β€” domain-adapted Tajik Whisper-large-v3 (CTranslate2 int8)

Domain-fine-tuned Tajik Whisper for dictation/protocol transcription. Public but gated=manual. Drop the chosen version folder into the Sado app as whisper-ct2/.

Versions

version where trained on test_synth WER FLEURS-150 WER real-dictation WER recommendation
v1 (2026-08-31, ckpt-370) repo root (model.bin, ...) 3.5 h synthetic in-domain + replay 2.1 11.4 12.1 βœ… recommended default
v2 (2026-09-03, ckpt-1318) v2-2026-09-03/ 32.3 h synthetic in-domain (10x) + replay 2.2 11.7 13.0 no measurable gain over v1

All three numbers per version come from the same evaluation harness (digit-normalized both sides, beam=1); base model before domain adaptation scored 17.4 / 10.8 / 21.5. The v2 experiment tested whether 10x more synthetic domain data helps: it does not β€” v1 remains the deployed default.

Base: continued FT of our Tajik Whisper-large-v3 (checkpoint-14450). Format: CTranslate2 int8 (faster-whisper compatible).

Downloads last month
5
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support