TheAnimatrix/indic-transcribe-flex-gguf

Indic-Transcribe-Flex (GGUF, Q8_0)

Built with Indic-Transcribe-Flex from Bodhan AI / AI4Bharat.

A GGUF conversion of bodhan-ai/indic-transcribe-flex for transcribe.cpp and Hexium, where it is a built-in model.

• Source: the published nemo/indic_transcribe_flex.nemo (Hugging Face revision ecba9ba).
• Converted with scripts/convert-indic-canary.py --default-mode mixed, quantized with transcribe-quantize --quant Q8_0 (tensors: F32, F16, Q8_0). 1.41 GB.
• Default output is mixed script: Indian-language words in their own script, English words and digits in Latin. --no-itn gives native script; a -Latn language (e.g. te-Latn) gives romanized output.
• Needs transcribe.cpp with the Indic-Transcribe canary patch (branch hexium of the fork above).

transcribe-cli -m indic-transcribe-flex-Q8_0.gguf -l te clip.wav            # mixed
transcribe-cli -m indic-transcribe-flex-Q8_0.gguf -l te --no-itn clip.wav   # native script
transcribe-cli -m indic-transcribe-flex-Q8_0.gguf -l te-Latn clip.wav       # romanized

SHA-256: 807af7f8fe04cfa623fa5edb77e422f1db6ef7052ccd3be6064a43ae02dd3282

License

This model is a Derivative of Indic-Transcribe-Flex and is licensed under the Indic Open Model License v1.0 (LICENSE.md; plain-language summary in indic-open-license.md), exactly as the original. In short: free to use, modify and redistribute, including commercially; keep this credit and license; hosting it as an API for others needs Bodhan AI's written approval (see section 3); the prohibited uses in section 10 apply. Accuracy figures and intended use: see the original model card.