Request for .nemo checkpoint of SraVaani multilingual model
Hi team,
Thanks for releasing the SraVaani multilingual model.
I would like to convert it to GGUF format for use with parakeet.cpp, but I see the repository currently only has a .ts file. The conversion script requires a .nemo checkpoint.
Could you please upload the .nemo file for the multilingual model, or share it via an external link (e.g., Google Drive) if size is an issue?
If it's not yet available, please let me know. I appreciate your work on Indian language ASR!
Thanks,
Bhushan
Hi Bhushan,
Could you please send a request to agneedh@artpark.in with more details about your intended use case ?
This information will help us evaluate your request and determine how we can facilitate access.
Thank you.
Hi Sujith,
i have mailed request to agneedh@artpark.in
could we have access to
- The SraVaani 1.0 .nemo checkpoint matching the current released
model and tokenizer, rather than the older Vaani FastConformer model. - The trainable .nemo checkpoint/configuration for SraVaani 0.5 Live.
- The associated model/training YAML, exact NeMo revision or container,
and official export scripts. - If shareable, a final Lightning .ckpt containing optimizer and
scheduler state.
we are currently using IndicCorformer, but it doesn't support code mixing and English lexicons. We believe this model would work after quantisation and fine-tuning, for our low resource gujarati heavy code mixed asr application.