Community conversion: Vietnamese Parakeet CTC 0.6B GGUF (F16/Q8_0) for CrispASR
Hello NVIDIA Parakeet team and community,
I have prepared a community GGUF conversion ofnvidia/parakeet-ctc-0.6b-Vietnamese
for the CrispASR runtime.
The conversion is published by @BaoKhuong .
This is not an official NVIDIA release and involves no additional training or fine-tuning.
Model files
| Download | Size (bytes) |
|---|---|
parakeet-ctc-0.6b-vi-f16.gguf |
1,219,538,848 |
parakeet-ctc-0.6b-vi-q8_0.gguf |
649,865,888 |
- Community GGUF repository: https://huggingface.co/BaoKhuong/parakeet-ctc-0.6b-vietnamese-GGUF
- Upstream contribution PR: https://huggingface.co/nvidia/parakeet-ctc-0.6b-Vietnamese/discussions/8
Conversion and intended use
The GGUF files are derived from NVIDIA's original Vietnamese Parakeet CTC 0.6B
NeMo checkpoint using CrispASR's convert-stt-fastconformer-ctc-to-gguf.py
converter (F16), followed by crispasr-quantize (Q8_0).
These weights are intended for Vietnamese ASR and for experimental CTC
word alignment in CrispASR (for example, as the -am model when transcribing
with Qwen3-ASR). They are not Qwen3-ASR weights. Actual forced-alignment
compatibility, Vietnamese word boundaries, and timestamp accuracy should be
verified independently; no benchmark or validation is claimed in this post.
SHA256 (when exposed by Hub metadata)
parakeet-ctc-0.6b-vi-f16.gguf:a53cd6893e544a81720fa553f3544383dcdb086ef7d53ebbb042d7ec85fbc22aparakeet-ctc-0.6b-vi-q8_0.gguf:1241e31c307f6e38871c173a9ae307ff9cb58bacea8cb4ee61375b65c29827a0
Attribution and licensing
Original model and training: NVIDIA. GGUF conversion: @BaoKhuong , using
CrispASR open-source conversion tools. The underlying model remains subject
to the NVIDIA Open Model License.
Please review the license terms before using or redistributing the weights.
Feedback on runtime compatibility and Vietnamese timestamp quality is welcome.