cstr/parakeet-rnnt-0.6b-GGUF

parakeet-rnnt-0.6b — GGUF

GGUF conversions of nvidia/parakeet-rnnt-0.6b for use with CrispASR.

Files

File	Size	Quantization
parakeet-rnnt-0.6b-q4_k.gguf	~447 MB	Q4_K (recommended)
parakeet-rnnt-0.6b-f16.gguf	~1.2 GB	F16 (full precision)

Usage

crispasr --backend parakeet \
  -m parakeet-rnnt-0.6b-q4_k.gguf \
  -f audio.wav

Or let CrispASR auto-download:

crispasr --backend parakeet-rnnt-0.6b -f audio.wav

Architecture

• Model: standard RNN-Transducer (no TDT duration head)
• Encoder: 24-layer FastConformer, 80-mel input, d_model=1024
• Decoder: RNN predictor (hidden=640) + joint network (hidden=640)
• Vocab: 1024-token BPE (lowercase English)
• Sample rate: 16 kHz

The RNNT decoder is auto-detected at runtime via n_tdt_durations==0.

License

nvidia/parakeet-rnnt-0.6b is released under the CC BY 4.0 license.

Provenance and EU AI Act Art. 53 note

• Upstream model: nvidia/parakeet-rnnt-0.6b — published by nvidia.
• Upstream licence: cc-by-4.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.