Nemotron 3.5 ASR Streaming 0.6B — Q5_K GGUF
Q5_K conversion from NVIDIA's original model, pinned ea30d66debe3740a08b573244286791d423d6b3e.
- File:
nemotron-3.5-asr-streaming-0.6b.q5_k.gguf - Size: 528,061,920 bytes (503.6 MiB).
- Conversion: quantized from an F16 conversion of the original NeMo checkpoint. The official Q8_0 GGUF supplies metadata, tensor ordering and protected tensors; Q5_K weights are derived from F16. Matrices whose row widths do not fit 256-element K-quant blocks stay F16; other protected tensors retain their original types. This is a mixed-precision Q5_K file, not a Q5_K_M preset.
- Runtime: Use NeMo-Speech.cpp v0.2.0
nemo-speech transcribe audio.wav \
--model nemotron-3.5-asr-streaming-0.6b.q5_k.gguf \
--language ja-JP