---
license: apache-2.0
language: [zh, en]
tags: [automatic-speech-recognition, zipformer2, transducer, ggml, gguf, rapidspeech, streaming-asr]
library_name: rapidspeech.cpp
---

# X-ASR zh-en 960ms streaming zipformer2 transducer — GGUF

GGUF conversion of [GilgameshWind/X-ASR-zh-en](https://huggingface.co/GilgameshWind/X-ASR-zh-en)
(k2-fsa / sherpa-onnx streaming zipformer2 transducer, 960 ms chunk) for the
[RapidSpeech.cpp](https://github.com/vieenrose/RapidSpeech.cpp) ggml runtime
(branch `zipformer`), targeting CUDA on the Jetson Nano gen1 (sm_53).

| File | Size | Notes |
|---|---|---|
| `x-asr-zh-en-960ms-f16.gguf`  | 292 MB | f16 weights, 922 tensors |
| `x-asr-zh-en-960ms-q4_k.gguf` | 85 MB  | Q4_K (block-32 fallback where ne0 % 256 != 0) |

arch `XAsrZipformer2`. 16 kHz, 80-dim kaldi fbank (povey window). Vocab 5000 (zh+en, punctuation).
Converted with `scripts/xasr/convert_xasr_to_gguf.py`. encoder_embed forward validated vs
sherpa-onnx (corr 0.999998); full encoder/transducer port in progress.