Griot Nano 1 Onnx

--- language: - ak - dag - ee - en - gaa license: cc-by-nc-sa-4.0 library_name: onnx pipeline_tag: automatic-speech-recognition tags: - onnx - onnxruntime - conformer - ctc - cpu --- # Griot Nano 1 — ONNX CPU Edition This is the unquantized FP32 ONNX Runtime distribution of [Qlerqly/griot-nano-1](https://huggingface.co/Qlerqly/griot-nano-1). It contains the same 153M-parameter Conformer CTC acoustic model and is intended for local CPU transcription without PyTorch. ## Run locally ```bash uvx --from huggingface-hub hf download Qlerqly/griot-nano-1-onnx \ --local-dir griot-nano-1-onnx cd griot-nano-1-onnx uv run --with-requirements requirements.txt python inference.py \ --model-dir . --audio recording.wav # Live microphone transcription uv run --with-requirements requirements.txt python inference.py \ --model-dir . --live ``` Use `--threads N` to control ONNX Runtime CPU threads. ## Recommended: KenLM decoding For best contextual accuracy, we encourage using this runtime together with [Qlerqly/griot-nano-1-kenlm](https://huggingface.co/Qlerqly/griot-nano-1-kenlm). It adds n-gram beam-search guidance without another neural network, keeping the overhead low relative to neural language-model rescoring. ```bash # From the directory containing griot-nano-1-onnx uvx --from huggingface-hub hf download Qlerqly/griot-nano-1-kenlm \ --local-dir griot-nano-1-kenlm cd griot-nano-1-onnx uv run --with-requirements requirements-kenlm.txt python inference.py \ --model-dir . --audio recording.wav \ --kenlm ../griot-nano-1-kenlm/multilingual.arpa --beam-width 25 ``` This runs ONNX Runtime and KenLM as one CPU transcription pipeline. The suggested starting values are beam width `25`, LM weight `0.5`, and word insertion bonus `1.0`; override the latter two with `--lm-alpha` and `--lm-beta`. ## Test results These are the canonical model's raw greedy-CTC results. The exporter performs a numerical PyTorch-versus-ONNX check recorded in `onnx_verification.json`. | Language | Samples | CER | WER | |---|---:|---:|---:| | ak | 21,435 | 19.27% | 41.67% | | dag | 5,307 | 7.04% | 23.75% | | ee | 6,820 | 9.66% | 30.76% | | en | 56,364 | 12.91% | 26.05% | | ga | 9,621 | 8.14% | 15.83% | | **Combined** | **99,547** | **13.66%** | **29.19%** | ## Format - ONNX opset 18 - FP32 weights - Dynamic batch and time dimensions - ONNX Runtime `CPUExecutionProvider` - No quantiz...

Source context: 4 downloads · 0 likes · Pipeline automatic-speech-recognition · Library onnx · Repo Qlerqly/griot-nano-1-onnx