| # Minimal suggested environment for the released scripts. | |
| # | |
| # This is intentionally not a full pip freeze / exact reproduction of the | |
| # authors' local environment. CUDA, PyTorch and NeMo stacks can be | |
| # platform-specific, so this file lists only the main direct dependencies | |
| # needed to understand and run the code in a typical setup. | |
| # | |
| # Core training / .nemo evaluation dependencies | |
| nemo_toolkit[asr]==2.6.2 | |
| numpy | |
| torch | |
| tqdm | |
| soundfile | |
| jiwer | |
| sentencepiece | |
| omegaconf | |
| # Used by eval_fleurs.py | |
| datasets | |
| # Used by onnx_eval/eval_fleurs_onnx.py | |
| onnxruntime-genai | |
| # Note: the WER normalization code is imported from a clone of | |
| # huggingface/open_asr_leaderboard via OPEN_ASR_LB_ROOT or PYTHONPATH. |