louisbertson/moore-bilstm-ssl

    ---
    library_name: pytorch
    language:
    - mos
    tags:
    - audio
    - speech
    - moore
    - pytorch
    - self-supervised-learning
    ---

    # louisbertson/moore-bilstm-ssl

    This repository contains a checkpoint exported from the Mooré audio training pipeline in the companion codebase.

    ## Model

    - model name: `bilstm_ssl`
    - task: `ssl_pretraining`
    - checkpoint file: `best.pt`
    - dataset repo: `louisbertson/moore-audio-standardized`

    ## Evaluation

    - model_name: bilstm_ssl

• validation_masked_mse: 1.046602

• test_masked_mse: 1.128022

• training_time_seconds: 17.957690

• peak_gpu_memory_mb: 0.000000

• throughput_samples_per_second: 0.111373

• inference_latency_per_10s_audio_seconds: 0.700198

• total_parameters: 16911056

• model_size_mb: 64.526966

    ## Files
  
    - `training_config.yaml`: config used by the repository entry points
    - `best.pt`: PyTorch checkpoint from this training run
    - `metrics.json`: exported evaluation metrics when available