diff --git a/ultravox/training/configs/release_config.yaml b/ultravox/training/configs/release_config.yaml index 776e238..6ebdd98 100644 --- a/ultravox/training/configs/release_config.yaml +++ b/ultravox/training/configs/release_config.yaml @@ -14,49 +14,50 @@ train_sets: - name: librispeech-other-continuation - name: peoplespeech-clean-continuation weight: 8 - - name: common-voice-en-continuation + - name: commonvoice-en-continuation weight: 8 - - name: common-voice-ar-continuation + - name: commonvoice-ar-continuation weight: 0.2 - - name: common-voice-de-continuation + - name: commonvoice-de-continuation weight: 4 - - name: common-voice-es-continuation + - name: commonvoice-es-continuation weight: 3 - - name: common-voice-fr-continuation + - name: commonvoice-fr-continuation weight: 4 - - name: common-voice-it-continuation + - name: commonvoice-it-continuation weight: 1.2 - - name: common-voice-ja-continuation + - name: commonvoice-ja-continuation weight: 0.1 - - name: common-voice-pt-continuation + - name: commonvoice-pt-continuation weight: 0.2 - - name: common-voice-ru-continuation + - name: commonvoice-ru-continuation weight: 0.2 - name: librispeech-clean-transcription - name: librispeech-other-transcription - - name: peoplespeech-transcription + - name: peoplespeech-clean-transcription weight: 0.8 - - name: common-voice-en-transcription + - name: commonvoice-en-transcription weight: 0.8 - - name: common-voice-ar-transcription + - name: commonvoice-ar-transcription weight: 0.02 - - name: common-voice-de-transcription + - name: commonvoice-de-transcription weight: 0.4 - - name: common-voice-es-transcription + - name: commonvoice-es-transcription weight: 0.3 - - name: common-voice-fr-transcription + - name: commonvoice-fr-transcription weight: 0.4 - - name: common-voice-it-transcription + - name: commonvoice-it-transcription weight: 0.12 - - name: common-voice-ja-transcription + - name: commonvoice-ja-transcription weight: 0.01 - - name: common-voice-pt-transcription + - name: commonvoice-pt-transcription weight: 0.02 - - name: common-voice-ru-transcription + - name: commonvoice-ru-transcription weight: 0.02 # Temporarily remove heysquad_human from val_sets as it causes the training to fail. -val_sets: ["peoplespeech-clean"] +val_sets: + - name: peoplespeech batch_size: 24 max_steps: 14400 # x8x24 = 2,764,800