forked from espnet/espnet
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request espnet#5832 from Darshan7575/master
WIP: Espnet Multi-convformer implementation
- Loading branch information
Showing
10 changed files
with
1,175 additions
and
1 deletion.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
82 changes: 82 additions & 0 deletions
82
egs2/librispeech/asr1/conf/tuning/train_asr_multiconvformer_conv_fusion.yaml
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,82 @@ | ||
| # Trained with A100 (80 GB) x 2 GPUs. It takes 110 minutes per epoch. | ||
| encoder: multiconv_conformer | ||
| encoder_conf: | ||
| output_size: 512 | ||
| attention_heads: 8 | ||
| selfattention_layer_type: rel_selfattn | ||
| pos_enc_layer_type: rel_pos | ||
| rel_pos_type: latest | ||
| cgmlp_linear_units: 3072 | ||
| multicgmlp_type: concat_fusion | ||
| multicgmlp_kernel_sizes: 7,15,23,31 | ||
| use_linear_after_conv: false | ||
| gate_activation: identity | ||
| num_blocks: 18 | ||
| dropout_rate: 0.1 | ||
| positional_dropout_rate: 0.1 | ||
| attention_dropout_rate: 0.1 | ||
| input_layer: conv2d | ||
| layer_drop_rate: 0.1 | ||
| linear_units: 1024 | ||
| positionwise_layer_type: linear | ||
| macaron_style: true | ||
| use_cnn_module: true | ||
|
|
||
| decoder: transformer | ||
| decoder_conf: | ||
| attention_heads: 8 | ||
| linear_units: 2048 | ||
| num_blocks: 6 | ||
| dropout_rate: 0.1 | ||
| positional_dropout_rate: 0.1 | ||
| self_attention_dropout_rate: 0.1 | ||
| src_attention_dropout_rate: 0.1 | ||
|
|
||
| model_conf: | ||
| ctc_weight: 0.3 | ||
| lsm_weight: 0.1 | ||
| length_normalized_loss: false | ||
|
|
||
| frontend_conf: | ||
| n_fft: 512 | ||
| hop_length: 160 | ||
|
|
||
| use_amp: true | ||
| unused_parameters: true | ||
| num_workers: 4 | ||
| batch_type: numel | ||
| batch_bins: 70000000 | ||
| accum_grad: 2 | ||
| max_epoch: 80 | ||
| patience: none | ||
| init: none | ||
| best_model_criterion: | ||
| - - valid | ||
| - acc | ||
| - max | ||
| keep_nbest_models: 10 | ||
| nbest_averaging_interval: 10 | ||
|
|
||
| optim: adam | ||
| optim_conf: | ||
| lr: 0.002 | ||
| weight_decay: 0.000001 | ||
| scheduler: warmuplr | ||
| scheduler_conf: | ||
| warmup_steps: 40000 | ||
|
|
||
| specaug: specaug | ||
| specaug_conf: | ||
| apply_time_warp: true | ||
| time_warp_window: 5 | ||
| time_warp_mode: bicubic | ||
| apply_freq_mask: true | ||
| freq_mask_width_range: | ||
| - 0 | ||
| - 27 | ||
| num_freq_mask: 2 | ||
| apply_time_mask: true | ||
| time_mask_width_ratio_range: | ||
| - 0. | ||
| - 0.05 | ||
| num_time_mask: 10 |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
84 changes: 84 additions & 0 deletions
84
egs2/librispeech_100/asr1/conf/tuning/train_asr_multiconvformer_conv_fusion_linear1024.yaml
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,84 @@ | ||
| # Trained with A100 (80 GB) x 1 GPUs. It takes 15 minutes per epoch. | ||
| encoder: multiconv_conformer | ||
| encoder_conf: | ||
| output_size: 256 | ||
| attention_heads: 4 | ||
| selfattention_layer_type: rel_selfattn | ||
| pos_enc_layer_type: rel_pos | ||
| rel_pos_type: latest | ||
| cgmlp_linear_units: 1024 | ||
| multicgmlp_type: concat_fusion | ||
| multicgmlp_kernel_sizes: 7,15,23,31 | ||
| multicgmlp_merge_conv_kernel: 31 | ||
| use_linear_after_conv: false | ||
| gate_activation: identity | ||
| num_blocks: 12 | ||
| dropout_rate: 0.1 | ||
| positional_dropout_rate: 0.1 | ||
| attention_dropout_rate: 0.1 | ||
| input_layer: conv2d | ||
| layer_drop_rate: 0.0 | ||
| linear_units: 1024 | ||
| positionwise_layer_type: linear | ||
| macaron_style: true | ||
| use_cnn_module: true | ||
|
|
||
| decoder: transformer | ||
| decoder_conf: | ||
| attention_heads: 4 | ||
| linear_units: 2048 | ||
| num_blocks: 6 | ||
| dropout_rate: 0.1 | ||
| positional_dropout_rate: 0.1 | ||
| self_attention_dropout_rate: 0.1 | ||
| src_attention_dropout_rate: 0.1 | ||
| layer_drop_rate: 0.0 | ||
|
|
||
| model_conf: | ||
| ctc_weight: 0.3 | ||
| lsm_weight: 0.1 | ||
| length_normalized_loss: false | ||
|
|
||
| frontend_conf: | ||
| n_fft: 512 | ||
| win_length: 400 | ||
| hop_length: 160 | ||
|
|
||
| seed: 2022 | ||
| num_workers: 4 | ||
| batch_type: numel | ||
| batch_bins: 16000000 | ||
| accum_grad: 4 | ||
| max_epoch: 70 | ||
| patience: none | ||
| init: none | ||
| best_model_criterion: | ||
| - - valid | ||
| - acc | ||
| - max | ||
| keep_nbest_models: 10 | ||
| use_amp: true | ||
|
|
||
| optim: adam | ||
| optim_conf: | ||
| lr: 0.002 | ||
| weight_decay: 0.000001 | ||
| scheduler: warmuplr | ||
| scheduler_conf: | ||
| warmup_steps: 15000 | ||
|
|
||
| specaug: specaug | ||
| specaug_conf: | ||
| apply_time_warp: true | ||
| time_warp_window: 5 | ||
| time_warp_mode: bicubic | ||
| apply_freq_mask: true | ||
| freq_mask_width_range: | ||
| - 0 | ||
| - 27 | ||
| num_freq_mask: 2 | ||
| apply_time_mask: true | ||
| time_mask_width_ratio_range: | ||
| - 0. | ||
| - 0.05 | ||
| num_time_mask: 5 |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
78 changes: 78 additions & 0 deletions
78
egs2/slurp_entity/asr1/conf/tuning/train_asr_multiconv_e12_mlp3072_linear2048_layerdrop.yaml
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,78 @@ | ||
| # network architecture | ||
| # encoder related | ||
| encoder: multiconv_conformer | ||
| encoder_conf: | ||
| output_size: 512 | ||
| attention_heads: 8 | ||
| selfattention_layer_type: rel_selfattn | ||
| pos_enc_layer_type: rel_pos | ||
| rel_pos_type: latest | ||
| cgmlp_linear_units: 3072 | ||
| multicgmlp_type: concat_fusion | ||
| multicgmlp_kernel_sizes: 7,15,23,31 | ||
| multicgmlp_merge_conv_kernel: 31 | ||
| use_linear_after_conv: false | ||
| gate_activation: identity | ||
| num_blocks: 12 # Maybe we can increase the size by 1 to match e-branchformer | ||
| dropout_rate: 0.1 | ||
| positional_dropout_rate: 0.1 | ||
| attention_dropout_rate: 0.1 | ||
| input_layer: conv2d | ||
| layer_drop_rate: 0.1 | ||
| linear_units: 1152 | ||
| positionwise_layer_type: linear | ||
| macaron_style: true | ||
| use_cnn_module: true | ||
|
|
||
|
|
||
| decoder: transformer | ||
| decoder_conf: | ||
| attention_heads: 8 | ||
| linear_units: 2048 | ||
| num_blocks: 6 | ||
| dropout_rate: 0.1 | ||
| positional_dropout_rate: 0.1 | ||
| self_attention_dropout_rate: 0.1 | ||
| src_attention_dropout_rate: 0.1 | ||
| layer_drop_rate: 0.2 | ||
|
|
||
| optim: adam | ||
| optim_conf: | ||
| lr: 0.001 | ||
| weight_decay: 0.000001 | ||
| scheduler: warmuplr | ||
| scheduler_conf: | ||
| warmup_steps: 35000 | ||
|
|
||
| unused_parameters: true | ||
| batch_type: folded | ||
| batch_size: 64 | ||
| accum_grad: 1 | ||
| max_epoch: 60 | ||
| best_model_criterion: | ||
| - - valid | ||
| - acc | ||
| - max | ||
| keep_nbest_models: 10 | ||
|
|
||
| model_conf: | ||
| ctc_weight: 0.3 | ||
| lsm_weight: 0.1 | ||
| length_normalized_loss: false | ||
| extract_feats_in_collect_stats: false # Note: "False" means during collect stats (stage 10), generating dummy stats files rather than extract_feats by forward frontend. | ||
|
|
||
| specaug: specaug | ||
| specaug_conf: | ||
| apply_time_warp: true | ||
| time_warp_window: 5 | ||
| time_warp_mode: bicubic | ||
| apply_freq_mask: true | ||
| freq_mask_width_range: | ||
| - 0 | ||
| - 30 | ||
| num_freq_mask: 2 | ||
| apply_time_mask: true | ||
| time_mask_width_range: | ||
| - 0 | ||
| - 40 | ||
| num_time_mask: 2 |
Oops, something went wrong.