-
Notifications
You must be signed in to change notification settings - Fork 474
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Feat: Add support for Whisper architecture
Notes: The layer template for the decoder is marked as optional since, for distilled or turbo models, the number of decoder layers is less than the number of encoder layers, and we have no way to separate them. Signed-off-by: sagewe <[email protected]>
- Loading branch information
Showing
1 changed file
with
189 additions
and
0 deletions.
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
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,189 @@ | ||
| { | ||
| "model_type": "whisper", | ||
| "architectures": [ | ||
| "WhisperForConditionalGeneration" | ||
| ], | ||
| "pre_weights": [ | ||
| { | ||
| "name": "model.encoder.conv1.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.conv1.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.conv2.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.conv2.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.embed_positions.weight" | ||
| }, | ||
| { | ||
| "name": "model.decoder.embed_tokens.weight" | ||
| }, | ||
| { | ||
| "name": "model.decoder.embed_positions.weight" | ||
| } | ||
| ], | ||
| "post_weights": [ | ||
| { | ||
| "name": "model.encoder.layer_norm.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layer_norm.bias" | ||
| }, | ||
| { | ||
| "name": "model.decoder.layer_norm.weight" | ||
| }, | ||
| { | ||
| "name": "model.decoder.layer_norm.bias" | ||
| } | ||
| ], | ||
| "num_layers_config_key": "num_hidden_layers", | ||
| "layer_templates": { | ||
| "weights": [ | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.k_proj.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.v_proj.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.v_proj.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.q_proj.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.q_proj.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.out_proj.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn.out_proj.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn_layer_norm.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.self_attn_layer_norm.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.fc1.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.fc1.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.fc2.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.fc2.bias" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.final_layer_norm.weight" | ||
| }, | ||
| { | ||
| "name": "model.encoder.layers.${layer_index}.final_layer_norm.bias" | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.k_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.v_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.v_proj.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.q_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.q_proj.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.out_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn.out_proj.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn_layer_norm.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.self_attn_layer_norm.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.k_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.v_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.v_proj.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.q_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.q_proj.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.out_proj.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn.out_proj.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn_layer_norm.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.encoder_attn_layer_norm.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.fc1.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.fc1.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.fc2.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.fc2.bias", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.final_layer_norm.weight", | ||
| "optional": true | ||
| }, | ||
| { | ||
| "name": "model.decoder.layers.${layer_index}.final_layer_norm.bias", | ||
| "optional": true | ||
| } | ||
| ] | ||
| } | ||
| } |