huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 10:38:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make the whisper model cloneable (#1200)

Browse Source 
* Add a quantized variant of llama2.c

* Clippy fixes.

* Make the whisper model cloneable.
This commit is contained in:
Laurent Mazare
2023-10-27 17:59:19 +02:00
committed by GitHub
parent b3181455d5
commit 85bea43e5b
2 changed files with 11 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
candle-transformers/src/models/whisper/model.rs
View File

@ -9,7 +9,7 @@ fn embedding(vocab_size: usize, hidden_size: usize, vb: VarBuilder) -> Result<Em
//
// We wrap the `Linear` layer here to add some tracing so that it's easier to profile the resulting
// model.
#[derive(Debug)]
#[derive(Debug, Clone)]
pub struct Linear {
inner: candle_nn::Linear,
span: tracing::Span,
@ -53,6 +53,7 @@ fn layer_norm(size: usize, vb: VarBuilder) -> Result<LayerNorm> {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L62
#[derive(Debug, Clone)]
struct MultiHeadAttention {
query: Linear,
key: Linear,
@ -162,6 +163,7 @@ impl MultiHeadAttention {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L111
#[derive(Debug, Clone)]
struct ResidualAttentionBlock {
attn: MultiHeadAttention,
attn_ln: LayerNorm,
@ -241,6 +243,7 @@ fn sinusoids(length: usize, channels: usize) -> Result<Tensor> {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L143
#[derive(Debug, Clone)]
pub struct AudioEncoder {
conv1: Conv1d,
conv2: Conv1d,
@ -316,6 +319,7 @@ impl AudioEncoder {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L176
#[derive(Debug, Clone)]
pub struct TextDecoder {
token_embedding: Embedding,
positional_embedding: Tensor,
@ -380,6 +384,7 @@ impl TextDecoder {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L221
#[derive(Debug, Clone)]
pub struct Whisper {
pub encoder: AudioEncoder,
pub decoder: TextDecoder,

5
candle-transformers/src/models/whisper/quantized_model.rs
View File

@ -19,6 +19,7 @@ fn conv1d(
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L62
#[derive(Debug, Clone)]
struct MultiHeadAttention {
query: Linear,
key: Linear,
@ -128,6 +129,7 @@ impl MultiHeadAttention {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L111
#[derive(Debug, Clone)]
struct ResidualAttentionBlock {
attn: MultiHeadAttention,
attn_ln: LayerNorm,
@ -206,6 +208,7 @@ fn sinusoids(length: usize, channels: usize) -> Result<Tensor> {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L143
#[derive(Debug, Clone)]
pub struct AudioEncoder {
conv1: Conv1d,
conv2: Conv1d,
@ -281,6 +284,7 @@ impl AudioEncoder {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L176
#[derive(Debug, Clone)]
pub struct TextDecoder {
token_embedding: Embedding,
positional_embedding: Tensor,
@ -347,6 +351,7 @@ impl TextDecoder {
}
// https://github.com/openai/whisper/blob/f572f2161ba831bae131364c3bffdead7af6d210/whisper/model.py#L221
#[derive(Debug, Clone)]
pub struct Whisper {
pub encoder: AudioEncoder,
pub decoder: TextDecoder,