Remove the unused pragma and properly apply the bias. (#1147)

candle-transformers/src/models/blip.rs

@@ -1,4 +1,3 @@
-#![allow(unused)]
 use super::blip_text;
 use super::with_tracing::{conv2d, linear, Conv2d, Linear};
 use candle::{Module, Result, Tensor, D};
@@ -65,7 +64,6 @@ struct VisionEmbeddings {
  class_embedding: Tensor,
  patch_embedding: Conv2d,
  position_embedding: Tensor,
- num_positions: usize,
 }
 
 impl VisionEmbeddings {
@@ -91,7 +89,6 @@ impl VisionEmbeddings {
  class_embedding,
  patch_embedding,
  position_embedding,
- num_positions,
  })
  }
 }
@@ -117,8 +114,6 @@ struct Attention {
  qkv: Linear,
  projection: Linear,
  scale: f64,
- embed_dim: usize,
- head_dim: usize,
  num_heads: usize,
 }
 
@@ -134,8 +129,6 @@ impl Attention {
  qkv,
  projection,
  scale,
- embed_dim,
- head_dim,
  num_heads,
  })
  }

candle-transformers/src/models/blip_text.rs

@@ -1,5 +1,4 @@
-#![allow(unused)]
-use super::with_tracing::{linear, linear_no_bias, Embedding, Linear};
+use super::with_tracing::{linear, Embedding, Linear};
 use candle::{Module, Result, Tensor, D};
 use candle_nn::{layer_norm, LayerNorm, VarBuilder};
 
@@ -63,7 +62,6 @@ struct TextSelfAttention {
  query: Linear,
  key: Linear,
  value: Linear,
- all_head_size: usize,
  attention_head_size: usize,
  num_attention_heads: usize,
  attention_scale: f64,
@@ -87,7 +85,6 @@ impl TextSelfAttention {
  query,
  key,
  value,
- all_head_size,
  attention_head_size,
  num_attention_heads,
  attention_scale,
@@ -301,12 +298,12 @@ impl TextEncoder {
 }
 
 #[derive(Debug, Clone)]
-struct TextPooler {
+pub struct TextPooler {
  dense: Linear,
 }
 
 impl TextPooler {
- fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+ pub fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
  let dense = linear(cfg.hidden_size, cfg.hidden_size, vb.pp("dense"))?;
  Ok(Self { dense })
  }
@@ -352,19 +349,15 @@ impl Module for TextPredictionHeadTransform {
 struct TextLMPredictionHead {
  transform: TextPredictionHeadTransform,
  decoder: Linear,
- bias: Tensor,
 }
 
 impl TextLMPredictionHead {
  fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
  let transform = TextPredictionHeadTransform::new(cfg, vb.pp("transform"))?;
- let decoder = linear_no_bias(cfg.hidden_size, cfg.vocab_size, vb.pp("decoder"))?;
+ let weight = vb.get((cfg.vocab_size, cfg.hidden_size), "decoder.weight")?;
  let bias = vb.get(cfg.vocab_size, "bias")?;
- Ok(Self {
- transform,
- decoder,
- bias,
- })
+ let decoder = Linear::from_weights(weight, Some(bias));
+ Ok(Self { transform, decoder })
  }
 }
 
@@ -396,7 +389,7 @@ impl Module for TextOnlyMLMHead {
 struct TextModel {
  embeddings: TextEmbeddings,
  encoder: TextEncoder,
- pooler: Option<TextPooler>,
+ // We do not need the pooler for caption generation
 }
 
 impl TextModel {
@@ -406,7 +399,6 @@ impl TextModel {
  Ok(Self {
  embeddings,
  encoder,
- pooler: None,
  })
  }
 

candle-transformers/src/models/with_tracing.rs

@@ -32,6 +32,14 @@ pub struct Linear {
  span: tracing::Span,
 }
 
+impl Linear {
+ pub fn from_weights(weights: Tensor, bias: Option<Tensor>) -> Self {
+ let inner = candle_nn::Linear::new(weights, bias);
+ let span = tracing::span!(tracing::Level::TRACE, "linear");
+ Self { inner, span }
+ }
+}
+
 pub fn linear(d1: usize, d2: usize, vb: VarBuilder) -> Result<Linear> {
  let inner = candle_nn::linear(d1, d2, vb)?;
  let span = tracing::span!(tracing::Level::TRACE, "linear");