Update CI to build without defaults (guillaume-be#299)

* Avoid attention copy, fix remote feature

* revert conversation change

* Updated CI for build without defaults

.github/workflows/continuous-integration.yml

@@ -21,6 +21,21 @@ jobs:
         with:
           command: build
 
+  build-no-defaults:
+    name: Build no defaults
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - uses: actions-rs/toolchain@v1
+        with:
+          profile: minimal
+          toolchain: stable
+          override: true
+      - uses: actions-rs/cargo@v1
+        with:
+          command: build
+          args: --no-default-features
+
   build-windows:
     name: Build Windows
     runs-on: windows-latest

src/albert/encoder.rs

@@ -138,7 +138,7 @@ impl AlbertLayerGroup {
             hidden_state = temp.0;
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/bart/decoder.rs

@@ -323,7 +323,7 @@ impl BartDecoder {
                 hidden_states.push(hidden_state.as_ref().copy());
             };
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(value) = &mut next_decoder_cache {
                 value[layer_idx] = temp.2

src/bart/encoder.rs

@@ -233,7 +233,7 @@ impl BartEncoder {
             hidden_state = temp.0;
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/bert/encoder.rs

@@ -292,7 +292,7 @@ impl BertEncoder {
             hidden_state = Some(layer_output.hidden_state);
             attention_weights = layer_output.attention_weights;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().unwrap().copy());

src/deberta/encoder.rs

@@ -287,7 +287,7 @@ impl DebertaEncoder {
             hidden_state = Some(layer_output.0);
             attention_weights = layer_output.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().unwrap().copy());

src/deberta_v2/encoder.rs

@@ -333,7 +333,7 @@ impl DebertaV2Encoder {
             }
             attention_weights = layer_output.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(output_states.as_ref().unwrap().copy());

src/distilbert/transformer.rs

@@ -164,7 +164,7 @@ impl Transformer {
             hidden_state = Some(temp.0);
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().unwrap().copy());

src/gpt2/gpt2_model.rs

@@ -470,10 +470,10 @@ impl Gpt2Model {
                 layer.forward_t(&hidden_state, past.as_ref(), attention_mask.as_ref(), train);
             hidden_state = temp.0;
             if let Some(presents) = all_presents.borrow_mut() {
-                presents.push(temp.1.as_ref().copy());
+                presents.push(temp.1);
             };
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(temp.2.as_ref().unwrap().copy());
+                attentions.push(temp.2.unwrap());
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/gpt_neo/gpt_neo_model.rs

@@ -425,7 +425,7 @@ impl GptNeoModel {
             attention_weights = temp.1;
             next_cache[layer_idx] = temp.2;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(x.as_ref().unwrap().copy());

src/longformer/encoder.rs

@@ -340,10 +340,11 @@ impl LongformerEncoder {
             attention_weights = temp.1;
             global_attention_weights = temp.2;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().transpose(1, 2));
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()).transpose(1, 2));
             };
             if let Some(global_attentions) = all_global_attentions.borrow_mut() {
-                global_attentions.push(global_attention_weights.as_ref().unwrap().transpose(2, 3));
+                global_attentions
+                    .push(std::mem::take(&mut global_attention_weights.unwrap()).transpose(2, 3));
             };
             if let Some(all_hidden_states) = all_hidden_states.borrow_mut() {
                 all_hidden_states.push(x.as_ref().unwrap().copy());

src/m2m_100/decoder.rs

@@ -174,7 +174,7 @@ impl M2M100Decoder {
                 hidden_states.push(hidden_state.as_ref().copy());
             };
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(value) = &mut next_decoder_cache {
                 value[layer_idx] = temp.2

src/m2m_100/encoder.rs

@@ -108,7 +108,7 @@ impl M2M100Encoder {
             hidden_state = temp.0;
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/mbart/decoder.rs

@@ -308,7 +308,7 @@ impl MBartDecoder {
                 hidden_states.push(hidden_state.as_ref().copy());
             };
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(value) = &mut next_decoder_cache {
                 value[layer_idx] = temp.2

src/mbart/encoder.rs

@@ -217,7 +217,7 @@ impl MBartEncoder {
             hidden_state = temp.0;
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/mobilebert/encoder.rs

@@ -491,7 +491,7 @@ impl MobileBertEncoder {
             x = Some(temp.0);
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(x.as_ref().unwrap().copy());

src/openai_gpt/openai_gpt_model.rs

@@ -260,7 +260,7 @@ impl OpenAiGptModel {
             let temp = layer.forward_t(&hidden_state, attention_mask.as_ref(), train);
             hidden_state = temp.0;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(temp.1.as_ref().unwrap().copy());
+                attentions.push(temp.1.unwrap());
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/pegasus/decoder.rs

@@ -168,7 +168,7 @@ impl PegasusDecoder {
                 hidden_states.push(hidden_state.as_ref().copy());
             };
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(value) = &mut next_decoder_cache {
                 value[layer_idx] = temp.2

src/pegasus/encoder.rs

@@ -110,7 +110,7 @@ impl PegasusEncoder {
             hidden_state = temp.0;
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(hidden_state.as_ref().copy());

src/pipelines/keywords_extraction/pipeline.rs

@@ -21,9 +21,11 @@
 /// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 /// SOFTWARE.
 use crate::pipelines::keywords_extraction::tokenizer::StopWordsTokenizer;
+#[cfg(feature = "remote")]
+use crate::pipelines::sentence_embeddings::SentenceEmbeddingsModelType;
 use crate::pipelines::sentence_embeddings::{
-    SentenceEmbeddingsConfig, SentenceEmbeddingsModel, SentenceEmbeddingsModelType,
-    SentenceEmbeddingsSentenceBertConfig, SentenceEmbeddingsTokenizerConfig,
+    SentenceEmbeddingsConfig, SentenceEmbeddingsModel, SentenceEmbeddingsSentenceBertConfig,
+    SentenceEmbeddingsTokenizerConfig,
 };
 use crate::{Config, RustBertError};
 use regex::Regex;

src/prophetnet/encoder.rs

@@ -195,7 +195,7 @@ impl ProphetNetEncoder {
             x = Some(temp.0);
             attention_weights = temp.1;
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(hidden_states) = all_hidden_states.borrow_mut() {
                 hidden_states.push(x.as_ref().unwrap().transpose(0, 1));

src/t5/encoder.rs

@@ -509,7 +509,7 @@ impl T5Stack {
                 hidden_states.push(hidden_state.as_ref().copy().transpose(0, 1));
             };
             if let Some(attentions) = all_attentions.borrow_mut() {
-                attentions.push(attention_weights.as_ref().unwrap().copy());
+                attentions.push(std::mem::take(&mut attention_weights.unwrap()));
             };
             if let Some(value) = &mut next_cache {
                 value[layer_idx] = block_output.cache