Replace hf models with dummies

.github/workflows/ci.yml

@@ -60,11 +60,6 @@ jobs:
       - name: Project setup
         uses: ./.github/actions/project-setup
 
-      - name: Install test dependencies
-        run: make setup
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN }}
-
       - name: Install llvm-tools-preview
         run: rustup component add llvm-tools-preview
 
@@ -116,11 +111,6 @@ jobs:
         with:
           toolchain: nightly
 
-      - name: Install test dependencies
-        run: make setup
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN }}
-
       - run: rustup component add --toolchain nightly-x86_64-unknown-linux-gnu clippy
 
       - name: Run Clippy

.gitignore

@@ -4,8 +4,6 @@
 # cargo tarpaulin
 *.profraw
 
-models/
-
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[codz]

encoderfile/src/build_cli/config.rs

@@ -350,7 +350,7 @@ mod tests {
         let cfg = EncoderfileConfig {
             name: "my-cool-model".into(),
             version: "1.0".into(),
-            path: ModelPath::Directory("../models/embedding".into()),
+            path: ModelPath::Directory("../models/dummy_electra_token_embeddings".into()),
             model_type: ModelType::Embedding,
             output_path: Some(base.clone()),
             cache_dir: Some(base.clone()),

encoderfile/src/build_cli/model.rs

@@ -18,13 +18,23 @@ impl ModelTypeExt for crate::common::ModelType {
             Self::Embedding => validate_embedding_model(model),
             Self::SequenceClassification => validate_sequence_classification_model(model),
             Self::TokenClassification => validate_token_classification_model(model),
-            Self::SentenceEmbedding => validate_embedding_model(model),
+            Self::SentenceEmbedding => validate_sentence_embedding_model(model),
         }?;
 
         PlannedAsset::from_asset_source(AssetSource::File(path), AssetKind::ModelWeights)
     }
 }
 
+fn validate_sentence_embedding_model(model: Session) -> Result<()> {
+    let shape = get_outp_dim(model.outputs.as_slice(), "last_hidden_state")?;
+
+    if shape.len() != 2 {
+        bail!("Model must return tensor of shape [batch_size, n_labels]")
+    }
+
+    Ok(())
+}
+
 fn validate_embedding_model(model: Session) -> Result<()> {
     let shape = get_outp_dim(model.outputs.as_slice(), "last_hidden_state")?;
 

encoderfile/src/build_cli/tokenizer.rs

@@ -257,7 +257,7 @@ mod tests {
         let config = EncoderfileConfig {
             name: "my-model".into(),
             version: "0.0.1".into(),
-            path: ModelPath::Directory("../models/embedding".into()),
+            path: ModelPath::Directory("../models/dummy_electra_token_embeddings".into()),
             model_type: ModelType::Embedding,
             output_path: None,
             cache_dir: None,
@@ -296,7 +296,7 @@ mod tests {
         let config = EncoderfileConfig {
             name: "my-model".into(),
             version: "0.0.1".into(),
-            path: ModelPath::Directory("../models/embedding".into()),
+            path: ModelPath::Directory("../models/dummy_electra_token_embeddings".into()),
             model_type: ModelType::Embedding,
             output_path: None,
             cache_dir: None,
@@ -334,7 +334,7 @@ mod tests {
 
     #[test]
     fn test_validate_tokenizer_no_config() {
-        let path = ModelPath::Directory("../models/dummy_token_classifier".into());
+        let path = ModelPath::Directory("../models/dummy_electra_token_classifier".into());
 
         let explicit_path = ModelPath::Paths {
             model_config_path: path.model_config_path().unwrap(),

encoderfile/src/build_cli/transforms/validation/embedding.rs

@@ -64,7 +64,9 @@ mod tests {
         EncoderfileConfig {
             name: "my-model".to_string(),
             version: "0.0.1".to_string(),
-            path: ModelPath::Directory(std::path::PathBuf::from("models/embedding")),
+            path: ModelPath::Directory(std::path::PathBuf::from(
+                "models/dummy_electra_token_embeddings",
+            )),
             model_type: ModelType::Embedding,
             cache_dir: None,
             output_path: None,
@@ -77,7 +79,8 @@ mod tests {
     }
 
     fn test_model_config() -> ModelConfig {
-        let config_json = include_str!("../../../../../models/embedding/config.json");
+        let config_json =
+            include_str!("../../../../../models/dummy_electra_token_embeddings/config.json");
 
         serde_json::from_str(config_json).unwrap()
     }

encoderfile/src/build_cli/transforms/validation/mod.rs

@@ -111,7 +111,9 @@ mod tests {
         EncoderfileConfig {
             name: "my-model".to_string(),
             version: "0.0.1".to_string(),
-            path: ModelPath::Directory(std::path::PathBuf::from("models/embedding")),
+            path: ModelPath::Directory(std::path::PathBuf::from(
+                "models/dummy_electra_token_embeddings",
+            )),
             model_type: ModelType::Embedding,
             cache_dir: None,
             output_path: None,
@@ -124,7 +126,8 @@ mod tests {
     }
 
     fn test_model_config() -> ModelConfig {
-        let config_json = include_str!("../../../../../models/embedding/config.json");
+        let config_json =
+            include_str!("../../../../../models/dummy_electra_token_embeddings/config.json");
 
         serde_json::from_str(config_json).unwrap()
     }
@@ -156,7 +159,9 @@ mod tests {
         let encoderfile_config = EncoderfileConfig {
             name: "my-model".to_string(),
             version: "0.0.1".to_string(),
-            path: ModelPath::Directory(std::path::PathBuf::from("models/embedding")),
+            path: ModelPath::Directory(std::path::PathBuf::from(
+                "models/dummy_electra_token_embeddings",
+            )),
             model_type: ModelType::Embedding,
             cache_dir: None,
             output_path: None,
@@ -169,7 +174,7 @@ mod tests {
 
         let model_config_str = include_str!(concat!(
             "../../../../../models/",
-            "embedding",
+            "dummy_electra_token_embeddings",
             "/config.json"
         ));
 
@@ -197,7 +202,7 @@ mod tests {
 
         let model_config_str = include_str!(concat!(
             "../../../../../models/",
-            "embedding",
+            "dummy_electra_token_embeddings",
             "/config.json"
         ));
 

encoderfile/src/build_cli/transforms/validation/sentence_embedding.rs

@@ -67,7 +67,9 @@ mod tests {
         EncoderfileConfig {
             name: "my-model".to_string(),
             version: "0.0.1".to_string(),
-            path: ModelPath::Directory(std::path::PathBuf::from("models/embedding")),
+            path: ModelPath::Directory(std::path::PathBuf::from(
+                "models/dummy_electra_sequence_embeddings",
+            )),
             model_type: ModelType::SentenceEmbedding,
             cache_dir: None,
             output_path: None,
@@ -80,7 +82,8 @@ mod tests {
     }
 
     fn test_model_config() -> ModelConfig {
-        let config_json = include_str!("../../../../../models/embedding/config.json");
+        let config_json =
+            include_str!("../../../../../models/dummy_electra_sequence_embeddings/config.json");
 
         serde_json::from_str(config_json).unwrap()
     }

encoderfile/src/build_cli/transforms/validation/sequence_classification.rs

@@ -79,7 +79,7 @@ mod tests {
 
     fn test_model_config() -> ModelConfig {
         let config_json =
-            include_str!("../../../../../models/dummy_sequence_classifier/config.json");
+            include_str!("../../../../../models/dummy_electra_sequence_classifier/config.json");
 
         serde_json::from_str(config_json).unwrap()
     }

encoderfile/src/build_cli/transforms/validation/token_classification.rs

@@ -77,7 +77,8 @@ mod tests {
     }
 
     fn test_model_config() -> ModelConfig {
-        let config_json = include_str!("../../../../../models/dummy_token_classifier/config.json");
+        let config_json =
+            include_str!("../../../../../models/dummy_electra_token_classifier/config.json");
 
         serde_json::from_str(config_json).unwrap()
     }

encoderfile/src/dev_utils/mod.rs

@@ -10,9 +10,9 @@ use parking_lot::Mutex;
 use std::str::FromStr;
 use std::{fs::File, io::BufReader};
 
-const EMBEDDING_DIR: &str = "../models/embedding";
-const SEQUENCE_CLASSIFICATION_DIR: &str = "../models/sequence_classification";
-const TOKEN_CLASSIFICATION_DIR: &str = "../models/token_classification";
+const EMBEDDING_DIR: &str = "../models/dummy_electra_token_embeddings";
+const SEQUENCE_CLASSIFICATION_DIR: &str = "../models/dummy_electra_sequence_classifier";
+const TOKEN_CLASSIFICATION_DIR: &str = "../models/dummy_electra_token_classifier";
 
 pub fn get_state<T: ModelTypeSpec>(dir: &str) -> AppState<T> {
     let config = Config {

encoderfile/tests/integration/test_build.rs

@@ -46,7 +46,7 @@ encoderfile:
     )
 }
 
-const MODEL_ASSETS_PATH: &str = "../models/dummy_token_classifier";
+const MODEL_ASSETS_PATH: &str = "../models/dummy_electra_token_classifier";
 
 #[tokio::test]
 async fn test_build_encoderfile() -> Result<()> {
@@ -56,7 +56,7 @@ async fn test_build_encoderfile() -> Result<()> {
         .canonicalize()
         .expect("Failed to canonicalize temp path");
 
-    let tmp_model_path = path.join("models").join("dummy_token_classifier");
+    let tmp_model_path = path.join("models").join("dummy_electra_token_classifier");
 
     let ef_config_path = path.join("encoderfile.yml");
     let encoderfile_path = path.join(BINARY_NAME);

encoderfile/tests/integration/test_inspect.rs

@@ -38,7 +38,7 @@ encoderfile:
     )
 }
 
-const MODEL_ASSETS_PATH: &str = "../models/token_classification";
+const MODEL_ASSETS_PATH: &str = "../models/dummy_electra_token_classifier";
 
 #[test]
 fn test_inspect_encoderfile() -> Result<()> {
@@ -48,7 +48,7 @@ fn test_inspect_encoderfile() -> Result<()> {
         .canonicalize()
         .expect("Failed to canonicalize temp path");
 
-    let tmp_model_path = path.join("models").join("token_classification");
+    let tmp_model_path = path.join("models").join("dummy_electra_token_classifier");
 
     let ef_config_path = path.join("encoderfile.yml");
     let encoderfile_path = path.join(BINARY_NAME);

encoderfile/tests/test_grpc.rs

@@ -46,6 +46,8 @@ macro_rules! test_grpc_service {
                     .unwrap()
                     .into_inner();
 
+                println!("Model metadata: {:?}", response);
+
                 if $has_labels {
                     assert!(!response.id2label.is_empty(), "id2label is an empty dict")
                 } else {

encoderfile/tests/test_model_validation.rs

@@ -4,50 +4,44 @@ use encoderfile::{build_cli::model::ModelTypeExt as _, common::ModelType};
 
 #[test]
 pub fn test_embedding() {
-    let path = PathBuf::from("../models/embedding/model.onnx");
+    let path = PathBuf::from("../models/dummy_electra_token_embeddings/model.onnx");
 
     assert!(ModelType::Embedding.validate_model(&path).is_ok());
-    assert!(
-        ModelType::SequenceClassification
-            .validate_model(&path)
-            .is_err()
-    );
     assert!(
         ModelType::TokenClassification
             .validate_model(&path)
             .is_err()
     );
-    assert!(ModelType::SentenceEmbedding.validate_model(&path).is_ok())
 }
 
 #[test]
-pub fn test_sequence_classification() {
-    let path = PathBuf::from("../models/dummy_sequence_classifier/model.onnx");
+pub fn test_token_classification() {
+    let path = PathBuf::from("../models/dummy_electra_token_classifier/model.onnx");
 
     assert!(ModelType::Embedding.validate_model(&path).is_err());
+    assert!(ModelType::TokenClassification.validate_model(&path).is_ok());
+}
+
+#[test]
+pub fn test_sequence_embeddings() {
+    let path = PathBuf::from("../models/dummy_electra_sequence_embeddings/model.onnx");
+
+    assert!(ModelType::SentenceEmbedding.validate_model(&path).is_ok());
     assert!(
         ModelType::SequenceClassification
-            .validate_model(&path)
-            .is_ok()
-    );
-    assert!(
-        ModelType::TokenClassification
             .validate_model(&path)
             .is_err()
     );
-    assert!(ModelType::SentenceEmbedding.validate_model(&path).is_err())
 }
 
 #[test]
-pub fn test_token_classification() {
-    let path = PathBuf::from("../models/dummy_token_classifier/model.onnx");
+pub fn test_sequence_classification() {
+    let path = PathBuf::from("../models/dummy_electra_sequence_classifier/model.onnx");
 
-    assert!(ModelType::Embedding.validate_model(&path).is_err());
+    assert!(ModelType::SentenceEmbedding.validate_model(&path).is_err());
     assert!(
         ModelType::SequenceClassification
             .validate_model(&path)
-            .is_err()
+            .is_ok()
     );
-    assert!(ModelType::TokenClassification.validate_model(&path).is_ok());
-    assert!(ModelType::SentenceEmbedding.validate_model(&path).is_err())
 }

models/dummy_electra_sequence_classifier/config.json

@@ -0,0 +1,47 @@
+{
+  "architectures": [
+    "DummySequenceClassifier"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "embedding_size": 128,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 256,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2",
+    "3": "LABEL_3",
+    "4": "LABEL_4",
+    "5": "LABEL_5",
+    "6": "LABEL_6"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 1024,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2,
+    "LABEL_3": 3,
+    "LABEL_4": 4,
+    "LABEL_5": 5,
+    "LABEL_6": 6
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "mozilla-ai/test-dummy-sequence-classifier",
+  "num_attention_heads": 4,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "summary_activation": "gelu",
+  "summary_last_dropout": 0.1,
+  "summary_type": "first",
+  "summary_use_proj": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.55.4",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

models/dummy_electra_sequence_classifier/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}