🥅🏷️ Use more defined struct for threshold (#66)

* 🥅🏷️ Use more defined struct for threshold

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>

* 🏷️ Update threshold to f64

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>

* 📝 Update TODOs

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>

---------

Signed-off-by: Evaline Ju <69598118+evaline-ju@users.noreply.github.com>

Author: evaline-ju

TODOs.md

@@ -1,14 +1,8 @@
 # TODOs
-- [ ] Design detector map config and implement loading this config file
 - [ ] Add TLS support for orchestrator server
 - [ ] Implement Health probes
-- [ ] Add request models (objects for request and response)
 - [ ] Add unit tests
-- [ ] Add request validation for classification with text generation endpoint
 - [ ] Add request validation for streaming classification with text generation endpoint
-- [ ] Host API in swagger pages in the repo
 - [ ] Tokenization REST API and client will need to be updated for bidirectional streaming if/when available for REST use
 - [ ] There is currently NO WAY for us to know the prefix ID required by TGIS for inferencing on tuned prompts
 - [ ] Add configurable request timeouts for all clients
-- [ ] [OSS public] Copyright headers
-- [ ] [OSS public] CI - github actions

src/config.rs

@@ -80,7 +80,7 @@ pub struct DetectorConfig {
     /// ID of chunker that this detector will use
     pub chunker_id: String,
     /// Default threshold with which to filter detector results by score
-    pub default_threshold: f32,
+    pub default_threshold: f64,
 }
 
 /// Overall orchestrator server configuration

src/models.rs

@@ -3,11 +3,12 @@
 use crate::pb;
 use std::collections::HashMap;
 
-// TODO: When detector API is updated, consider if fields
-// like 'threshold' can be named options instead of the
-// use a generic HashMap with Values here
-// ref. https://github.com/foundation-model-stack/fms-guardrails-orchestrator/issues/37
-pub type DetectorParams = HashMap<String, serde_json::Value>;
+/// Parameters relevant to each detector
+#[derive(Clone, Debug, PartialEq, serde::Serialize, serde::Deserialize)]
+pub struct DetectorParams {
+    /// Threshold with which to filter detector results by score
+    pub threshold: Option<f64>,
+}
 
 /// User request to orchestrator
 #[derive(Clone, Debug, serde::Serialize, serde::Deserialize)]

src/orchestrator.rs

@@ -323,15 +323,12 @@ async fn handle_chunk_task(
 async fn handle_detection_task(
     ctx: Arc<Context>,
     detector_id: String,
-    default_threshold: f32,
+    default_threshold: f64,
     detector_params: DetectorParams,
     chunks: Vec<Chunk>,
 ) -> Result<Vec<TokenClassificationResult>, Error> {
     let detector_id = detector_id.clone();
-    let threshold = detector_params
-        .get("threshold")
-        .and_then(|v| v.as_f64())
-        .unwrap_or(default_threshold as f64);
+    let threshold = detector_params.threshold.unwrap_or(default_threshold);
     let contents = chunks.iter().map(|chunk| chunk.text.clone()).collect();
     let request = ContentAnalysisRequest::new(contents);
     debug!(