update proto, generated types, and add new builders

Author: coszio

proto/collections.proto

@@ -425,7 +425,7 @@ enum RecommendStrategy {
   // examples, its score is then chosen from the `max(max_pos_score, max_neg_score)`.
   // If the `max_neg_score` is chosen then it is squared and negated.
   BestScore = 1;
-  
+
   // Uses custom search objective. Compares against all inputs, sums all the scores.
   // Scores against positive vectors are added, against negatives are subtracted.
   SumScores = 2;
@@ -631,13 +631,40 @@ message PowExpression {
 
 message DecayParamsExpression {
     // The variable to decay
-    Expression x = 1; 
+    Expression x = 1;
     // The target value to start decaying from. Defaults to 0.
-    optional Expression target = 2; 
+    optional Expression target = 2;
     // The scale factor of the decay, in terms of `x`. Defaults to 1.0. Must be a non-zero positive number.
-    optional float scale = 3; 
+    optional float scale = 3;
     // The midpoint of the decay. Defaults to 0.5. Output will be this value when `|x - target| == scale`.
-    optional float midpoint = 4; 
+    optional float midpoint = 4;
+}
+
+message NearestInputWithMmr {
+    // The vector to search for nearest neighbors.
+    VectorInput nearest = 1;
+
+    // Perform MMR (Maximal Marginal Relevance) reranking after search,
+    // using the same vector in this query to calculate relevance.
+    Mmr mmr = 2;
+}
+
+// Maximal Marginal Relevance (MMR) algorithm for re-ranking the points.
+message Mmr {
+    // Tunable parameter for the MMR algorithm.
+    // Determines the balance between diversity and relevance.
+    //
+    // A higher value favors diversity (dissimilarity to selected results),
+    // while a lower value favors relevance (similarity to the query vector).
+    //
+    // Must be in the range [0, 1].
+    // Default value is 0.5.
+    optional float diversity = 2;
+
+    // The maximum number of candidates to consider for re-ranking.
+    //
+    // If not specified, the `limit` value is used.
+    optional uint32 candidates_limit = 3;
 }
 
 message Query {
@@ -650,6 +677,7 @@ message Query {
     Fusion fusion = 6; // Fuse the results of multiple prefetches.
     Sample sample = 7; // Sample points from the collection.
     Formula formula = 8; // Score boosting via an arbitrary formula
+    NearestInputWithMmr nearest_with_mmr = 9; // Search nearest neighbors, but re-rank based on the Maximal Marginal Relevance algorithm.
   }
 }
 
@@ -830,7 +858,7 @@ message UpdateBatchPoints {
 message PointsOperationResponse {
   UpdateResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message UpdateResult {
@@ -887,25 +915,25 @@ message GroupsResult {
 message SearchResponse {
   repeated ScoredPoint result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message QueryResponse {
   repeated ScoredPoint result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message QueryBatchResponse {
   repeated BatchResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message QueryGroupsResponse {
   GroupsResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message BatchResult {
@@ -915,26 +943,26 @@ message BatchResult {
 message SearchBatchResponse {
   repeated BatchResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message SearchGroupsResponse {
   GroupsResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message CountResponse {
   CountResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message ScrollResponse {
   optional PointId next_page_offset = 1; // Use this offset for the next query
   repeated RetrievedPoint result = 2;
   double time = 3; // Time spent to process
-  optional HardwareUsage usage = 4;
+  optional Usage usage = 4;
 }
 
 message CountResult {
@@ -953,42 +981,43 @@ message RetrievedPoint {
 message GetResponse {
   repeated RetrievedPoint result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message RecommendResponse {
   repeated ScoredPoint result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message RecommendBatchResponse {
   repeated BatchResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message DiscoverResponse {
   repeated ScoredPoint result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message DiscoverBatchResponse {
   repeated BatchResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message RecommendGroupsResponse {
   GroupsResult result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message UpdateBatchResponse {
   repeated UpdateResult result = 1;
   double time = 2; // Time spent to process
+  optional Usage usage = 3;
 }
 
 message FacetResponse {
@@ -999,13 +1028,13 @@ message FacetResponse {
 message SearchMatrixPairsResponse {
   SearchMatrixPairs result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 message SearchMatrixOffsetsResponse {
   SearchMatrixOffsets result = 1;
   double time = 2; // Time spent to process
-  optional HardwareUsage usage = 3;
+  optional Usage usage = 3;
 }
 
 // ---------------------------------------------
@@ -1080,6 +1109,7 @@ message Match {
     RepeatedIntegers integers = 6; // Match multiple integers
     RepeatedIntegers except_integers = 7; // Match any other value except those integers
     RepeatedStrings except_keywords = 8; // Match any other value except those keywords
+    string phrase = 9; // Match phrase text
   }
 }
 
@@ -1166,6 +1196,27 @@ message GeoPoint {
   double lat = 2;
 }
 
+// ---------------------------------------------
+// ----------- Measurements collector ----------
+// ---------------------------------------------
+message Usage {
+  optional HardwareUsage hardware = 1;
+  optional InferenceUsage inference = 2;
+}
+
+
+// ---------------------------------------------
+// ------------ Inference measurements ----------
+// ---------------------------------------------
+
+message InferenceUsage {
+  map<string, ModelUsage> models = 1;
+}
+
+message ModelUsage {
+  uint64 tokens = 1;
+}
+
 // ---------------------------------------------
 // ------------ Hardware measurements ----------
 // ---------------------------------------------

proto/points.proto

@@ -4,26 +4,47 @@ use crate::qdrant::*;
 pub struct BinaryQuantizationBuilder {
     /// If true - quantized vectors always will be stored in RAM, ignoring the config of main storage
     pub(crate) always_ram: Option<Option<bool>>,
+    pub(crate) encoding: Option<Option<i32>>,
+    pub(crate) query_encoding: Option<Option<BinaryQuantizationQueryEncoding>>,
 }
 
 impl BinaryQuantizationBuilder {
     /// If true - quantized vectors always will be stored in RAM, ignoring the config of main storage
-    #[allow(unused_mut)]
     pub fn always_ram(self, value: bool) -> Self {
         let mut new = self;
-        new.always_ram = Option::Some(Option::Some(value));
+        new.always_ram = Some(Some(value));
+        new
+    }
+
+    /// Binary quantization encoding method
+    pub fn encoding(self, value: impl Into<BinaryQuantizationEncoding>) -> Self {
+        let mut new = self;
+        let encoding: BinaryQuantizationEncoding = value.into();
+        new.encoding = Some(Some(encoding.into()));
+        new
+    }
+
+    /// Asymmetric quantization configuration allows a query to have different quantization than stored vectors.
+    /// It can increase the accuracy of search at the cost of performance.
+    pub fn query_encoding(self, value: impl Into<BinaryQuantizationQueryEncoding>) -> Self {
+        let mut new = self;
+        new.query_encoding = Some(Some(value.into()));
         new
     }
 
     fn build_inner(self) -> Result<BinaryQuantization, BinaryQuantizationBuilderError> {
         Ok(BinaryQuantization {
             always_ram: self.always_ram.unwrap_or_default(),
+            encoding: self.encoding.unwrap_or_default(),
+            query_encoding: self.query_encoding.unwrap_or_default(),
         })
     }
     /// Create an empty builder, with all fields set to `None` or `PhantomData`.
     fn create_empty() -> Self {
         Self {
-            always_ram: core::default::Default::default(),
+            always_ram: Default::default(),
+            encoding: Default::default(),
+            query_encoding: Default::default(),
         }
     }
 }
@@ -57,8 +78,6 @@ impl BinaryQuantizationBuilder {
     }
 }
 
-// src/builders/binary_quantization_builder.rs
-
 #[non_exhaustive]
 #[derive(Debug)]
 pub enum BinaryQuantizationBuilderError {
@@ -96,3 +115,11 @@ impl From<String> for BinaryQuantizationBuilderError {
         Self::ValidationError(error)
     }
 }
+
+impl BinaryQuantizationQueryEncoding {
+    pub fn new_setting(setting: binary_quantization_query_encoding::Setting) -> Self {
+        Self {
+            variant: Some(binary_quantization_query_encoding::Variant::Setting(setting.into())),
+        }
+    }
+}