WIP: Adding logits processors to protocol

proto/generate.proto

@@ -50,6 +50,13 @@ message ClearCacheRequest {
 /// Empty response
 message ClearCacheResponse {}
 
+message LogitsProcessorParameters {
+    // The name of the processor to apply
+    string name = 1;
+    // The parameters to pass to the processor
+    repeated string parameters = 2;
+}
+
 message NextTokenChooserParameters {
     /// exponential scaling output probability distribution
     float temperature = 1;
@@ -67,6 +74,8 @@ message NextTokenChooserParameters {
     float repetition_penalty = 7;
     /// token watermarking using "A Watermark for Large Language Models"
     bool watermark = 8;
+    /// Optional Logits Processors definitions
+    repeated LogitsProcessorParameters logits_processors = 9;
 }
 
 message StoppingCriteriaParameters {

router/client/src/client.rs

@@ -125,6 +125,7 @@ impl Client {
                     seed: 0,
                     repetition_penalty: 1.2,
                     watermark: true,
+                    logits_processors: vec![],
                 }),
                 stopping_parameters: Some(StoppingCriteriaParameters {
                     max_new_tokens: max_total_tokens - truncate,

router/src/health.rs

@@ -44,6 +44,7 @@ impl Health {
                     seed: 0,
                     repetition_penalty: 1.0,
                     watermark: false,
+                    logits_processors: vec![],
                 }),
                 stopping_parameters: Some(StoppingCriteriaParameters {
                     max_new_tokens: 1,

router/src/validation.rs

@@ -279,6 +279,7 @@ impl Validation {
             do_sample,
             seed,
             watermark,
+            logits_processors: vec![],
         };
         let stopping_parameters = StoppingCriteriaParameters {
             max_new_tokens,