feat(router): add max_total_tokens and empty_input validation

router/src/main.rs

@@ -20,6 +20,12 @@ use tracing_subscriber::{EnvFilter, Layer};
 struct Args {
     #[clap(default_value = "128", long, env)]
     max_concurrent_requests: usize,
+    #[clap(default_value = "512", long, env)]
+    max_max_new_tokens: u32,
+    #[clap(default_value = "4", long, env)]
+    max_stop_sequences: usize,
+    #[clap(default_value = "1512", long, env)]
+    max_total_tokens: usize,
     #[clap(default_value = "1000", long, env)]
     max_input_length: usize,
     #[clap(default_value = "32", long, env)]
@@ -46,7 +52,10 @@ fn main() -> Result<(), std::io::Error> {
     // Pattern match configuration
     let Args {
         max_concurrent_requests,
+        max_max_new_tokens,
+        max_stop_sequences,
         max_input_length,
+        max_total_tokens,
         max_batch_size,
         max_waiting_tokens,
         port,
@@ -92,7 +101,10 @@ fn main() -> Result<(), std::io::Error> {
             // Run server
             server::run(
                 max_concurrent_requests,
+                max_max_new_tokens,
+                max_stop_sequences,
                 max_input_length,
+                max_total_tokens,
                 max_batch_size,
                 max_waiting_tokens,
                 sharded_client,

router/src/server.rs

@@ -291,7 +291,10 @@ async fn generate_stream(
 #[allow(clippy::too_many_arguments)]
 pub async fn run(
     max_concurrent_requests: usize,
+    max_max_new_tokens: u32,
+    max_stop_sequences: usize,
     max_input_length: usize,
+    max_total_tokens: usize,
     max_batch_size: usize,
     max_waiting_tokens: usize,
     client: ShardedClient,
@@ -333,7 +336,14 @@ pub async fn run(
     struct ApiDoc;
 
     // Create state
-    let validation = Validation::new(validation_workers, tokenizer, max_input_length);
+    let validation = Validation::new(
+        validation_workers,
+        tokenizer,
+        max_max_new_tokens,
+        max_stop_sequences,
+        max_input_length,
+        max_total_tokens,
+    );
     let infer = Infer::new(
         client,
         validation,

router/src/validation.rs

@@ -1,3 +1,4 @@
+use crate::validation::ValidationError::EmptyInput;
 /// Payload validation logic
 use crate::{GenerateParameters, GenerateRequest};
 use rand::rngs::ThreadRng;
@@ -8,9 +9,6 @@ use tokenizers::tokenizer::Tokenizer;
 use tokio::sync::{mpsc, oneshot};
 use tracing::{instrument, Span};
 
-const MAX_MAX_NEW_TOKENS: u32 = 512;
-const MAX_STOP_SEQUENCES: usize = 4;
-
 /// Validation
 #[derive(Debug, Clone)]
 pub struct Validation {
@@ -19,7 +17,14 @@ pub struct Validation {
 }
 
 impl Validation {
-    pub(crate) fn new(workers: usize, tokenizer: Tokenizer, max_input_length: usize) -> Self {
+    pub(crate) fn new(
+        workers: usize,
+        tokenizer: Tokenizer,
+        max_max_new_tokens: u32,
+        max_stop_sequences: usize,
+        max_input_length: usize,
+        max_total_tokens: usize,
+    ) -> Self {
         // Create channel
         let (validation_sender, validation_receiver) = mpsc::channel(128);
 
@@ -27,7 +32,10 @@ impl Validation {
         tokio::spawn(validation_task(
             workers,
             tokenizer,
+            max_max_new_tokens,
+            max_stop_sequences,
             max_input_length,
+            max_total_tokens,
             validation_receiver,
         ));
 
@@ -61,7 +69,10 @@ impl Validation {
 async fn validation_task(
     workers: usize,
     tokenizer: Tokenizer,
+    max_max_new_tokens: u32,
+    max_stop_sequences: usize,
     max_input_length: usize,
+    max_total_tokens: usize,
     mut receiver: mpsc::Receiver<ValidationRequest>,
 ) {
     let mut workers_senders = Vec::with_capacity(workers);
@@ -75,7 +86,14 @@ async fn validation_task(
 
         // Spawn worker
         tokio::task::spawn_blocking(move || {
-            validation_worker(tokenizer_clone, max_input_length, worker_receiver)
+            validation_worker(
+                tokenizer_clone,
+                max_max_new_tokens,
+                max_stop_sequences,
+                max_input_length,
+                max_total_tokens,
+                worker_receiver,
+            )
         });
     }
 
@@ -95,7 +113,10 @@ async fn validation_task(
 /// the tokenizer
 fn validation_worker(
     tokenizer: Tokenizer,
+    max_max_new_tokens: u32,
+    max_stop_sequences: usize,
     max_input_length: usize,
+    max_total_tokens: usize,
     mut receiver: mpsc::Receiver<ValidationRequest>,
 ) {
     // Seed rng
@@ -106,7 +127,16 @@ fn validation_worker(
         parent_span.in_scope(|| {
             response_tx
                 .send(
-                    validate(request, &tokenizer, max_input_length, &mut rng).map_err(|err| {
+                    validate(
+                        request,
+                        &tokenizer,
+                        max_max_new_tokens,
+                        max_stop_sequences,
+                        max_input_length,
+                        max_total_tokens,
+                        &mut rng,
+                    )
+                    .map_err(|err| {
                         tracing::error!("{err}");
                         err
                     }),
@@ -119,7 +149,10 @@ fn validation_worker(
 fn validate(
     request: GenerateRequest,
     tokenizer: &Tokenizer,
+    max_max_new_tokens: u32,
+    max_stop_sequences: usize,
     max_input_length: usize,
+    max_total_tokens: usize,
     rng: &mut ThreadRng,
 ) -> Result<ValidGenerateRequest, ValidationError> {
     let GenerateParameters {
@@ -161,13 +194,13 @@ fn validate(
         }
     }?;
 
-    if max_new_tokens == 0 || max_new_tokens > MAX_MAX_NEW_TOKENS {
-        return Err(ValidationError::MaxNewTokens(MAX_MAX_NEW_TOKENS));
+    if max_new_tokens == 0 || max_new_tokens > max_max_new_tokens {
+        return Err(ValidationError::MaxNewTokens(max_max_new_tokens));
     }
 
-    if stop_sequences.len() > MAX_STOP_SEQUENCES {
+    if stop_sequences.len() > max_stop_sequences {
         return Err(ValidationError::StopSequence(
-            MAX_STOP_SEQUENCES,
+            max_stop_sequences,
             stop_sequences.len(),
         ));
     }
@@ -178,13 +211,23 @@ fn validate(
         Some(seed) => seed,
     };
 
+    // Check if inputs is empty
+    if request.inputs.is_empty() {
+        return Err(EmptyInput);
+    }
+
     // Get the number of tokens in the input
     match tokenizer.encode(request.inputs.clone(), true) {
         Ok(encoding) => {
             let input_length = encoding.len();
-
+            let total_tokens = input_length + max_new_tokens as usize;
             if input_length > max_input_length {
-                Err(ValidationError::InputLength(input_length, max_input_length))
+                Err(ValidationError::InputLength(max_input_length, input_length))
+            } else if total_tokens > max_total_tokens {
+                Err(ValidationError::MaxTotalTokens(
+                    max_total_tokens,
+                    total_tokens,
+                ))
             } else {
                 // Return ValidGenerateRequest
                 let parameters = NextTokenChooserParameters {
@@ -238,8 +281,12 @@ pub enum ValidationError {
     TopK,
     #[error("max_new_tokens must be strictly positive and <= {0}")]
     MaxNewTokens(u32),
-    #[error("inputs must have less than {1} tokens. Given: {0}")]
+    #[error("input tokens + max_new_tokens must be <= {0}. Given {1}")]
+    MaxTotalTokens(usize, usize),
+    #[error("inputs must have less than {0} tokens. Given: {1}")]
     InputLength(usize, usize),
+    #[error("inputs cannot be empty")]
+    EmptyInput,
     #[error("stop supports up to {0} stop sequences. Given: {1}")]
     StopSequence(usize, usize),
     #[error("tokenizer error {0}")]