feat: add auto-truncate arg (huggingface#224)

Author: OlivierDehaene

router/src/http/server.rs

@@ -159,7 +159,7 @@ async fn predict(
             let permit = infer.try_acquire_permit().map_err(ErrorResponse::from)?;
             let (prompt_tokens, tokenization, queue, inference, predictions) = predict_inner(
                 inputs,
-                req.truncate,
+                req.truncate.unwrap_or(info.auto_truncate),
                 req.raw_scores,
                 infer.0,
                 info.0,
@@ -208,7 +208,7 @@ async fn predict(
                 let local_info = info.clone();
                 futures.push(predict_inner(
                     input,
-                    req.truncate,
+                    req.truncate.unwrap_or(info.auto_truncate),
                     req.raw_scores,
                     local_infer.0,
                     local_info.0,
@@ -370,7 +370,7 @@ async fn rerank(
             futures.push(rerank_inner(
                 req.query.clone(),
                 text.clone(),
-                req.truncate,
+                req.truncate.unwrap_or(info.auto_truncate),
                 req.raw_scores,
                 local_infer.0,
             ))
@@ -478,7 +478,12 @@ async fn embed(
 
             let permit = infer.try_acquire_permit().map_err(ErrorResponse::from)?;
             let response = infer
-                .embed_pooled(input, req.truncate, req.normalize, permit)
+                .embed_pooled(
+                    input,
+                    req.truncate.unwrap_or(info.auto_truncate),
+                    req.normalize,
+                    permit,
+                )
                 .await
                 .map_err(ErrorResponse::from)?;
 
@@ -531,11 +536,12 @@ async fn embed(
             for input in inputs {
                 compute_chars += input.count_chars();
 
+                let truncate = req.truncate.unwrap_or(info.auto_truncate);
                 let local_infer = infer.clone();
                 futures.push(async move {
                     let permit = local_infer.acquire_permit().await;
                     local_infer
-                        .embed_pooled(input, req.truncate, req.normalize, permit)
+                        .embed_pooled(input, truncate, req.normalize, permit)
                         .await
                 })
             }
@@ -634,7 +640,7 @@ async fn embed_sparse(
 
             let permit = infer.try_acquire_permit().map_err(ErrorResponse::from)?;
             let response = infer
-                .embed_sparse(input, req.truncate, permit)
+                .embed_sparse(input, req.truncate.unwrap_or(info.auto_truncate), permit)
                 .await
                 .map_err(ErrorResponse::from)?;
 
@@ -687,12 +693,11 @@ async fn embed_sparse(
             for input in inputs {
                 compute_chars += input.count_chars();
 
+                let truncate = req.truncate.unwrap_or(info.auto_truncate);
                 let local_infer = infer.clone();
                 futures.push(async move {
                     let permit = local_infer.acquire_permit().await;
-                    let response = local_infer
-                        .embed_sparse(input, req.truncate, permit)
-                        .await?;
+                    let response = local_infer.embed_sparse(input, truncate, permit).await?;
                     Ok((sparsify(response.results), response.metadata))
                 })
             }
@@ -782,7 +787,7 @@ async fn embed_all(
 
             let permit = infer.try_acquire_permit().map_err(ErrorResponse::from)?;
             let response = infer
-                .embed_all(input, req.truncate, permit)
+                .embed_all(input, req.truncate.unwrap_or(info.auto_truncate), permit)
                 .await
                 .map_err(ErrorResponse::from)?;
 
@@ -835,10 +840,11 @@ async fn embed_all(
             for input in inputs {
                 compute_chars += input.count_chars();
 
+                let truncate = req.truncate.unwrap_or(info.auto_truncate);
                 let local_infer = infer.clone();
                 futures.push(async move {
                     let permit = local_infer.acquire_permit().await;
-                    local_infer.embed_all(input, req.truncate, permit).await
+                    local_infer.embed_all(input, truncate, permit).await
                 })
             }
             let results = join_all(futures)

router/src/http/types.rs

@@ -196,9 +196,8 @@ impl<'__s> ToSchema<'__s> for PredictInput {
 #[derive(Deserialize, ToSchema)]
 pub(crate) struct PredictRequest {
     pub inputs: PredictInput,
-    #[serde(default)]
-    #[schema(default = "false", example = "false")]
-    pub truncate: bool,
+    #[schema(default = "false", example = "false", nullable = true)]
+    pub truncate: Option<bool>,
     #[serde(default)]
     #[schema(default = "false", example = "false")]
     pub raw_scores: bool,
@@ -226,8 +225,8 @@ pub(crate) struct RerankRequest {
     #[schema(example = json!(["Deep Learning is ..."]))]
     pub texts: Vec<String>,
     #[serde(default)]
-    #[schema(default = "false", example = "false")]
-    pub truncate: bool,
+    #[schema(default = "false", example = "false", nullable = true)]
+    pub truncate: Option<bool>,
     #[serde(default)]
     #[schema(default = "false", example = "false")]
     pub raw_scores: bool,
@@ -322,8 +321,8 @@ pub(crate) struct OpenAICompatResponse {
 pub(crate) struct EmbedRequest {
     pub inputs: Input,
     #[serde(default)]
-    #[schema(default = "false", example = "false")]
-    pub truncate: bool,
+    #[schema(default = "false", example = "false", nullable = true)]
+    pub truncate: Option<bool>,
     #[serde(default = "default_normalize")]
     #[schema(default = "true", example = "true")]
     pub normalize: bool,
@@ -341,8 +340,8 @@ pub(crate) struct EmbedResponse(pub Vec<Vec<f32>>);
 pub(crate) struct EmbedSparseRequest {
     pub inputs: Input,
     #[serde(default)]
-    #[schema(default = "false", example = "false")]
-    pub truncate: bool,
+    #[schema(default = "false", example = "false", nullable = true)]
+    pub truncate: Option<bool>,
 }
 
 #[derive(Serialize, ToSchema)]
@@ -358,8 +357,8 @@ pub(crate) struct EmbedSparseResponse(pub Vec<Vec<SparseValue>>);
 pub(crate) struct EmbedAllRequest {
     pub inputs: Input,
     #[serde(default)]
-    #[schema(default = "false", example = "false")]
-    pub truncate: bool,
+    #[schema(default = "false", example = "false", nullable = true)]
+    pub truncate: Option<bool>,
 }
 
 #[derive(Serialize, ToSchema)]

router/src/lib.rs

@@ -51,6 +51,7 @@ pub async fn run(
     max_batch_tokens: usize,
     max_batch_requests: Option<usize>,
     max_client_batch_size: usize,
+    auto_truncate: bool,
     hf_api_token: Option<String>,
     hostname: Option<String>,
     port: u16,
@@ -236,6 +237,7 @@ pub async fn run(
         tokenization_workers,
         max_batch_requests,
         max_client_batch_size,
+        auto_truncate,
         version: env!("CARGO_PKG_VERSION"),
         sha: option_env!("VERGEN_GIT_SHA"),
         docker_label: option_env!("DOCKER_LABEL"),
@@ -428,6 +430,7 @@ pub struct Info {
     pub max_batch_requests: Option<usize>,
     #[cfg_attr(feature = "http", schema(example = "32"))]
     pub max_client_batch_size: usize,
+    pub auto_truncate: bool,
     #[cfg_attr(feature = "http", schema(example = "4"))]
     pub tokenization_workers: usize,
     /// Router Info

router/src/main.rs

@@ -73,6 +73,12 @@ struct Args {
     #[clap(default_value = "32", long, env)]
     max_client_batch_size: usize,
 
+    /// Automatically truncate inputs that are longer than the maximum supported size
+    ///
+    /// Unused for gRPC servers
+    #[clap(long, env)]
+    auto_truncate: bool,
+
     /// Your HuggingFace hub token
     #[clap(long, env)]
     #[redact(partial)]
@@ -117,7 +123,7 @@ struct Args {
     #[clap(long, env)]
     otlp_endpoint: Option<String>,
 
-    // Unused for gRPC servers
+    /// Unused for gRPC servers
     #[clap(long, env)]
     cors_allow_origin: Option<Vec<String>>,
 }
@@ -143,6 +149,7 @@ async fn main() -> Result<()> {
         args.max_batch_tokens,
         args.max_batch_requests,
         args.max_client_batch_size,
+        args.auto_truncate,
         args.hf_api_token,
         Some(args.hostname),
         args.port,