refactor(config): increase rate limit to 1200, set 6000 for llama-cpp…

…-server based config (#3482)

crates/http-api-bindings/src/rate_limit.rs

@@ -11,28 +11,23 @@ use futures::stream::BoxStream;
 use ratelimit::Ratelimiter;
 use tabby_inference::{ChatCompletionStream, CompletionOptions, CompletionStream, Embedding};
 
-fn new_rate_limiter(rpm: u64) -> anyhow::Result<Ratelimiter> {
+fn new_rate_limiter(rpm: u64) -> Ratelimiter {
     Ratelimiter::builder(rpm, Duration::from_secs(60))
         .max_tokens(rpm)
         .initial_available(rpm)
         .build()
-        .map_err(|e| {
-            anyhow::anyhow!(
-                "Failed to create ratelimiter, please check the rate limit configuration: {}",
-                e,
-            )
-        })
+        .expect("Failed to create RateLimiter, please check the HttpModelConfig.rate_limit configuration")
 }
 
 pub struct RateLimitedEmbedding {
     embedding: Box<dyn Embedding>,
     rate_limiter: Ratelimiter,
 }
 
-pub fn new_embedding(embedding: Box<dyn Embedding>, rpm: u64) -> impl Embedding {
+pub fn new_embedding(embedding: Box<dyn Embedding>, request_per_minute: u64) -> impl Embedding {
     RateLimitedEmbedding {
         embedding,
-        rate_limiter: new_rate_limiter(rpm).unwrap(),
+        rate_limiter: new_rate_limiter(request_per_minute),
     }
 }
 
@@ -57,10 +52,13 @@ pub struct RateLimitedCompletion {
     rate_limiter: Ratelimiter,
 }
 
-pub fn new_completion(completion: Box<dyn CompletionStream>, rpm: u64) -> impl CompletionStream {
+pub fn new_completion(
+    completion: Box<dyn CompletionStream>,
+    request_per_minute: u64,
+) -> impl CompletionStream {
     RateLimitedCompletion {
         completion,
-        rate_limiter: new_rate_limiter(rpm).unwrap(),
+        rate_limiter: new_rate_limiter(request_per_minute),
     }
 }
 
@@ -86,10 +84,13 @@ pub struct RateLimitedChatStream {
     rate_limiter: Ratelimiter,
 }
 
-pub fn new_chat(completion: Box<dyn ChatCompletionStream>, rpm: u64) -> impl ChatCompletionStream {
+pub fn new_chat(
+    completion: Box<dyn ChatCompletionStream>,
+    request_per_minute: u64,
+) -> impl ChatCompletionStream {
     RateLimitedChatStream {
         completion,
-        rate_limiter: new_rate_limiter(rpm).unwrap(),
+        rate_limiter: new_rate_limiter(request_per_minute),
     }
 }
 

crates/llama-cpp-server/src/lib.rs

@@ -9,7 +9,7 @@ use futures::stream::BoxStream;
 use serde::Deserialize;
 use supervisor::LlamaCppSupervisor;
 use tabby_common::{
-    config::{HttpModelConfigBuilder, LocalModelConfig, ModelConfig},
+    config::{HttpModelConfigBuilder, LocalModelConfig, ModelConfig, RateLimit, RateLimitBuilder},
     registry::{parse_model_id, ModelRegistry, GGML_MODEL_PARTITIONED_PREFIX},
 };
 use tabby_inference::{ChatCompletionStream, CompletionOptions, CompletionStream, Embedding};
@@ -46,6 +46,7 @@ impl EmbeddingServer {
 
         let config = HttpModelConfigBuilder::default()
             .api_endpoint(Some(api_endpoint(server.port())))
+            .rate_limit(build_rate_limit_config())
             .kind("llama.cpp/embedding".to_string())
             .build()
             .expect("Failed to create HttpModelConfig");
@@ -95,6 +96,7 @@ impl CompletionServer {
     async fn new_with_supervisor(server: Arc<LlamaCppSupervisor>) -> Self {
         let config = HttpModelConfigBuilder::default()
             .api_endpoint(Some(api_endpoint(server.port())))
+            .rate_limit(build_rate_limit_config())
             .kind("llama.cpp/completion".to_string())
             .build()
             .expect("Failed to create HttpModelConfig");
@@ -142,6 +144,7 @@ impl ChatCompletionServer {
     async fn new_with_supervisor(server: Arc<LlamaCppSupervisor>) -> Self {
         let config = HttpModelConfigBuilder::default()
             .api_endpoint(Some(api_endpoint(server.port())))
+            .rate_limit(build_rate_limit_config())
             .kind("openai/chat".to_string())
             .model_name(Some("local".into()))
             .build()
@@ -320,3 +323,10 @@ async fn resolve_prompt_info(model_id: &str) -> PromptInfo {
         }
     }
 }
+
+fn build_rate_limit_config() -> RateLimit {
+    RateLimitBuilder::default()
+        .request_per_minute(6000)
+        .build()
+        .expect("Failed to create RateLimit")
+}

crates/tabby-common/src/config.rs

@@ -289,7 +289,6 @@ pub struct HttpModelConfig {
     #[builder(default)]
     pub api_key: Option<String>,
 
-    #[builder(default)]
     #[serde(default)]
     pub rate_limit: RateLimit,
 
@@ -354,7 +353,7 @@ pub struct RateLimit {
 impl Default for RateLimit {
     fn default() -> Self {
         Self {
-            request_per_minute: 600,
+            request_per_minute: 1200,
         }
     }
 }