refactor: Upgrade async-openai (#1693)

Cargo.lock

@@ -199,10 +199,11 @@ checksum = "4288f83726785267c6f2ef073a3d83dc3f9b81464e9f99898240cced85fce35a"
 
 [[package]]
 name = "async-openai"
-version = "0.27.2"
+version = "0.29.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2d126927c78e1562d7e8473008ac8b082318c04d69e3a83e3495a563f8b84a66"
+checksum = "31acf814d6b499e33ec894bb0fd7ddaf2665b44fbdd42b858d736449271fde0c"
 dependencies = [
+ "async-openai-macros",
  "backoff",
  "base64 0.22.1",
  "bytes",
@@ -222,6 +223,17 @@ dependencies = [
  "tracing",
 ]
 
+[[package]]
+name = "async-openai-macros"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0289cba6d5143bfe8251d57b4a8cac036adf158525a76533a7082ba65ec76398"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.100",
+]
+
 [[package]]
 name = "async-stream"
 version = "0.3.6"

Cargo.toml

@@ -46,6 +46,7 @@ dynamo-tokens = { path = "lib/tokens", version = "0.3.1" }
 # External dependencies
 anyhow = { version = "1" }
 async-nats = { version = "0.40", features = ["service"] }
+async-openai = { version = "0.29.0" }
 async-stream = { version = "0.3" }
 async-trait = { version = "0.1" }
 async_zmq = { version = "0.4.0" }

launch/dynamo-run/Cargo.toml

@@ -45,7 +45,7 @@ tracing = { workspace = true }
 tracing-subscriber = { workspace = true }
 uuid = { workspace = true }
 
-async-openai = { version = "0.27.2" }
+async-openai = { workspace = true }
 clap = { version = "4.5", features = ["derive", "env"] }
 dialoguer = { version = "0.11", default-features = false, features = ["editor", "history"] }
 futures-util = { version = "0.3" }

lib/bindings/python/Cargo.lock

@@ -180,10 +180,11 @@ checksum = "4288f83726785267c6f2ef073a3d83dc3f9b81464e9f99898240cced85fce35a"
 
 [[package]]
 name = "async-openai"
-version = "0.27.2"
+version = "0.29.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2d126927c78e1562d7e8473008ac8b082318c04d69e3a83e3495a563f8b84a66"
+checksum = "31acf814d6b499e33ec894bb0fd7ddaf2665b44fbdd42b858d736449271fde0c"
 dependencies = [
+ "async-openai-macros",
  "backoff",
  "base64 0.22.1",
  "bytes",
@@ -203,6 +204,17 @@ dependencies = [
  "tracing",
 ]
 
+[[package]]
+name = "async-openai-macros"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0289cba6d5143bfe8251d57b4a8cac036adf158525a76533a7082ba65ec76398"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.100",
+]
+
 [[package]]
 name = "async-stream"
 version = "0.3.6"

lib/bindings/python/Cargo.toml

@@ -44,7 +44,7 @@ dynamo-runtime = { path = "../../runtime" }
 anyhow = { version = "1" }
 async-stream = { version = "0.3" }
 async-trait = { version = "0.1" }
-async-openai = "0.27.2"
+async-openai = { version = "0.29.0" }
 futures = { version = "0.3" }
 once_cell = { version = "1.20.3" }
 serde = { version = "1" }

lib/engines/mistralrs/Cargo.toml

@@ -34,7 +34,7 @@ dynamo-runtime = { workspace = true }
 dynamo-llm = { workspace = true }
 
 anyhow = { workspace = true }
-async-openai = "0.27.2"
+async-openai = { workspace = true }
 async-stream = { workspace = true }
 async-trait = { workspace = true }
 candle-core = { version = "0.8.0" }

lib/llm/Cargo.toml

@@ -71,7 +71,7 @@ uuid = { workspace = true }
 xxhash-rust = { workspace = true }
 
 akin = "0.4.0"
-async-openai = "0.27.2"
+async-openai = { workspace = true }
 blake3 = "1"
 bytemuck = "1.22"
 candle-core = { version = "0.8.0" }

lib/llm/src/preprocessor/prompt/template/oai.rs

@@ -98,15 +98,12 @@ impl OAIChatLikeRequest for NvCreateCompletionRequest {
 
     fn extract_tokens(&self) -> Option<TokenInput> {
         match &self.inner.prompt {
-            async_openai::types::Prompt::IntegerArray(tokens) => Some(TokenInput::Single(
-                tokens.iter().map(|&t| t as u32).collect(),
-            )),
-            async_openai::types::Prompt::ArrayOfIntegerArray(arrays) => Some(TokenInput::Batch(
-                arrays
-                    .iter()
-                    .map(|arr| arr.iter().map(|&t| t as u32).collect())
-                    .collect(),
-            )),
+            async_openai::types::Prompt::IntegerArray(tokens) => {
+                Some(TokenInput::Single(tokens.clone()))
+            }
+            async_openai::types::Prompt::ArrayOfIntegerArray(arrays) => {
+                Some(TokenInput::Batch(arrays.clone()))
+            }
             _ => None,
         }
     }