refactor: rename triton_distributed to dynemo (#22)

Co-authored-by: Graham King <grahamk@nvidia.com>

lib/llm/Cargo.toml

@@ -18,11 +18,11 @@ version = "0.2.1"
 edition = "2021"
 authors = ["NVIDIA"]
 license = "Apache-2.0"
-homepage = "https://github.com/triton-inference-server/triton_distributed"
-repository = "https://github.com/triton-inference-server/triton_distributed"
+homepage = "https://github.com/dynemo-ai/dynemo"
+repository = "https://github.com/dynemo-ai/dynemo.git"
 
 [package]
-name = "triton-distributed-llm"
+name = "dynemo-llm"
 version.workspace = true
 edition.workspace = true
 authors.workspace = true
@@ -43,7 +43,7 @@ vulkan = ["llama-cpp-2/vulkan"]
 
 [workspace.dependencies]
 # local or crates.io
-triton-distributed-runtime = { version = "0.2.0", path = "../runtime" }
+dynemo-runtime = { version = "0.2.0", path = "../runtime" }
 
 # crates.io
 anyhow = { version = "1" }
@@ -66,7 +66,7 @@ strum = { version = "0.27", features = ["derive"] }
 [dependencies]
 
 # repo
-triton-distributed-runtime = { workspace = true }
+dynemo-runtime = { workspace = true }
 
 # workspace
 anyhow = { workspace = true }

lib/llm/src/backend.rs

@@ -34,7 +34,7 @@ use futures::stream::{self, StreamExt};
 use tracing as log;
 
 use crate::model_card::model::{ModelDeploymentCard, TokenizerKind};
-use triton_distributed_runtime::{
+use dynemo_runtime::{
     pipeline::{
         async_trait, AsyncEngineContextProvider, ManyOut, Operator, ResponseStream,
         ServerStreamingEngine, SingleIn,

lib/llm/src/engines/llamacpp.rs

@@ -22,6 +22,11 @@ use std::{
 use anyhow::Context;
 use async_stream::stream;
 use async_trait::async_trait;
+use dynemo_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
+use dynemo_runtime::pipeline::error as pipeline_error;
+use dynemo_runtime::pipeline::{Error, ManyOut, SingleIn};
+use dynemo_runtime::protocols::annotated::Annotated;
+use dynemo_runtime::CancellationToken;
 use llama_cpp_2::{
     context::{params::LlamaContextParams, LlamaContext},
     llama_backend::LlamaBackend,
@@ -30,11 +35,6 @@ use llama_cpp_2::{
     sampling::LlamaSampler,
     token::LlamaToken,
 };
-use triton_distributed_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
-use triton_distributed_runtime::pipeline::error as pipeline_error;
-use triton_distributed_runtime::pipeline::{Error, ManyOut, SingleIn};
-use triton_distributed_runtime::protocols::annotated::Annotated;
-use triton_distributed_runtime::CancellationToken;
 
 use crate::backend::ExecutionContext;
 use crate::protocols::common::llm_backend::{BackendInput, LLMEngineOutput};

lib/llm/src/engines/mistralrs.rs

@@ -28,10 +28,10 @@ use mistralrs::{
 };
 use tokio::sync::mpsc::channel;
 
-use triton_distributed_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
-use triton_distributed_runtime::pipeline::error as pipeline_error;
-use triton_distributed_runtime::pipeline::{Error, ManyOut, SingleIn};
-use triton_distributed_runtime::protocols::annotated::Annotated;
+use dynemo_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
+use dynemo_runtime::pipeline::error as pipeline_error;
+use dynemo_runtime::pipeline::{Error, ManyOut, SingleIn};
+use dynemo_runtime::protocols::annotated::Annotated;
 
 use crate::protocols::openai::chat_completions::{
     NvCreateChatCompletionRequest, NvCreateChatCompletionStreamResponse,

lib/llm/src/engines/sglang.rs

@@ -17,8 +17,8 @@ use std::path::Path;
 use std::sync::Arc;
 
 use crate::backend::ExecutionContext;
-use triton_distributed_runtime::pipeline::error as pipeline_error;
-use triton_distributed_runtime::CancellationToken;
+use dynemo_runtime::pipeline::error as pipeline_error;
+use dynemo_runtime::CancellationToken;
 
 mod worker;
 

lib/llm/src/engines/sglang/engine.rs

@@ -19,10 +19,10 @@ use async_stream::stream;
 use async_trait::async_trait;
 
 use crate::protocols::common::llm_backend::{BackendInput, LLMEngineOutput};
-use triton_distributed_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
-use triton_distributed_runtime::pipeline::{Error, ManyOut, SingleIn};
-use triton_distributed_runtime::protocols::annotated::Annotated;
-use triton_distributed_runtime::runtime::CancellationToken;
+use dynemo_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
+use dynemo_runtime::pipeline::{Error, ManyOut, SingleIn};
+use dynemo_runtime::protocols::annotated::Annotated;
+use dynemo_runtime::runtime::CancellationToken;
 
 use crate::engines::MultiNodeConfig;
 

lib/llm/src/engines/sglang/worker.rs

@@ -37,8 +37,8 @@ use tokio::sync::mpsc::Sender;
 use tokio::{io::AsyncBufReadExt, sync::mpsc::error::SendError};
 use tokio::{io::AsyncReadExt as _, task::JoinHandle};
 
-use triton_distributed_runtime::protocols::annotated::Annotated;
-use triton_distributed_runtime::runtime::CancellationToken;
+use dynemo_runtime::protocols::annotated::Annotated;
+use dynemo_runtime::runtime::CancellationToken;
 
 use crate::engines::sglang::MultiGPUConfig;
 use crate::engines::MultiNodeConfig;

lib/llm/src/engines/trtllm.rs

@@ -16,7 +16,7 @@
 use std::sync::Arc;
 
 use crate::backend::ExecutionContext;
-use triton_distributed_runtime::pipeline::error as pipeline_error;
+use dynemo_runtime::pipeline::error as pipeline_error;
 
 pub mod executor;
 

lib/llm/src/engines/trtllm/executor/engine.rs

@@ -15,12 +15,12 @@
 
 use anyhow::{Error, Result};
 use async_trait::async_trait;
+use dynemo_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
+use dynemo_runtime::pipeline::{ManyOut, SingleIn};
+use dynemo_runtime::protocols::annotated::Annotated;
 use futures::stream;
 use tokio::sync::mpsc;
 use tokio_util::sync::CancellationToken;
-use triton_distributed_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
-use triton_distributed_runtime::pipeline::{ManyOut, SingleIn};
-use triton_distributed_runtime::protocols::annotated::Annotated;
 
 use super::Executor;
 use crate::protocols::common::llm_backend::{BackendInput, LLMEngineOutput};

lib/llm/src/engines/vllm.rs

@@ -19,8 +19,8 @@ use std::pin::Pin;
 use std::sync::Arc;
 use std::task::{Context, Poll};
 
-use triton_distributed_runtime::pipeline::error as pipeline_error;
-use triton_distributed_runtime::CancellationToken;
+use dynemo_runtime::pipeline::error as pipeline_error;
+use dynemo_runtime::CancellationToken;
 
 use crate::backend::ExecutionContext;
 use crate::engines::MultiNodeConfig;

lib/llm/src/engines/vllm/engine.rs

@@ -21,10 +21,10 @@ use async_trait::async_trait;
 use crate::engines::vllm::worker;
 use crate::engines::MultiNodeConfig;
 use crate::protocols::common::llm_backend::{BackendInput, LLMEngineOutput};
-use triton_distributed_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
-use triton_distributed_runtime::pipeline::{Error, ManyOut, SingleIn};
-use triton_distributed_runtime::protocols::annotated::Annotated;
-use triton_distributed_runtime::runtime::CancellationToken;
+use dynemo_runtime::engine::{AsyncEngine, AsyncEngineContextProvider, ResponseStream};
+use dynemo_runtime::pipeline::{Error, ManyOut, SingleIn};
+use dynemo_runtime::protocols::annotated::Annotated;
+use dynemo_runtime::runtime::CancellationToken;
 
 pub struct VllmEngine {
     cancel_token: CancellationToken,

lib/llm/src/engines/vllm/ray.rs

@@ -24,7 +24,7 @@ use tokio::select;
 use tokio::time;
 use tracing;
 
-use triton_distributed_runtime::CancellationToken;
+use dynemo_runtime::CancellationToken;
 
 /// Default is 16 seconds, we make it a bit shorter
 const RAY_STOP_TIMEOUT_SECS: u32 = 10;

lib/llm/src/engines/vllm/worker.rs

@@ -19,6 +19,8 @@ use std::{
 };
 
 use async_zmq::{SinkExt, StreamExt};
+use dynemo_runtime::protocols::annotated::Annotated;
+use dynemo_runtime::CancellationToken;
 use pyo3::{
     prelude::*,
     types::{IntoPyDict, PyBytes, PyString},
@@ -26,8 +28,6 @@ use pyo3::{
 use tokio::sync::mpsc::Sender;
 use tokio::task::JoinHandle;
 use tokio::{io::AsyncBufReadExt, sync::mpsc::error::SendError};
-use triton_distributed_runtime::protocols::annotated::Annotated;
-use triton_distributed_runtime::CancellationToken;
 
 use crate::engines::MultiNodeConfig;
 use crate::protocols::common::llm_backend::LLMEngineOutput;

lib/llm/src/http/service/discovery.rs

@@ -18,7 +18,7 @@ use std::sync::Arc;
 use serde::{Deserialize, Serialize};
 use tokio::sync::mpsc::Receiver;
 
-use triton_distributed_runtime::{
+use dynemo_runtime::{
     protocols::{self, annotated::Annotated},
     raise,
     transports::etcd::{KeyValue, WatchEvent},

lib/llm/src/http/service/openai.rs

@@ -48,7 +48,7 @@ use crate::types::{
     Annotated,
 };
 
-use triton_distributed_runtime::pipeline::{AsyncEngineContext, Context};
+use dynemo_runtime::pipeline::{AsyncEngineContext, Context};
 
 #[derive(Serialize, Deserialize)]
 pub(crate) struct ErrorResponse {
@@ -91,7 +91,7 @@ impl ErrorResponse {
         )
     }
 
-    /// The OAI endpoints call an [`triton_distributed_runtime::engine::AsyncEngine`] which are specialized to return
+    /// The OAI endpoints call an [`dynemo_runtime::engine::AsyncEngine`] which are specialized to return
     /// an [`anyhow::Error`]. This method will convert the [`anyhow::Error`] into an [`HttpError`].
     /// If successful, it will return the [`HttpError`] as an [`ErrorResponse::internal_server_error`]
     /// with the details of the error.
@@ -516,7 +516,7 @@ pub fn list_models_router(
     path: Option<String>,
 ) -> (Vec<RouteDoc>, Router) {
     // TODO: Why do we have this endpoint?
-    let custom_path = path.unwrap_or("/triton/alpha/list-models".to_string());
+    let custom_path = path.unwrap_or("/dynemo/alpha/list-models".to_string());
     let doc_for_custom = RouteDoc::new(axum::http::Method::GET, &custom_path);
 
     // Standard OpenAI compatible list models endpoint

lib/llm/src/kv_router.rs

@@ -14,11 +14,11 @@
 // limitations under the License.
 
 use anyhow::Result;
+use dynemo_runtime::{component::Component, DistributedRuntime};
 use futures::stream::StreamExt;
 use std::{sync::Arc, time::Duration};
 use tokio_util::sync::CancellationToken;
 use tracing;
-use triton_distributed_runtime::{component::Component, DistributedRuntime};
 
 pub mod indexer;
 pub mod protocols;
@@ -62,7 +62,7 @@ impl KvRouter {
     }
 
     pub async fn new(
-        nats_client: triton_distributed_runtime::transports::nats::Client,
+        nats_client: dynemo_runtime::transports::nats::Client,
         service_name: String,
         kv_subject: String,
     ) -> Result<Arc<Self>> {
@@ -135,7 +135,7 @@ impl KvRouter {
 }
 
 async fn collect_endpoints(
-    nats_client: triton_distributed_runtime::transports::nats::Client,
+    nats_client: dynemo_runtime::transports::nats::Client,
     service_name: String,
     ep_tx: tokio::sync::mpsc::Sender<ProcessedEndpoints>,
     cancel: CancellationToken,

lib/llm/src/kv_router/publisher.rs

@@ -15,11 +15,7 @@
 
 use crate::kv_router::{indexer::RouterEvent, protocols::*, KV_EVENT_SUBJECT};
 use async_trait::async_trait;
-use futures::stream;
-use std::sync::Arc;
-use tokio::sync::mpsc;
-use tracing as log;
-use triton_distributed_runtime::{
+use dynemo_runtime::{
     component::Component,
     pipeline::{
         network::Ingress, AsyncEngine, AsyncEngineContextProvider, ManyOut, ResponseStream,
@@ -28,6 +24,10 @@ use triton_distributed_runtime::{
     protocols::annotated::Annotated,
     DistributedRuntime, Error, Result,
 };
+use futures::stream;
+use std::sync::Arc;
+use tokio::sync::mpsc;
+use tracing as log;
 
 pub struct KvEventPublisher {
     tx: mpsc::UnboundedSender<KvCacheEvent>,

lib/llm/src/kv_router/worker.rs

@@ -19,12 +19,12 @@ pub use crate::kv_router::protocols::ForwardPassMetrics;
 
 use anyhow::Result;
 use derive_builder::Builder;
-use triton_distributed_runtime::pipeline::network::{
+use dynemo_runtime::pipeline::network::{
     ingress::push_endpoint::PushEndpoint,
     PushWorkHandler,
 };
 
-use triton_distributed_runtime::transports::nats::{self, ServiceExt};
+use dynemo_runtime::transports::nats::{self, ServiceExt};
 
 use tokio::sync::watch;
 use tokio_util::sync::CancellationToken;

lib/llm/src/lib.rs

@@ -13,9 +13,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-//! # Triton LLM
+//! # Dynemo LLM
 //!
-//! The `triton-llm` crate is a Rust library that provides a set of traits and types for building
+//! The `dynemo-llm` crate is a Rust library that provides a set of traits and types for building
 //! distributed LLM inference solutions.
 
 pub mod backend;

lib/llm/src/model_card/model.rs

@@ -37,7 +37,7 @@ use std::time::Duration;
 
 use derive_builder::Builder;
 
-use triton_distributed_runtime::slug::Slug;
+use dynemo_runtime::slug::Slug;
 
 pub const BUCKET_NAME: &str = "mdc";