refactor: rename triton_distributed to dynemo (#22)

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

lib/bindings/python/examples/hello_world/client.py

@@ -17,12 +17,12 @@ import asyncio
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
-    await init(runtime, "triton-init")
+    await init(runtime, "dynemo")
 
 
 async def init(runtime: DistributedRuntime, ns: str):

lib/bindings/python/examples/hello_world/run.py

@@ -21,7 +21,7 @@ import uvloop
 from client import init as client_init
 from server import init as server_init
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 
 def random_string(length=10):
@@ -29,7 +29,7 @@ def random_string(length=10):
     return "".join(random.choices(chars, k=length))
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     ns = random_string()
     task = asyncio.create_task(server_init(runtime, ns))

lib/bindings/python/examples/hello_world/server.py

@@ -17,7 +17,7 @@ import asyncio
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 
 class RequestHandler:
@@ -31,9 +31,9 @@ class RequestHandler:
             yield char
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
-    await init(runtime, "triton-init")
+    await init(runtime, "dynemo")
 
 
 async def init(runtime: DistributedRuntime, ns: str):

lib/bindings/python/examples/pipeline/backend.py

@@ -17,7 +17,7 @@ import asyncio
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 uvloop.install()
 
@@ -29,7 +29,7 @@ class RequestHandler:
             yield char
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     component = runtime.namespace("examples/pipeline").component("backend")
     await component.create_service()

lib/bindings/python/examples/pipeline/frontend.py

@@ -17,7 +17,7 @@ import asyncio
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 uvloop.install()
 
@@ -32,7 +32,7 @@ class RequestHandler:
             yield output.get("data")
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     # client to the next component - in this case the middle component
     next = (

lib/bindings/python/examples/pipeline/middle.py

@@ -17,7 +17,7 @@ import asyncio
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 uvloop.install()
 
@@ -32,7 +32,7 @@ class RequestHandler:
             yield output.get("data")
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     # client to backend
     backend = (

lib/bindings/python/examples/pipeline/pipeline.py

@@ -17,12 +17,12 @@ import asyncio
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 uvloop.install()
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     """
     # Pipeline Example

lib/bindings/python/examples/typed/client.py

@@ -17,18 +17,16 @@ import asyncio
 
 from protocol import Request
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     """
     Instantiate a `backend` client and call the `generate` endpoint
     """
     # get endpoint
-    endpoint = (
-        runtime.namespace("triton-init").component("backend").endpoint("generate")
-    )
+    endpoint = runtime.namespace("dynemo").component("backend").endpoint("generate")
 
     # create client
     client = await endpoint.client()

lib/bindings/python/examples/typed/server.py

@@ -19,11 +19,7 @@ import asyncio
 import uvloop
 from protocol import Request, Response
 
-from triton_distributed.runtime import (
-    DistributedRuntime,
-    triton_endpoint,
-    triton_worker,
-)
+from dynemo.runtime import DistributedRuntime, dynemo_endpoint, dynemo_worker
 
 uvloop.install()
 
@@ -33,19 +29,19 @@ class RequestHandler:
     Request handler for the generate endpoint
     """
 
-    @triton_endpoint(Request, Response)
+    @dynemo_endpoint(Request, Response)
     async def generate(self, request):
         for char in request.data:
             yield char
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     """
     Instantiate a `backend` component and serve the `generate` endpoint
     A `Component` can serve multiple endpoints
     """
-    component = runtime.namespace("triton-init").component("backend")
+    component = runtime.namespace("dynemo").component("backend")
     await component.create_service()
 
     endpoint = component.endpoint("generate")

lib/bindings/python/rust/engine.rs

@@ -15,8 +15,7 @@
 
 use std::sync::Arc;
 
-pub use serde::{Deserialize, Serialize};
-pub use triton_distributed_runtime::{
+pub use dynemo_runtime::{
     error,
     pipeline::{
         async_trait, AsyncEngine, AsyncEngineContextProvider, Data, ManyOut, ResponseStream,
@@ -25,6 +24,7 @@ pub use triton_distributed_runtime::{
     protocols::annotated::Annotated,
     Error, Result,
 };
+pub use serde::{Deserialize, Serialize};
 
 use pyo3::prelude::*;
 use pyo3_async_runtimes::TaskLocals;
@@ -192,7 +192,6 @@ where
                                 // tell the python async generator to stop generating
                                 // right now, this is impossible as we are not passing the context to the python async generator
                                 // todo: add task-local context to the python async generator
-                                // see: https://github.com/triton-inference-server/triton_distributed/issues/130
                                 ctx.stop_generating();
                                 let msg = format!("critical error: invalid response object from python async generator; application-logic-mismatch: {}", e);
                                 tracing::error!(request_id, "{}", msg);

lib/bindings/python/rust/lib.rs

@@ -25,14 +25,14 @@ use std::{fmt::Display, sync::Arc};
 use tokio::sync::Mutex;
 use tracing_subscriber::FmtSubscriber;
 
-use triton_distributed_runtime::{
+use dynemo_runtime::{
     self as rs,
     pipeline::{EngineStream, ManyOut, SingleIn},
     protocols::annotated::Annotated as RsAnnotated,
     traits::DistributedRuntimeProvider,
 };
 
-use triton_distributed_llm::{self as llm_rs};
+use dynemo_llm::{self as llm_rs};
 
 mod engine;
 mod llm;

lib/bindings/python/rust/llm/backend.rs

@@ -19,7 +19,7 @@ use crate::llm::model_card::ModelDeploymentCard;
 use llm_rs::protocols::common::llm_backend::{BackendInput, BackendOutput};
 use llm_rs::types::Annotated;
 
-use triton_distributed_runtime::pipeline::{Operator, ServiceBackend, ServiceFrontend, Source};
+use dynemo_runtime::pipeline::{Operator, ServiceBackend, ServiceFrontend, Source};
 
 use crate::engine::PythonAsyncEngine;
 

lib/bindings/python/rust/llm/preprocessor.rs

@@ -27,9 +27,9 @@ use llm_rs::{
     },
 };
 
-use triton_distributed_runtime::pipeline::{Operator, ServiceFrontend, Source};
+use dynemo_runtime::pipeline::{Operator, ServiceFrontend, Source};
 
-use triton_distributed_runtime::pipeline::{ManyOut, SegmentSink, SingleIn};
+use dynemo_runtime::pipeline::{ManyOut, SegmentSink, SingleIn};
 
 #[pyclass]
 pub(crate) struct OAIChatPreprocessor {

lib/bindings/python/src/triton_distributed/_core.pyi → lib/bindings/python/src/dynemo/_core.pyi

@@ -26,7 +26,7 @@ RequestHandler = Callable[[JsonLike], AsyncGenerator[JsonLike, None]]
 
 class DistributedRuntime:
     """
-    The runtime object for a distributed NOVA applications
+    The runtime object for dynemo applications
     """
 
     ...

lib/bindings/python/src/triton_distributed/llm/__init__.py → lib/bindings/python/src/dynemo/llm/__init__.py

@@ -13,5 +13,5 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from triton_distributed._core import KvMetricsPublisher as KvMetricsPublisher
-from triton_distributed._core import KvRouter as KvRouter
+from dynemo._core import KvMetricsPublisher as KvMetricsPublisher
+from dynemo._core import KvRouter as KvRouter

lib/bindings/python/tests/soak.py

@@ -19,7 +19,7 @@ import string
 
 import uvloop
 
-from triton_distributed.runtime import DistributedRuntime, triton_worker
+from dynemo.runtime import DistributedRuntime, dynemo_worker
 
 # Soak Test
 #
@@ -31,7 +31,7 @@ from triton_distributed.runtime import DistributedRuntime, triton_worker
 # could still eventually be a problem.
 
 
-@triton_worker()
+@dynemo_worker()
 async def worker(runtime: DistributedRuntime):
     ns = random_string()
     task = asyncio.create_task(server_init(runtime, ns))