chore: Version bump to 0.4.0 (#2179)

Cargo.lock

@@ -1847,7 +1847,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-engine-llamacpp"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "async-stream",
  "dynamo-llm",
@@ -1859,7 +1859,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-engine-mistralrs"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "async-openai",
@@ -1877,7 +1877,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-llm"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "ahash",
  "akin",
@@ -1954,7 +1954,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-run"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "async-openai",
@@ -1983,7 +1983,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-runtime"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "arc-swap",
@@ -2039,7 +2039,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-tokens"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "bytemuck",
  "derive-getters",
@@ -3808,7 +3808,7 @@ checksum = "d750af042f7ef4f724306de029d18836c26c1765a54a6a3f094cbd23a7267ffa"
 
 [[package]]
 name = "libdynamo_llm"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "async-once-cell",
@@ -4105,7 +4105,7 @@ dependencies = [
 
 [[package]]
 name = "metrics"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "axum 0.6.20",
  "clap 4.5.40",
@@ -5921,7 +5921,7 @@ dependencies = [
 
 [[package]]
 name = "router"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "clap 4.5.40",
  "dynamo-llm",

Cargo.toml

@@ -15,7 +15,7 @@ members = [
 resolver = "3"
 
 [workspace.package]
-version = "0.3.2"
+version = "0.4.0"
 edition = "2021"
 description = "Dynamo Inference Framework"
 authors = ["NVIDIA Inc. <sw-dl-dynamo@nvidia.com>"]
@@ -26,9 +26,9 @@ keywords = ["llm", "genai", "inference", "nvidia", "distributed", "dynamo"]
 
 [workspace.dependencies]
 # Local crates
-dynamo-runtime = { path = "lib/runtime", version = "0.3.2" }
-dynamo-llm = { path = "lib/llm", version = "0.3.2" }
-dynamo-tokens = { path = "lib/tokens", version = "0.3.2" }
+dynamo-runtime = { path = "lib/runtime", version = "0.4.0" }
+dynamo-llm = { path = "lib/llm", version = "0.4.0" }
+dynamo-tokens = { path = "lib/tokens", version = "0.4.0" }
 
 # External dependencies
 anyhow = { version = "1" }

deploy/helm/chart/Chart.yaml

@@ -17,5 +17,5 @@ apiVersion: v2
 name: dynamo-graph
 description: A Helm chart to deploy a Dynamo graph on Kubernetes
 type: application
-version: 0.3.2
-appVersion: 0.3.2
+version: 0.4.0
+appVersion: 0.4.0

docs/support_matrix.md

@@ -58,9 +58,9 @@ If you are using a **GPU**, the following GPU models and architectures are suppo
 
 | **Python Package** | **Version**   | glibc version                        | CUDA Version |
 | :----------------- | :------------ | :----------------------------------- | :----------- |
-| ai-dynamo          | 0.3.2         | >=2.28                               |              |
-| ai-dynamo-runtime  | 0.3.2         | >=2.28 (Python 3.12 has known issues)|              |
-| NIXL               | 0.4.0         | >=2.27                               | >=11.8       |
+| ai-dynamo          | 0.4.0         | >=2.28                               |              |
+| ai-dynamo-runtime  | 0.4.0         | >=2.28 (Python 3.12 has known issues)|              |
+| NIXL               | 0.5.0         | >=2.27                               | >=11.8       |
 
 ### Build Dependency
 

lib/bindings/python/Cargo.lock

@@ -1169,7 +1169,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-llm"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "ahash",
  "akin",
@@ -1235,7 +1235,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-py3"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "async-openai",
@@ -1262,7 +1262,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-runtime"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "arc-swap",

lib/bindings/python/Cargo.toml

@@ -19,7 +19,7 @@
 
 [package]
 name = "dynamo-py3"
-version = "0.3.2"
+version = "0.4.0"
 edition = "2021"
 authors = ["NVIDIA"]
 license = "Apache-2.0"

lib/runtime/examples/Cargo.lock

@@ -683,7 +683,7 @@ dependencies = [
 
 [[package]]
 name = "dynamo-runtime"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "anyhow",
  "arc-swap",
@@ -1060,7 +1060,7 @@ checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
 
 [[package]]
 name = "hello_world"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "dynamo-runtime",
 ]
@@ -2548,7 +2548,7 @@ dependencies = [
 
 [[package]]
 name = "service_metrics"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "dynamo-runtime",
  "futures",
@@ -2724,7 +2724,7 @@ dependencies = [
 
 [[package]]
 name = "system_metrics"
-version = "0.3.2"
+version = "0.4.0"
 dependencies = [
  "dynamo-runtime",
  "futures",

lib/runtime/examples/Cargo.toml

@@ -22,7 +22,7 @@ members = [
 resolver = "3"
 
 [workspace.package]
-version = "0.3.2"
+version = "0.4.0"
 edition = "2021"
 authors = ["NVIDIA"]
 license = "Apache-2.0"

pyproject.toml

@@ -15,7 +15,7 @@
 
 [project]
 name = "ai-dynamo"
-version = "0.3.2"
+version = "0.4.0"
 description = "Distributed Inference Framework"
 readme = "README.md"
 authors = [
@@ -25,7 +25,7 @@ license = { text = "Apache-2.0" }
 license-files = ["LICENSE"]
 requires-python = ">=3.10"
 dependencies = [
-    "ai-dynamo-runtime==0.3.2",
+    "ai-dynamo-runtime==0.4.0",
     "pytest>=8.3.4",
     "types-psutil>=7.0.0.20250218",
     "kubernetes>=32.0.1,<33.0.0",