Bumped Dynamo version to 0.4.1

Author: dmitry-tokarev-nv

Cargo.lock

@@ -17,7 +17,7 @@ members = [
 resolver = "3"
 
 [workspace.package]
-version = "0.4.0+post0"
+version = "0.4.1"
 edition = "2021"
 description = "Dynamo Inference Framework"
 authors = ["NVIDIA Inc. <sw-dl-dynamo@nvidia.com>"]
@@ -28,10 +28,10 @@ keywords = ["llm", "genai", "inference", "nvidia", "distributed", "dynamo"]
 
 [workspace.dependencies]
 # Local crates
-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" }
-dynamo-async-openai = { path = "lib/async-openai", version = "0.4.0", features = ["byot", "rustls"]}
+dynamo-runtime = { path = "lib/runtime", version = "0.4.1" }
+dynamo-llm = { path = "lib/llm", version = "0.4.1" }
+dynamo-tokens = { path = "lib/tokens", version = "0.4.1" }
+dynamo-async-openai = { path = "lib/async-openai", version = "0.4.1", features = ["byot", "rustls"]}
 
 # External dependencies
 anyhow = { version = "1" }

Cargo.toml

@@ -58,17 +58,17 @@ If you are using a **GPU**, the following GPU models and architectures are suppo
 
 | **Python Package** | **Version**   | glibc version                        | CUDA Version |
 | :----------------- | :------------ | :----------------------------------- | :----------- |
-| 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       |
+| ai-dynamo          | 0.4.1         | >=2.28                               |              |
+| ai-dynamo-runtime  | 0.4.1         | >=2.28 (Python 3.12 has known issues)|              |
+| NIXL               | 0.4.1         | >=2.27                               | >=11.8       |
 
 ### Build Dependency
 
 | **Build Dependency** | **Version**                                                                      |
 | :------------------- | :------------------------------------------------------------------------------- |
 | **Base Container**   | [25.03](https://catalog.ngc.nvidia.com/orgs/nvidia/containers/cuda-dl-base/tags) |
-| **TensorRT-LLM**     | 1.0.0rc²                                                                         |
-| **NIXL**             | 0.4.0                                                                            |
+| **TensorRT-LLM**     | 1.0.0rc4                                                                         |
+| **NIXL**             | 0.4.1                                                                            |
 
 > [!Important]
 > ² Specific versions of TensorRT-LLM supported by Dynamo are subject to change.

docs/support_matrix.md

@@ -19,7 +19,7 @@
 
 [package]
 name = "dynamo-py3"
-version = "0.4.0+post0"
+version = "0.4.1"
 edition = "2021"
 authors = ["NVIDIA"]
 license = "Apache-2.0"

lib/bindings/python/Cargo.lock

@@ -16,7 +16,7 @@
 
 [project]
 name = "ai-dynamo-runtime"
-version = "0.4.0.post0"
+version = "0.4.1"
 description = "Dynamo Inference Framework Runtime"
 readme = "README.md"
 authors = [

lib/bindings/python/Cargo.toml

@@ -22,7 +22,7 @@ members = [
 resolver = "3"
 
 [workspace.package]
-version = "0.4.0+post0"
+version = "0.4.1"
 edition = "2021"
 authors = ["NVIDIA"]
 license = "Apache-2.0"

lib/bindings/python/pyproject.toml

@@ -3,7 +3,7 @@
 
 [project]
 name = "ai-dynamo"
-version = "0.4.0.post0"
+version = "0.4.1"
 description = "Distributed Inference Framework"
 readme = "README.md"
 authors = [
@@ -13,7 +13,7 @@ license = { text = "Apache-2.0" }
 license-files = ["LICENSE"]
 requires-python = ">=3.10"
 dependencies = [
-    "ai-dynamo-runtime==0.4.0.post0",
+    "ai-dynamo-runtime==0.4.1",
     "pytest>=8.3.4",
     "types-psutil>=7.0.0.20250218",
     "kubernetes>=32.0.1,<33.0.0",