Skip to content

GitLab

"benchmarks/vscode:/vscode.git/clone" did not exist on "09c7b73caf8efd67e2ddd59a6d25b0931673ec4d"
Unverified Commit 9a021885 authored by Dmitry Tokarev's avatar Dmitry Tokarev Committed by GitHub
Browse files

chore: Bumped Dynamo version to 0.4.1 (#2545)

parent 177d662f
Show whitespace changes
Inline Side-by-side
Showing with 62 additions and 64 deletions
Cargo.lock
View file @ 9a021885
...@@ -240,7 +240,7 @@ checksum = "4288f83726785267c6f2ef073a3d83dc3f9b81464e9f99898240cced85fce35a" ...@@ -240,7 +240,7 @@ checksum = "4288f83726785267c6f2ef073a3d83dc3f9b81464e9f99898240cced85fce35a"
[[package]] [[package]]
name = "async-openai-macros" name = "async-openai-macros"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"proc-macro2", "proc-macro2",
"quote", "quote",
...@@ -1836,7 +1836,7 @@ dependencies = [ ...@@ -1836,7 +1836,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-async-openai" name = "dynamo-async-openai"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"async-openai-macros", "async-openai-macros",
"backoff", "backoff",
...@@ -1862,7 +1862,7 @@ dependencies = [ ...@@ -1862,7 +1862,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-engine-llamacpp" name = "dynamo-engine-llamacpp"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"async-stream", "async-stream",
"dynamo-llm", "dynamo-llm",
...@@ -1874,7 +1874,7 @@ dependencies = [ ...@@ -1874,7 +1874,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-engine-mistralrs" name = "dynamo-engine-mistralrs"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"async-stream", "async-stream",
...@@ -1892,7 +1892,7 @@ dependencies = [ ...@@ -1892,7 +1892,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-llm" name = "dynamo-llm"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"ahash", "ahash",
"akin", "akin",
...@@ -1975,7 +1975,7 @@ dependencies = [ ...@@ -1975,7 +1975,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-run" name = "dynamo-run"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"async-stream", "async-stream",
...@@ -2004,7 +2004,7 @@ dependencies = [ ...@@ -2004,7 +2004,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-runtime" name = "dynamo-runtime"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"arc-swap", "arc-swap",
...@@ -2061,7 +2061,7 @@ dependencies = [ ...@@ -2061,7 +2061,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-tokens" name = "dynamo-tokens"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"bytemuck", "bytemuck",
"derive-getters", "derive-getters",
...@@ -3818,7 +3818,7 @@ checksum = "1171693293099992e19cddea4e8b849964e9846f4acee11b3948bcc337be8776" ...@@ -3818,7 +3818,7 @@ checksum = "1171693293099992e19cddea4e8b849964e9846f4acee11b3948bcc337be8776"
[[package]] [[package]]
name = "libdynamo_llm" name = "libdynamo_llm"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"async-once-cell", "async-once-cell",
...@@ -4130,7 +4130,7 @@ dependencies = [ ...@@ -4130,7 +4130,7 @@ dependencies = [
[[package]] [[package]]
name = "metrics" name = "metrics"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"axum 0.8.4", "axum 0.8.4",
"clap 4.5.42", "clap 4.5.42",
...@@ -6037,7 +6037,7 @@ dependencies = [ ...@@ -6037,7 +6037,7 @@ dependencies = [
[[package]] [[package]]
name = "router" name = "router"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"clap 4.5.42", "clap 4.5.42",
"dynamo-llm", "dynamo-llm",
......
Cargo.toml
View file @ 9a021885
...@@ -17,7 +17,7 @@ members = [ ...@@ -17,7 +17,7 @@ members = [
resolver = "3" resolver = "3"
[workspace.package] [workspace.package]
version = "0.4.0+post0" version = "0.4.1"
edition = "2021" edition = "2021"
description = "Dynamo Inference Framework" description = "Dynamo Inference Framework"
authors = ["NVIDIA Inc. <sw-dl-dynamo@nvidia.com>"] authors = ["NVIDIA Inc. <sw-dl-dynamo@nvidia.com>"]
...@@ -28,10 +28,10 @@ keywords = ["llm", "genai", "inference", "nvidia", "distributed", "dynamo"] ...@@ -28,10 +28,10 @@ keywords = ["llm", "genai", "inference", "nvidia", "distributed", "dynamo"]
[workspace.dependencies] [workspace.dependencies]
# Local crates # Local crates
dynamo-runtime = { path = "lib/runtime", version = "0.4.0" } dynamo-runtime = { path = "lib/runtime", version = "0.4.1" }
dynamo-llm = { path = "lib/llm", version = "0.4.0" } dynamo-llm = { path = "lib/llm", version = "0.4.1" }
dynamo-tokens = { path = "lib/tokens", version = "0.4.0" } dynamo-tokens = { path = "lib/tokens", version = "0.4.1" }
dynamo-async-openai = { path = "lib/async-openai", version = "0.4.0", features = ["byot", "rustls"]} dynamo-async-openai = { path = "lib/async-openai", version = "0.4.1", features = ["byot", "rustls"]}
# External dependencies # External dependencies
anyhow = { version = "1" } anyhow = { version = "1" }
......
docs/support_matrix.md
View file @ 9a021885
...@@ -58,17 +58,17 @@ If you are using a **GPU**, the following GPU models and architectures are suppo ...@@ -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 | | **Python Package** | **Version** | glibc version | CUDA Version |
| :----------------- | :------------ | :----------------------------------- | :----------- | | :----------------- | :------------ | :----------------------------------- | :----------- |
| ai-dynamo | 0.4.0 | >=2.28 | | | ai-dynamo | 0.4.1 | >=2.28 | |
| ai-dynamo-runtime | 0.4.0 | >=2.28 (Python 3.12 has known issues)| | | ai-dynamo-runtime | 0.4.1 | >=2.28 (Python 3.12 has known issues)| |
| NIXL | 0.5.0 | >=2.27 | >=11.8 | | NIXL | 0.4.1 | >=2.27 | >=11.8 |
### Build Dependency ### Build Dependency
| **Build Dependency** | **Version** | | **Build Dependency** | **Version** |
| :------------------- | :------------------------------------------------------------------------------- | | :------------------- | :------------------------------------------------------------------------------- |
| **Base Container** | [25.03](https://catalog.ngc.nvidia.com/orgs/nvidia/containers/cuda-dl-base/tags) | | **Base Container** | [25.03](https://catalog.ngc.nvidia.com/orgs/nvidia/containers/cuda-dl-base/tags) |
| **TensorRT-LLM** | 1.0.0rc² | | **TensorRT-LLM** | 1.0.0rc4 |
| **NIXL** | 0.4.0 | | **NIXL** | 0.4.1 |
> [!Important] > [!Important]
> ² Specific versions of TensorRT-LLM supported by Dynamo are subject to change. > ² Specific versions of TensorRT-LLM supported by Dynamo are subject to change.
......
lib/bindings/python/Cargo.lock
View file @ 9a021885
...@@ -198,37 +198,9 @@ version = "0.5.4" ...@@ -198,37 +198,9 @@ version = "0.5.4"
source = "registry+https://github.com/rust-lang/crates.io-index" source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4288f83726785267c6f2ef073a3d83dc3f9b81464e9f99898240cced85fce35a" checksum = "4288f83726785267c6f2ef073a3d83dc3f9b81464e9f99898240cced85fce35a"
[[package]]
name = "async-openai"
version = "0.29.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "0864223dd927e9418f2fdb7fd773b34482518596c17693ff2cdb216261d22913"
dependencies = [
"async-openai-macros",
"backoff",
"base64 0.22.1",
"bytes",
"derive_builder",
"eventsource-stream",
"futures",
"rand 0.9.2",
"reqwest",
"reqwest-eventsource",
"secrecy",
"serde",
"serde_json",
"thiserror 2.0.15",
"tokio",
"tokio-stream",
"tokio-util",
"tracing",
]
[[package]] [[package]]
name = "async-openai-macros" name = "async-openai-macros"
version = "0.1.0" version = "0.4.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "0289cba6d5143bfe8251d57b4a8cac036adf158525a76533a7082ba65ec76398"
dependencies = [ dependencies = [
"proc-macro2", "proc-macro2",
"quote", "quote",
...@@ -1224,15 +1196,38 @@ dependencies = [ ...@@ -1224,15 +1196,38 @@ dependencies = [
"bytemuck", "bytemuck",
] ]
[[package]]
name = "dynamo-async-openai"
version = "0.4.1"
dependencies = [
"async-openai-macros",
"backoff",
"base64 0.22.1",
"bytes",
"derive_builder",
"eventsource-stream",
"futures",
"rand 0.9.2",
"reqwest",
"reqwest-eventsource",
"secrecy",
"serde",
"serde_json",
"thiserror 2.0.15",
"tokio",
"tokio-stream",
"tokio-util",
"tracing",
]
[[package]] [[package]]
name = "dynamo-llm" name = "dynamo-llm"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"ahash", "ahash",
"akin", "akin",
"anyhow", "anyhow",
"async-nats", "async-nats",
"async-openai",
"async-stream", "async-stream",
"async-trait", "async-trait",
"async_zmq", "async_zmq",
...@@ -1249,6 +1244,7 @@ dependencies = [ ...@@ -1249,6 +1244,7 @@ dependencies = [
"derive-getters", "derive-getters",
"derive_builder", "derive_builder",
"dialoguer", "dialoguer",
"dynamo-async-openai",
"dynamo-runtime", "dynamo-runtime",
"either", "either",
"erased-serde", "erased-serde",
...@@ -1298,15 +1294,15 @@ dependencies = [ ...@@ -1298,15 +1294,15 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-py3" name = "dynamo-py3"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"async-openai",
"async-stream", "async-stream",
"async-trait", "async-trait",
"cudarc 0.16.6", "cudarc 0.16.6",
"derive-getters", "derive-getters",
"dlpark", "dlpark",
"dynamo-async-openai",
"dynamo-llm", "dynamo-llm",
"dynamo-runtime", "dynamo-runtime",
"either", "either",
...@@ -1329,7 +1325,7 @@ dependencies = [ ...@@ -1329,7 +1325,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-runtime" name = "dynamo-runtime"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"arc-swap", "arc-swap",
......
lib/bindings/python/Cargo.toml
View file @ 9a021885
...@@ -19,7 +19,7 @@ ...@@ -19,7 +19,7 @@
[package] [package]
name = "dynamo-py3" name = "dynamo-py3"
version = "0.4.0+post0" version = "0.4.1"
edition = "2021" edition = "2021"
authors = ["NVIDIA"] authors = ["NVIDIA"]
license = "Apache-2.0" license = "Apache-2.0"
......
lib/bindings/python/pyproject.toml
View file @ 9a021885
...@@ -16,7 +16,7 @@ ...@@ -16,7 +16,7 @@
[project] [project]
name = "ai-dynamo-runtime" name = "ai-dynamo-runtime"
version = "0.4.0.post0" version = "0.4.1"
description = "Dynamo Inference Framework Runtime" description = "Dynamo Inference Framework Runtime"
readme = "README.md" readme = "README.md"
authors = [ authors = [
......
lib/runtime/examples/Cargo.lock
View file @ 9a021885
...@@ -648,7 +648,7 @@ dependencies = [ ...@@ -648,7 +648,7 @@ dependencies = [
[[package]] [[package]]
name = "dynamo-runtime" name = "dynamo-runtime"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"arc-swap", "arc-swap",
...@@ -1020,7 +1020,7 @@ checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea" ...@@ -1020,7 +1020,7 @@ checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
[[package]] [[package]]
name = "hello_world" name = "hello_world"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"dynamo-runtime", "dynamo-runtime",
] ]
...@@ -2492,7 +2492,7 @@ dependencies = [ ...@@ -2492,7 +2492,7 @@ dependencies = [
[[package]] [[package]]
name = "service_metrics" name = "service_metrics"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"dynamo-runtime", "dynamo-runtime",
"futures", "futures",
...@@ -2668,7 +2668,7 @@ dependencies = [ ...@@ -2668,7 +2668,7 @@ dependencies = [
[[package]] [[package]]
name = "system_metrics" name = "system_metrics"
version = "0.4.0+post0" version = "0.4.1"
dependencies = [ dependencies = [
"anyhow", "anyhow",
"dynamo-runtime", "dynamo-runtime",
...@@ -2881,6 +2881,8 @@ dependencies = [ ...@@ -2881,6 +2881,8 @@ dependencies = [
"bytes", "bytes",
"futures-core", "futures-core",
"futures-sink", "futures-sink",
"futures-util",
"hashbrown",
"pin-project-lite", "pin-project-lite",
"tokio", "tokio",
] ]
......
lib/runtime/examples/Cargo.toml
View file @ 9a021885
...@@ -22,7 +22,7 @@ members = [ ...@@ -22,7 +22,7 @@ members = [
resolver = "3" resolver = "3"
[workspace.package] [workspace.package]
version = "0.4.0+post0" version = "0.4.1"
edition = "2021" edition = "2021"
authors = ["NVIDIA"] authors = ["NVIDIA"]
license = "Apache-2.0" license = "Apache-2.0"
......
pyproject.toml
View file @ 9a021885
...@@ -3,7 +3,7 @@ ...@@ -3,7 +3,7 @@
[project] [project]
name = "ai-dynamo" name = "ai-dynamo"
version = "0.4.0.post0" version = "0.4.1"
description = "Distributed Inference Framework" description = "Distributed Inference Framework"
readme = "README.md" readme = "README.md"
authors = [ authors = [
...@@ -13,7 +13,7 @@ license = { text = "Apache-2.0" } ...@@ -13,7 +13,7 @@ license = { text = "Apache-2.0" }
license-files = ["LICENSE"] license-files = ["LICENSE"]
requires-python = ">=3.10" requires-python = ">=3.10"
dependencies = [ dependencies = [
"ai-dynamo-runtime==0.4.0.post0", "ai-dynamo-runtime==0.4.1",
"pytest>=8.3.4", "pytest>=8.3.4",
"types-psutil>=7.0.0.20250218", "types-psutil>=7.0.0.20250218",
"kubernetes>=32.0.1,<33.0.0", "kubernetes>=32.0.1,<33.0.0",
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment