pvijayakrish · Sep 17, 2026 · Sep 17, 2026 · Sep 17, 2026 · Sep 17, 2026
@@ -56,7 +56,7 @@ keywords = ["llm", "genai", "inference", "nvidia", "distributed"]
 
 [workspace.dependencies]
 # Published separately by https://github.com/ai-dynamo/aisimulate.
-aisimulate-core = { version = "=0.12.0-dev.2" }
+aisimulate-core = { version = "=0.12.0" }
 
 # Local crates
 dynamo-runtime = { path = "lib/runtime", version = "1.5.0" }

@@ -39,7 +39,7 @@ classifiers = [
 ]
 
 dependencies = [
-    "aisimulate==0.12.0.dev2; python_version >= '3.11' and python_version < '3.14'",
+    "aisimulate==0.12.0; python_version >= '3.11' and python_version < '3.14'",
     "aiperf==0.12.0",
     "matplotlib",
     "networkx",

@@ -2,4 +2,4 @@
 # SPDX-License-Identifier: Apache-2.0
 
 # Published AISimulate wheel staged into planner and framework runtime wheelhouses.
-aisimulate==0.12.0.dev2
+aisimulate==0.12.0
@@ -635,16 +635,19 @@ COPY deploy/inference-gateway/sidecar/ /opt/dynamo/deploy/inference-gateway/side
 {% if target == "planner" or (target == "runtime" and framework in ("vllm", "sglang", "trtllm")) %}
 COPY container/deps/requirements.aisimulate.txt /opt/dynamo/container/deps/requirements.aisimulate.txt
 
-# AI Simulate is released separately as an abi3 wheel. Stage the exact published
-# wheel consumed by ai-dynamo instead of rebuilding it from vendored source.
-# Download only this distribution; runtime images own dependency installation
-# through their requirements files and local wheels.
+# AI Simulate is released separately as an abi3 wheel. Its public PyPI artifact
+# is a small resolver sdist because the full wheel is too large for that registry;
+# stage the actual wheel directly from NVIDIA's package index. Download only this
+# distribution; runtime images own dependency installation through their
+# requirements files and local wheels.
 RUN --mount=type=cache,id=uv-root-{{ context.dynamo.uv_version }},target=/root/.cache/uv,sharing=shared \
     export UV_CACHE_DIR=/root/.cache/uv && \
     source ${VIRTUAL_ENV}/bin/activate && \
     python -m pip download \
         --only-binary=:all: \
         --no-deps \
+        --no-index \
+        --find-links https://pypi.nvidia.com/aisimulate/ \
         --dest /opt/dynamo/dist \
         --requirement /opt/dynamo/container/deps/requirements.aisimulate.txt
 {% endif %}

@@ -22,7 +22,7 @@ Use Python 3.11 through 3.13. Install AISimulate, which provides the retained `a
 sizing command:
 
 ```bash
-python3 -m pip install "aisimulate==0.1.0.dev2"
+python3 -m pip install "aisimulate==0.12.0"
 ```
 
 Collect:

@@ -71,7 +71,7 @@ Use Python 3.11 through 3.13.
 
 ```bash
 # Install the distribution that provides the compatibility command
-python3 -m pip install "aisimulate==0.1.0.dev2"
+python3 -m pip install "aisimulate==0.12.0"
 
 # Optional: check whether the model/system/backend is covered
 aiconfigurator cli support \

@@ -42,15 +42,15 @@ path. They have no adapter migration.
 AISimulate requires Python 3.11 through 3.13. Dynamo itself still supports Python 3.10, but the
 `aisimulate` dependency and its CLI are not installed in a Python 3.10 environment.
 
-The `dynamo-planner` image installs the published `aisimulate==0.1.0.dev2` wheel from its local
-wheelhouse. Dynamo builds `aisimulate-core==0.1.0-dev.2` from crates.io instead of vendoring the
+The `dynamo-planner` image installs the published `aisimulate==0.12.0` wheel from its local
+wheelhouse. Dynamo builds `aisimulate-core==0.12.0` from crates.io instead of vendoring the
 AISimulate source tree.
 
 For Dynamo source development, install the published AISimulate wheel, Dynamo, and the Planner
 dependencies from the Dynamo repository root:
 
 ```bash
-python3 -m pip install "aisimulate==0.1.0.dev2"
+python3 -m pip install "aisimulate==0.12.0"
 python3 -m pip install --no-deps -e .
 python3 -m pip install -r container/deps/requirements.planner.txt
 ```
@@ -18,8 +18,8 @@ AISimulate requires Python 3.11 through 3.13. The `ai-dynamo` package remains in
 3.10, but the AISimulate dependency and CLI are not installed there.
 
 The supported prebuilt environment is the `dynamo-planner` image. The image stages and installs the
-published `aisimulate==0.1.0.dev2` wheel alongside the Dynamo wheels. Dynamo's Rust workspace
-resolves `aisimulate-core==0.1.0-dev.2` from crates.io.
+published `aisimulate==0.12.0` wheel alongside the Dynamo wheels. Dynamo's Rust workspace
+resolves `aisimulate-core==0.12.0` from crates.io.
 
 For Dynamo source development, install the published AISimulate wheel and build the matching
 Dynamo bindings:
@@ -30,7 +30,7 @@ cd lib/bindings/python
 maturin develop --uv --release --features aic-forward-pass
 cd ../../..
 python3 -m pip install --no-deps -e .
-python3 -m pip install "aisimulate==0.1.0.dev2"
+python3 -m pip install "aisimulate==0.12.0"
 python3 -m pip install -r container/deps/requirements.planner.txt
 ```
 

@@ -65,7 +65,8 @@ provide one config as a strict pin or several configs as a custom projection poo
 
 ## Configuration
 
-Use this reference configuration:
+The experiment used the following configuration shape. It documents the retired
+`aisimulate==0.1.0.dev1` API and is not accepted by AISimulate 0.12.0:
 
 ```yaml
 search_space:

@@ -10,10 +10,10 @@ subtitle: Experimental Sweeper search over Planner policies for MiniMax-M2.5 on
 > treat them as production capacity guidance or a performance commitment. Sweeper's search behavior
 > and output may change without a standard deprecation period.
 
-Enable the planner (SLA mode) on the same deployment as the
-[router experiment](router-end-to-end-latency-sweep.md), then use Sweeper's smart sweep to **find the
-planner configuration that maximizes goodput-per-GPU** on the Mooncake toolagent trace, and
-compare against the static deployment and the default planner.
+This experiment enabled the planner (SLA mode) on the same deployment as the
+[router experiment](router-end-to-end-latency-sweep.md), then used Sweeper's smart sweep to find the
+planner configuration that maximized goodput-per-GPU on the Mooncake toolagent trace and compared
+it against the static deployment and the default planner.
 
 ## Setup
 
@@ -50,8 +50,10 @@ scales the fleet to an average of ~13.5 GPUs instead of pinning 32.
 
 ## Sweep config
 
-`Sweeper.run`, `goal.target = goodput_per_gpu`. Deployment + router + workload + SLA
-fixed; only the planner knobs swept (full trace, open-loop):
+The experiment used `Sweeper.run` with `goal.target = goodput_per_gpu`. It fixed the deployment,
+router, workload, and SLA, and swept only the planner settings using the full open-loop trace. The
+configuration below documents the retired `aisimulate==0.1.0.dev1` API and is not accepted by
+AISimulate 0.12.0:
 
 ```yaml
 search_space:

@@ -12,7 +12,7 @@ subtitle: Run a backend-neutral sweep with an injected replay runtime
 Install AISimulate:
 
 ```bash
-python3 -m pip install "aisimulate==0.1.0.dev2"
+python3 -m pip install "aisimulate==0.12.0"
 ```
 
 Sweeper requires a `RunnerFactory` supplied by the application that owns replay execution:

@@ -10,9 +10,9 @@ subtitle: Experimental Sweeper search over KV router settings for MiniMax-M2.5 o
 > treat them as production capacity guidance or a performance commitment. Sweeper's search behavior
 > and output may change without a standard deprecation period.
 
-Reproduce the DynoSim router experiment, then use Sweeper's smart sweep to **find the
-KV-router configuration that minimizes mean end-to-end latency** on the Mooncake toolagent
-trace, and validate the winner against the default-router baseline on the full trace.
+This experiment reproduced the DynoSim router experiment, then used Sweeper's smart sweep to find
+the KV-router configuration that minimized mean end-to-end latency on the Mooncake toolagent trace
+and validated the winner against the default-router baseline on the full trace.
 
 ## Setup
 
@@ -32,9 +32,10 @@ concurrency) so any change in e2e latency is attributable to the **router** alon
 
 ## Sweep config
 
-Searched with `Sweeper.run` (Vizier GP-bandit), `goal.target = e2e_latency`. Search
-runs on a 2k-request subset (fast, representative — same long-context mix); the winner is
-then validated on the full trace.
+The experiment searched with `Sweeper.run` (Vizier GP-bandit) and
+`goal.target = e2e_latency`. It ran on a 2k-request subset with the same long-context mix, then
+validated the winner on the full trace. The configuration below documents the retired
+`aisimulate==0.1.0.dev1` API and is not accepted by AISimulate 0.12.0.
 
 ```yaml
 search_space:

@@ -27,7 +27,7 @@ forward-pass timing model.
   retained `aiconfigurator` sizing command:
 
   ```bash
-  python3 -m pip install "aisimulate==0.1.0.dev2"
+  python3 -m pip install "aisimulate==0.12.0"
   ```
 - Your checkpoint's HuggingFace ID, GPU system (for example `h200_sxm`), backend (`vllm`, `sglang`, or `trtllm`), the total GPUs you can allocate, and your SLA: Time To First Token (TTFT), Time Per Output Token (TPOT), and typical input/output sequence lengths (ISL/OSL).
 

@@ -60,7 +60,7 @@ request-trace-s3 = ["dynamo-llm/request-trace-s3"]
 # former AIC crate surface. The `python` feature is required for
 # AicEngineBuilder; this extension supplies the interpreter, so the dependency
 # must not enable `extension-module`.
-aisimulate-core = { version = "=0.12.0-dev.2", optional = true, features = ["python"] }
+aisimulate-core = { version = "=0.12.0", optional = true, features = ["python"] }
 dynamo-runtime = { path = "../../runtime" }
 dynamo-mocker = { path = "../../mocker" }
 dynamo-parsers = { version = "8.1.0" }

@@ -14,7 +14,7 @@ license-files = ["LICENSE"]
 requires-python = ">=3.10"
 dependencies = [
     "ai-dynamo-runtime==1.5.0",
-    "aisimulate==0.12.0.dev2; python_version >= '3.11' and python_version < '3.14'",
+    "aisimulate==0.12.0; python_version >= '3.11' and python_version < '3.14'",
     "aiohttp>=3.14.3,<4.0",
     "transformers>=4.56.0",
     "kubernetes>=32.0.1,<33.0.0",

@@ -93,7 +93,7 @@ def test_no_manifest_installs_retired_aic_distributions() -> None:
         assert all(package["name"] != "aiconfigurator-core" for package in packages)
     assert features["aic-forward-pass"] == ["dep:aisimulate-core"]
     assert dependencies["aisimulate-core"] == {
-        "version": "=0.12.0-dev.2",
+        "version": "=0.12.0",
         "optional": True,
         "features": ["python"],
     }

@@ -145,7 +145,10 @@ def test_container_stages_the_published_aisimulate_wheel() -> None:
         "--requirement /opt/dynamo/container/deps/requirements.aisimulate.txt"
         in wheel_builder
     )
+    assert "--only-binary=:all:" in wheel_builder
     assert "--no-deps" in wheel_builder
+    assert "--no-index" in wheel_builder
+    assert "--find-links https://pypi.nvidia.com/aisimulate/" in wheel_builder
     assert "COPY aisimulate" not in wheel_builder
     assert "/opt/dynamo/aisimulate" not in wheel_builder
     assert not (ROOT / "aisimulate").exists()