fix(aic): adopt published AIC 0.11 dev artifacts

[jasonqinzhou]

Summary

• adopt the paired published dev releases aiconfigurator==0.11.0.dev20260728 and aiconfigurator-core==0.11.0.dev20260728
• pin both Rust consumers to the published crates.io release aiconfigurator-core==0.11.0-dev20260728, with registry checksums recorded in both lockfiles
• use aiconfigurator_core for Runtime, Mocker, Router, and Planner modeling while retaining upper aiconfigurator for Profiler and AI Simulate orchestration/generator surfaces
• migrate the Python bindings to the typed AicEngineBuilder API and remove pre-0.11 compatibility/fallback paths
• keep acceptance-rate scheduling in Dynamo while forwarding nonzero nextn to AIC core for MTP verification cost
• align Rapid with the published Task-v2 _execute_tasks six-value return contract, including per-task outcomes
• extend dependency, clean-wheel, native-engine, replay, Planner, frontend-image, and AIC feature test/Clippy coverage

Packaging

The Python dev releases are index-resolvable from PyPI. Upper aiconfigurator pins the exact matching aiconfigurator-core release; AI Simulate declares both because it directly imports both package layers. The core package uses its published wheel-stub flow to install the platform wheel from the NVIDIA package index.

The Rust crate is published and non-yanked on crates.io. Dynamo no longer embeds an AIC Git revision or builds AIC from source in its package metadata or container flow. The Planner image installs the standalone aisimulate wheel together with the paired AIC packages.

Reviewer guide

1. Dependency contract: Cargo.toml, both Cargo lockfiles, root/benchmark/AI Simulate Python metadata, and container/deps/requirements.planner.txt
2. Runtime boundary: lib/bindings/python/rust/llm/aic_callback.rs, dynamo._internal.aic, Mocker perf-model configuration, replay, and scheduler adapters
3. Upper-package consumers: Rapid/Profiler Task-v2 integration and AI Simulate generator/core imports
4. Packaging and gates: Planner/frontend Dockerfiles, dependency consistency, clean-wheel smoke, integration tests, and AIC-triggered CI paths

Validation

• clean Python 3.10 installation of both exact published Python dev packages
• dependency-consistency checks across Python metadata, Cargo manifests, and both lockfiles
• published Task-v2 contract integration coverage
• cargo check --locked -p dynamo-mocker --features aic-forward-pass
• cargo check --locked --manifest-path lib/bindings/python/Cargo.toml --features aic-forward-pass
• cargo fmt --manifest-path lib/bindings/python/Cargo.toml -- --check
• rendered multi-arch Planner, Runtime, and Frontend Dockerfiles; verified the Planner builds and installs the standalone AI Simulate wheel
• affected pre-commit hooks and git diff --check

Full PR CI is running on the current signed head. Merge remains gated on the required current-head reviews.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[devin-ai-integration]

Devin Review found 1 potential issue.

[coderabbitai]

Walkthrough

Changes

The AIC speculative-decoding configuration now uses scalar nextn_accepted values derived from legacy accept-rate inputs. Python, Rust, mock-engine, planner, cache-key, dependency, and test paths were updated accordingly.

Speculative acceptance configuration

Layer / File(s)	Summary
Legacy acceptance conversion lib/bindings/python/src/dynamo/_internal/aic.py, lib/bindings/python/rust/llm/aic_callback.rs, lib/bindings/python/tests/test_aic_capacity.py	Legacy conditional accept rates are converted into scalar nextn_accepted values, forwarded to AIC model configuration, and covered by helper and session tests.
Mock-engine speculative configuration lib/mocker/src/common/engine_perf.rs, Cargo.toml, lib/bindings/python/Cargo.toml	Mock AIC configuration parses and validates nextn_accepted, stores it in SpeculativeConfig, updates raw-iteration and native configuration paths, and pins updated AIC dependency revisions.
Planner adapter integration components/src/dynamo/planner/core/perf_model/rust_adapter.py, components/src/dynamo/planner/tests/unit/test_rust_perf_adapter.py	Planner configuration and model cache keys conditionally use nextn and nextn_accepted, with corresponding expectations updated in unit tests.

Estimated code review effort: 3 (Moderate) | ~25 minutes

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 64.29% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The PR description covers the work, but it does not follow the required template or include the required Related Issues section.	Reformat it to the repository template and add the required Related Issues section with a linked issue or a confirmed no-issue checkbox.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title is concise and describes the main change: adopting published AIC dev artifacts.

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 3

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@lib/bindings/python/rust/llm/aic_callback.rs`:
- Around line 116-128: Update the nextn conversion in the callback logic to use
u32::try_from instead of narrowing with as u32, and return a PyValueError when
the usize value exceeds u32::MAX. Perform this validation before the nextn > 0
check and before calling _nextn_accepted_from_accept_rates, while preserving the
existing zero and valid-value behavior.

In `@lib/bindings/python/tests/test_aic_capacity.py`:
- Around line 255-258: Update the pytest.raises match in
test_nextn_accepted_rejects_unsupported_mocker_depth to use a raw regex with
both literal dots escaped, ensuring the assertion validates the exact “nextn
must be 1..=5” message.

In `@lib/mocker/src/common/engine_perf.rs`:
- Around line 1136-1144: Update the speculative configuration construction in
the args-to-config flow to ignore aic_nextn values of zero before creating
SpeculativeConfig. Preserve the existing conversion and error propagation for
positive values, ensuring aic_nextn=0 produces no speculative config and matches
into_aic_config’s dense-mode behavior.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: b82d3afb-55e0-4487-a823-3e4278e26fb1

📥 Commits

Reviewing files that changed from the base of the PR and between cfdc8d5 and fab8967.

⛔ Files ignored due to path filters (2)

• Cargo.lock is excluded by !**/*.lock
• lib/bindings/python/Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (8)

• Cargo.toml
• components/src/dynamo/planner/core/perf_model/rust_adapter.py
• components/src/dynamo/planner/tests/unit/test_rust_perf_adapter.py
• lib/bindings/python/Cargo.toml
• lib/bindings/python/rust/llm/aic_callback.rs
• lib/bindings/python/src/dynamo/_internal/aic.py
• lib/bindings/python/tests/test_aic_capacity.py
• lib/mocker/src/common/engine_perf.rs

Replacing this request-changes review with a neutral review containing both notes inline.

[PeaBrane]

Two inline notes: the Python/Rust dependency alignment is the actionable contract issue; the FPM note only requests a TODO documenting a dormant integration boundary.

[jasonqinzhou]

/ok to test d19f515

[jasonqinzhou]

Rechecked the exact current head 70e3176fe98e537bbec211478f1ca83e44704a5d.

The two blocking regressions found against the prior d19f5156 head are already addressed by 70e3176f:

• the pre-0.11 compatibility probe for aiconfigurator.sdk.engine was removed, so the supported core-only ai-dynamo[mocker] path now enters AicEngineBuilder directly;
• the planner image once again installs /opt/dynamo/wheelhouse/aisimulate*.whl, matching the wheel-builder output and preventing the Spica suite from being absent in the shipped planner image.

Focused validation on this head:

• cargo check --locked --manifest-path lib/bindings/python/Cargo.toml --features aic-forward-pass
• cargo fmt --manifest-path lib/bindings/python/Cargo.toml -- --check
• rendered the multi-arch planner Dockerfile and verified both the AI Simulate wheel build and install lines
• git diff --check

I also replied to and resolved the outdated benchmark dependency thread: AIC core is pinned to 0.11.0.dev20260728; aiperf==0.10.0 is a separate package/version. All inline review threads are resolved.

[tedzhouhk]

/ok to test 70e3176

[jthomson04]

approving the one frontend change.

[saturley-hall]

confirmed that the planner container installed the dev version of aiconfigurator-core from the compliance CSV