Make vllm.cpp v0.0.2 builds deterministic

[zatevakhin]

Changes

• make backend selection deterministic and consume all vendored Triton AOT targets
• curate packaged native sources and include required ROCm syntax-check inputs
• reuse four-mode link tests for extracted crates and update bounded package limits

Testing

• nix develop -c just sys
• CARGO_NET_OFFLINE=true nix develop -c just link-modes
• CARGO_NET_OFFLINE=true nix develop -c just package-test

Summary by CodeRabbit

• New Features

  • Added support for token completion, transcription, embeddings, video generation, video muxing, and server operations.
  • Expanded model configuration with device selection and GPU/KV-cache memory controls.
  • Improved compatibility across CPU, Vulkan, Apple, Metal, MLX, Triton, HIP, and other backends.

• Bug Fixes

  • Added earlier detection and clearer handling of native ABI mismatches.
  • Improved validation of packaged native libraries and exported API symbols.

• Documentation

  • Updated ABI, native version, supported architectures, configuration, provenance, and licensing documentation.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: f3343582-12f8-4024-b95e-d48340f02329

📥 Commits

Reviewing files that changed from the base of the PR and between 28ac8a7 and 57f3786.

📒 Files selected for processing (18)

• .github/workflows/platforms.yml
• Justfile
• vllm-cpp-sys/Cargo.toml
• vllm-cpp-sys/README.md
• vllm-cpp-sys/THIRD_PARTY.md
• vllm-cpp-sys/build.rs
• vllm-cpp-sys/licenses/DOCTEST-MIT.txt
• vllm-cpp-sys/src/bindings.rs
• vllm-cpp-sys/src/build_config.rs
• vllm-cpp-sys/src/lib.rs
• vllm-cpp-sys/tests/build_config.rs
• vllm-cpp-sys/tests/layout.c
• vllm-cpp-sys/tests/layout.rs
• vllm-cpp-sys/tests/symbols.rs
• vllm-cpp-sys/vllm.cpp
• vllm-cpp/src/engine.rs
• vllm-cpp/src/lib.rs
• vllm-cpp/tests/safe_api.rs

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.

---

📝 Walkthrough

Walkthrough

The update pins vllm.cpp to v0.0.2 and ABI version 17. It adds C APIs for completion, transcription, embeddings, video, and server operations. It updates build configuration, packaging validation, ABI tests, and Rust-side ABI checks.

Changes

ABI 17 integration

Layer / File(s)	Summary
Native ABI surface .github/workflows/platforms.yml, vllm-cpp-sys/vllm.cpp, vllm-cpp-sys/src/bindings.rs, vllm-cpp-sys/src/lib.rs, vllm-cpp-sys/README.md	The native revision and Rust declarations now target ABI 17 with 35 exported C symbols and new completion, transcription, embedding, video, and server APIs.
Build configuration and architecture support .github/workflows/platforms.yml, vllm-cpp-sys/src/build_config.rs, vllm-cpp-sys/tests/build_config.rs, vllm-cpp-sys/README.md	CMake definitions now cover additional backends and build controls. Triton AOT validation accepts all supported CUDA architecture combinations.
Package contents and validation Justfile, vllm-cpp-sys/Cargo.toml, vllm-cpp-sys/build.rs, vllm-cpp-sys/THIRD_PARTY.md, vllm-cpp-sys/licenses/DOCTEST-MIT.txt	Package inclusion, exclusion, licensing, provenance, size diagnostics, and extracted-package validation now match the updated native tree.
ABI layout and symbol conformance vllm-cpp-sys/tests/layout.c, vllm-cpp-sys/tests/layout.rs, vllm-cpp-sys/tests/symbols.rs, Justfile, vllm-cpp/tests/safe_api.rs	Tests validate ABI 17, structure layouts, field offsets, callbacks, 35 symbols, server behavior, native errors, and version 0.0.2.
Safe ABI-checked initialization vllm-cpp/src/engine.rs, vllm-cpp/src/lib.rs	Model-parameter initialization now checks the runtime ABI immediately before calling the native default initializer.

Estimated code review effort: 4 (Complex) | ~45 minutes

Merge Risk: ⚪ Minimal · up to 57f37

The PR makes build and packaging behavior deterministic while retaining the existing ABI-check path; no actionable merge-blocking risk remains after normal checks and review.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 48.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 25 functions across 12 files. (6 skipped:…	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 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.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: making vllm.cpp v0.0.2 builds deterministic.

Full details: Docstring Coverage

Explanation

Docstring coverage is 48.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 25 functions across 12 files. (6 skipped: 6 unsupported.)

• Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch stack/v0.0.2-02-build-package

---

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