[franciscojavierarceo]

Overview

active upstream review: vllm-project/vllm#63901. this fork PR keeps the isolated native-stack diff and previous comments. all layers were rebased onto 3d81a510f9; the fork's pinned base was advanced to that commit.

preserve developer instructions and media order when an HF template falls back from developer to system messages.

upstream review order: 1/3: renderer vllm-project/vllm#63901 → 2/3: non-streaming vllm-project/vllm#63902 → 3/3: streaming vllm-project/vllm#63903. all three upstream review PRs target main; later diffs include earlier layers.

Claims

• preserve string and OpenAI-array content shapes, including multimodal parts, when consolidating system messages.
• carry media source order through rendering so placeholders stay aligned with extracted media.
• keep the change in the shared Chat Completions/Responses renderer; this layer adds no endpoint.

Validation

Latest upstream synchronization — October 28

current head: 90cc52cdd0; rebased onto upstream f4fd1ab68e. the fork keeps its native stacked bases; upstream PRs still target main.

• resolved the shared error-handler overlap with upstream's missing-LoRA 404 change. both NotFound and ResponseNotFound retain their existing wire envelopes; the LoRA/gRPC and Responses HTTP regressions pass.
• cargo test --locked -p vllm-chat --lib -- --test-threads=1: 438 passed as part of the full-stack run.
• strict Clippy for server/chat/text/CLI, applicable pre-commit hooks, and git diff --check: passed. cargo nextest remains unavailable.
• range-diff shows only the error-handler conflict adaptation; all other feature/fix patches are unchanged. all branch tips passed merge-tree checks against f4fd1ab68e.
• the old umbrella [Rust Frontend] Add OpenAI Responses API endpoint vllm-project/vllm#58783 was synchronized without rewriting its original history. its source tree equals the tested complete stack and it stays draft.

the model-backed HTTP/Codex evidence below predates this synchronization (complete-stack head 3db4c9abf0); it was not repeated after this rebase. today's evidence covers compilation and deterministic wire/route/renderer regressions, not new model qualification. CI reruns on these new heads are separate from the earlier successful workflows. agentic-api was not changed or pushed.

Earlier review and runtime evidence

reviewed again on October 27; renderer head: 699ef82150. no new renderer change was needed.

• cargo test --locked -p vllm-chat --lib -- --test-threads=1: 430 passed.
• the final stack's full server/chat/text suite passed 1,107 tests (one existing text test ignored).
• strict Clippy, applicable pre-commit hooks, formatting, and git diff --check: passed.
• all three tips merge cleanly with upstream f8f7634cb4 using git merge-tree --write-tree origin/main <layer>. the fork's isolated stack bases are preserved.

commands ran from rust/ with OPENSSL_DIR pointing to the local OpenSSL build. cargo nextest was unavailable. tests ran serially after the previous parallel chat run exposed a shared audio-metrics collector collision; the isolated test and complete serial suites passed.

fresh model-backed HTTP and real Codex checks on the revised complete stack are recorded in vllm-project#63903. these are text/function-tool interoperability checks, not a multimodal evaluation. deterministic renderer/media-extraction regressions cover string and OpenAI-array content formats; fresh model-backed multimodal qualification remains outstanding.

Details

a plain-string merge can lose policies for array-based templates; moving system/developer messages can also reorder placeholders without moving their media. preserve the effective content representation and its source order.

the rebase retains upstream's request-template fuel controls and uses the normal template constructor in the new upstream fuel-test fixture. the renderer changes also affect Chat Completions.

these are the existing vllm-project#58783 layers opened upstream at @BugenZhao's request, rather than competing implementations. the original authored history and discussion remain in vllm-project#58783. @gcanlin's vllm-project#53380 implementation is credited in the endpoint and streaming commits. duplicate checks found vllm-project#63497 affects DeepSeek renderers and vllm-project#56391 affects Python's HF renderer; neither covers this Rust HF path. the fork layer retains the isolated review history.

diff against the new base: 4 files, +274 / -10. Codex assistance was used. human review of every changed line and independent validation remain required before merge.

---

Pull Request Checklist

• I used vLLM's /pr-checklist skill. (Mandatory for agents, optional for humans).

• AI assistance was used during the creation of this PR.

• Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.

• Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.

• Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.

• Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.

BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing (anything written below this line will be removed by GitHub Actions)

[Copilot]

Copilot review overview

🔵 Needs a closer look

It changes shared multimodal rendering behavior without model-backed multimodal validation.

Review effort: Balanced
Findings: None

What changed in this PR

Preserves developer instructions and multimodal ordering when HF templates fall back to system messages.

Changes:

• Detects developer-role support and consolidates unsupported developer messages.
• Preserves content shape and media ordering.
• Adds string and OpenAI-array regression coverage.

File	Description
rust/​src/​chat/​src/​renderer/​hf/​template.rs	Caches developer-role support detection.
rust/​src/​chat/​src/​renderer/​hf/​mod.rs	Implements fallback consolidation.
rust/​src/​chat/​src/​renderer/​hf/​media.rs	Reorders media sources with placeholders.
rust/​src/​chat/​src/​multimodal.rs	Tests multimodal alignment.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.