[https://nvbugs/6442074][fix] Unbreak main: DSparkWorker forward rename + test stub _forward_impl

[dongfengy]

Main is broken since d973974 (the SpecWorkerBase.__init_subclass__ guard from #16382) crossed in flight with two later additions:

• dspark.py defines forward → import-time TypeError for import tensorrt_llm (via speculative/utils.py module-level import)
• test_rejection_buffers_guard.py's _Worker stub lacks _forward_impl → TypeError at instantiation (3 tests)

This PR carries both fixes. The dspark rename is cherry-picked from #16577 — full credit @brnguyen2. Swept current main: these are the only two violations. Neither fix can pass CI alone (each PR's CI merge still contains the other breakage), hence one combined PR.

Test plan

• Sweep of current main for all SpecWorkerBase subclasses / forward overrides — only these two sites
• Local H100 validation: import tensorrt_llm + dspark import chain OK; test_rejection_buffers_guard.py 16/16; test_force_accepted_tokens.py 22/22

Links

• Bug: https://nvbugs/6442074
• Supersedes/combines: [https://nvbugs/6442074][fix] DSparkWorker: rename forward to _forward_impl #16577

[dongfengy]

/bot run --disable-fail-fast

[coderabbitai]

📝 Walkthrough

Walkthrough

Renames DSparkWorker.forward(...) to _forward_impl(...) without changing its method body, and adds _forward_impl(...) to the rejection-buffer guard test worker stub.

Changes

DSpark dispatch alignment

Layer / File(s)	Summary
Worker implementation entry point tensorrt_llm/_torch/speculative/dspark.py, tests/unittest/_torch/speculative/test_rejection_buffers_guard.py	DSparkWorker now defines _forward_impl(...); the test worker stub implements the same method and raises NotImplementedError.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#16382: Introduces the related speculative-decoding worker entry-point refactor.
• NVIDIA/TensorRT-LLM#16577: Contains the same DSparkWorker.forward(...) to _forward_impl(...) rename.

Suggested reviewers: junyixu-nv

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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 matches the PR scope and follows the required ticket/type pattern.
Description check	✅ Passed	The description explains the bug, fix, and testing, though it uses custom headings and omits the checklist checkbox.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[tensorrt-cicd]

PR_Github #60199 [ run ] triggered by Bot. Commit: 0dbe52f Link to invocation

[dongfengy]

Local validation on H100 (viking-prod-235) at this PR's head 0dbe52f5, full source build:

1. Import chain restored: import tensorrt_llm + from tensorrt_llm._torch.speculative.utils import get_spec_worker + from tensorrt_llm._torch.speculative.dspark import DSparkWorker — all OK (this is the path that currently throws TypeError on main).
2. The 3 CI-failing tests fixed: tests/unittest/_torch/speculative/test_rejection_buffers_guard.py — 16/16 passed.
3. Regression: test_force_accepted_tokens.py — 22/22 passed.

Given main is import-broken for everyone, appreciate a fast review — @brnguyen2's dspark rename rides in here with authorship preserved.

[tensorrt-cicd]

PR_Github #60199 [ run ] completed with state FAILURE. Commit: 0dbe52f
/LLM/main/L0_MergeRequest_PR pipeline #48569 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[chzblych]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #60218 [ run ] triggered by Bot. Commit: 0dbe52f Link to invocation

[chzblych]

/bot skip --comment "Skip the rerun pipeline and merge the PR to unblock main. The only failing test is perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL], which is a known failure and is waived in #16582"

[zhaoyangwang-nvidia]

Please get this PR merged ASAP

[tensorrt-cicd]

PR_Github #60220 [ skip ] triggered by Bot. Commit: 0dbe52f Link to invocation

[tensorrt-cicd]

PR_Github #60218 [ run ] completed with state ABORTED. Commit: 0dbe52f

Link to invocation

[tensorrt-cicd]

PR_Github #60220 [ skip ] completed with state SUCCESS. Commit: 0dbe52f
Skipping testing for commit 0dbe52f

Link to invocation