[None][chore] BREAKING: Remove AutoDeploy from TensorRT-LLM

[bmarimuthu-nv]

Summary

• BREAKING: remove the in-tree AutoDeploy backend, examples, documentation, and tests, including backend selection in serving, the LLM API, checkpoint loading, and benchmarking
• remove dependent packaging, CI inventory, scanning, and ownership configuration
• address review feedback: preserve chunked-prefill validation coverage; remove unused loader/serve arguments, unreachable legacy logits and request-validation branches, unused V-stride overrides, and the orphaned non-update FLA implementation
• preserve the live FLA update path and default attention stride calculations
• rebase onto main at fffdfba72, retaining upstream OpenEngine, guided-decoding, performance, and speculative-decoding changes while removing their obsolete AutoDeploy references

Validation

The deletion and review fixes at a264233ad passed the following local checks. The two subsequent MiniMax test-fixture repairs in ef428052a also pass all applicable file pre-commit checks:

• applicable pre-commit checks pass, including formatting, lint, YAML/TOML, generated configuration checks, and test inventory validation
• repository lightweight mypy passes for its 17 configured source files
• all 62 modified Python sources relative to main compile
• no tracked AutoDeploy identifiers or unresolved merge entries remain
• diff whitespace checks pass; commits have DCO sign-offs

CI and test limitations

Pipeline #60745 completed both x86_64 and SBSA builds successfully. Its progress report exposed two stale MiniMax test fixtures: the shared-expert helper rejected lora_params, and the LoRA composition fixture lacked routed_output_is_global. Both fixtures are repaired in ef428052a, with existing numerical and LoRA-forwarding assertions preserved. The compliance scan separately aborted because its Jenkins agent failed to launch (jnlp exit 255).

CI with fail-fast disabled has been dispatched on ef428052a; runtime results remain pending. No failures have been waived.

The local environment lacks PyTorch and compiled TensorRT-LLM bindings. Runtime tests, bindings-aware type checking, golden-manifest verification, and C++/GPU validation therefore remain dependent on CI or a matching build environment.

E2E output

No inference output generated locally.

[bmarimuthu-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #75212 [ run ] triggered by Bot. Commit: a91e1d6 Link to invocation

[tensorrt-cicd]

PR_Github #75184 [ run ] completed with state ABORTED. Commit: fc7d147

Link to invocation

[bmarimuthu-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #75224 [ run ] triggered by Bot. Commit: 65736da Link to invocation

[tensorrt-cicd]

PR_Github #75212 [ run ] completed with state ABORTED. Commit: a91e1d6

Link to invocation

[bmarimuthu-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #75238 [ run ] triggered by Bot. Commit: 75725ca Link to invocation

[tensorrt-cicd]

PR_Github #75224 [ run ] completed with state ABORTED. Commit: 65736da

Link to invocation

[tensorrt-cicd]

PR_Github #75238 [ run ] completed with state SUCCESS. Commit: 75725ca
/LLM/main/L0_MergeRequest_PR pipeline #61988 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

[bmarimuthu-nv]

/bot run

[tensorrt-cicd]

PR_Github #75323 [ run ] triggered by Bot. Commit: 75725ca Link to invocation

[tensorrt-cicd]

PR_Github #75323 [ run ] completed with state SUCCESS. Commit: 75725ca
/LLM/main/L0_MergeRequest_PR pipeline #62069 completed with status: 'SUCCESS'

CI Report

Link to invocation