Skip to content

[None][docs] fix dead example links in legacy speculative-decoding - #18955

Open
imitater-dou wants to merge 2 commits into
NVIDIA:mainfrom
imitater-dou:docs/fix-legacy-speculative-decoding-dead-links
Open

[None][docs] fix dead example links in legacy speculative-decoding#18955
imitater-dou wants to merge 2 commits into
NVIDIA:mainfrom
imitater-dou:docs/fix-legacy-speculative-decoding-dead-links

Conversation

@imitater-dou

@imitater-dou imitater-dou commented Sep 9, 2026

Copy link
Copy Markdown

Summary

  • docs/source/legacy/advanced/speculative-decoding.md still linked to deleted example trees: examples/draft_target_model/README.md, examples/ngram/run_dtm_ngram.py, examples/medusa/README.md, examples/redrafter/README.md, examples/eagle/README.md, and examples/lookahead/README.md (all 404 on main).
  • Retarget Draft/Target and NGram how-tos to docs/source/features/speculative-decoding.md (and keep the live examples/ngram/README.md pointer).
  • Mark Medusa / Lookahead / ReDrafter / legacy EAGLE example trees as removed; point readers at the current speculative-decoding feature doc. Live telemetry speculative_config.decoding_type is AUTO, DFlash, DSpark, Draft_Target, Eagle3, Eagle, MTP, NGram, PARD, SA, SaveState, User_Provided (no Medusa/Lookahead).

Test plan

  • Confirmed 404 for the six deleted paths on HEAD 07a58bd
  • Confirmed examples/ngram/README.md still exists
  • Confirmed current feature doc has no Medusa/Lookahead sections
  • Diff limited to one legacy docs file

Signed-off-by: ADou ikun3.1415927@gmail.com

Dev Engineer Review

  • Updated legacy speculative-decoding documentation links.
  • Redirected Draft/Target, NGram, Medusa, Lookahead, ReDrafter, and EAGLE guidance to current documentation.
  • Retained the live examples/ngram/README.md link.
  • Redirected disaggregated-serving guidance to disagg-serving.md.
  • Marked removed decoding examples and documented the EAGLE3 draft_vocab_size default and warning.
  • No source-code, API, configuration, or runtime behavior changes.

Per-File QA Perspective

  • docs/source/legacy/advanced/speculative-decoding.md: Verify that redirected links resolve, removed examples are clearly identified, the NGram README remains valid, and the EAGLE3 guidance is accurate. No executable behavior changes require regression testing.

QA Engineer Review

No test changes.

Retarget Draft/Target, NGram, Medusa, ReDrafter, EAGLE, and Lookahead
how-to pointers away from deleted example trees
(examples/draft_target_model, run_dtm_ngram.py, medusa, redrafter,
eagle, lookahead) to docs/source/features/speculative-decoding.md.
Telemetry decoding_type no longer lists Medusa/Lookahead.

Signed-off-by: ADou <ikun3.1415927@gmail.com>
@coderabbitai

coderabbitai Bot commented Sep 9, 2026

Copy link
Copy Markdown
Contributor

Review Change StackReview Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 09c8a266-18b1-40b0-b1ed-2f152f02ca14

📥 Commits

Reviewing files that changed from the base of the PR and between 8a45f28 and d5c19c4.

📒 Files selected for processing (1)
  • docs/source/legacy/advanced/speculative-decoding.md
🚧 Files skipped from review as they are similar to previous changes (1)
  • docs/source/legacy/advanced/speculative-decoding.md

Included review availability: Your plan provides up to 12 included reviews per hour; 10 remain after this review.


Walkthrough

The legacy speculative decoding guide now links to current documentation, identifies removed examples and decoding types, and documents EAGLE3 draft_vocab_size behavior.

Changes

Speculative decoding documentation

Layer / File(s) Summary
Legacy guidance and current links
docs/source/legacy/advanced/speculative-decoding.md
The guide links Draft/Target, NGram, ReDrafter, EAGLE, Medusa, and Lookahead references to current documentation. It records removed examples and decoding types and corrects the EAGLE3 disaggregated-serving link.
EAGLE3 configuration note
docs/source/legacy/advanced/speculative-decoding.md
The guide states that omitted draft_vocab_size defaults to vocab_size with a warning. Different draft vocabularies require explicit configuration.

Priority: ⬇️ Low

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

Merge Risk: ⚪ Minimal · up to d5c19

The legacy speculative-decoding guide redirects obsolete example links to current documentation and marks removed content without any identified merge-readiness risk.

🚥 Pre-merge checks | ✅ 5
✅ Passed checks (5 passed)
Check name Status Explanation
Title check ✅ Passed The title clearly and concisely identifies the documentation fix for dead links in the legacy speculative-decoding page. The [docs] type is understandable, although the template lists [doc] as the sta…
Description check ✅ Passed The description explains the problem, the solution, and the verification steps. It includes relevant test coverage, but it omits the repository's standard Description heading and PR Checklist section.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0…
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.
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

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

…gg-serving.md

Signed-off-by: ADou <ikun3.1415927@gmail.com>

@brnguyen2 brnguyen2 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving — the comments below are optional touch-ups, not blockers.

Content checks out: examples/{draft_target_model,medusa,redrafter,eagle,lookahead} and examples/ngram/run_dtm_ngram.py are gone, examples/ngram/README.md is present, and the Draft/Target, NGram, and EAGLE 3 sections you point at exist in docs/source/features/speculative-decoding.md (which has no Medusa/Lookahead sections).

One nit beyond the inline: these could be repo-relative Markdown links (e.g. ../../features/speculative-decoding.md) so they resolve within the built docs instead of bouncing readers to GitHub. The file already used absolute URLs, so not a blocker.

## Lookahead Decoding

Lookahead decoding algorithm operates through two parallel computation branches within the same model: a lookahead branch that generates n-grams using a fixed-sized 2D window, and a verification branch that validates promising n-gram candidates. This approach eliminates the necessity for additional model training or fine-tuning and can be enabled for any autoregressive model. Refer to the [Lookahead decoding README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/lookahead/README.md) for information about building and running the model.
Lookahead decoding algorithm operates through two parallel computation branches within the same model: a lookahead branch that generates n-grams using a fixed-sized 2D window, and a verification branch that validates promising n-gram candidates. This approach eliminates the necessity for additional model training or fine-tuning and can be enabled for any autoregressive model. The legacy `examples/lookahead` tree has been removed, and Lookahead is no longer available as a decoding type. This section is retained for historical context only. For current speculative decoding how-tos, see [Speculative Decoding](https://github.com/NVIDIA/TensorRT-LLM/blob/main/docs/source/features/speculative-decoding.md). No newline at end of file

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The trailing newline at end of file was dropped (the base version had one). end-of-file-fixer in .pre-commit-config.yaml will rewrite this — run pre-commit run --files docs/source/legacy/advanced/speculative-decoding.md and re-commit so CI's pre-commit check passes.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants