feat(patterns) author the Speculative Sampling entry in family 17-ai-agentic - #441
Merged
Conversation
…agentic Rebases PR #403's entry onto current main. the branch predated a large share of the family 17 work that has since landed, so its own copies of the generated dashboards, indexes, and family README conflicted on every file except the entry itself. Pulled the content file forward unchanged and regenerated every dependent file fresh. Speculative Sampling, also known as Assisted Generation. Leviathan, Kalman, and Matias 2023 (Google Research) and Chen and colleagues 2023 (DeepMind) introduced the pattern independently. a small draft model proposes a run of tokens, the target model verifies the whole run in one forward pass, and a modified rejection sampling scheme guarantees the accepted output distribution is mathematically identical to sampling from the target model alone. Regenerates the family 17 index, BY-PROBLEM and BY-LANGUAGE, the root README and catalogue-status dashboards. Verified. check-structure.py 890/890, check-prose.py 931/931, markdownlint-cli2 0 issues, check-code.py --strict 2846 compiled with 0 failed, check-duplicates.py finds no new collision, validate-refs.py --strict resolves all 4995 citations including the entry's own arXiv, vLLM, and TensorRT-LLM sources.
The prior commit ran gen-catalogue-status.py before validate-refs.py --strict had finished its probe pass, so references_checked locked in 5063 instead of the 5067 the cache actually held once probing completed. CI's own fresh run of the generator caught the mismatch. Re-running validate-refs.py first, then the generator last, brings README, dist/catalogue-status.json, and docs/PROGRESS.md back in sync with the real cache state.
Owner
Author
|
REVIEWED. Reviewed the Speculative Sampling entry against its cited primary sources (Leviathan/Kalman/Matias 2023, Chen et al. 2023) before rebasing it forward from #403. All 6 gates plus check-duplicates pass on the full repository; the catalogue-status staleness from the first push (references_checked 5063 vs 5067, caused by running the generator before validate-refs.py --strict finished settling .ref-cache.json) was root-caused and fixed in a follow-up commit, now matching CI's independently computed value exactly. Squash-merging. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Rebases #403's entry onto current main. That branch predated a large
share of the family 17 work that has since landed, so its own copies
of the generated dashboards, indexes, and family README conflicted
against everything except the entry itself. Pulled the content file
forward unchanged and regenerated every dependent file fresh against
the current repository state.
Adds Speculative Sampling (also known as Assisted Generation or
Draft-Target sequential-token sampling) to family 17-ai-agentic.
Leviathan, Kalman, and Matias 2023 (Google Research) and Chen and
colleagues 2023 (DeepMind) introduced the pattern independently, both
around exploiting spare GPU compute during the memory-bandwidth-bound
autoregressive generation phase. a small draft model proposes a run
of tokens, the target model verifies the whole run in a single
forward pass, and a modified rejection sampling scheme guarantees the
accepted output distribution is mathematically identical to sampling
from the target model alone.
Verification
check-structure.py. 890/890 entries passcheck-prose.py. 931/931 files passmarkdownlint-cli2. 0 issuescheck-code.py --strict. 2846 compiled, 0 failed (Python,TypeScript, Go code samples)
check-duplicates.py --check. no new collision, including againstthe historical-proposal pass that landed in feat(tools) activate Pass 5 historical proposal collision detection #440
validate-refs.py --strict. all 4995 citations resolve, includingthe entry's own new arXiv (2211.17192, 2302.01318), vLLM, and
TensorRT-LLM sources
Test plan
the new entry
sources before rebasing it forward