---
cited_by:
  - '[[thoughts/inference/replayssm-serving#5-pr-reading-order|replayssm §5]]'
  - '[[thoughts/papers/rusanovsky2026lilicorrlightweightlikelihoodcorrelation|reading-map §2.3]]'
  - '[[thoughts/inference/speculative-drafting#3-vllm-status-at-a-glance|reading-map §3]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/models/qwen3_dflash.py
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/layers/logits_processor.py
  - file:///Users/aarnphm/workspace/vllm/vllm/config/vllm.py
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/models/nemotron_h.py
merged: '2026-09-25'
number: 57934
source:
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/model_executor/models/lilicorr.py
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/v1/worker/gpu/spec_decode/dflash2/speculator.py
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/v1/worker/gpu/spec_decode/lilicorr/speculator.py
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/model_executor/models/qwen3_dflash.py
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/docs/features/speculative_decoding/lilicorr.md
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/model_executor/layers/logits_processor.py
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/config/vllm.py
  - https://github.com/vllm-project/vllm/blob/73a78e6f1f38e280986b81e0f2a9aa5e1ee6fe47/vllm/model_executor/models/nemotron_h.py
state: merged
tags:
  - vllm
  - vllm-pr
  - specdec
title: '#57934 [SpecDecode] Add LiLiCorr drafter'
topic: specdec
url: https://github.com/vllm-project/vllm/pull/57934
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/57934
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/57934.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Add `LiLiCorrDraftModel` to the existing MRV2 DFlash path. The exported SGLang head reranks global candidate lattices using target embeddings, full-vocabulary log probabilities, draft hidden states and the normalized last committed target… · full dump: [[thoughts/inference/vllm-prs/57934|#57934]]

