vllm-project/speculators

[RFC]: Investigate whether HF-native MTP support can simplify our MTP finetune + stitch pathway 12 days ago

good first issue RFC

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Save the training launch command to the checkpoint dir about 1 month ago

enhancement good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

[RFC]: Add DFlash model converter about 2 months ago

good first issue RFC

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

[RFC]: Support Qwen3 base architecture for eagle3 & p-eagle about 2 months ago

good first issue RFC

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Reuse verifier rope parameters in drafter 2 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Add default activation 2 months ago

good first issue RFC

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Support configurable loss function via training args 3 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Subtask 2: Model Implementation 3 months ago

enhancement good first issue help wanted stale mtp

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Subtask 4: Optimization & Documentation 4 months ago

good first issue help wanted keep-open mtp

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

[RFC]: Drop bad samples during preprocessing 4 months ago

good first issue keep-open RFC data generation

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Subtask 3: Training & Validation 4 months ago

enhancement good first issue help wanted mtp

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Subtask 1: Data Preparation 4 months ago

enhancement good first issue help wanted mtp

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

"Save best" and other improvements to model checkpointing 5 months ago

good first issue training

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Simplify quality check development commands and CI 6 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

[RFC]: FastMTP-style Fine-tuning for Speculators 6 months ago

enhancement help wanted keep-open RFC mtp

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Incorrect length estimates for data sample files 7 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Create `Eagle3DraftModel`s which use the full verifier vocab 7 months ago

enhancement good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Feature Request: Enable Initial Base Support for MLP Speculators 7 months ago

enhancement good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Update assistant pattern detection logic 8 months ago

enhancement good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Update `scripts/build_vocab_mapping.py` to take target model name / path 8 months ago

enhancement good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Add datagen and training scripts to documentation 9 months ago

documentation good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Update links that reference old repo organization 9 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Combine `CONTRIBUTING.md` and `DEVELOPING.md` 9 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Remove `.MAINTAINERS` 9 months ago

good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Feature Request: Enable Initial Base Support for Independent Speculators 11 months ago

enhancement good first issue

vllm-project/speculators

644

A unified library for building, evaluating, and storing speculative decoding algorithms for LLM inference in vLLM

Python

Open Issues Need Help