# deepseek-ai/DeepSpec

DeepSpec: a full-stack codebase for training and evaluating speculative decoding algorithms

Repository: https://github.com/deepseek-ai/DeepSpec
Canonical: https://ross.abutalabs.com/products/deepspec
Language: Python
License: MIT
License Family: permissive
Last push: 2026-07-09T04:37:14+00:00

## Health v2 (maintenance only)
Score: 54/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 91, release rhythm 35, longevity 4
- inputs: {"age_days": 68, "days_push": 55, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 7041, forks 660 (observed 2026-08-28T04:09:55.332366+00:00)

## What it is
DeepSpec is a full-stack Python codebase from DeepSeek for training and evaluating draft models used in speculative decoding of large language models. It provides data preparation pipelines, training scripts, evaluation benchmarks, and released checkpoints for algorithms like Eagle3.

## Use cases
- train a draft model for speculative decoding
- evaluate speculative decoding acceptance rates on benchmarks
- speed up LLM inference with a trained draft model
- prepare target model output caches for draft training
- reproduce speculative decoding research results

## When to choose
- you want to train or benchmark speculative decoding draft models
- you need released Eagle3-style checkpoints for Qwen or Gemma targets
- you have multi-GPU hardware and want a complete training-to-evaluation pipeline

## When to avoid
- you only need to run inference without speculative decoding
- you lack GPU resources or the large storage required for target caches
- you need a plug-and-play inference engine rather than a training framework

## Facets
- artifact type: library
- maturity: active
- function: llm-training, llm-inference, machine-learning, benchmarking, etl
- domain: large-language-models, machine-learning, deep-learning, developer-tools
- platform: python
- tags: speculative-decoding, draft-models, eagle3, model-training, inference-optimization, llm-acceleration, linux, gpu, docker

## Member repositories
- deepseek-ai/DeepSpec (main) score 54

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:55.332366+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:40:32.986168+00:00, confidence not recorded.
  - readme: https://github.com/deepseek-ai/DeepSpec (fetched 2026-08-28T04:09:55.332366+00:00, sha 2a6ccc8a3326)
- Data as of 2026-08-30T08:39:29.467469+00:00.