# HobbitLong/RepDistiller

[ICLR 2020] Contrastive Representation Distillation (CRD), and benchmark of recent knowledge distillation methods

Repository: https://github.com/HobbitLong/RepDistiller
Canonical: https://ross.abutalabs.com/products/repdistiller
Language: Python
License: BSD-2-Clause
License Family: permissive
Last push: 2023-10-16T19:21:54+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2508, "days_push": 1052, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2437, forks 398 (observed 2026-08-28T04:06:51.486478+00:00)

## What it is
RepDistiller is a PyTorch research codebase implementing Contrastive Representation Distillation (CRD) from ICLR 2020. It also benchmarks 12 state-of-the-art knowledge distillation methods for training smaller student networks from larger teacher models.

## Use cases
- compress a large neural network into a smaller student model
- compare knowledge distillation methods on CIFAR/ImageNet
- reproduce the CRD paper results
- train a student network with contrastive representation distillation
- experiment with combining KD loss with other distillation objectives
- download pretrained teacher models for distillation experiments

## When to choose
- you need a unified PyTorch framework to benchmark multiple knowledge distillation techniques
- you want to implement or extend CRD for model compression research
- you need pretrained teacher checkpoints and reproducible distillation pipelines on CIFAR datasets

## When to avoid
- you need production-ready model compression tooling rather than research code
- you require support for the latest PyTorch versions with active maintenance
- you want distillation for NLP or transformer models, since the codebase focuses on CNN image classifiers

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, benchmarking
- domain: machine-learning, deep-learning
- platform: python
- tags: knowledge-distillation, pytorch, model-compression, contrastive-learning, research-code, iclr-2020, algorithms, linux, gpu

## Member repositories
- HobbitLong/RepDistiller (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:51.486478+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 2026-08-30T02:30:59.473966+00:00, confidence not recorded.
  - readme: https://github.com/HobbitLong/RepDistiller (fetched 2026-08-28T04:06:51.486478+00:00, sha 42195451cf58)
- Data as of 2026-08-30T08:39:29.467469+00:00.