# Zefan-Cai/KVCache-Factory

Unified KV Cache Compression Methods for Auto-Regressive Models

Repository: https://github.com/Zefan-Cai/KVCache-Factory
Canonical: https://ross.abutalabs.com/products/kvcache-factory
Language: Python
License: MIT
License Family: permissive
Topics: kv-cache, kv-cache-compression, llm
Last push: 2026-08-13T07:35:40+00:00

## Health v2 (maintenance only)
Score: 68/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 35, longevity 58
- inputs: {"age_days": 819, "days_push": 20, "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 1375, forks 179 (observed 2026-08-28T04:04:33.106115+00:00)

## What it is
KVCache-Factory is a unified Python framework implementing many KV cache compression, retrieval, merging, and quantization methods for long-context LLM inference. It provides a single evaluation interface across methods like PyramidKV, SnapKV, H2O, Quest, MiniCache, and KIVI.

## Use cases
- compress kv cache for long-context llm inference
- benchmark kv cache eviction methods on longbench
- reduce gpu memory for llama or mistral inference
- compare pyramidkv vs snapkv vs h2o
- quantize kv cache with kivi or kvquant
- offload kv cache to cpu for 70b models
- evaluate needle-in-a-haystack retrieval under cache compression

## When to choose
- you need to experiment with or compare many KV cache compression methods under one interface
- you are running long-context inference on limited GPU memory
- you want reproducible LongBench-style evaluation of cache baselines

## When to avoid
- you need production serving with high throughput rather than research evaluation
- your model architecture is not Llama or Mistral style
- you need broad coverage of the newest methods without checking runner support

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, machine-learning, benchmarking
- domain: large-language-models, deep-learning, machine-learning
- platform: python
- tags: kv-cache, kv-cache-compression, long-context, inference-optimization, pyramidkv, quantization, attention, gpu, linux

## Member repositories
- Zefan-Cai/KVCache-Factory (main) score 68

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:33.106115+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 2026-08-30T04:40:33.354072+00:00, confidence not recorded.
  - readme: https://github.com/Zefan-Cai/KVCache-Factory (fetched 2026-08-28T04:04:33.106115+00:00, sha cc33c1e68a88)
- Data as of 2026-08-30T08:39:29.467469+00:00.