# dvmazur/mixtral-offloading

Run Mixtral-8x7B models in Colab or consumer desktops

Repository: https://github.com/dvmazur/mixtral-offloading
Canonical: https://ross.abutalabs.com/products/mixtral-offloading
Language: Python
License: MIT
License Family: permissive
Topics: colab-notebook, deep-learning, google-colab, language-model, llm, mixture-of-experts, offloading, pytorch, quantization
Last push: 2024-04-08T08:40:22+00:00

## Health v2 (maintenance only)
Score: 26/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 70
- inputs: {"age_days": 992, "days_push": 877, "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 2332, forks 226 (observed 2026-08-28T04:06:37.856694+00:00)

## What it is
A Python library enabling efficient inference of Mixtral-8x7B mixture-of-experts language models on limited hardware like Google Colab or consumer desktops. It combines mixed HQQ quantization with per-expert GPU offloading and LRU caching to fit the model into combined GPU and CPU memory.

## Use cases
- run mixtral 8x7b on google colab
- run mixtral on a consumer gpu with limited vram
- quantize mixture-of-experts model for inference
- offload llm experts between cpu and gpu
- demo notebook for mixtral inference

## When to choose
- you want to run Mixtral-8x7B on hardware with limited GPU memory
- you want a Colab-ready demo of quantized MoE inference
- you need HQQ-based mixed quantization with expert offloading

## When to avoid
- you need a stable CLI or production inference server
- you need speculative expert prefetching or other techniques not yet implemented
- you need support for quantization methods beyond HQQ

## Facets
- artifact type: library
- maturity: experimental
- function: llm-inference, machine-learning, deep-learning
- domain: large-language-models, machine-learning, deep-learning
- platform: python, cloud
- tags: mixture-of-experts, quantization, hqq, offloading, mixtral, colab, pytorch, gpu

## Member repositories
- dvmazur/mixtral-offloading (main) score 26

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:37.856694+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 2026-08-30T02:37:56.730554+00:00, confidence not recorded.
  - readme: https://github.com/dvmazur/mixtral-offloading (fetched 2026-08-28T04:06:37.856694+00:00, sha fc047d9cfa02)
- Data as of 2026-08-30T08:39:29.467469+00:00.