RepoRadar
Leave a Tip
Item detail

leyten/shard

Shard is an Apache-2.0 inference engine that splits frontier-scale LLMs across GPUs on different machines and streams activations between them, with a published proof path for running GLM-5.2 744B over scattered prosumer GPUs at usable speed.

Open source →Back to radar
Score8.4
Popularity53.0
Riskconditional
TierGold
Score breakdown
Usefulness8.0
Novelty9.0
Momentum6.0
Maturity7.7
Open-source/build8.4
Evidence7.2
Workflow potential8.8
Setup ease6.2

Popularity is tracked separately. Support, ads, sponsorships, and tips never affect these signals.

Why it matters

Useful for advanced inference builders who want to test whether distributed, cross-machine serving can unlock models that do not fit on one node: start by reading the proof docs and reproducing a smaller multi-GPU run before planning any serious WAN deployment.

Who should use it

inference engineersdistributed systems buildershomelab experimentersteams exploring frontier-model serving costs

Who should skip it

Skip if the source link, docs, or setup requirements do not match your workflow.

Risk explanation

Shard is aimed at multi-machine inference and assumes several GPUs plus reliable networking, so treat it as an infrastructure experiment rather than a drop-in local-model upgrade.

Evidence links

Closest alternatives / related signals

inferencedistributed-systemspipeline-parallelgpullm-serving