Use the release boundary
Read the tagged artifact, runtime identity, method, measured results, corrections, and limits together.
Dated public evidence
2026 Public Activity Record
This page indexes dated public work that can be checked against releases and source material. It is an activity record, not a claim that every community interaction is a distinct person served.
Page published: July 28, 2026. Last updated: July 30, 2026.
Technical publication record
Versioned public outputs
Reproducible GFX906 Qwen3.6 MI50 runtime release
The v0.1.0 release published the first named LocalAIServers GFX906 Qwen3.6 MI50 reproducibility artifact and its stated benchmark boundary.
Inspect v0.1.0ROCm 7.2 Dense/MoE active-contract release
The v0.2.0 release published a named boundary for dense and mixture-of-experts GFX906 work, including runtime identity, measured points, correction context, and limitations.
Inspect v0.2.0Public deploy reproduction check and package
The v0.2.1 package pinned corrected public reproduction documentation and scripts and included the dated public deploy reproduction check.
Inspect v0.2.1Stand-alone GGUF/HF reproduction package
The vNext release established its own tagged reproduction boundary, public inputs, validation tables, instructions, and claim limits for GGUF and Hugging Face model-package paths.
Inspect the vNext releaseFirst publicly documented third-party vNext reproduction claim
A community member reported using LocalAIServers' vNext benchmark harness on a separate four-MI50 system and published exact GGUF and Hugging Face benchmark ladders, with both strict gates reported as passed. On the builder's own Upgrade of my AI server post, the same community member publicly credited Joe, LocalAIServers' Founder / Director, with helping the build and getting vNext running. The builder also reported that the fan-shroud setup kept all four MI50 GPUs at 52C or below throughout the benchmark run.
This is the same contributor, system, and run, not a second reproduction. The temperature is user-reported, not independent thermal validation. The reproduction claim is publicly documented, not formal validation, peer review, proof of artifact-hash equivalence, or an identical reproduction.
Read the public benchmark report Read the attribution and thermal report
Community evidence
Dated aggregate demand signals
LocalAIServers publishes current aggregate community metrics as a machine-readable snapshot. The record includes the public member count and operator-supplied Reddit Insights totals for views, published posts, and published comments.
These figures describe reach and discussion activity. They are not unique-person counts, training completion counts, or private beneficiary records. The JSON includes its update timestamp and caveats so a citation can be tied to a specific snapshot.
Reporting method
How to read this record
Use the timestamp
Cite the metrics JSON with its last_updated_utc value and preserve its aggregate-demand caveat.
Do not inflate outcomes
Public artifacts and discussion are evidence of work and reach; they do not establish individualized outcomes that were not measured.
This dated record is limited to public evidence that can be checked directly. Community reach metrics are aggregate platform measurements, not beneficiary or outcome counts.