Structured Run publisher

Publish evidence, not just a performance claim.

This form creates a manual Run. The open-source CLI adds machine capture, artifact hashes, redaction and standardized evaluation adapters.

You are starting a reproduction of run_orion_m3max_coding. The complete production flow preloads the original manifest and links the resulting Run automatically.

Model and task

Tie the claim to a canonical model, exact use case and runtime.

Run title

Model

Task

Runtime

Runtime version

Quantization

Hardware

Do not include hostnames, serial numbers, IP addresses or account names.

Hardware name

System memory (GB)

VRAM / unified GPU memory (GB)

Peak memory (GB)

Performance and reproduction

Metrics remain visibly self-reported until evidence or machine capture is attached.

Tokens per second

Time to first token (ms)

Exact command

Evidence URL (optional)

Notes

By publishing, you confirm that secrets and personal machine identifiers have been removed. Published Runs are immutable; corrections create a revision.

Prefer machine capture?

pip install -e cli/modelverse-cli
modelverse inspect
modelverse init run.json
modelverse validate run.json
modelverse publish run.json

Privacy checklist

Remove access tokens, API keys and environment secrets.
Remove hostnames, usernames, home paths, IP addresses and serial numbers.
Do not upload private prompts, datasets or outputs without authorization.
Link the exact model and dataset revisions wherever possible.