Objective
The Rust browser model lab is the smallest credible path from site theory to an executable demonstration. It does not need real model weights at first. It can start with mock .slm metadata, adapter manifests, eval cases, diagnostics JSON, and release packets.
Components
.slmmodel card and checksum;- tokenizer identity;
- tensor-layout digest;
- adapter-delta package manifest;
- local eval case file;
- diagnostics capture;
- lineage record;
- release packet.
Flow
PROCEDURE browser_lab_demo(model_package, adapter_package, eval_cases)
VERIFY_PACKAGE_MANIFEST(model_package)
VERIFY_ADAPTER_IDENTITY(adapter_package, model_package)
evidence <- RUN_OR_SIMULATE_EVAL_CASES(model_package, adapter_package, eval_cases)
diagnostics <- CAPTURE_RUNTIME_DIAGNOSTICS()
lineage <- WRITE_LINEAGE_NODE(model_package, adapter_package, evidence)
packet <- BUILD_RELEASE_PACKET(lineage, diagnostics)
RETURN packet
END PROCEDURESuccess criterion
A reader can understand exactly how a local candidate moves from package identity to eval evidence to release packet without needing a cloud database.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.