Defines the source manifest, path containment, tensor identity, byte counts, checksums, finite-value checks, and conversion outputs required before trained weights enter SLM1.
Implementation evidence: this topic is grounded in the reviewed GGUF.MiRust.com source snapshot. It documents observed code and artifacts without claiming broad deployment, model quality, or production readiness.
Source layout
A line-based source.manifest declares schema version, model family and shape, tokenizer, f32 source dtype, source identity, license and dataset fields, training summary, dimensions, tensor count, and one file record per required tensor.
Path safety
Tensor file paths are resolved under the declared source directory. Absolute paths and traversal outside that root are rejected.
Byte and value validation
Every tensor file must match declared dimensions, exact byte length, and custom checksum. Each f32 value must be finite.
Conversion
The converter emits f32, q8_0, or q4_0 SLM1 plus a provenance sidecar. Conversion proves format transformation, not task quality or safety.
Scope
This starter page defines the questions, boundaries, evidence, and failure modes that should be recorded before a capability is presented as supported.
Engineering considerations
- Identify the source, version, target environment, and owner.
- Separate observed values from estimates and externally reported values.
- Record trade-offs, unsupported cases, and fallback behavior.
- Link performance statements to a compatible benchmark methodology.
Verification questions
- What exact artifact, revision, backend, and environment were reviewed?
- Which assumptions could change the result?
- Which data should be retained so another engineer can reproduce the conclusion?