Model card:
Model details
Section titled “Model details”- Developer:
, as part of . - Architecture: <tl_arch and size: layers, hidden size, heads, KV heads, vocabulary, context length, parameters>.
- Training: <data (link the DATASHEET), tokens, steps, precision, seed, the train spec path and its sha256>.
- Release:
, models/<model_id>/<version>/, MANIFEST sha256. - Third-party components: <weights or tokenizers you did not train, with their licenses (for example SmolLM2-135M-Instruct, Apache-2.0)>.
- License:
.
Intended use
Section titled “Intended use”- Primary uses:
. - Out of scope: <uses the model must not serve; the usage policy rules that enforce them (gateway/policy.v1.yaml rule ids)>.
Evaluation
Section titled “Evaluation”| Suite | Metric | Value (95% CI) | Release threshold |
|---|---|---|---|
| <mean (low, high)> |
<Copy the numbers from evals/summary.json of this release; say which subjects and seeds produced them.>
Bias, risks, and limitations
Section titled “Bias, risks, and limitations”- <Measured bias results (ethics.04) and what they mean.>
-
- <Limitations of size, data, and language coverage.>
<Sources and licenses, from ledger.json of this release: every source must
allow train. PII handling (scrubbed counts from the corpus manifest).
Decontamination against the eval sets.>