docs/usermanuals/en/training/lora.md: broken example — target_modules: "all-linear" is a valid form ("broadest — every Linear layer"). (DOC-D3B06-011) #245

Wave 1 · theme D1-10: Documentation-only fixes · highest severity High · 1 finding(s) · effort docs fix

Summary

An operator who copies the "broadest" option gets a config error and cannot train.

Primary finding

DOC-D3B06-011 — docs/usermanuals/en/training/lora.md: broken example — target_modules: "all-linear" is a valid form ("broadest — every Linear layer").

severity High

Where. docs/usermanuals/en/training/lora.md:97 (Turkish mirror: docs/usermanuals/tr/training/lora.md:95) · type: broken example

The document says. target_modules: "all-linear" is a valid form ("broadest — every Linear layer").

What the code does. LoraConfigModel.target_modules is typed List[str] (forgelm/config.py:384-387) with extra=forbid and no string coercion. A scratch config with target_modules: "all-linear" fails --dry-run with lora.target_modules Input should be a valid list [type=list_type] (exit 1, run in this audit).

Code references. forgelm/config.py:384-387

User impact. An operator who copies the "broadest" option gets a config error and cannot train.

Proposed correction. Delete the "all-linear" line, or write it as a list. Add: "target_modules must be a YAML list; the PEFT shorthand string all-linear is rejected by the schema."

Turkish mirror. the Turkish page has the same error; fix both in the same PR

Acceptance criteria

• Each statement is re-checked against the code at fix time.
• Documentation listed above is corrected in English and in its Turkish mirror in the same PR.
• Every finding in this issue is resolved or explicitly closed with a reason.

_{Found by the October 2026 ForgeLM code and documentation review against commit cab2947. Code links point at that commit; line numbers may have moved since. Finding IDs (CR-…, DOC-…, TECH-…) are stable references for this backlog.}