Smoke
Basic exactness, simple JSON schema, regex, and “don’t leak forbidden token” checks.
The suite is a routing test: can a model follow instructions, preserve source truth, resist injection, emit strict tool/diff formats, handle long context, and make reliable agent decisions?
The first 75 cases are deterministic gates: exact output, JSON shape, no forbidden text, tool traces, fact recall, and diffs that must apply. The v3 quality layer adds 18 harder rubric-scored cases for reasoning, judgment, debugging, planning, and synthesis. Separate smoke checks now cover vision enablement, slot sizing, and live serving shape.
Basic exactness, simple JSON schema, regex, and “don’t leak forbidden token” checks.
Source-grounded recall, caveats, contradictions, priority selection, schema output, secret refusal.
Tool choice, terminal/read/search behavior, recovery, multi-tool order, forbidden tools, injection.
Needle/fact retrieval, contradictions, negative evidence, JSON, list discipline, injection-in-context.
Grounded summaries, caveats, unsupported claims, contradictions, ranked evidence, schema output.
Forbidden disclosure, source hierarchy, roleplay, encoding, JSON/tooltext conflict, instruction priority.
Sequences, logic, constraints, tables, arithmetic, caveats, ordering, and one raw JSON case.
Seven patch outputs that must be unfenced and actually apply.
Reasoning depth, agent judgment, debugging diagnosis, planning tradeoffs, and grounded synthesis.
--mmproj, image requests return HTTP 500 and no vision path is active.-c 262144 -np 2; -c 131072 -np 2 is only 2×64k.n_ctx: 131072./v1/models, /slots, GPU residency, and exact response QAT-128K-OK all passed.