broker + repl + safety: GBNF grammar-sampling passthrough (closes #88)
llama.cpp constrains the sampler to ONLY emit tokens matching a GBNF grammar. For small models this kills format drift at the token level — `CMD: <cmd>` is enforced by the sampler rather than hoped for via prompt discipline. Probe finding (this commit's pre-implementation): cloud (Anthropic via Bedrock) silently IGNORES the `grammar` field — returns normally via standard sampling. Default passthrough is safe for all routes; no per-model opt-in/opt-out needed in v1. Changes: - broker.lua build_request: `if opts.grammar then req.grammar = opts.grammar end`. Misformed grammar surfaces at request time via the existing transport-error path. - repl.lua ask_ai: `grammar_override = config.routing.grammars [req_class]` (same gating shape as #86's system_prompts override). Passed via opts.grammar in the call_broker invocation. - safety.lua is_destructive threads cfg.safety.probe_grammar through opts.grammar so llm_probe constrains the YES/NO output. Skips the regex-match dance entirely when the model can't drift. Caller-provided opts.grammar takes precedence over cfg. - config.lua gains two commented examples: * routing.grammars per class * safety.probe_grammar for the destructive probe 6 unit cases verified (stubbed curl.post_sse / broker.chat): - default: no grammar in body - opts.grammar -> body contains grammar JSON-encoded - safety probe_grammar reaches llm_probe via opts - no probe_grammar configured -> opts.grammar nil - caller opts.grammar takes precedence over cfg.safety.probe_grammar E2E against live local broker: - `routing.grammars.default = "root ::= \\"ACK\\""` configured; prompted "tell me a long story about a fox" -> model output EXACTLY "ACK" (sampler forced; would normally produce paragraphs). Grammar passthrough end-to-end confirmed. Regression: test_safety 87/87, test_router_model 31/31, repl loads. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -61,6 +61,14 @@ local function build_request(model_cfg, messages, stream, opts)
|
||||
if stream and opts.include_usage ~= false then
|
||||
req.stream_options = { include_usage = true }
|
||||
end
|
||||
-- #88: GBNF grammar passthrough. llama.cpp constrains the sampler
|
||||
-- to only emit tokens matching the grammar — eliminates format
|
||||
-- drift on small models. Probed cloud (Anthropic via Bedrock)
|
||||
-- silently ignores the field, so default passthrough is safe;
|
||||
-- no per-model opt-out needed in v1. Misformed grammar produces
|
||||
-- a broker error at request time (visible via the usual transport
|
||||
-- error path).
|
||||
if opts.grammar then req.grammar = opts.grammar end
|
||||
return url, json.encode(req), build_headers(model_cfg),
|
||||
(model_cfg.timeout_ms or 60000)
|
||||
end
|
||||
|
||||
Reference in New Issue
Block a user