Method SetDefaults

core/config/model_config.go:1213–1294

(opts ...ConfigLoaderOption)

}

func (cfg *ModelConfig) SetDefaults(opts ...ConfigLoaderOption) {
	lo := &LoadOptions{}
	lo.Apply(opts...)

	ctx := lo.ctxSize
	threads := lo.threads
	f16 := lo.f16
	debug := lo.debug

	// Cloud-proxy: normalise empty Mode so downstream consumers
	// switch on two concrete values only. Validate accepts empty too,
	// but SetDefaults is the chokepoint that runs before any
	// inference path reads cfg.Proxy.Mode.
	if cfg.Proxy.Mode == "" {
		cfg.Proxy.Mode = ProxyModePassthrough
	}

	// When templating is delegated to the backend (use_tokenizer_template),
	// the backend also owns tool-call grammar generation and parsing. Sending
	// a LocalAI-generated grammar alongside overrides the backend's native
	// (name-first) tool pipeline and makes it stream the tool-call JSON back as
	// plain content (issue #10052). The GGUF auto-import path already couples
	// these two flags; enforce it here so gallery and hand-written configs that
	// set use_tokenizer_template directly stay consistent.
	if cfg.TemplateConfig.UseTokenizerTemplate {
		cfg.FunctionsConfig.GrammarConfig.NoGrammar = true
	}

	// Apply model-family-specific inference defaults before generic fallbacks.
	// This ensures gallery-installed and runtime-loaded models get optimal parameters.
	ApplyInferenceDefaults(cfg, cfg.Name, cfg.Model)

	// Apply serving-policy defaults (device-independent): cross-request prefix
	// caching. Propagates to distributed nodes via the model options.
	ApplyServingDefaults(cfg)

	// Generic fallback defaults (sampling params + runtime flags), applied after
	// the model-family / hardware / serving tiers above. Only fills unset values.
	ApplyGenericDefaults(cfg)

	trueV := true
	falseV := false

	if threads == 0 {
		// Threads can't be 0
		threads = 4
	}

	if cfg.Threads == nil {
		cfg.Threads = &threads
	}

	if cfg.F16 == nil {
		cfg.F16 = &f16
	}

	if cfg.Debug == nil {
		cfg.Debug = &falseV