Documentation
¶
Overview ¶
Package presets owns reusable inference configuration presets and persistence.
Index ¶
Constants ¶
const ( SortPrice = "price" SortLatency = "latency" SortThroughput = "throughput" SortSpread = "spread" )
Provider sort orders a preset may store. They mirror the OpenRouter wire values; "throughput" routes by measured latency, Starport's closest signal. "spread" is the Starport extension that balances traffic across the routes whose ranking metric sits inside one band.
const ( // StorageSchemaVersion identifies the only preset schema. StorageSchemaVersion = 1 // StoragePrefix is the preset v1 namespace. StoragePrefix = "presets:v1:name:" )
const ( // RevisionPrefix is the immutable preset revision namespace. Each // save stores its record here beside the head, keyed by name and a // zero-padded revision number so a prefix scan lists in order. RevisionPrefix = "presets:v1:revision:" )
Variables ¶
var ( // ErrRepositoryRequired reports an absent preset storage adapter. ErrRepositoryRequired = errors.New("preset storage is required") // ErrNotFound reports an absent preset. ErrNotFound = errors.New("preset not found") // ErrConflict reports a preset revision conflict. ErrConflict = errors.New("preset revision conflict") // ErrCorruptRecord reports invalid durable preset data. ErrCorruptRecord = errors.New("preset record is invalid") // ErrNameImmutable reports an attempted preset-name change. ErrNameImmutable = errors.New("preset name is immutable") )
var ErrInvalidPreset = errors.New("invalid preset")
ErrInvalidPreset reports a preset that fails validation.
Functions ¶
This section is empty.
Types ¶
type Config ¶ added in v1.1.0
type Config struct {
Model string `json:"model,omitempty"`
Models []string `json:"models,omitempty"`
Provider *ProviderPreferences `json:"provider,omitempty"`
System string `json:"system,omitempty"`
Temperature *float32 `json:"temperature,omitempty"`
TopP *float32 `json:"top_p,omitempty"`
PresencePenalty *float32 `json:"presence_penalty,omitempty"`
FrequencyPenalty *float32 `json:"frequency_penalty,omitempty"`
MaxTokens *int `json:"max_tokens,omitempty"`
Seed *int `json:"seed,omitempty"`
Stop []string `json:"stop,omitempty"`
}
Config is the typed preset subset a request may inherit: model selection, provider preferences, a system prompt, and sampling overrides. Fields the request supplies win over preset fields.
type Preset ¶
type Preset struct {
Name string `json:"name"`
Description string `json:"description,omitempty"`
Config Config `json:"config"`
CreatedAt time.Time `json:"created_at"`
UpdatedAt time.Time `json:"updated_at"`
}
Preset is one reusable inference configuration. The name is the reference: requests select a preset with "@preset/<name>" or a "preset" body field.
type ProviderPreferences ¶ added in v1.1.0
type ProviderPreferences struct {
Order []string `json:"order,omitempty"`
Only []string `json:"only,omitempty"`
Ignore []string `json:"ignore,omitempty"`
AllowFallbacks *bool `json:"allow_fallbacks,omitempty"`
// Sort selects the route ordering; empty keeps the server default.
Sort string `json:"sort,omitempty"`
// Price caps are USD per million tokens, matching the wire shape.
// Zero means no cap.
MaxPromptPricePer1M float64 `json:"max_prompt_price_per_1m,omitempty"`
MaxCompletionPricePer1M float64 `json:"max_completion_price_per_1m,omitempty"`
}
ProviderPreferences is the preset-owned provider routing policy.
type Record ¶
Record is one versioned preset repository value. Revision carries one meaning with two uses: it is the preset's place in its immutable history, and because every save is a new revision, the same number is the optimistic-concurrency token an update must name. The two never diverge by construction.
type Repository ¶
type Repository interface {
Create(context.Context, Preset) (Record, error)
Get(context.Context, string) (Record, error)
List(context.Context, int) ([]Record, error)
Update(context.Context, Preset, uint64) (Record, error)
Delete(context.Context, string, uint64) error
// History answers stored revisions newest-first, up to limit.
History(ctx context.Context, name string, limit int) ([]Record, error)
// GetRevision answers one pinned revision verbatim.
GetRevision(ctx context.Context, name string, revision uint64) (Record, error)
// Rollback saves a new head revision that copies an old one. The
// expected revision names the head the caller read, like Update.
Rollback(ctx context.Context, name string, toRevision, expectedRevision uint64) (Record, error)
}
Repository is the durable preset contract. Every save writes an immutable revision snapshot beside the head, so a request can pin a revision and an operator can roll back to one.
Source Files