Skip to content

Project NOMAD is an offline-first knowledge and education server. Wikipedia, thousands of books, courses, maps, and optional local AI, all running on hardware you own with no internet required.

TypeScript
3.4 KB
Raw Blame
export type NomadOllamaModel = {
  id: string
  name: string
  description: string
  estimated_pulls: string
  model_last_updated: string
  first_seen: string
  tags: NomadOllamaModelTag[]
}

export type NomadOllamaModelTag = {
  name: string
  size: string
  context: string
  input: string
  cloud: boolean
  thinking: boolean
}

export type NomadOllamaModelAPIResponse = {
  success: boolean
  message: string
  models: NomadOllamaModel[]
}

export type OllamaChatMessage = {
  role: 'system' | 'user' | 'assistant'
  content: string
}

export type OllamaChatRequest = {
  model: string
  messages: OllamaChatMessage[]
  stream?: boolean
  sessionId?: number
  // Effective thinking preference for this request (per-model override or global default).
  think?: boolean
  collection?: string
}

export type ModelVisionCapability = 'supported' | 'unsupported' | 'unknown'

export type OllamaChatResponse = {
  model: string
  created_at: string
  message: {
    role: string
    content: string
  }
  done: boolean
}

export type NomadInstalledModel = {
  name: string
  size: number
  digest?: string
  details?: Record<string, any>
  // Whether the model supports "thinking" (set by the installed-models endpoint enrichment).
  thinking?: boolean
  // Capability detection is authoritative for Ollama and llama.cpp. Other OpenAI-compatible
  // backends report "unknown" so the user can still try an image-capable model.
  vision?: ModelVisionCapability
}

export type NomadChatResponse = {
  message: { content: string; thinking?: string }
  done: boolean
  model: string
  // 'length' means the answer hit the generation cap and was cut off (#1342).
  done_reason?: string
}

/**
 * The four sampler profiles offered as "Response Style" in Settings > Models.
 *
 * `auto` is the default and the only one that defers to the model: a value the
 * model author baked into their modelfile wins over NOMAD's baseline, key by
 * key. `focused` and `creative` are deliberate user overrides and replace those
 * values wholesale. `off` sends no sampler settings at all, reproducing the
 * pre-setting behaviour and giving a bug report somewhere clean to start.
 */
export type ResponseStyle = 'auto' | 'focused' | 'creative' | 'off'

/**
 * Resolved sampler settings for one chat request.
 *
 * Every field is optional because "unset" is meaningful: it means the backend
 * (or the model's own modelfile) decides, which is exactly what `off` wants and
 * what a missing key under `auto` falls back to.
 */
export type SamplerProfile = {
  temperature?: number
  /**
   * Nucleus cut for the native transport. The presets leave this at 1.0 and let
   * min_p do the truncating: min_p is relative to the top token's probability,
   * so it tightens on a confident distribution and loosens on an uncertain one,
   * where a fixed nucleus does neither. Stacking both mostly means the weaker
   * cut never fires.
   */
  topP?: number
  topK?: number
  /** The lever this whole setting exists for. Ollama's default is 0 (disabled). */
  minP?: number
  repeatPenalty?: number
  /**
   * `top_p` to send on the OpenAI-compatible transport, which has no min_p.
   *
   * Without this the compat path would inherit `topP: 1.0` from a profile whose
   * truncation lives entirely in min_p, and end up sampling the raw distribution
   * — strictly worse than the 0.9 Ollama's /v1 endpoint defaults to. So the
   * compat path substitutes a real nucleus value instead of dropping to nothing.
   */
  compatTopP?: number
}