SPB Git forge

spb/llm-api

Public
0commits 0branches 0releases
0 Bsize
maindefault branch
—last push
5.5 KB · 226 lines typescript
Raw Blame History
1export type Compat = "compatible" | "compatible_with_restrictions" | "experimental" | "not_recommended" | "incompatible";23export interface Model {4  id: string;5  name: string;6  family: string | null;7  provider: string | null;8  architecture: string | null;9  model_type: string | null;10  parameter_count: number | null;11  active_parameter_count: number | null;12  quantization: string | null;13  quant_bits: number | null;14  runtime: "mlx" | "llamacpp";15  format: string;16  path: string;17  weights_file: string | null;18  disk_size_bytes: number;19  weights_bytes: number;20  estimated_ram_gb: number | null;21  kv_bytes_per_token: number | null;22  recommended_context: number | null;23  max_context: number | null;24  task: string;25  vision: boolean;26  embedding: boolean;27  reranker: boolean;28  thinking: boolean;29  tools: boolean;30  size_class: string | null;31  installed: boolean;32  enabled: boolean;33  favorite: boolean;34  pinned: boolean;35  verified: boolean;36  compatible: boolean;37  compatibility_status: Compat;38  compatibility_reason: string | null;39  tags: string[];40  repository: string | null;41  notes: string | null;42  overrides: Record<string, unknown>;43  created_at: number;44  updated_at: number;45  last_loaded_at: number | null;46  last_used_at: number | null;47  load_count: number;48  request_count: number;49  tokens_generated: number;50  min_load_ms: number | null;51  avg_load_ms: number | null;52  max_load_ms: number | null;53  last_load_ms: number | null;54  avg_tps: number | null;55  first_token_latency_ms: number | null;56  status: string;57  loaded: boolean;58  aliases?: string[];59  worker?: Worker;60  progress?: Progress;61}6263export interface Worker {64  model_id: string;65  name: string;66  runtime: string;67  status: string;68  port: number;69  pid: number;70  context: number;71  started_at: number;72  ready_at: number | null;73  last_used: number;74  in_flight: number;75  estimate_gb: number;76  measured_gb: number;77  warm: Record<string, unknown>;78  error: string | null;79  requests: number;80  pinned: boolean;81  elapsed_seconds: number;82  load_ms?: number;83}8485export interface Progress {86  status: string;87  started: number;88  elapsed_seconds: number;89  [k: string]: unknown;90}9192export interface ManagerState {93  loaded: Worker[];94  progress: Record<string, Progress>;95  waiting: Record<string, number>;96  switching: boolean;97  stats: { requests: number; tokens: number; loads: number; unloads: number; evictions: number; errors: number };98  resident_gb: number;99  runtimes: Record<string, boolean>;100}101102export interface Telemetry {103  ts: number;104  mem_total_gb: number;105  mem_used_gb: number;106  mem_available_gb: number;107  mem_wired_gb: number | null;108  mem_compressed_gb: number | null;109  mem_pressure_percent: number | null;110  mem_pressure_level: string;111  swap_used_gb: number;112  swap_total_gb: number;113  cpu_percent: number;114  cpu_per_core: number[];115  load_avg: number[];116  gpu_percent: number | null;117  gpu_renderer_percent: number | null;118  gpu_memory_gb: number | null;119  thermal_state: string;120  thermal_cpu_speed_limit: number | null;121  disk_total_gb: number;122  disk_used_gb: number;123  disk_free_gb: number;124  uptime_seconds: number;125  process_rss_gb: number;126  worker_rss_gb?: number;127  loaded_model?: string | null;128  loaded_models?: string[];129  app_uptime_seconds?: number;130}131132export interface Hardware {133  chip: string;134  memory_gb: number;135  cpu_cores: number;136  performance_cores: number | null;137  efficiency_cores: number | null;138  gpu_cores: number | null;139  os: string;140  os_version: string;141  hostname: string;142  apple_silicon: boolean;143  disk_total_gb: number;144  python: string;145}146147export interface Job {148  id: string;149  kind: string;150  title: string;151  payload: Record<string, unknown>;152  status: "queued" | "running" | "completed" | "failed" | "cancelled";153  progress: number;154  detail: Record<string, unknown>;155  result: unknown;156  error: string | null;157  created_at: number;158  started_at: number | null;159  finished_at: number | null;160}161162export interface ApiKey {163  id: number;164  name: string;165  prefix: string;166  scopes: string[];167  created_at: number;168  last_used_at: number | null;169  request_count: number;170  revoked_at: number | null;171}172173export interface Benchmark {174  id: number;175  model_id: string;176  created_at: number;177  load_ms: number | null;178  prompt_tokens: number | null;179  prompt_tps: number | null;180  generation_tokens: number | null;181  generation_tps: number | null;182  ttft_ms: number | null;183  peak_memory_gb: number | null;184  avg_memory_gb: number | null;185  cpu_percent: number | null;186  gpu_percent: number | null;187  thermal_state: string | null;188  context: number | null;189  runtime: string | null;190  params: Record<string, unknown> | string | null;191}192193export interface Candidate {194  repo_id: string;195  runtime: string;196  family: string;197  base_model: string;198  name: string;199  task: string;200  quantization: string | null;201  parameter_count: number | null;202  download_bytes: number | null;203  estimated_ram_gb: number | null;204  size_class: string | null;205  compatibility_status: Compat;206  compatibility_reason: string | null;207  downloads: number;208  likes: number;209  last_modified: string;210  duplicate_of: string | null;211  installed: number;212  selected: number;213  dismissed: number;214  score: number;215}216217export interface SystemInfo {218  hardware: Hardware;219  telemetry: Telemetry;220  manager: ManagerState;221  policy: { max_model_memory_gb: number; absolute_max_memory_gb: number; max_simultaneous_models: number; macos_reserve_gb: number; min_free_disk_gb: number };222  version: string;223  started_at: number;224  runtimes: Record<string, string | null>;225}226