@simulatte/doppler 0.1.9 → 0.2.0

package/src/client/doppler-provider/types.d.ts

@@ -1,127 +0,0 @@
-import type { ExtensionBridgeClient } from '../../bridge/index.js';
-import type { InferencePipeline, KVCacheSnapshot } from '../../inference/pipelines/text.js';
-import type { LoRAManifest } from '../../adapters/lora-loader.js';
-import type { RDRRManifest } from '../../formats/rdrr/index.js';
-
-export declare const DOPPLER_PROVIDER_VERSION: string;
-
-export interface TextModelConfig {
-  numLayers: number;
-  hiddenSize: number;
-  intermediateSize: number;
-  numHeads: number;
-  numKVHeads: number;
-  headDim: number;
-  vocabSize: number;
-  maxSeqLen: number;
-  quantization: string;
-}
-
-export interface InferredAttentionParams {
-  numHeads: number;
-  numKVHeads: number;
-  headDim: number;
-}
-
-export interface ModelEstimate {
-  weightsBytes: number;
-  kvCacheBytes: number;
-  totalBytes: number;
-  modelConfig: TextModelConfig;
-}
-
-export interface LoadProgressEvent {
-  stage: 'connecting' | 'manifest' | 'estimate' | 'warming' | 'downloading' | 'loading';
-  message: string;
-  estimate?: ModelEstimate;
-}
-
-export interface GenerateOptions {
-  maxTokens?: number;
-  temperature?: number;
-  topP?: number;
-  topK?: number;
-  stopTokens?: number[];
-  stopSequences?: string[];
-  useChatTemplate?: boolean;
-  onToken?: (token: string) => void;
-}
-
-export interface ChatMessage {
-  role: 'system' | 'user' | 'assistant';
-  content: string | Array<Record<string, unknown>>;
-}
-
-export interface ChatResponse {
-  content: string;
-  usage: {
-    promptTokens: number;
-    completionTokens: number;
-    totalTokens: number;
-  };
-}
-
-export interface DopplerCapabilitiesType {
-  available: boolean;
-  HAS_MEMORY64: boolean;
-  HAS_SUBGROUPS: boolean;
-  HAS_F16: boolean;
-  IS_UNIFIED_MEMORY: boolean;
-  TIER_LEVEL: number;
-  TIER_NAME: string;
-  MAX_MODEL_SIZE: number;
-  initialized: boolean;
-  currentModelId: string | null;
-  kernelsWarmed: boolean;
-  kernelsTuned: boolean;
-  lastModelEstimate: ModelEstimate | null;
-  bridgeClient?: ExtensionBridgeClient | null;
-  localPath?: string | null;
-}
-
-export interface DopplerProviderInterface {
-  name: string;
-  displayName: string;
-  isLocal: boolean;
-  init(): Promise<boolean>;
-  loadModel(
-    modelId: string,
-    modelUrl?: string | null,
-    onProgress?: ((event: LoadProgressEvent) => void) | null,
-    localPath?: string | null
-  ): Promise<boolean>;
-  unloadModel(): Promise<void>;
-  chat(messages: ChatMessage[], options?: GenerateOptions): Promise<ChatResponse>;
-  stream(messages: ChatMessage[], options?: GenerateOptions): AsyncGenerator<string>;
-  prefillKV(prompt: string, options?: GenerateOptions): Promise<KVCacheSnapshot>;
-  generateWithPrefixKV(prefix: KVCacheSnapshot, prompt: string, options?: GenerateOptions): AsyncGenerator<string>;
-  loadLoRAAdapter(adapter: LoRAManifest | RDRRManifest | string): Promise<void>;
-  activateLoRAFromTrainingOutput(
-    trainingOutput:
-      | string
-      | {
-        adapter?: LoRAManifest | RDRRManifest | string;
-        adapterManifest?: LoRAManifest | RDRRManifest;
-        adapterManifestJson?: string;
-        adapterManifestUrl?: string;
-        adapterManifestPath?: string;
-      }
-      | null
-      | undefined
-  ): Promise<{
-    activated: boolean;
-    adapterName: string | null;
-    source: string | null;
-    reason: string | null;
-  }>;
-  unloadLoRAAdapter(): Promise<void>;
-  getActiveLoRA(): string | null;
-  getPipeline(): InferencePipeline | null;
-  getCurrentModelId(): string | null;
-  getCapabilities(): DopplerCapabilitiesType;
-  getModels(): Promise<string[]>;
-  getAvailableModels(): Promise<string[]>;
-  destroy(): Promise<void>;
-}
-
-export declare const DopplerCapabilities: DopplerCapabilitiesType;

package/src/client/doppler-provider/types.js

@@ -1,17 +0,0 @@
-export { DOPPLER_PROVIDER_VERSION } from '../../version.js';
-
-export const DopplerCapabilities = {
-  available: false,
-  HAS_MEMORY64: false,
-  HAS_SUBGROUPS: false,
-  HAS_F16: false,
-  IS_UNIFIED_MEMORY: false,
-  TIER_LEVEL: 1,
-  TIER_NAME: '',
-  MAX_MODEL_SIZE: 0,
-  initialized: false,
-  currentModelId: null,
-  kernelsWarmed: false,
-  kernelsTuned: false,
-  lastModelEstimate: null,
-};

package/src/client/doppler-provider.d.ts

@@ -1,46 +0,0 @@
-export {
-  // Version
-  DOPPLER_PROVIDER_VERSION,
-
-  // Types
-  type TextModelConfig,
-  type InferredAttentionParams,
-  type ModelEstimate,
-  type LoadProgressEvent,
-  type GenerateOptions,
-  type ChatMessage,
-  type ChatResponse,
-  type DopplerCapabilitiesType,
-  type DopplerProviderInterface,
-
-  // Capability flags
-  DopplerCapabilities,
-
-  // Model management
-  initDoppler,
-  loadModel,
-  unloadModel,
-  loadLoRAAdapter,
-  activateLoRAFromTrainingOutput,
-  unloadLoRAAdapter,
-  getActiveLoRA,
-  getAvailableModels,
-  destroyDoppler,
-  getPipeline,
-  getCurrentModelId,
-
-  // Generation
-  generate,
-  prefillKV,
-  generateWithPrefixKV,
-  formatGemmaChat,
-  formatLlama3Chat,
-  formatGptOssChat,
-  formatChatMessages,
-  buildChatPrompt,
-  dopplerChat,
-
-  // Main provider
-  DopplerProvider,
-  default,
-} from './doppler-provider.js';

package/src/client/doppler-provider.js

@@ -1,36 +0,0 @@
-// Types and interfaces
-export {
-  DOPPLER_PROVIDER_VERSION,
-  DopplerCapabilities,
-} from './doppler-provider/types.js';
-
-// Model management
-export {
-  initDoppler,
-  loadModel,
-  unloadModel,
-  loadLoRAAdapter,
-  activateLoRAFromTrainingOutput,
-  unloadLoRAAdapter,
-  getActiveLoRA,
-  getAvailableModels,
-  destroyDoppler,
-  getPipeline,
-  getCurrentModelId,
-} from './doppler-provider/model-manager.js';
-
-// Generation
-export {
-  generate,
-  prefillKV,
-  generateWithPrefixKV,
-  formatGemmaChat,
-  formatLlama3Chat,
-  formatGptOssChat,
-  formatChatMessages,
-  buildChatPrompt,
-  dopplerChat,
-} from './doppler-provider/generation.js';
-
-// Main provider
-export { DopplerProvider, default } from './doppler-provider/provider.js';

package/src/client/doppler-registry.d.ts

@@ -1,23 +0,0 @@
-export interface QuickstartRegistryEntry {
-  modelId: string;
-  aliases: string[];
-  modes: string[];
-  hf: {
-    repoId: string;
-    revision: string | null;
-    path: string;
-  } | null;
-}
-
-export declare function listQuickstartModels(): Promise<Array<{
-  modelId: string;
-  aliases: string[];
-  modes: string[];
-}>>;
-
-export declare function resolveQuickstartModel(model: string): Promise<QuickstartRegistryEntry>;
-
-export declare function buildQuickstartModelBaseUrl(
-  entry: QuickstartRegistryEntry,
-  options?: { cdnBasePath?: string }
-): string;

package/src/client/doppler-registry.js

@@ -1,86 +0,0 @@
-import { getCdnBasePath } from '../storage/download-types.js';
-import { buildHfResolveBaseUrl } from '../utils/hf-resolve-url.js';
-import { loadJson } from '../utils/load-json.js';
-
-let registryPromise = null;
-
-function normalizeEntry(entry) {
-  if (!entry || typeof entry !== 'object') {
-    return null;
-  }
-  const modelId = typeof entry.modelId === 'string' ? entry.modelId.trim() : '';
-  if (!modelId) {
-    return null;
-  }
-  return {
-    modelId,
-    aliases: Array.isArray(entry.aliases)
-      ? entry.aliases.filter((alias) => typeof alias === 'string' && alias.trim().length > 0)
-      : [],
-    modes: Array.isArray(entry.modes)
-      ? entry.modes.filter((mode) => typeof mode === 'string' && mode.trim().length > 0)
-      : [],
-    hf: entry.hf && typeof entry.hf === 'object'
-      ? {
-        repoId: typeof entry.hf.repoId === 'string' ? entry.hf.repoId.trim() : '',
-        revision: typeof entry.hf.revision === 'string' && entry.hf.revision.trim().length > 0
-          ? entry.hf.revision.trim()
-          : null,
-        path: typeof entry.hf.path === 'string' ? entry.hf.path.trim() : '',
-      }
-      : null,
-  };
-}
-
-async function loadRegistry() {
-  if (!registryPromise) {
-    registryPromise = loadJson(
-      './doppler-registry.json',
-      import.meta.url,
-      'Failed to load Doppler quickstart registry'
-    ).then((raw) => {
-      const entries = Array.isArray(raw?.models)
-        ? raw.models.map(normalizeEntry).filter(Boolean)
-        : [];
-      return { models: entries };
-    });
-  }
-  return registryPromise;
-}
-
-export async function listQuickstartModels() {
-  const registry = await loadRegistry();
-  return registry.models.map((entry) => ({
-    modelId: entry.modelId,
-    aliases: [...entry.aliases],
-    modes: [...entry.modes],
-  }));
-}
-
-export async function resolveQuickstartModel(model) {
-  const requested = typeof model === 'string' ? model.trim() : '';
-  if (!requested) {
-    throw new Error('Quickstart model id is required.');
-  }
-
-  const registry = await loadRegistry();
-  const resolved = registry.models.find((entry) => (
-    entry.modelId === requested || entry.aliases.includes(requested)
-  ));
-  if (resolved) {
-    return resolved;
-  }
-
-  const available = registry.models.map((entry) => entry.modelId).join(', ');
-  throw new Error(`Unknown quickstart model "${requested}". Available: ${available}`);
-}
-
-export function buildQuickstartModelBaseUrl(entry, options = {}) {
-  if (!entry?.hf?.repoId || !entry?.hf?.path) {
-    throw new Error(`Quickstart model "${entry?.modelId ?? 'unknown'}" does not have a hosted Hugging Face source.`);
-  }
-  const cdnBasePath = typeof options.cdnBasePath === 'string' && options.cdnBasePath.length > 0
-    ? options.cdnBasePath
-    : getCdnBasePath();
-  return buildHfResolveBaseUrl(entry.hf, { cdnBasePath });
-}

package/src/client/doppler-registry.json

@@ -1,40 +0,0 @@
-{
-  "version": 1,
-  "source": "models/catalog.json",
-  "models": [
-    {
-      "modelId": "gemma-3-270m-it-q4k-ehf16-af32",
-      "aliases": [
-        "gemma3-270m",
-        "google/gemma-3-270m-it",
-        "gemma-3-270m-it-wq4k-ef16",
-        "gemma-3-270m-it-wq4k-ef16-hf16",
-        "gemma-3-270m-it-wq4k-ef16-hf16-f32"
-      ],
-      "modes": [
-        "run"
-      ],
-      "hf": {
-        "repoId": "Clocksmith/rdrr",
-        "revision": "ca6f0dbdf3882d3893a65cf48f2bb6f1520df162",
-        "path": "models/gemma-3-270m-it-q4k-ehf16-af32"
-      }
-    },
-    {
-      "modelId": "google-embeddinggemma-300m-q4k-ehf16-af32",
-      "aliases": [
-        "embeddinggemma-300m",
-        "google/embeddinggemma-300m",
-        "google-embeddinggemma-300m-wq4k-ef16"
-      ],
-      "modes": [
-        "embedding"
-      ],
-      "hf": {
-        "repoId": "Clocksmith/rdrr",
-        "revision": "7e79c466d54455bd370c81685956ea9abae0fd30",
-        "path": "models/google-embeddinggemma-300m-q4k-ehf16-af32"
-      }
-    }
-  ]
-}

package/src/config/README.md

@@ -1,69 +0,0 @@
-# DOPPLER Config System
-
-Implementation notes for `src/config/*`.
-
-Runtime-facing config behavior is canonical in
-[../../docs/config.md](../../docs/config.md).
-
-## Scope
-
-This directory owns:
-- schema/default definitions
-- preset loading and merge helpers
-- kernel-path registry loading
-- config validation utilities
-
-## Directory map
-
-```text
-src/config/
-├── runtime.js                   # runtime get/set and resolved config access
-├── loader.js                    # preset/schema loading utilities
-├── merge.js                     # canonical merge behavior
-├── kernel-path-loader.js        # kernel-path registry/path resolution
-├── schema/                      # schemas + defaults
-└── presets/                     # runtime/model/platform/kernel-path presets
-```
-
-## Maintainer rules
-
-- Keep runtime behavior config-first; avoid hardcoded fallbacks in runtime paths.
-- Treat `registry.json` as source-of-truth for kernel-path identity/lifecycle.
-- Keep merge semantics centralized in config utilities (no ad-hoc deep merges in runners).
-- Preserve `null` vs `undefined` semantics required by schema contracts.
-
-## Internal workflows
-
-Validate config/preset integrity:
-
-```bash
-npm run kernels:check
-npm run support:matrix:check
-npm run onboarding:check:strict
-```
-
-Inspect kernel-path registry IDs:
-
-```bash
-node -e "const fs=require('node:fs');const r=JSON.parse(fs.readFileSync('src/config/presets/kernel-paths/registry.json','utf8'));console.log(r.entries.map((e)=>e.id).join('\n'));"
-```
-
-Inspect legacy aliases:
-
-```bash
-node -e "const fs=require('node:fs');const r=JSON.parse(fs.readFileSync('src/config/presets/kernel-paths/registry.json','utf8'));console.log(r.entries.filter((e)=>e.status==='legacy').map((e)=>`${e.id} -> ${e.aliasOf}`).join('\n'));"
-```
-
-## When adding a model family
-
-1. Add model preset in `src/config/presets/models/`.
-2. Add/adjust kernel-path presets in `src/config/presets/kernel-paths/`.
-3. Register IDs/status in `src/config/presets/kernel-paths/registry.json`.
-4. Wire converter defaults if runtime behavior must be emitted into manifest.
-5. Add tests for merge/selection behavior and run validation commands.
-
-## Related
-
-- [../../docs/config.md](../../docs/config.md)
-- [../../docs/style/config-style-guide.md](../../docs/style/config-style-guide.md)
-- [../../docs/conversion-runtime-contract.md](../../docs/conversion-runtime-contract.md)

package/src/config/backward-registry-loader.d.ts

@@ -1,3 +0,0 @@
-import type { BackwardRegistrySchema } from './schema/backward-registry.schema.js';
-
-export declare function loadBackwardRegistry(): BackwardRegistrySchema;

package/src/config/backward-registry-loader.js

@@ -1,23 +0,0 @@
-import { loadJson } from '../utils/load-json.js';
-import { validateBackwardRegistry } from './schema/backward-registry.schema.js';
-
-function deepFreeze(value, seen = new WeakSet()) {
-  if (!value || typeof value !== 'object' || seen.has(value)) {
-    return value;
-  }
-  seen.add(value);
-  for (const entry of Object.values(value)) {
-    deepFreeze(entry, seen);
-  }
-  return Object.freeze(value);
-}
-
-const backwardRegistryData = deepFreeze(
-  validateBackwardRegistry(
-    await loadJson('./kernels/backward-registry.json', import.meta.url, 'Failed to load json')
-  )
-);
-
-export function loadBackwardRegistry() {
-  return backwardRegistryData;
-}

package/src/config/execution-contract-check.d.ts

@@ -1,82 +0,0 @@
-import type {
-  ExecutionV0ContractArtifact,
-} from './execution-v0-contract-check.js';
-import type {
-  ExecutionV0GraphContractArtifact,
-} from './execution-v0-graph-contract-check.js';
-
-export interface ExecutionContractStepFacts {
-  id: string;
-  phase: 'prefill' | 'decode' | 'both';
-  opClass: 'attention' | 'embed' | 'norm' | 'projection' | 'residual' | 'sample' | 'other';
-}
-
-export interface ExecutionContractSessionFacts {
-  layout: 'contiguous' | 'paged' | 'tiered' | 'bdpa';
-  disableCommandBatching: boolean;
-  decodeBatchSize: number;
-  headDim: number;
-  kvLen: number;
-  coldQuantMode: 'none' | 'int8' | 'int4';
-}
-
-export interface ExecutionContractFacts {
-  modelId: string;
-  session: ExecutionContractSessionFacts;
-  steps: ExecutionContractStepFacts[];
-}
-
-export interface ExecutionContractCheckResult {
-  id: string;
-  ok: boolean;
-}
-
-export interface ExecutionContractValidationResult {
-  ok: boolean;
-  errors: string[];
-  checks: ExecutionContractCheckResult[];
-}
-
-export interface ManifestExecutionContractValidationResult extends ExecutionContractValidationResult {
-  facts: ExecutionContractFacts;
-}
-
-export interface ExecutionContractArtifact {
-  schemaVersion: 1;
-  source: 'doppler';
-  ok: boolean;
-  checks: ExecutionContractCheckResult[];
-  errors: string[];
-  session: ExecutionContractSessionFacts | null;
-  steps: {
-    total: number;
-    attention: number;
-    attentionPhases: {
-      prefill: number;
-      decode: number;
-      both: number;
-    };
-  } | null;
-  executionV0?: {
-    kernelProfiles: ExecutionV0ContractArtifact | null;
-    graph: ExecutionV0GraphContractArtifact | null;
-  };
-}
-
-export declare function sanitizeLeanModuleName(value: unknown): string;
-
-export declare function extractExecutionContractFacts(
-  manifest: Record<string, unknown>
-): ExecutionContractFacts;
-
-export declare function validateExecutionContractFacts(
-  facts: ExecutionContractFacts
-): ExecutionContractValidationResult;
-
-export declare function validateManifestExecutionContract(
-  manifest: Record<string, unknown>
-): ManifestExecutionContractValidationResult;
-
-export declare function buildExecutionContractArtifact(
-  manifest: Record<string, unknown>
-): ExecutionContractArtifact | null;