Functions

aggregateCompletion

function aggregateCompletion(raw: AsyncIterable<{ tokenId: number; scores: Map<number, number> }>, tokenizer: TokenizerAdapter, request: CompletionRequest, promptIds: readonly number[], eosTokenIds: readonly number[] = tokenizer.eosTokenIds): AsyncGenerator<CompletionStreamChunk>

Convert a raw stream to compatibility chunks with input and output word relevance.

Source: js/src/completions.ts:96

createClientWithTokenizer

function createClientWithTokenizer(options: InveraClientOptions, tokenizer: TokenizerAdapter): InveraClient

Create a high-level client with an advanced host-managed tokenizer.

Source: js/src/client.ts:222

createHuggingFaceTokenizer

function createHuggingFaceTokenizer(profile: ModelProfile): Promise<TokenizerAdapter>

Create a Node.js tokenizer adapter using the release-pinned Hugging Face tokenizer.

Source: js/src/completions.ts:56

decodeRawAttributionChunk

function decodeRawAttributionChunk(payload: unknown): { tokenId: number; scores: Map<number, number> }

Decode a streamed raw-attribution object into token relevance positions.

Source: js/src/completions.ts:18

parseQwenToolCalls

function parseQwenToolCalls(chunks: AsyncIterable<CompletionStreamChunk>): AsyncGenerator<CompletionStreamChunk>

Parse Qwen XML tool-call markers without exposing parser mechanics to Core users.

Source: js/src/completions.ts:150

streamEvidence

function streamEvidence(chunks: AsyncIterable<CompletionStreamChunk>, request: CompletionRequest, options: EvidenceOptions = {}): AsyncGenerator<EvidenceStreamEvent>

Add claim and tool-call evidence to an attributed completion stream.

Source: js/src/evidence.ts:51