@axflow/models/dist/ollama/generation.mjs

Zero-dependency, modular SDK for building robust natural language applications

// src/ollama/generation.ts
import { HttpError, POST } from "@axflow/models/shared";
var OLLAMA_URL = "http://127.0.0.1:11434/api/generate";
async function streamBytes(request, options) {
  const url = options.apiUrl || OLLAMA_URL;
  const body = JSON.stringify(request);
  const response = await POST(url, {
    headers: options.headers || {},
    body,
    fetch: options.fetch,
    signal: options.signal
  });
  if (!response.body) {
    throw new HttpError("Expected response body to be a ReadableStream", response);
  }
  return response.body;
}
function noop(chunk) {
  return chunk;
}
function chunkToToken(chunk) {
  if (!chunk.done) {
    return chunk.response;
  } else {
    return "";
  }
}
async function stream(request, options) {
  const byteStream = await streamBytes(request, options);
  return byteStream.pipeThrough(new OllamaDecoderStream(noop));
}
async function streamTokens(request, options) {
  const byteStream = await streamBytes(request, options);
  return byteStream.pipeThrough(new OllamaDecoderStream(chunkToToken));
}
var OllamaGeneration = class {
  static streamBytes = streamBytes;
  static stream = stream;
  static streamTokens = streamTokens;
};
var OllamaDecoderStream = class _OllamaDecoderStream extends TransformStream {
  static parseChunk(line) {
    line = line.trim();
    if (line.length === 0) {
      return null;
    }
    try {
      return JSON.parse(line);
    } catch (e) {
      throw new Error(`Malformed streaming data from Ollama: ${JSON.stringify(line)}`);
    }
  }
  static transformer(map) {
    let buffer = [];
    const decoder = new TextDecoder();
    return (bytes, controller) => {
      const chunk = decoder.decode(bytes);
      for (let i = 0, len = chunk.length; i < len; ++i) {
        const isSeparator = chunk[i] === "\n";
        if (!isSeparator) {
          buffer.push(chunk[i]);
          continue;
        }
        const parsedChunk = _OllamaDecoderStream.parseChunk(buffer.join(""));
        if (parsedChunk) {
          controller.enqueue(map(parsedChunk));
        }
        buffer = [];
      }
    };
  }
  constructor(map) {
    super({ transform: _OllamaDecoderStream.transformer(map) });
  }
};
export {
  OllamaGeneration
};