lmc/index.ts

78 lines
1.9 KiB
TypeScript

import os from "node:os";
import { parseArgs } from "node:util";
import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { type ModelMessage, streamText } from "ai";
import { renderStream } from "./src/render";
const home = os.homedir();
const convoFile = Bun.file(`${home}/.aicl_convo.json`);
const { values: args, positionals } = parseArgs({
args: Bun.argv,
options: {
model: {
type: "string",
},
new: {
type: "boolean",
},
},
strict: true,
allowPositionals: true,
});
async function main() {
const provider = createOpenAICompatible({
name: "llama.cpp",
apiKey: "local",
baseURL: "http://127.0.0.1:8080/v1",
includeUsage: true, // Include usage information in streaming responses
});
const messages: ModelMessage[] = [];
if (!args.new) {
try {
const convo = await convoFile.json();
messages.push(...convo);
} catch {}
}
const prompt = positionals[2] ?? "Introduce yourself.";
messages.push({ role: "user", content: prompt });
const { fullStream, textStream, text, totalUsage } = streamText({
model: provider(args.model ?? "gemma4-26b-a4b"),
messages,
});
const start = Date.now();
const streamRender = renderStream(fullStream);
const [width] = process.stdout.getWindowSize();
const glow = Bun.spawn({
cmd: ["glow", "-w", `${width}`, "-"],
stdin: textStream,
stdout: "inherit",
});
await streamRender;
const stop = Date.now();
await glow.exited;
messages.push({ role: "assistant", content: await text });
await convoFile.write(JSON.stringify(messages));
const { inputTokens, outputTokens } = await totalUsage;
const context = (inputTokens ?? 0) / 1000;
const output = (outputTokens ?? 0) / 1000;
const time = (stop - start) / 1000;
process.stderr.write(
`\u001b[2mContext: ${context.toFixed(1)}k, Output: ${output.toFixed(1)}k, Time: ${time.toFixed(2)}s\u001b[22m\n`,
);
}
await main();