refactor: only write reasoning to file; count line overflow better

This commit is contained in:
mitchell 2026-04-16 04:00:33 -04:00
parent 8facbb0de2
commit 993fb143a2
2 changed files with 13 additions and 15 deletions

View file

@ -6,9 +6,6 @@ import { renderStream } from "./src/render";
const home = os.homedir();
const convoFile = Bun.file(`${home}/.aicl_convo.json`);
const reasoningFile = Bun.file(`${home}/.aicl_reasoning.md`);
const responsePath = `${home}/.aicl_response.md`;
const responseFile = Bun.file(responsePath);
const provider = createOpenAICompatible({
name: "llama.cpp",
@ -27,8 +24,8 @@ try {
const prompt = Bun.argv[2] ?? "Introduce yourself.";
messages.push({ role: "user", content: prompt });
const { fullStream, textStream, reasoningText, text, totalUsage } = streamText({
model: provider("qwen3.5-35b-a3b"),
const { fullStream, textStream, text, totalUsage } = streamText({
model: provider("gemma4-26b-a4b"),
messages,
});
@ -43,20 +40,15 @@ const glow = Bun.spawn({
stdout: "inherit",
});
const reasoning = await reasoningText;
const response = await text;
await streamRender;
const stop = Date.now();
messages.push({ role: "assistant", content: response });
await convoFile.write(JSON.stringify(messages));
await reasoningFile.write(reasoning ?? "");
await responseFile.write(response);
await streamRender;
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;