32 lines
840 B
JavaScript
32 lines
840 B
JavaScript
import delay from 'delay';
|
|
|
|
fetch('http://localhost:11434/api/generate', {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: 'deepseek-r1:latest',
|
|
prompt: 'Hello! Tell me a long long joke.',
|
|
stream: true,
|
|
think: false,
|
|
}),
|
|
})
|
|
.then((res) => {
|
|
const reader = res.body.getReader();
|
|
const decoder = new TextDecoder();
|
|
|
|
reader.read().then(async ({ done, value }) => {
|
|
while (!done) {
|
|
// Convert chunk to string
|
|
const chunk = decoder.decode(value, { stream: true });
|
|
|
|
// Optionally, print each chunk as it arrives
|
|
const line = JSON.parse(chunk);
|
|
process.stdout.write(line.response ?? '');
|
|
|
|
// Read the next chunk
|
|
({ done, value } = await reader.read());
|
|
}
|
|
});
|
|
})
|
|
.catch(console.error);
|