mindcraft/src/models/groq.js

96 lines
3.1 KiB
JavaScript
Raw Normal View History

2024-06-18 19:48:16 -07:00
import Groq from 'groq-sdk'
2024-06-18 17:56:49 -07:00
import { getKey } from '../utils/keys.js';
2024-06-18 17:56:04 -07:00
2025-02-17 15:49:52 -08:00
// THIS API IS NOT TO BE CONFUSED WITH GROK!
// Go to grok.js for that. :)
2024-08-25 13:16:32 -07:00
2025-02-17 15:49:52 -08:00
// Umbrella class for everything under the sun... That GroqCloud provides, that is.
2024-08-31 15:29:34 -07:00
export class GroqCloudAPI {
2025-02-17 15:49:52 -08:00
2025-02-04 13:02:57 -06:00
constructor(model_name, url, params) {
2025-02-17 15:49:52 -08:00
2024-08-25 13:16:32 -07:00
this.model_name = model_name;
this.url = url;
this.params = params || {};
2025-02-04 13:02:57 -06:00
2025-02-17 15:49:52 -08:00
// Remove any mention of "tools" from params:
if (this.params.tools)
delete this.params.tools;
// This is just a bit of future-proofing in case we drag Mindcraft in that direction.
// I'm going to do a sneaky ReplicateAPI theft for a lot of this, aren't I?
if (this.url)
2024-08-31 15:29:34 -07:00
console.warn("Groq Cloud has no implementation for custom URLs. Ignoring provided URL.");
2025-02-17 15:49:52 -08:00
2024-08-31 15:29:34 -07:00
this.groq = new Groq({ apiKey: getKey('GROQCLOUD_API_KEY') });
2025-02-17 15:49:52 -08:00
2025-02-08 22:41:07 -08:00
}
2025-02-17 15:49:52 -08:00
2025-03-15 17:25:11 -05:00
async sendRequest(turns, systemMessage, stop_seq = null) {
// Construct messages array
let messages = [{"role": "system", "content": systemMessage}].concat(turns);
2024-08-25 13:16:32 -07:00
2025-03-15 17:25:11 -05:00
let res = null;
try {
console.log("Awaiting Groq response...");
// Handle deprecated max_tokens parameter
if (this.params.max_tokens) {
console.warn("GROQCLOUD WARNING: A profile is using `max_tokens`. This is deprecated. Please move to `max_completion_tokens`.");
this.params.max_completion_tokens = this.params.max_tokens;
delete this.params.max_tokens;
}
if (!this.params.max_completion_tokens) {
this.params.max_completion_tokens = 4000;
}
let completion = await this.groq.chat.completions.create({
"messages": messages,
"model": this.model_name || "llama-3.3-70b-versatile",
"stream": false,
"stop": stop_seq,
...(this.params || {})
});
res = completion.choices[0].message.content;
2025-02-17 15:49:52 -08:00
2025-03-15 17:25:11 -05:00
res = res.replace(/<think>[\s\S]*?<\/think>/g, '').trim();
}
2024-08-25 13:16:32 -07:00
catch(err) {
if (err.message.includes("content must be a string")) {
res = "Vision is only supported by certain models.";
} else {
console.log(this.model_name);
res = "My brain disconnected, try again.";
}
2024-08-25 13:16:32 -07:00
console.log(err);
}
return res;
}
2025-01-27 17:29:14 +09:00
async sendVisionRequest(messages, systemMessage, imageBuffer) {
const imageMessages = messages.filter(message => message.role !== 'system');
2025-01-27 17:29:14 +09:00
imageMessages.push({
role: "user",
content: [
{ type: "text", text: systemMessage },
{
type: "image_url",
image_url: {
url: `data:image/jpeg;base64,${imageBuffer.toString('base64')}`
}
}
]
});
return this.sendRequest(imageMessages);
}
2025-02-17 15:49:52 -08:00
async embed(_) {
2025-02-17 16:47:45 -06:00
throw new Error('Embeddings are not supported by Groq.');
2024-06-18 19:10:56 -07:00
}
}