(00:00:00) – How batch size affects token cost and speed
(00:32:09) – How MoE models are laid out across a GPU racks
(00:47:12) – How pipeline parallelism moves model layers across racks
(01:03:37) – Why Ilya said, “As we now know, pipelining is not wise.”
(01:18:59) – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal
(01:33:02) – Deducing long context memory costs from API pricing
(02:04:02) – Convergent evolution between neural nets and cryptography
| #!/usr/bin/env python3 | |
| """ | |
| Human quality transcripts from audio files using | |
| AssemblyAI for transcription and Google's Gemini for enhancement. | |
| Requirements: | |
| - AssemblyAI API key (https://www.assemblyai.com/) | |
| - Google API key (https://aistudio.google.com/) | |
| - Python packages: assemblyai, google-generativeai, pydub |
| import 'dart:math'; | |
| import 'package:flutter/material.dart'; | |
| import 'package:flutter/rendering.dart'; | |
| void main() { | |
| runApp(MyApp()); | |
| } | |
| class MyApp extends StatelessWidget { | |
| @override |
| import kotlinx.coroutines.CoroutineScope | |
| import kotlinx.coroutines.coroutineScope | |
| import kotlinx.coroutines.delay | |
| import kotlinx.coroutines.launch | |
| import java.lang.reflect.InvocationHandler | |
| import java.lang.reflect.InvocationTargetException | |
| import java.lang.reflect.Method | |
| import java.lang.reflect.Proxy | |
| import kotlin.coroutines.Continuation | |
| import kotlin.coroutines.cancellation.CancellationException |
Last updated March 28, 2021
There are now two ways to approach this:
- Using gpg and generating keys
- Using Kryptonite by krypt.co
This Gist explains how to do this using gpg in a step-by-step fashion. Kryptonite is actually wickedly easy to use-but you will still need to follow the instructions
If anybody needs animated webP support with Expo (Custom Dev Client) for the native <Image /> and <FastImage /> (read comments):
// create a file like plugins/withAnimatedWebPSupport.js -> this is for the native <Image />
const {I’m upgrading from a Mid 2014 MacBook Pro, so this isn’t a fair comparison to recent Intel machines, but if you’re like me and were waiting for a MacBook with a decent keyboard, you’ll see a big speed boost.
Non-scientific comparison - time to compile my ClojureScript project
- Mid 2014 MacBook Pro - 14s
- M1 Pro MacBook Pro (under Rosetta) - 17s
- M1 Pro MacBook Pro (native) - 6s
Adaptive bed mesh is merged into klipper master branch. You can use this feature without this custom macro. Official klipper adaptive bed mesh
- This macro will dynamically changing the bed mesh area based on the size of the parts will be printed. The fw will only probe on the area that the part will be printed (plus mesh_area_offset value)