FEI Hao suxue

YouTube Transcript SKILL

Use this skill to fetch the transcript of a YouTube video, with or without timestamps.

Use this skill with Claude (by extracting it to .claude/skills/) or with any other agent using Skillz.

Note: This skill is unlikely to run successfully on the Claude web app, since access to YouTube is blocked. Use it with Claude Code or other local agents.

Claude Code Hooks for working with `uv`

by Glenn Matlin / glennmatlin on all socials

Installation

Download and copy all files in this gist to ~/.claude/
Move the .py files to ~/.claude/hooks
Restart Claude Code.

Recently, I learned that some of the top reward models on RewardBench were trained on a preference dataset that has unintentional contamination with the benchmark. The dataset, Skyworks Preferences 80k contains contamination by mixing a Magpie dataset in. Magpie is a new method for having language models generate instructions by prompting them with an empty chat template. The source for the Skyworks dataset that was contaminated is Argilla/magpie-ultra-v0.1, generated with Llama 3.1 405B Instruct. I would never expect a Magpie dataset to be contaminated.

What seems likely is that Meta trained on some these prompts, but the exact provenance of each prompt needs more example. For example, we learned that some of the prompts we used in our LLMBar subsets they got from popular training sets like Al

You are doing JSON APIs wrong.

When you use JSON to call an API - not a REST API, but something like JSON-RPC - you will usually want to encode one of several possible messages.

Your request body looks like this:

{
 "type": "MessageWithA",

58 bytes of CSS to look great nearly everywhere

When making this website, i wanted a simple, reasonable way to make it look good on most displays. Not counting any minimization techniques, the following 58 bytes worked well for me:

main {
  max-width: 38rem;
  padding: 2rem;
  margin: auto;
}

ANSI Escape Sequences

Standard escape codes are prefixed with Escape:

Ctrl-Key: ^[
Octal: \033
Unicode: \u001b
Hexadecimal: \x1B
Decimal: 27

	# =============================================
	# Git Worktree 快速管理助手
	# 用法：
	# ga <branch-name> # 在平行目录创建新 worktree + 新分支，并自动 cd 进去
	# gd # 删除当前 worktree 和对应分支（带确认）
	#
	# 依赖：
	# - gum[](https://github.com/charmbracelet/gum) 用于美观的确认对话框
	# - mise (可选，如果你用 mise 管理工具版本)
	# =============================================

	import torch
	from triton.testing import do_bench
	from torch.nn.attention.flex_attention import create_block_mask, flex_attention, noop_mask

	torch.manual_seed(0)

	import torch
	torch.set_default_device('cuda')

	def sliding_window(b, h, q_idx, kv_idx):

	#include <iostream>
	#include <fstream>
	#include <string>
	#include <vector>
	#include <unordered_map>
	#include <boost/algorithm/string.hpp>
	#include <utf8proc.h>

	//https://unicode.org/reports/tr15/#Norm_Forms
	//https://ssl.icu-project.org/apiref/icu4c/uchar_8h.html

	from __future__ import absolute_import, division, print_function

	import argparse
	import glob
	import logging
	import os
	import random

	import numpy as np
	import torch