jaderinoo

FP8 draft stack for Qwen3.5-family MTP on vLLM

modelopt NVFP4 checkpoints exclude mtp* from quantization, so the entire MTP speculative-decoding stack runs as BF16 passthrough: ~850 MiB of draft-layer weights + mtp.fc, plus the 40960-row draft LM head (~420 MiB) — all re-read on every draft forward (3× per decode step at num_speculative_tokens=3).

This gist converts the whole draft stack to FP8 e4m3 (per-tensor weight scales, dynamic per-token activations, CUTLASS scaled_mm). The drafter only proposes tokens — the target model verifies every one — so **outputs are bit-identical

karpathy

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

jasonsnell

use AppleScript version "2.4" -- Yosemite (10.10) or later
use scripting additions

# Sample Keyboard Maestro macros at 
# <https://sixcolors.com/wp-content/uploads/2022/01/Podcast-Noter-Macros.kmmacros.zip>

on run argv
	try
		set theNote to (item 1 of argv)
	on error

alanyee

"""Using boto3, instead of awscli, to tail logs. Based on https://github.com/aws/aws-cli/blob/v2/awscli/customizations/logs/tail.py"""

from collections import defaultdict
import sys
import time

import boto3
from botocore.exceptions import ClientError, ParamValidationError

SLEEP = 5

lancethomps

function run(input, parameters) {
  const appNames = [];
  const skipAppNames = [];
  const verbose = true;

  const scriptName = 'close_notifications_applescript';

  const CLEAR_ALL_ACTION = 'Clear All';
  const CLEAR_ALL_ACTION_TOP = 'Clear';
  const CLOSE_ACTION = 'Close';

qistoph

# Based on gist https://gist.github.com/staaldraad/4c4c80800ce15b6bef1c1186eaa8da9f
# - added TCP states

awk 'BEGIN{states["01"]="TCP_ESTABLISHED"
states["02"]="TCP_SYN_SENT"
states["03"]="TCP_SYN_RECV"
states["04"]="TCP_FIN_WAIT1"
states["05"]="TCP_FIN_WAIT2"
states["06"]="TCP_TIME_WAIT"
states["07"]="TCP_CLOSE"

benkehoe

# MIT No Attribution
#
# Copyright 2022 Ben Kehoe
#
# Permission is hereby granted, free of charge, to any person obtaining a copy of this
# software and associated documentation files (the "Software"), to deal in the Software
# without restriction, including without limitation the rights to use, copy, modify,
# merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
# permit persons to whom the Software is furnished to do so.
#

JacksonChen666

# It is recommended that you save this script as an application, so you don't have to open Script editor.
(* 
It all started when I wanted to join zoom faster than a human could do with no prompt like how I expected like this.
I first discovered zoom.us's URL scheme to join meeting, but when I found out it doesn't work, I started to make this program.
After this all of stuff, I finally have something that is constantly updated.
After this pandemic is over, development might slow down or become discontinued.
*)
global meetingNames, meetingIDs, meetingPwds, wname
# Compiling the app will reset all your saved meetings!
property meetingNames : {}

karanlyons

private rule Macho
{
    meta:
        description = "private rule to match Mach-O binaries (copied from Apple's XProtect)"
    condition:
        uint32(0) == 0xfeedface or uint32(0) == 0xcefaedfe or uint32(0) == 0xfeedfacf or uint32(0) == 0xcffaedfe or uint32(0) == 0xcafebabe or uint32(0) == 0xbebafeca
}

rule ZoomDaemon
{

ttscoff

This has been replaced by a gem

Use gem install bunchcli to install, then run bunch -h for a list of commands.