JosephCatrambone

import numpy
import random

class Graph:
	def __init__(self):
		self.nodes = list()
		self.last_tape = None
	def forward(self, variables, output_node):
		for n in self.nodes:
			n.forward(variables)

JosephCatrambone

//use serde::{Serialize, Deserialize};
use std::collections::HashMap;

// If you have serde, serialization is available with this derive chain:
//#[derive(Clone, Debug, Default, Serialize, Deserialize)]
#[derive(Clone, Debug, Default)]
pub struct DecisionTree {
	feature: usize,
	threshold: f32,

JosephCatrambone

# Sample Usage:
#	var day_features = [
#		# freezing, raining, foggy, sunny
#		[0, 1, 0, 0],
#		[0, 0, 0, 1],
#		[0, 0, 0, 0],
#		[0, 0, 1, 0],
#		[1, 0, 0, 1],
#		[1, 1, 0, 0],
#		[1, 0, 1, 0],

JosephCatrambone

#include <BleKeyboard.h>
#include <TinyPICO.h>

TinyPICO tinypico = TinyPICO();
BleKeyboard keyboard; // keyboard("Name", "manufacturer", battery_level_100)

// a .-
// b -...
// c -.-.
// d -..

JosephCatrambone

class UnionFind:
  def __init__(self):
    self.parents = dict()
  
  def add(self, node_id:int):
    self.parents[node_id] = node_id
    
  def union(self, node_a:int, node_b:int):
    min_parent = min(self.parents[node_a], self.parents[node_b])
    self.parents[node_a] = min_parent

JosephCatrambone

mod ml_thread;

use gdnative::prelude::{godot_print, methods, Method, NativeClass, Node as GDNode, InitHandle, godot_init};
use ml_thread::start_language_model_thread;
use std::sync::mpsc::{channel, Receiver, RecvError, Sender, SendError};


const MAX_INPUT_LENGTH: usize = 512;
const BATCH_SIZE: usize = 1;

JosephCatrambone

import itertools
import sys

import numpy
import torch
from scipy.cluster.vq import kmeans, kmeans2
from transformers import AutoTokenizer, GPT2Model

def magnitude(x):
    return numpy.dot(x, x)**0.5

JosephCatrambone

def square_iterator(x: int, y: int):
    """Generate all the points on a square 2d lattice, spiraling out from the center.  Modified from a solution by Nikita Rybak. https://stackoverflow.com/a/3706260"""
    dx = 1
    dy = 0
    segment_length = 1
    leg_iterations = 0

    while True:
        x += dx
        y += dy

JosephCatrambone

Overview:

• MEND is a way to edit models to change output in a way that is local, reliable, and general.
• "Local" means unrelated output is not changed. "Reliable" means the model takes the desired corrections. "General" meaning variations on similar questions which would need correction also are corrected.
• Works even on very large models.

Differences to Prior Art:

Open Questions:

• They say it's a method for transforming raw fine-tuning gradients into a targeted parameter update. That would seem to suggest a new model results. At the same time, they say that the edits are applied to the model's weights at test time, which says the base MEND models need to be kept. Which is it?
• If they're producing a layer's parameter edit as an output, why do they care about producing the low-rank embedding of the deltas? Accumulate and apply them.