Hugging Face's logo

Datasets:
AmiriHayes
/
sae_token_programs

Tasks:
Feature Extraction
Modalities:
Tabular
Formats:
csv
Size:
< 1K
Tags:
interpretability
sparse-autoencoder
gpt2
program-synthesis
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
sae_token_programs / programs /layer08 /08 /L08_F08000-08099.py
AmiriHayes's picture
AmiriHayes
Upload folder using huggingface_hub (part 7)
d8e5318 verified
Raw History Blame Contribute Delete
161 kB
"""GPT-2 SAE last-token programs, layer 8 -- features 08000 to 08099 (100 programs).
One function per SAE feature. Grouped 100 to a file so the set stays browsable;
the programs themselves are unchanged.
"""
def prog_L08F08000(input_tokens):
"""46 token families, 60 surface forms (train F1 of the uncapped optimum: 0.616).
Fires at 8,301 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: terminology related to static values and operations in programming or technology"""
# --- end external annotation ---
fire = {
# dynam (404 firings)
' dynamic', ' Dynamic', 'Dynamic',
# stat (315 firings)
' static', 'static', ' Static', 'Static',
# defens (310 firings)
' defensive',
# garbage (239 firings)
' garbage',
# class (186 firings)
' classical', ' Classical',
# surplu (176 firings)
' surplus',
# freel (174 firings)
' freelance',
# runtime (142 firings)
' runtime', ' Runtime', 'Runtime', 'runtime',
# stray (137 firings)
' stray',
# rent (117 firings)
' rental',
# acoust (99 firings)
' acoustic',
# rc (97 firings)
' RC', ' rc',
# ricky (93 firings)
' Ricky',
# res (92 firings)
' RES', 'RES',
# fly (91 firings)
' Flyers',
# repetit (90 firings)
' repetitive',
# outgo (89 firings)
' outgoing',
# ambient (86 firings)
' ambient',
# rapid (85 firings)
' Rapid',
# overhead (84 firings)
' overhead',
# newborn (81 firings)
' newborn',
# gravel (79 firings)
' gravel',
# offline (78 firings)
'Offline',
# turnov (75 firings)
' turnover',
# dispos (73 firings)
' disposable',
# idle (72 firings)
' idle',
# resp (65 firings)
' resp',
# reptil (62 firings)
' reptiles',
# intrins (56 firings)
' intrinsic',
# spontane (56 firings)
' spontaneous',
# lifeles (55 firings)
' lifeless',
# reflex (53 firings)
' reflex', ' Reflex',
# stationary (50 firings)
' stationary',
# inco (46 firings)
' inco',
# ---------------------------------------------------------------- (46 firings)
'----------------------------------------------------------------',
# redundant (46 firings)
' redundant',
# discretionary (42 firings)
' discretionary',
# rip (41 firings)
'RIP', ' RIP',
# undead (40 firings)
' undead',
# dynamical (38 firings)
' dynamically',
# inert (37 firings)
' inert',
# anarchy (35 firings)
' Anarchy',
# supplementary (33 firings)
' supplementary', ' Supplementary',
# ynam (26 firings)
'ynamic',
# airst (23 firings)
' airst',
# animate (23 firings)
'animate',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'dynam-family', 1
def prog_L08F08001(input_tokens):
"""8 token families, 14 surface forms (train F1 of the uncapped optimum: 0.137).
Fires at 175,135 of 36,786,566 positions.
Fires at 0.48% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases indicating possession or association"""
# --- end external annotation ---
fire = {
# - (9,466 firings)
'-',
# p (1,592 firings)
' P', 'p', 'P',
# b (1,484 firings)
' B', 'B', 'b',
# k (1,473 firings)
'k', ' K',
# 1 (972 firings)
' 1',
# v (894 firings)
' V', 'v',
# h (546 firings)
'h',
# w (521 firings)
'w',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '--family', 1
def prog_L08F08002(input_tokens):
"""38 token families, 50 surface forms (train F1 of the uncapped optimum: 0.423).
Fires at 20,032 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to JavaScript and its related functionalities"""
# --- end external annotation ---
fire = {
# javascript (955 firings)
' Javascript', ' JavaScript', ' javascript', 'javascript',
# privacy (612 firings)
' privacy',
# cbs (538 firings)
'CBS',
# enable (494 firings)
' enable',
# weekday (479 firings)
' weekday',
# newslett (468 firings)
' newsletters', 'Newsletter',
# brows (441 firings)
' browser', 'Browser',
# what (263 firings)
' Whats',
# pastebin (207 firings)
' Pastebin',
# unlock (204 firings)
' unlocks',
# read (194 firings)
' Reader', 'Reader',
# regist (184 firings)
' register',
# tweet (181 firings)
' Tweet',
# log (179 firings)
' log',
# html (163 firings)
' HTML',
# ph (161 firings)
'PH',
# load (155 firings)
'Loading', ' Loading',
# inbox (149 firings)
' inbox',
# ctv (136 firings)
'CTV',
# ent (125 firings)
' Enter',
# copyright (116 firings)
'Copyright',
# php (113 firings)
'php',
# merit (97 firings)
'Merit',
# npr (89 firings)
'NPR',
# wik (86 firings)
'Wik', ' Wik', 'wik',
# email (83 firings)
'Email',
# taboola (79 firings)
'taboola',
# google (73 firings)
'google', ' google',
# content (72 firings)
'Content', 'content',
# attribu (70 firings)
' Attribution',
# untitl (70 firings)
'Untitled',
# iframe (60 firings)
'iframe',
# width (59 firings)
' width',
# nex (53 firings)
' Nex',
# transcript (53 firings)
' Transcript',
# wiki (53 firings)
'Wiki',
# user (52 firings)
'User',
# spoil (51 firings)
'Spoiler', ' Spoiler',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'javascript-family', 1
def prog_L08F08003(input_tokens):
"""4 token families, 10 surface forms (train F1 of the uncapped optimum: 0.066).
Fires at 90,641 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases that express inclusivity and diversity"""
# --- end external annotation ---
fire = {
# or (3,568 firings)
' or', ' Or',
# no (1,506 firings)
' no', ' No', ' NO',
# and (1,007 firings)
' And', 'And', ' AND',
# ; (573 firings)
';', ' ;',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'or-family', 1
def prog_L08F08004(input_tokens):
"""2 token families, 3 surface forms (train F1 of the uncapped optimum: 0.154).
Fires at 102,124 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to ongoing events or situations in various locations"""
# --- end external annotation ---
fire = {
# in (46,665 firings)
' in', ' IN',
# inside (468 firings)
' inside',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'in-family', 1
def prog_L08F08005(input_tokens):
"""32 token families, 38 surface forms (train F1 of the uncapped optimum: 0.604).
Fires at 8,056 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: forms of the prefix "ab," which suggest a theme of separation or negation"""
# --- end external annotation ---
fire = {
# ab (3,306 firings)
'ab', ' ab', ' Ab', 'Ab', 'AB', ' AB',
# reb (252 firings)
' reb',
# aw (186 firings)
' aw',
# ur (180 firings)
' ur',
# ob (167 firings)
' ob',
# ec (125 firings)
' ec',
# apost (121 firings)
' apost', ' Apost',
# unab (111 firings)
' unab',
# od (100 firings)
' od',
# unb (96 firings)
' unb',
# wa (78 firings)
' wa',
# asc (76 firings)
' asc',
# wee (75 firings)
' wee',
# arche (75 firings)
' arche',
# kab (73 firings)
' Kab',
# hay (68 firings)
' hay',
# encro (57 firings)
' encro',
# receptor (55 firings)
' receptors',
# forb (54 firings)
' forb',
# q (47 firings)
' q',
# dab (46 firings)
' dab',
# ce (42 firings)
' ce',
# arg (37 firings)
' arg',
# ib (32 firings)
' ib',
# aber (31 firings)
' aber',
# ah (31 firings)
' ah',
# ala (31 firings)
' Alas',
# nob (30 firings)
'nob',
# esche (29 firings)
' esche',
# alk (27 firings)
' alk',
# abbre (25 firings)
' abbre',
# abhor (21 firings)
' abhor',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ab-family', 1
def prog_L08F08006(input_tokens):
"""47 token families, 54 surface forms (train F1 of the uncapped optimum: 0.700).
Fires at 15,832 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions of the name "Albert" and related references to Einstein"""
# --- end external annotation ---
fire = {
# alberta (765 firings)
' Alberta',
# albert (575 firings)
' Albert', 'Albert',
# oscar (541 firings)
' Oscar',
# bert (534 firings)
'bert',
# indian (345 firings)
' Indians',
# ton (295 firings)
'TON',
# abdullah (281 firings)
' Abdullah',
# alice (269 firings)
' Alice', 'Alice',
# aw (238 firings)
' AW', ' aw',
# dale (218 firings)
' Dale',
# eagle (203 firings)
' Eagle', ' eagle',
# einstein (199 firings)
' Einstein',
# alfr (197 firings)
' Alfred',
# aby (193 firings)
'aby',
# bowie (192 firings)
' Bowie',
# moh (190 firings)
' Moh',
# arlington (189 firings)
' Arlington',
# ralph (183 firings)
' Ralph',
# abraham (141 firings)
' Abraham',
# vet (137 firings)
'Veter',
# irv (136 firings)
' Irving',
# harold (132 firings)
' Harold',
# ustra (126 firings)
'ustration',
# ab (120 firings)
' AB',
# herbert (117 firings)
' Herbert',
# albion (111 firings)
' Albion',
# alberto (108 firings)
' Alberto',
# ر (107 firings)
'ر',
# athletic (105 firings)
' Athletics',
# alph (99 firings)
' Alph', 'alph',
# calvin (94 firings)
' Calvin',
# algeria (90 firings)
' Algeria',
# annie (88 firings)
' Annie',
# woody (82 firings)
' Woody',
# fant (78 firings)
' Fant', ' FANT',
# ali (78 firings)
'Ali', ' ali',
# ira (73 firings)
' IRA',
# berto (71 firings)
'berto',
# elvi (70 firings)
' Elvis',
# urs (62 firings)
' Urs',
# abbrevi (62 firings)
' abbrevi',
# idol (61 firings)
' Idol',
# edmund (60 firings)
' Edmund',
# mist (59 firings)
'mist',
# mathemat (57 firings)
' mathematic',
# algebra (56 firings)
' algebra',
# amer (51 firings)
' americ',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'alberta-family', 1
def prog_L08F08007(input_tokens):
"""9 token families, 22 surface forms (train F1 of the uncapped optimum: 0.154).
Fires at 353,971 of 36,786,566 positions.
Fires at 0.96% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: proper nouns and names related to individuals and organizations"""
# --- end external annotation ---
fire = {
# � (10,491 firings)
'�',
# s (4,611 firings)
's', 'és', 'ís',
# ( (3,191 firings)
' (',
# ) (3,051 firings)
')',
# " (2,470 firings)
'"',
# a (1,327 firings)
'a', 'ña', 'ía', 'ça',
# 2 (1,250 firings)
' 2', '2',
# i (1,181 firings)
'i',
# man (906 firings)
'man', ' Man', 'Man', 'mans', ' Mans', ' MAN', ' Mansion', ' manic',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '�-family', 1
def prog_L08F08008(input_tokens):
"""28 token families, 66 surface forms (train F1 of the uncapped optimum: 0.493).
Fires at 139,004 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: nouns related to specific categories or themes, particularly those indicating classifications or identifiers"""
# --- end external annotation ---
fire = {
# police (3,801 firings)
'Police', ' Police',
# form (2,535 firings)
'Former', ' Former', 'Form',
# people (1,494 firings)
'People', ' People',
# media (1,089 firings)
'Media',
# research (1,084 firings)
'Researchers', ' Researchers', 'Research',
# ear (813 firings)
'Early', ' Early', 'Ear', 'earable',
# scientist (789 firings)
'Scientists', ' Scientists',
# no (684 firings)
'No',
# offic (629 firings)
'Officers', 'Official', ' Officers', ' Official',
# much (603 firings)
'Much', ' Much',
# official (562 firings)
'Officials', ' Officials',
# thousand (557 firings)
'Thousands', ' Thousands',
# head (538 firings)
'Head', ' Head',
# high (500 firings)
'High', 'Higher', ' Highly',
# state (498 firings)
'State', ' Statements',
# hundred (495 firings)
'Hundreds', ' Hundreds',
# author (472 firings)
'Authorities', ' Authorities', ' Authors',
# using (471 firings)
'Using', ' Using',
# loc (464 firings)
'Local', ' Local',
# work (456 firings)
'Work', 'Working', ' Working',
# everyth (454 firings)
' Everything', 'Everything',
# fan (448 firings)
'Fans', ' Fans', 'Fan',
# noth (432 firings)
'Nothing', ' Nothing',
# student (420 firings)
'Students', ' Students', 'Student',
# memb (418 firings)
'Members', ' Members', 'Member',
# recent (385 firings)
'Recent', ' Recent',
# investigator (361 firings)
'Investigators', ' Investigators',
# sourc (356 firings)
'Sources', ' Sources',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'police-family', 1
def prog_L08F08009(input_tokens):
"""7 token families, 19 surface forms (train F1 of the uncapped optimum: 0.549).
Fires at 15,732 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "end" or its variations, indicating conclusions or conclusions of events"""
# --- end external annotation ---
fire = {
# end (8,524 firings)
' end', ' ended', ' ending', ' ends', 'End', ' endings', ' Ending', ' Ends', ' END', ' Ended',
# conclud (429 firings)
' concluded', ' concludes', ' concluding',
# finale (287 firings)
' finale',
# culminat (152 firings)
' culminated', ' culminating',
# clos (99 firings)
' closes',
# expir (93 firings)
' expires',
# terminate (52 firings)
' terminate',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'end-family', 1
def prog_L08F08010(input_tokens):
"""12 token families, 19 surface forms (train F1 of the uncapped optimum: 0.110).
Fires at 873,616 of 36,786,566 positions.
Fires at 2.37% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: key terms related to investigations and legal proceedings"""
# --- end external annotation ---
fire = {
# the (68,431 firings)
' the',
# in (19,829 firings)
' in',
# thi (5,862 firings)
' this',
# his (5,784 firings)
' his',
# he (4,820 firings)
' he', ' HE', 'HE',
# accord (3,517 firings)
' according', ' Accord',
# their (3,347 firings)
' their',
# last (2,768 firings)
' last', ' LAST',
# there (2,610 firings)
' there', ' THERE',
# her (2,550 firings)
' her', ' Hers',
# had (2,513 firings)
' had',
# just (2,363 firings)
' just', ' JUST',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'the-family', 1
def prog_L08F08011(input_tokens):
"""9 token families, 18 surface forms (train F1 of the uncapped optimum: 0.085).
Fires at 163,585 of 36,786,566 positions.
Fires at 0.44% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: sequences of letters and numbers typically associated with identifiers or codes"""
# --- end external annotation ---
fire = {
# : (2,360 firings)
':', ' :',
# ) (1,456 firings)
')', ' )',
# / (1,104 firings)
'/', ' /',
# $ (641 firings)
' $', '$',
# - (640 firings)
' -',
# ! (590 firings)
'!', ' !',
# pic (420 firings)
' pic', 'pic', ' Pic', 'Pic',
# | (417 firings)
' |', '|',
# gmt (411 firings)
' GMT',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ':-family', 1
def prog_L08F08012(input_tokens):
"""48 token families, 51 surface forms (train F1 of the uncapped optimum: 0.668).
Fires at 12,365 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions of prestigious awards, specifically the Nobel and Pulitzer Prizes"""
# --- end external annotation ---
fire = {
# olymp (949 firings)
' Olympic',
# prize (525 firings)
' Prize',
# nobel (464 firings)
' Nobel',
# oscar (360 firings)
' Oscar',
# billionaire (261 firings)
' billionaire',
# osama (220 firings)
' Osama',
# marathon (218 firings)
' Marathon',
# phd (200 firings)
' PhD',
# emmy (169 firings)
' Emmy',
# grammy (164 firings)
' Grammy',
# lottery (142 firings)
' lottery',
# nob (136 firings)
'nob', 'Nob', ' Nob',
# noble (134 firings)
' Noble',
# swat (133 firings)
' SWAT',
# knox (128 firings)
' Knox',
# pulitz (127 firings)
' Pulitzer',
# ted (124 firings)
' TED',
# pharmaceut (124 firings)
' pharmaceutical',
# hugo (123 firings)
' Hugo',
# doctor (122 firings)
' doctoral', 'doctoral',
# wembley (120 firings)
' Wembley',
# sharia (120 firings)
' Sharia',
# europa (119 firings)
' Europa',
# guin (115 firings)
' Guinness',
# tur (104 firings)
' Turing',
# klux (103 firings)
' Klux',
# heisman (102 firings)
' Heisman',
# naacp (99 firings)
' NAACP',
# volkswagen (97 firings)
' Volkswagen',
# smithsonian (89 firings)
' Smithsonian',
# parkinson (86 firings)
' Parkinson',
# oprah (85 firings)
' Oprah',
# honorary (85 firings)
' honorary',
# institut (82 firings)
' Institutes',
# oslo (81 firings)
' Oslo',
# playboy (80 firings)
' Playboy',
# lancet (80 firings)
' Lancet',
# hawk (79 firings)
' Hawking',
# laureate (67 firings)
' laureate',
# brook (63 firings)
' Brookings',
# chomsky (57 firings)
' Chomsky',
# para (55 firings)
' Paraly',
# omsday (53 firings)
'omsday',
# kyoto (52 firings)
' Kyoto',
# macarthur (50 firings)
' MacArthur',
# laure (43 firings)
' laure',
# doomsday (41 firings)
' Doomsday',
# fluorescent (40 firings)
' fluorescent',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'olymp-family', 1
def prog_L08F08013(input_tokens):
"""3 token families, 6 surface forms (train F1 of the uncapped optimum: 0.827).
Fires at 35,448 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "many" and related quantities"""
# --- end external annotation ---
fire = {
# many (25,087 firings)
' many', ' Many', 'Many', 'many',
# numer (686 firings)
' numerous',
# countles (523 firings)
' countless',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'many-family', 1
def prog_L08F08014(input_tokens):
"""19 token families, 45 surface forms (train F1 of the uncapped optimum: 0.212).
Fires at 199,291 of 36,786,566 positions.
Fires at 0.54% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to the band "The Who" and its associated contexts"""
# --- end external annotation ---
fire = {
# s (2,287 firings)
's',
# trump (1,491 firings)
' Trump', 'Trump', ' trump', 'trump', 'TRUMP',
# unit (1,320 firings)
' United', 'United', 'Unit', 'united',
# o (1,216 firings)
'o', 'O',
# company (1,121 firings)
' company', ' Company',
# getty (999 firings)
'Getty', ' Getty',
# ap (952 firings)
' AP', 'AP',
# a (857 firings)
'a',
# on (829 firings)
'on', 'ON',
# is (799 firings)
'is', 'IS', ' IS',
# g (646 firings)
'G',
# m (640 firings)
'M',
# may (631 firings)
' May', 'May', ' Mayer',
# reut (610 firings)
'Reuters', ' Reuters', ' REUTERS',
# time (572 firings)
' Time', ' TIME', 'Time', 'TIME',
# ia (565 firings)
'IA', 'ia',
# more (530 firings)
' MORE',
# who (529 firings)
' Who', ' WHO', 'WHO',
# dead (499 firings)
' Dead', 'Dead', ' DEAD',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 's-family', 1
def prog_L08F08015(input_tokens):
"""20 token families, 22 surface forms (train F1 of the uncapped optimum: 0.339).
Fires at 25,121 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: characters or symbols from a specific non-Latin alphabet script"""
# --- end external annotation ---
fire = {
# .......... (528 firings)
' ..........',
# и (268 firings)
'и',
# а (241 firings)
'а',
# н (218 firings)
'н',
# е (214 firings)
'е',
# т (193 firings)
'т',
# р (143 firings)
'р',
# с (137 firings)
'с',
# к (131 firings)
'к',
# о� (113 firings)
'о�',
# о (112 firings)
'о',
# ي (97 firings)
'ي',
# ‎ (95 firings)
'\u200e',
# л (93 firings)
'л',
# ن (90 firings)
'ن',
# ال (79 firings)
' ال', 'ال',
# ا (70 firings)
'ا',
# в (68 firings)
'в',
# م (67 firings)
'م', ' م',
# ς (65 firings)
'ς',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '..........-family', 1
def prog_L08F08016(input_tokens):
"""5 token families, 7 surface forms (train F1 of the uncapped optimum: 0.891).
Fires at 10,333 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: specific dates, particularly in June"""
# --- end external annotation ---
fire = {
# june (7,314 firings)
' June', 'June',
# jun (778 firings)
' Jun', 'Jun',
# pdt (127 firings)
' PDT',
# une (69 firings)
'UNE',
# ileen (46 firings)
'ileen',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'june-family', 1
def prog_L08F08017(input_tokens):
"""3 token families, 4 surface forms (train F1 of the uncapped optimum: 0.042).
Fires at 38,219 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases that indicate positions or states of being at the forefront of situations or issues"""
# --- end external annotation ---
fire = {
# the (21,517 firings)
' the', ' THE',
# same (480 firings)
' same',
# top (227 firings)
' top',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'the-family', 1
def prog_L08F08018(input_tokens):
"""12 token families, 15 surface forms (train F1 of the uncapped optimum: 0.089).
Fires at 551,255 of 36,786,566 positions.
Fires at 1.50% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to decision-making processes within organizations"""
# --- end external annotation ---
fire = {
# the (34,216 firings)
' the',
# that (6,481 firings)
' that',
# from (4,234 firings)
' from',
# as (3,759 firings)
' as',
# be (3,506 firings)
' be',
# year (3,392 firings)
' year', ' years', 'year', ' yearly',
# his (2,690 firings)
' his',
# been (2,345 firings)
' been',
# its (1,692 firings)
' its',
# people (1,616 firings)
' people',
# when (1,493 firings)
' when',
# some (1,468 firings)
' some',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'the-family', 1
def prog_L08F08019(input_tokens):
"""8 token families, 12 surface forms (train F1 of the uncapped optimum: 0.125).
Fires at 91,263 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: punctuation or special characters that signal the end of statements"""
# --- end external annotation ---
fire = {
# ," (3,970 firings)
',"', ' ,"',
# ? (2,552 firings)
'?',
# — (2,431 firings)
' —',
# by (1,271 firings)
'By', ' By',
# ." (1,221 firings)
'."', ' ."',
# pic (466 firings)
' pic', 'pic',
# mr (399 firings)
' Mr',
# ", (337 firings)
'",',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ',"-family', 1
def prog_L08F08020(input_tokens):
"""7 token families, 10 surface forms (train F1 of the uncapped optimum: 0.147).
Fires at 207,003 of 36,786,566 positions.
Fires at 0.56% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to specific events or entities in popular culture"""
# --- end external annotation ---
fire = {
# ) (2,300 firings)
')',
# ? (1,302 firings)
'?', ' ?',
# ! (968 firings)
'!', ' !',
# 2017 (909 firings)
' 2017', '2017',
# ' (676 firings)
"'",
# 2 (654 firings)
' 2',
# 1 (642 firings)
' 1',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ')-family', 1
def prog_L08F08021(input_tokens):
"""12 token families, 22 surface forms (train F1 of the uncapped optimum: 0.522).
Fires at 19,928 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to significant events and their details"""
# --- end external annotation ---
fire = {
# event (6,749 firings)
' events', ' event', ' Events', ' Event', 'event', 'Events', 'events', ' EVENTS',
# happen (558 firings)
' happening',
# develop (402 firings)
' developments',
# unfold (376 firings)
' unfolded', ' unfold', ' unfolding', ' unfolds',
# tragedy (362 firings)
' tragedy',
# circumstanc (297 firings)
' circumstances',
# proceed (267 firings)
' proceedings',
# incident (258 firings)
' incidents',
# phenomena (136 firings)
' phenomena',
# trauma (72 firings)
' trauma',
# disast (68 firings)
' disasters',
# traged (65 firings)
' tragedies',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'event-family', 1
def prog_L08F08022(input_tokens):
"""30 token families, 59 surface forms (train F1 of the uncapped optimum: 0.330).
Fires at 29,754 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: actions related to reversing or undoing policies"""
# --- end external annotation ---
fire = {
# change (676 firings)
' change',
# halt (547 firings)
' halt', ' halted', ' halting',
# revers (483 firings)
' reversed', ' reversal', ' reversing', ' reversible',
# overturn (459 firings)
' overturn', ' overturned',
# reverse (427 firings)
' reverse', 'reverse', ' Reverse',
# correct (380 firings)
' correct', ' corrected', ' correcting', ' corrective', 'correct',
# shift (285 firings)
' shift', ' shifts',
# repeat (201 firings)
' repeat', ' repeating', ' repeats', ' Repeat', 'Repeat',
# stripp (189 firings)
' stripped', ' stripping',
# correc (164 firings)
'Correction', ' correction', ' Correction',
# accelerat (164 firings)
' accelerated', ' accelerating',
# restor (147 firings)
' restored', ' restoring', ' restores',
# alt (142 firings)
' alter',
# undo (127 firings)
' undo',
# alter (125 firings)
' altering', ' altered',
# reconsid (123 firings)
' reconsider',
# revok (122 firings)
' revoked',
# accelerate (116 firings)
' accelerate',
# turnaround (108 firings)
' turnaround',
# retract (104 firings)
' retracted', ' retract',
# rescind (103 firings)
' rescind',
# cancel (103 firings)
' cancel',
# flipp (101 firings)
' flipped',
# repeal (94 firings)
' repealed', ' repealing',
# affirm (89 firings)
' affirmed', ' affirm',
# reset (85 firings)
' reset',
# revert (85 firings)
' revert', ' reverted',
# erase (82 firings)
' erase',
# turn (79 firings)
'turn',
# offset (78 firings)
' offset', ' offsets',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'change-family', 1
def prog_L08F08023(input_tokens):
"""2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.156).
Fires at 167,101 of 36,786,566 positions.
Fires at 0.45% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: sentences that express uncertainty or raise questions"""
# --- end external annotation ---
fire = {
# . (97,093 firings)
'.',
# ? (7,715 firings)
'?',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '.-family', 1
def prog_L08F08024(input_tokens):
"""20 token families, 45 surface forms (train F1 of the uncapped optimum: 0.095).
Fires at 541,152 of 36,786,566 positions.
Fires at 1.47% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: official titles or entities within a given context"""
# --- end external annotation ---
fire = {
# , (40,329 firings)
',',
# a (24,402 firings)
' a', ' A',
# that (11,698 firings)
' that',
# the (10,814 firings)
' The', 'The',
# but (6,422 firings)
' but', ' But', 'But',
# i (4,700 firings)
' I',
# who (3,580 firings)
' who', 'who', 'Who',
# and (3,323 firings)
' And', 'and', 'And', 'anding', 'anders',
# which (3,259 firings)
' which', 'which', ' Which', 'Which',
# ? (3,251 firings)
'?',
# new (2,704 firings)
' new', ' newer',
# about (2,673 firings)
' about',
# there (2,621 firings)
' there', ' There',
# thi (2,080 firings)
' This', 'This',
# people (2,072 firings)
' people', ' People', 'People',
# like (1,893 firings)
' like', 'like',
# my (1,831 firings)
' my', ' My', 'My',
# because (1,639 firings)
' because', ' Because', 'Because', 'because',
# if (1,506 firings)
'If', ' If',
# it (1,467 firings)
' It',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ',-family', 1
def prog_L08F08025(input_tokens):
"""45 token families, 51 surface forms (train F1 of the uncapped optimum: 0.692).
Fires at 10,299 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to the name "Baba." """
# --- end external annotation ---
fire = {
# bab (766 firings)
' Bab', ' babies', ' bab',
# madd (232 firings)
' Madd',
# hos (228 firings)
' Hos',
# bah (226 firings)
' Bah', 'Bah',
# gab (225 firings)
' Gab', 'Gab',
# barn (174 firings)
' Barn',
# man (171 firings)
' Mans',
# rab (170 firings)
' Rab', ' rab',
# abram (147 firings)
' Abrams',
# wax (146 firings)
' wax',
# mum (132 firings)
' mum', ' Mum',
# mam (131 firings)
' Mam',
# buzz (131 firings)
' Buzz',
# incub (124 firings)
' incub',
# reb (114 firings)
'Reb',
# hod (107 firings)
' Hod',
# hom (103 firings)
' Homs',
# infant (102 firings)
' infants',
# bod (101 firings)
' Bod',
# kab (99 firings)
' Kab',
# kad (98 firings)
' Kad',
# pediatr (96 firings)
' pediatric',
# pedd (91 firings)
' pedd',
# dram (88 firings)
' dram',
# nab (87 firings)
' Nab',
# theat (86 firings)
' theat',
# lad (82 firings)
' lad',
# cyru (81 firings)
' Cyrus',
# preschool (78 firings)
' preschool',
# dup (75 firings)
' dup',
# nursery (73 firings)
' nursery',
# babylon (70 firings)
' Babylon',
# boll (69 firings)
' Boll',
# gob (69 firings)
' gob',
# tut (67 firings)
' Tut',
# condition (67 firings)
' conditional',
# glimp (64 firings)
' glimps',
# drib (60 firings)
' drib',
# dispen (59 firings)
' dispens',
# lull (56 firings)
' lull',
# pix (55 firings)
' Pix',
# bas (51 firings)
' BAS',
# dab (49 firings)
' dab',
# dazz (47 firings)
' dazz',
# dwind (45 firings)
' dwind',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'bab-family', 1
def prog_L08F08026(input_tokens):
"""13 token families, 16 surface forms (train F1 of the uncapped optimum: 0.051).
Fires at 123,689 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to authority figures and their actions or characteristics"""
# --- end external annotation ---
fire = {
# who (2,620 firings)
' who', 'who',
# have (2,544 firings)
' have',
# they (1,618 firings)
' they',
# we (1,439 firings)
' we',
# some (1,136 firings)
' some',
# how (856 firings)
' how',
# there (825 firings)
' there', ' THERE',
# make (627 firings)
' make',
# because (523 firings)
' because',
# any (510 firings)
' any',
# those (442 firings)
' those',
# why (314 firings)
' why', ' Why',
# oth (310 firings)
' others',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'who-family', 1
def prog_L08F08027(input_tokens):
"""2 token families, 3 surface forms (train F1 of the uncapped optimum: 0.087).
Fires at 69,654 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "email" in various contexts"""
# --- end external annotation ---
fire = {
# , (58,239 firings)
',', ' ,',
# ," (2,473 firings)
',"',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ',-family', 1
def prog_L08F08028(input_tokens):
"""17 token families, 38 surface forms (train F1 of the uncapped optimum: 0.092).
Fires at 481,028 of 36,786,566 positions.
Fires at 1.31% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: incidents of violence or severe crime"""
# --- end external annotation ---
fire = {
# a (17,457 firings)
' a',
# is (10,364 firings)
' is', ' Is',
# it (10,023 firings)
' it', ' It',
# i (6,782 firings)
' I', ' i',
# but (6,323 firings)
' but', ' But',
# the (5,421 firings)
' The',
# you (4,965 firings)
' you', ' You', 'you',
# we (3,602 firings)
' we', ' We',
# one (3,306 firings)
' one', ' One',
# thi (2,172 firings)
' This',
# some (2,112 firings)
' some', ' Some',
# say (2,086 firings)
' says', ' say', ' saying', ' Says', ' Saying',
# there (1,779 firings)
' there', ' There', ' THERE',
# subscribe (1,684 firings)
' Subscribe', 'Subscribe',
# m (1,329 firings)
'm', "'m",
# see (1,325 firings)
' see', 'see', ' Seeing',
# that (1,258 firings)
' That', 'That', ' thats',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'a-family', 1
def prog_L08F08029(input_tokens):
"""7 token families, 12 surface forms (train F1 of the uncapped optimum: 0.096).
Fires at 202,961 of 36,786,566 positions.
Fires at 0.55% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: keywords associated with operations or initiatives, particularly those labeled with specific names or codes"""
# --- end external annotation ---
fire = {
# : (4,159 firings)
':', ' :',
# ! (986 firings)
'!', ' !',
# against (757 firings)
' against', ' Against',
# [ (711 firings)
' [',
# 11 (671 firings)
'11',
# ] (651 firings)
']', ' ]',
# between (512 firings)
' between', ' Between',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ':-family', 1
def prog_L08F08030(input_tokens):
"""28 token families, 51 surface forms (train F1 of the uncapped optimum: 0.305).
Fires at 49,450 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to causes of incidents or issues"""
# --- end external annotation ---
fire = {
# reason (2,141 firings)
' reason', ' reasons', ' reasoning', 'Reason',
# respons (1,193 firings)
' responsible', ' responsibility',
# cause (1,101 firings)
' cause', ' Cause', 'cause', 'Cause',
# source (935 firings)
' source',
# origin (704 firings)
' origin', ' origins', 'Original', 'Origin', ' Origins',
# involv (524 firings)
' involved',
# caus (512 firings)
' causes', ' causal', ' Causes',
# victim (469 firings)
' victims',
# factor (382 firings)
' factors', ' factor', ' Factors',
# explana (378 firings)
' explanation',
# ident (340 firings)
' identity', ' identities',
# mot (277 firings)
' motive', ' motives',
# blame (259 firings)
' blame',
# purpose (257 firings)
' purpose',
# root (256 firings)
' root', ' roots',
# perpetrator (231 firings)
' perpetrators', ' perpetrator',
# mechanism (224 firings)
' mechanisms', ' mechanism',
# circumstanc (193 firings)
' circumstances',
# cure (181 firings)
' cure',
# culprit (179 firings)
' culprit',
# winn (179 firings)
' winners', ' Winners',
# inspira (176 firings)
' inspiration',
# kill (167 firings)
' killer', ' killers',
# tim (160 firings)
' timing',
# underly (146 firings)
' underlying',
# symptom (146 firings)
' symptoms', ' Symptoms',
# involve (144 firings)
' involvement',
# motiva (141 firings)
' motivation',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'reason-family', 1
def prog_L08F08031(input_tokens):
"""44 token families, 58 surface forms (train F1 of the uncapped optimum: 0.540).
Fires at 41,309 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: words related to upholstery or fabric"""
# --- end external annotation ---
fire = {
# ol (553 firings)
'OL',
# lo (437 firings)
'lo', 'LO',
# hol (420 firings)
'hol', ' hol', ' HOL', 'Hol',
# og (325 firings)
'OG',
# angel (317 firings)
' ANGEL',
# col (306 firings)
'COL', 'col', ' COL',
# ament (290 firings)
'ament', 'aments', 'amental',
# holocaust (274 firings)
' Holocaust',
# bur (273 firings)
'BUR',
# feld (255 firings)
'feld',
# rah (202 firings)
'rah',
# sidelin (194 firings)
' sidelines',
# olog (193 firings)
'olog', 'OLOG',
# ital (185 firings)
'ital',
# è (182 firings)
'è',
# bach (172 firings)
'bach',
# cycl (170 firings)
'cycl', 'cycles',
# outlaw (167 firings)
' outlaw',
# oga (165 firings)
'oga',
# oria (158 firings)
'oria',
# bra (152 firings)
'bra',
# fil (152 firings)
'fil', ' FIL', 'FIL',
# geneva (149 firings)
' Geneva',
# review (149 firings)
' Reviews', 'Reviewer',
# hoffman (148 firings)
' Hoffman',
# hole (146 firings)
'hole', ' Hole',
# frankfurt (145 firings)
' Frankfurt',
# park (143 firings)
' PARK',
# ollah (137 firings)
'ollah',
# guantanamo (136 firings)
' Guantanamo',
# theft (135 firings)
' Theft',
# wind (134 firings)
' winding',
# wich (133 firings)
'wich',
# hau (130 firings)
'haus',
# brom (125 firings)
' Brom',
# ov (121 firings)
'OV',
# ô (120 firings)
'ô',
# oland (120 firings)
'oland',
# duc (118 firings)
' Duc',
# rab (115 firings)
'rab',
# ual (112 firings)
'UAL',
# iba (112 firings)
'iba',
# olf (112 firings)
'olf',
# othe (104 firings)
'othe',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ol-family', 1
def prog_L08F08032(input_tokens):
"""1 token families, 4 surface forms (train F1 of the uncapped optimum: 0.504).
Fires at 44,532 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: repeated instances of the word "this" in various contexts"""
# --- end external annotation ---
fire = {
# thi (35,848 firings)
' this', 'this', ' THIS', 'THIS',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'thi-family', 1
def prog_L08F08033(input_tokens):
"""30 token families, 67 surface forms (train F1 of the uncapped optimum: 0.264).
Fires at 89,209 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: verbs related to sales or transactions"""
# --- end external annotation ---
fire = {
# spread (935 firings)
' spread', ' spreading', ' spreads', 'spread',
# pass (918 firings)
' passed', ' passes',
# launch (770 firings)
' launched', ' launches', ' launching',
# sold (671 firings)
' sold',
# sell (661 firings)
' sell', ' selling', ' sells', ' Selling',
# air (631 firings)
' aired', ' airing', ' airs',
# open (624 firings)
' opened', ' opens',
# continue (517 firings)
' continue', 'continue',
# work (514 firings)
' works',
# app (510 firings)
' apply', ' Apply', 'apply',
# operat (496 firings)
' operating', ' operates', ' operated',
# count (487 firings)
' count', ' counts', ' counted', ' counting',
# run (457 firings)
' runs', ' Runs',
# appl (437 firings)
' applies', ' applied',
# rank (435 firings)
' ranks', ' ranked', ' rank',
# carr (412 firings)
' carries', ' carried',
# fit (410 firings)
' fit', ' fits',
# pas (368 firings)
' pass',
# extend (361 firings)
' extend', ' extends', ' extending',
# weigh (360 firings)
' weighs', ' weighing', ' weigh', ' weighed',
# tend (324 firings)
' tend', ' tends', ' tended', ' tending',
# ref (322 firings)
' refers', ' refer',
# carry (311 firings)
' carry',
# play (282 firings)
' plays',
# operate (281 firings)
' operate',
# dat (281 firings)
' dates', ' dating',
# debut (265 firings)
' debuted',
# leak (250 firings)
' leaked', ' leaking',
# hold (234 firings)
' holds',
# redirect (229 firings)
' redirect',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'spread-family', 1
def prog_L08F08034(input_tokens):
"""19 token families, 23 surface forms (train F1 of the uncapped optimum: 0.144).
Fires at 58,216 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: specific URLs and web addresses"""
# --- end external annotation ---
fire = {
# ig (502 firings)
'ig',
# ad (492 firings)
'ad',
# am (484 firings)
'am', 'AM',
# amb (399 firings)
'amb',
# im (299 firings)
'im',
# ah (258 firings)
'ah',
# em (256 firings)
'em',
# ip (253 firings)
'ip',
# ot (243 firings)
'ot',
# ib (231 firings)
'ib', 'IB',
# od (227 firings)
'od', 'OD',
# ok (226 firings)
'ok', 'OK',
# om (209 firings)
'om',
# ag (195 firings)
'ag',
# ap (188 firings)
'ap',
# ow (184 firings)
'ow',
# ob (181 firings)
'ob',
# ep (162 firings)
'ep',
# iv (161 firings)
'iv',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ig-family', 1
def prog_L08F08035(input_tokens):
"""11 token families, 13 surface forms (train F1 of the uncapped optimum: 0.557).
Fires at 9,462 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to the letter 'K'"""
# --- end external annotation ---
fire = {
# k (4,680 firings)
'k', ' k',
# kg (238 firings)
' kg', 'kg',
# ke (169 firings)
' ke',
# ko (82 firings)
' ko',
# kosh (62 firings)
' kosher',
# kh (46 firings)
' kh',
# kung (42 firings)
' Kung',
# hk (33 firings)
' HK',
# kettle (33 firings)
' kettle',
# dh (27 firings)
' dh',
# ka (26 firings)
' ka',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'k-family', 1
def prog_L08F08036(input_tokens):
"""19 token families, 24 surface forms (train F1 of the uncapped optimum: 0.086).
Fires at 160,199 of 36,786,566 positions.
Fires at 0.44% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of processes related to approval and review"""
# --- end external annotation ---
fire = {
# in (18,152 firings)
' in',
# at (5,875 firings)
' at',
# dur (2,259 firings)
' during', 'during',
# aft (1,972 firings)
' after',
# when (1,741 firings)
' when',
# if (1,299 firings)
' if',
# before (971 firings)
' before',
# through (937 firings)
' through',
# until (804 firings)
' until', 'until',
# using (788 firings)
' using',
# next (726 firings)
' next',
# und (701 firings)
' under',
# via (677 firings)
' via',
# ago (641 firings)
' ago',
# lat (617 firings)
' later',
# 2017 (582 firings)
' 2017', '2017',
# without (581 firings)
' without', 'without',
# within (558 firings)
' within', 'within',
# follow (496 firings)
' following',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'in-family', 1
def prog_L08F08037(input_tokens):
"""19 token families, 27 surface forms (train F1 of the uncapped optimum: 0.393).
Fires at 12,804 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "together" and related phrases emphasizing unity or collaboration"""
# --- end external annotation ---
fire = {
# togeth (2,248 firings)
' together', ' Together', 'Together', 'together',
# combin (535 firings)
' combined', ' Combined',
# conjunc (235 firings)
' conjunction',
# collective (220 firings)
' collectively',
# confus (192 firings)
' confused',
# joint (173 firings)
' jointly',
# soon (169 firings)
'Soon', ' Soon',
# collabora (148 firings)
' collaboration',
# organ (124 firings)
'Organ',
# iti (106 firings)
'iti',
# coupl (99 firings)
' coupled',
# apart (94 firings)
' Apart', 'Apart',
# comb (86 firings)
'Comb', ' Comb',
# unit (73 firings)
' united',
# independent (56 firings)
' independently',
# group (47 firings)
'Group', ' grouped',
# friend (41 firings)
'Friends',
# individual (41 firings)
' individually',
# interconnect (40 firings)
' interconnected',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'togeth-family', 1
def prog_L08F08038(input_tokens):
"""49 token families, 57 surface forms (train F1 of the uncapped optimum: 0.719).
Fires at 18,242 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions of the Jets, a sports team"""
# --- end external annotation ---
fire = {
# jet (826 firings)
' Jets', ' Jet', 'Jet',
# cub (438 firings)
' Cubs',
# winnipeg (403 firings)
' Winnipeg',
# knick (351 firings)
' Knicks',
# sanchez (343 firings)
' Sanchez',
# met (283 firings)
' Mets',
# kushn (266 firings)
' Kushner',
# nj (237 firings)
' NJ',
# wis (222 firings)
' Wis',
# friedman (221 firings)
' Friedman',
# tsa (212 firings)
' TSA',
# rutg (205 firings)
' Rutgers',
# manitoba (201 firings)
' Manitoba',
# net (197 firings)
' Nets',
# futur (188 firings)
' futures',
# pluto (185 firings)
' Pluto',
# fut (174 firings)
' Fut',
# trent (171 firings)
' Trent',
# simpson (163 firings)
' Simpsons',
# morph (162 firings)
' morph', ' morphed', ' Morph', 'morph',
# matti (156 firings)
' Mattis',
# weis (154 firings)
' Weiss', 'Weiss',
# onz (141 firings)
'onz',
# schwartz (137 firings)
' Schwartz',
# tur (132 firings)
' Turing',
# cap (129 firings)
' Caps',
# bolton (129 firings)
' Bolton',
# hobbit (126 firings)
' Hobbit',
# shapiro (124 firings)
' Shapiro',
# went (124 firings)
' Went',
# toma (121 firings)
' Tomas',
# tweak (120 firings)
' tweak', ' tweaked',
# kirst (117 firings)
' Kirst',
# middl (109 firings)
' Middles',
# sik (106 firings)
' Sik',
# tina (105 firings)
' Tina',
# pet (105 firings)
' Peters',
# jit (95 firings)
'jit',
# itect (92 firings)
'itect',
# tick (92 firings)
' ticking', ' Tick',
# horizon (90 firings)
' Horizons',
# buc (82 firings)
' Bucs',
# ket (81 firings)
' ket',
# enezuel (78 firings)
'enezuel',
# hendrick (73 firings)
' Hendricks',
# kom (72 firings)
' Kom',
# neocon (70 firings)
' neocons',
# fitzpatrick (64 firings)
' Fitzpatrick',
# wig (63 firings)
' wig',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'jet-family', 1
def prog_L08F08039(input_tokens):
"""14 token families, 16 surface forms (train F1 of the uncapped optimum: 0.753).
Fires at 14,961 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions of a specific agency"""
# --- end external annotation ---
fire = {
# agency (6,200 firings)
' agency', ' Agency', 'agency',
# agenc (1,559 firings)
' agencies',
# administra (884 firings)
' Administration',
# corpora (324 firings)
' Corporation',
# airline (278 firings)
' airline',
# contractor (195 firings)
' contractor',
# aley (139 firings)
'aley',
# bureau (117 firings)
' bureau',
# directorate (91 firings)
' Directorate',
# metro (91 firings)
'Metro',
# rag (72 firings)
' rag',
# gency (68 firings)
'gency',
# enc (54 firings)
'encies',
# bureaucracy (47 firings)
' bureaucracy',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'agency-family', 1
def prog_L08F08040(input_tokens):
"""10 token families, 11 surface forms (train F1 of the uncapped optimum: 0.139).
Fires at 93,017 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: numerical data related to age ranges and statistics"""
# --- end external annotation ---
fire = {
# th (2,870 firings)
'th',
# year (2,363 firings)
'year',
# percent (649 firings)
' percent',
# point (535 firings)
' points', 'point',
# 00 (531 firings)
'00',
# block (445 firings)
' block',
# st (387 firings)
'st',
# minut (356 firings)
' minutes',
# nd (327 firings)
'nd',
# rd (317 firings)
'rd',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'th-family', 1
def prog_L08F08041(input_tokens):
"""35 token families, 46 surface forms (train F1 of the uncapped optimum: 0.574).
Fires at 12,724 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to various types of pathways or paths"""
# --- end external annotation ---
fire = {
# path (2,271 firings)
' path', ' paths', ' Path', 'path', 'Path', ' PATH', 'pathic', 'PATH',
# road (1,071 firings)
' road', 'Road',
# route (552 firings)
' route',
# pathway (206 firings)
' pathway', ' pathways',
# progres (172 firings)
' progression',
# trajectory (152 firings)
' trajectory',
# avenu (124 firings)
' avenues',
# hallway (114 firings)
' hallway',
# propag (104 firings)
' propag',
# corridor (104 firings)
' Corridor', ' corridor',
# prot (89 firings)
' prot',
# footstep (87 firings)
' footsteps',
# journey (86 firings)
' Journey',
# labyrinth (85 firings)
' labyrinth', ' Labyrinth',
# yll (82 firings)
'yll',
# maze (77 firings)
' maze',
# nem (75 firings)
' nem',
# psychopath (71 firings)
' psychopath',
# magnet (70 firings)
' magnet',
# christina (67 firings)
' Christina',
# roadmap (65 firings)
' roadmap',
# avenue (62 firings)
' avenue',
# nes (59 firings)
' Ness',
# tal (54 firings)
' Tales',
# fern (52 firings)
' Fern',
# doorway (50 firings)
' doorway',
# pathogen (48 firings)
' pathogens',
# pathfind (47 firings)
' Pathfinder',
# lineage (46 firings)
' lineage',
# morph (43 firings)
' morph',
# courage (42 firings)
' Courage',
# opath (39 firings)
'opath',
# herb (39 firings)
' herbal',
# pathology (38 firings)
' pathology',
# prism (32 firings)
' prism',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'path-family', 1
def prog_L08F08042(input_tokens):
"""15 token families, 19 surface forms (train F1 of the uncapped optimum: 0.086).
Fires at 354,368 of 36,786,566 positions.
Fires at 0.96% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: terms related to communication methods and professional correspondence"""
# --- end external annotation ---
fire = {
# the (37,798 firings)
' the',
# for (7,992 firings)
' for',
# with (5,681 firings)
' with',
# an (4,828 firings)
' an',
# about (2,393 firings)
' about',
# first (1,802 firings)
' first',
# accord (1,280 firings)
' according', ' accordance',
# what (1,193 firings)
' what',
# part (1,029 firings)
' part',
# help (991 firings)
' help', ' helping', ' helpful',
# how (991 firings)
' how',
# where (981 firings)
' where',
# through (948 firings)
' through',
# way (946 firings)
' way', ' ways',
# see (896 firings)
' see',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'the-family', 1
def prog_L08F08043(input_tokens):
"""7 token families, 10 surface forms (train F1 of the uncapped optimum: 0.068).
Fires at 187,920 of 36,786,566 positions.
Fires at 0.51% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to first experiences or encounters"""
# --- end external annotation ---
fire = {
# in (16,846 firings)
' in',
# on (9,158 firings)
' on',
# since (1,332 firings)
' since',
# against (1,139 firings)
' against',
# between (625 firings)
' between',
# see (531 firings)
' see', ' SEE',
# appear (509 firings)
' appearance', ' appeared', ' Appearance',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'in-family', 1
def prog_L08F08044(input_tokens):
"""4 token families, 7 surface forms (train F1 of the uncapped optimum: 0.090).
Fires at 130,327 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: monetary pledges and financial commitments"""
# --- end external annotation ---
fire = {
# : (5,529 firings)
':',
# $ (2,588 firings)
' $', '$',
# ic (766 firings)
'ic', 'IC',
# £ (410 firings)
' £', '£',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ':-family', 1
def prog_L08F08045(input_tokens):
"""44 token families, 47 surface forms (train F1 of the uncapped optimum: 0.489).
Fires at 13,854 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to the concept of "momentum" in various contexts"""
# --- end external annotation ---
fire = {
# momentum (461 firings)
' momentum',
# stall (252 firings)
' stalled', ' stall',
# senti (223 firings)
' sentiment',
# pois (204 firings)
' poised',
# stand (179 firings)
' standings',
# trac (161 firings)
' traction',
# trajectory (149 firings)
' trajectory',
# steam (142 firings)
' steam',
# fortun (138 firings)
' fortunes',
# hype (133 firings)
' hype',
# dynamic (119 firings)
' dynamics',
# propell (118 firings)
' propelled', ' propell',
# euph (117 firings)
' euph', ' Euph',
# ernaut (104 firings)
'ernaut',
# enthusiasm (99 firings)
' enthusiasm',
# motiva (95 firings)
' motivation',
# moment (92 firings)
' Moment',
# suspense (85 firings)
' suspense',
# buoy (83 firings)
' buoy',
# ling (80 firings)
' linger',
# unstopp (78 firings)
' unstoppable',
# urgency (75 firings)
' urgency',
# bandwagon (72 firings)
' bandwagon',
# optimism (72 firings)
' optimism',
# propel (68 firings)
' propel',
# sedi (65 firings)
' sediment',
# surg (63 firings)
' surging',
# emph (62 firings)
' emph',
# catalyst (62 firings)
' catalyst',
# foothold (61 firings)
' foothold',
# discord (61 firings)
' discord',
# suitcase (58 firings)
' suitcase',
# storylin (58 firings)
' storylines',
# strid (58 firings)
' strides',
# mindset (57 firings)
' mindset',
# impetu (52 firings)
' impetus',
# decisive (52 firings)
' decisively',
# galvan (51 firings)
' galvan',
# morale (51 firings)
' morale',
# heartbeat (49 firings)
' heartbeat',
# cohe (45 firings)
' cohesion',
# mongo (45 firings)
' Mongo',
# snowball (41 firings)
' snowball',
# hip (41 firings)
' hips',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'momentum-family', 1
def prog_L08F08046(input_tokens):
"""10 token families, 16 surface forms (train F1 of the uncapped optimum: 0.080).
Fires at 135,436 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: statements expressing opinions or observations about situations"""
# --- end external annotation ---
fire = {
# copyright (1,861 firings)
' copyright',
# going (974 firings)
' going', 'going',
# think (884 firings)
' think',
# very (859 firings)
' very', 'very', 'Very',
# lot (658 firings)
' lot',
# real (654 firings)
' really', 'real', 'really',
# don (609 firings)
' don',
# thing (560 firings)
' things',
# alway (450 firings)
' always', 'always',
# re (339 firings)
"'re",
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'copyright-family', 1
def prog_L08F08047(input_tokens):
"""1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.121).
Fires at 56,765 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases that include the conjunction "and." """
# --- end external annotation ---
fire = {
# and (38,671 firings)
' and',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'and-family', 1
def prog_L08F08048(input_tokens):
"""28 token families, 39 surface forms (train F1 of the uncapped optimum: 0.460).
Fires at 31,442 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: adverbs, particularly those ending in 'ly'"""
# --- end external annotation ---
fire = {
# ly (5,162 firings)
'ly', 'LY',
# ally (759 firings)
'ally', 'ALLY',
# ably (641 firings)
'ably',
# ate (403 firings)
'ately',
# ical (388 firings)
'ically',
# ily (325 firings)
'ily', 'ILY',
# ful (284 firings)
'fully', ' Fully',
# ent (268 firings)
'ently',
# ial (234 firings)
'ially',
# ive (201 firings)
'ively',
# ear (183 firings)
'earing', 'early',
# ual (168 firings)
'ually',
# body (161 firings)
'body',
# ative (156 firings)
'atively',
# ant (147 firings)
'antly',
# strict (144 firings)
' strictly',
# mediate (140 firings)
'mediate', 'mediately',
# inal (139 firings)
'inally', 'INAL',
# ential (138 firings)
'entially',
# general (117 firings)
'Generally', ' Generally',
# clear (111 firings)
'Clearly', ' Clearly',
# ari (104 firings)
'arily',
# omew (102 firings)
'omew',
# ere (94 firings)
'ERE',
# most (90 firings)
' Mostly', 'mostly',
# otal (88 firings)
'otally',
# zen (82 firings)
'zens',
# personal (79 firings)
'Personally', ' Personally',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ly-family', 1
def prog_L08F08049(input_tokens):
"""13 token families, 26 surface forms (train F1 of the uncapped optimum: 0.214).
Fires at 709,419 of 36,786,566 positions.
Fires at 1.93% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: topics related to immigration and legal issues surrounding it"""
# --- end external annotation ---
fire = {
# , (23,760 firings)
',',
# year (6,775 firings)
' year', ' years', 'years',
# city (2,680 firings)
' city', ' City',
# trump (2,673 firings)
' Trump',
# world (2,551 firings)
' world', ' World',
# state (2,486 firings)
' state', ' State', 'state',
# people (2,347 firings)
' people',
# elec (2,340 firings)
' election', 'election', ' Election',
# " (2,234 firings)
'"',
# country (2,184 firings)
' country',
# month (2,167 firings)
' month', ' months', ' Month',
# stat (2,124 firings)
' States', ' states', 'states',
# season (2,077 firings)
' season', ' seasons',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ',-family', 1
def prog_L08F08050(input_tokens):
"""10 token families, 19 surface forms (train F1 of the uncapped optimum: 0.090).
Fires at 253,687 of 36,786,566 positions.
Fires at 0.69% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to tables or structured data within a document"""
# --- end external annotation ---
fire = {
# � (16,102 firings)
'�',
# : (3,078 firings)
':', ' :',
# ) (1,878 firings)
')',
# b (799 firings)
' B', 'b', ' b',
# f (780 firings)
' F', 'f', ' f',
# down (771 firings)
' down', 'down', ' Down', ' DOWN', 'downs',
# ing (697 firings)
'ing',
# 2 (694 firings)
' 2',
# 1 (667 firings)
' 1',
# 4 (652 firings)
' 4',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '�-family', 1
def prog_L08F08051(input_tokens):
"""4 token families, 6 surface forms (train F1 of the uncapped optimum: 0.076).
Fires at 20,665 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: punctuation marks and dialogue indicators in conversational text"""
# --- end external annotation ---
fire = {
# : (5,316 firings)
':', ' :',
# > (227 firings)
'>', ' >',
# ): (92 firings)
'):',
# * (86 firings)
'*',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return ':-family', 1
def prog_L08F08052(input_tokens):
"""9 token families, 12 surface forms (train F1 of the uncapped optimum: 0.742).
Fires at 4,473 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions and references to Mac computers"""
# --- end external annotation ---
fire = {
# mac (2,337 firings)
' Mac', 'Mac', ' mac', 'mac',
# macbook (209 firings)
' MacBook',
# macron (195 firings)
' Macron',
# mex (95 firings)
'Mex',
# maco (70 firings)
' macOS',
# mcmast (70 firings)
' McMaster',
# macintosh (61 firings)
' Macintosh',
# oad (59 firings)
'oad',
# mech (31 firings)
' Mech',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'mac-family', 1
def prog_L08F08053(input_tokens):
"""38 token families, 53 surface forms (train F1 of the uncapped optimum: 0.685).
Fires at 9,851 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: occurrences of the substring "tra" across various contexts"""
# --- end external annotation ---
fire = {
# tra (1,119 firings)
' Tra', 'tra', ' tra', 'Tra', ' TRA',
# des (584 firings)
' Des',
# trait (377 firings)
' traits', ' trait',
# dom (321 firings)
' Dom', 'Dom',
# trail (274 firings)
' Trail', ' Trails',
# traff (239 firings)
' Traffic', ' traff', ' Traff',
# rus (202 firings)
' Russ', 'Russ',
# tre (199 firings)
'Tre', 'tre',
# derrick (152 firings)
' Derrick',
# wra (122 firings)
' wra', ' Wra', 'wra', 'Wra',
# dra (118 firings)
' Dra',
# pra (117 firings)
' pra', ' Pra',
# hoffman (116 firings)
' Hoffman',
# django (116 firings)
' Django',
# trauma (104 firings)
' trauma',
# taj (101 firings)
' Taj',
# sophie (88 firings)
' Sophie',
# traumat (85 firings)
' traumatic',
# theo (84 firings)
' Theo',
# sus (82 firings)
'Sus',
# skywalk (78 firings)
' Skywalker',
# dri (73 firings)
' Dri',
# ty (68 firings)
'Ty',
# trav (63 firings)
' travers',
# russell (60 firings)
'Russell',
# tw (52 firings)
'TW',
# traitor (51 firings)
' traitor',
# incent (49 firings)
'incent',
# behind (49 firings)
'Behind',
# koen (47 firings)
' Koen',
# trace (43 firings)
' Trace',
# traverse (43 firings)
' traverse',
# omorph (41 firings)
'omorph',
# rudolph (40 firings)
' Rudolph',
# rosso (38 firings)
'rosso',
# veterinary (34 firings)
' Veterinary',
# の (32 firings)
'の',
# benef (28 firings)
'Benef',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'tra-family', 1
def prog_L08F08054(input_tokens):
"""1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.096).
Fires at 41,227 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: the phrase "of the" used in various contexts"""
# --- end external annotation ---
fire = {
# of (36,630 firings)
' of',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'of-family', 1
def prog_L08F08055(input_tokens):
"""49 token families, 77 surface forms (train F1 of the uncapped optimum: 0.457).
Fires at 157,033 of 36,786,566 positions.
Fires at 0.43% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: titles or roles of notable individuals, particularly those involved in activism or professional achievements"""
# --- end external annotation ---
fire = {
# man (3,919 firings)
' man',
# candidate (2,368 firings)
' candidate', ' Candidate',
# ist (2,191 firings)
'ist',
# minist (2,021 firings)
' Minister', ' minister',
# offic (2,009 firings)
' officer', ' Officer',
# woman (1,844 firings)
' woman', 'woman',
# lead (1,819 firings)
' leader', 'leader',
# star (1,671 firings)
' star',
# student (1,617 firings)
' student',
# play (1,419 firings)
' player', ' Player',
# actor (1,384 firings)
' actor', 'Actor', ' Actor',
# professor (1,291 firings)
' professor', ' Professor', 'Professor',
# writ (1,235 firings)
' writer', 'writer',
# senator (1,192 firings)
' senator', ' Senator',
# activist (1,032 firings)
' activist',
# friend (1,030 firings)
' friend', 'friend',
# artist (989 firings)
' artist', ' Artist', 'artist',
# journalist (925 firings)
' journalist',
# champion (922 firings)
' champion', ' Champion',
# host (892 firings)
' host',
# own (851 firings)
' owner', 'owner',
# ceo (787 firings)
' CEO',
# mp (779 firings)
' MP',
# report (748 firings)
' reporter',
# rep (737 firings)
' Rep',
# midfield (688 firings)
' midfielder',
# resident (648 firings)
' resident',
# lawy (636 firings)
' lawyer',
# teach (636 firings)
' teacher', ' Teacher',
# quarterback (631 firings)
' quarterback',
# produc (608 firings)
' producer', ' Producer',
# mayor (582 firings)
' Mayor',
# veteran (565 firings)
' veteran', ' Veteran',
# actres (564 firings)
' actress', ' Actress',
# ologist (545 firings)
'ologist',
# strik (544 firings)
' striker',
# scientist (542 firings)
' scientist', ' Scientist',
# guard (540 firings)
' guard',
# analyst (521 firings)
' analyst', ' Analyst',
# advis (515 firings)
' adviser', ' Adviser',
# defend (507 firings)
' defender',
# expert (507 firings)
' expert', ' Expert',
# engine (490 firings)
' engineer', ' Engineer',
# soldi (480 firings)
' soldier', ' Soldier',
# legend (458 firings)
' legend', ' legends',
# politician (453 firings)
' politician',
# comedian (446 firings)
' comedian',
# photograph (419 firings)
' photographer',
# businessman (401 firings)
' businessman',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'man-family', 1
def prog_L08F08056(input_tokens):
"""12 token families, 14 surface forms (train F1 of the uncapped optimum: 0.097).
Fires at 489,238 of 36,786,566 positions.
Fires at 1.33% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to events or actions associated with criminal behavior and investigations"""
# --- end external annotation ---
fire = {
# a (22,726 firings)
' a',
# in (16,866 firings)
' in',
# by (6,289 firings)
' by',
# with (5,888 firings)
' with',
# as (4,977 firings)
' as',
# from (4,937 firings)
' from',
# an (3,765 firings)
' an',
# out (2,076 firings)
' out', 'outed',
# " (1,791 firings)
'"',
# aft (1,739 firings)
' after',
# u (1,344 firings)
' U',
# ' (1,341 firings)
"'", " '",
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'a-family', 1
def prog_L08F08057(input_tokens):
"""17 token families, 26 surface forms (train F1 of the uncapped optimum: 0.206).
Fires at 153,200 of 36,786,566 positions.
Fires at 0.42% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: terms related to organizations, locations, and associated rights and issues"""
# --- end external annotation ---
fire = {
# state (1,833 firings)
' State',
# univers (1,251 firings)
' University',
# city (1,120 firings)
' City',
# county (1,055 firings)
' County',
# new (984 firings)
' News',
# today (937 firings)
' TODAY', ' Today',
# park (729 firings)
' Park', ' Parks', ' Parking',
# police (709 firings)
' Police',
# day (581 firings)
' Day', ' Days',
# institute (544 firings)
' Institute',
# fc (531 firings)
' FC', 'FC', 'fc',
# avenue (488 firings)
' Avenue',
# riv (470 firings)
' River', ' Riv',
# cup (466 firings)
' Cup', 'cup',
# war (439 firings)
' War',
# army (411 firings)
' Army',
# tech (402 firings)
' Tech', ' TECH',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'state-family', 1
def prog_L08F08058(input_tokens):
"""18 token families, 41 surface forms (train F1 of the uncapped optimum: 0.304).
Fires at 100,613 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to the number two in various contexts"""
# --- end external annotation ---
fire = {
# two (12,672 firings)
' two', ' Two', 'Two', 'two', ' TWO',
# three (5,265 firings)
' three', ' Three', 'Three', 'three', ' THREE',
# few (3,062 firings)
' few',
# four (2,989 firings)
' four', 'Four',
# sever (2,668 firings)
' several', ' Several',
# five (2,585 firings)
' five', 'Five', ' Five',
# six (1,587 firings)
' six',
# seven (1,185 firings)
' seven', ' Seven', 'Seven', 'seven',
# dozen (1,024 firings)
' dozen', ' dozens', 'dozen', ' Dozens',
# eight (915 firings)
' eight', ' Eight',
# multiple (802 firings)
' multiple', 'multiple',
# hundred (490 firings)
' hundreds',
# nine (451 firings)
' nine',
# 200 (365 firings)
' 200',
# ten (349 firings)
' ten',
# twenty (331 firings)
' twenty', 'Twenty', ' Twenty',
# hundr (259 firings)
' hundred', ' Hundred',
# numer (252 firings)
' numerous',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'two-family', 1
def prog_L08F08059(input_tokens):
"""25 token families, 60 surface forms (train F1 of the uncapped optimum: 0.215).
Fires at 90,934 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases related to the movement of liquids or substances"""
# --- end external annotation ---
fire = {
# com (1,650 firings)
' coming', ' Coming', ' Comes',
# down (1,079 firings)
' down', ' DOWN',
# going (858 firings)
' going',
# fall (705 firings)
' fall', ' falling', ' falls', ' Falling',
# roll (680 firings)
' rolling', ' roll', ' rolled', ' rolls',
# runn (656 firings)
' running',
# arriv (618 firings)
' arrived', ' arriving',
# broke (569 firings)
' broke',
# pass (552 firings)
' passed', ' passes', ' Passed',
# flow (497 firings)
' flowing', ' flow', ' flows', ' flowed',
# fly (477 firings)
' flying', ' fly',
# spread (463 firings)
' spread', ' spreading', ' spreads',
# hang (446 firings)
' hanging', ' hangs', ' hang',
# fell (383 firings)
' fell',
# arr (355 firings)
' arrive', ' arrives',
# fir (297 firings)
' fired',
# pour (295 firings)
' pouring', ' poured', ' pour',
# float (289 firings)
' floating', ' float', ' floated', ' floats', ' Float',
# grow (264 firings)
' grow', ' grows',
# wait (263 firings)
' waiting', ' waits', ' Waiting',
# rain (259 firings)
' rain', ' raining', ' rains', ' Rain',
# direct (258 firings)
' directly',
# popp (251 firings)
' popping', ' popped',
# leak (246 firings)
' leaked', ' leaking',
# happen (228 firings)
' happening',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'com-family', 1
def prog_L08F08060(input_tokens):
"""50 token families, 58 surface forms (train F1 of the uncapped optimum: 0.606).
Fires at 10,781 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions of specific individuals, particularly those with the name "Abdul" or "Abdel." """
# --- end external annotation ---
fire = {
# african (728 firings)
' African',
# ul (648 firings)
'ul', ' UL',
# abdul (335 firings)
' Abdul',
# mal (181 firings)
'Mal', 'MAL',
# des (177 firings)
'Des',
# del (176 firings)
'Del',
# dil (167 firings)
' Dil',
# abdel (154 firings)
' Abdel',
# abd (148 firings)
' Abd',
# char (142 firings)
'CHAR', ' CHAR',
# dol (131 firings)
' Dol',
# arra (129 firings)
' arra',
# mald (117 firings)
' Mald',
# arabian (113 firings)
' Arabian',
# detect (105 firings)
'Detect', ' Detect',
# deaf (98 firings)
' deaf',
# dish (97 firings)
' Dish',
# methyl (93 firings)
' methyl', 'methyl',
# alban (92 firings)
' Alban',
# mohammad (92 firings)
' Mohammad',
# abel (92 firings)
'abel', ' Abel',
# faul (89 firings)
' Faul',
# aber (86 firings)
' Aber',
# exh (84 firings)
' exh',
# indec (84 firings)
' indec',
# ال (83 firings)
' ال', 'ال',
# pred (76 firings)
' Pred', 'Pred',
# mikhail (75 firings)
' Mikhail',
# fitz (72 firings)
' Fitz',
# sul (72 firings)
' Sul',
# alc (68 firings)
' Alc',
# desmond (68 firings)
' Desmond',
# oub (66 firings)
'oub',
# raf (63 firings)
'raf',
# ر (60 firings)
'ر',
# edmund (59 firings)
' Edmund',
# dianne (59 firings)
' Dianne',
# nikol (57 firings)
' Nikol',
# burr (57 firings)
' Burr',
# adal (55 firings)
'adal',
# abram (52 firings)
' Abram',
# bout (51 firings)
' Bout',
# ibur (51 firings)
'ibur',
# ethyl (50 firings)
'ethyl',
# alg (49 firings)
' Alger',
# duches (48 firings)
' Duchess',
# morbid (46 firings)
' morbid',
# inh (45 firings)
' inhal',
# jam (42 firings)
' Jamal',
# alm (34 firings)
' Alm',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'african-family', 1
def prog_L08F08061(input_tokens):
"""3 token families, 6 surface forms (train F1 of the uncapped optimum: 0.679).
Fires at 11,010 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "again." """
# --- end external annotation ---
fire = {
# again (6,737 firings)
' again', ' Again', 'again', 'Again',
# anew (36 firings)
' anew',
# reapp (35 firings)
' reapp',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'again-family', 1
def prog_L08F08062(input_tokens):
"""1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.371).
Fires at 140,944 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "in" across various contexts"""
# --- end external annotation ---
fire = {
# in (120,240 firings)
' in',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'in-family', 1
def prog_L08F08063(input_tokens):
"""17 token families, 19 surface forms (train F1 of the uncapped optimum: 0.142).
Fires at 68,097 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: prominent names and entities in various contexts, including politics, sports, and business"""
# --- end external annotation ---
fire = {
# trump (2,788 firings)
' Trump',
# obama (961 firings)
' Obama',
# clinton (665 firings)
' Clinton',
# police (448 firings)
' Police',
# court (442 firings)
' Court',
# commis (349 firings)
' Commission',
# china (323 firings)
'China',
# korea (319 firings)
' Korea',
# more (306 firings)
' MORE',
# sand (271 firings)
' Sanders',
# committee (252 firings)
' Committee',
# russia (237 firings)
'Russia',
# agency (220 firings)
' Agency',
# fbi (199 firings)
' FBI', 'FBI',
# israel (183 firings)
'Israel',
# lawmak (179 firings)
' lawmakers', ' lawmaker',
# service (173 firings)
' Service',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'trump-family', 1
def prog_L08F08064(input_tokens):
"""5 token families, 6 surface forms (train F1 of the uncapped optimum: 0.480).
Fires at 57,239 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "said" in various contexts"""
# --- end external annotation ---
fire = {
# said (29,472 firings)
' said',
# say (1,707 firings)
' saying',
# confirm (1,466 firings)
' confirmed',
# insist (243 firings)
' insisted', ' insisting',
# indicat (208 firings)
' indicated',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'said-family', 1
def prog_L08F08065(input_tokens):
"""6 token families, 7 surface forms (train F1 of the uncapped optimum: 0.130).
Fires at 84,150 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: questions and considerations related to decision-making processes"""
# --- end external annotation ---
fire = {
# what (2,994 firings)
' what',
# how (2,526 firings)
' how',
# wheth (2,524 firings)
' whether', 'whether',
# future (788 firings)
' future',
# transgend (272 firings)
' transgender',
# decide (238 firings)
' decide',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'what-family', 1
def prog_L08F08066(input_tokens):
"""1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.270).
Fires at 128,727 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: statements of being or existence in descriptive contexts"""
# --- end external annotation ---
fire = {
# is (53,438 firings)
' is',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'is-family', 1
def prog_L08F08067(input_tokens):
"""10 token families, 28 surface forms (train F1 of the uncapped optimum: 0.070).
Fires at 49,527 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to positions of authority or leadership in government"""
# --- end external annotation ---
fire = {
# of (13,108 firings)
' of', ' OF',
# chair (468 firings)
' chair', ' chaired', 'chair', ' chairs', 'chairs', 'Chair',
# chairman (451 firings)
' chairman', ' Chairman',
# committee (377 firings)
' Committee', ' committee',
# lead (352 firings)
' lead', ' leading', ' leads', 'lead',
# memb (328 firings)
' member', ' Member', 'member',
# house (287 firings)
' House',
# head (254 firings)
' heads', ' headed', ' heading', 'heading',
# senate (181 firings)
' Senate', ' senate',
# shadow (128 firings)
' shadow', ' Shadow',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'of-family', 1
def prog_L08F08068(input_tokens):
"""48 token families, 55 surface forms (train F1 of the uncapped optimum: 0.715).
Fires at 17,387 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to Nebraska and its associations"""
# --- end external annotation ---
fire = {
# ne (365 firings)
' NE', 'Ne',
# nebraska (359 firings)
' Nebraska',
# nep (321 firings)
' Nepal', ' Nep', ' nep',
# ense (283 firings)
'ense',
# hus (265 firings)
' Hus',
# noah (240 firings)
' Noah',
# agriculture (228 firings)
' Agriculture',
# mes (214 firings)
' Mes',
# indoor (202 firings)
' indoor',
# milo (178 firings)
' Milo',
# wendy (170 firings)
' Wendy',
# dwarf (168 firings)
' dwarf', ' Dwarf',
# wes (166 firings)
' Wes',
# mesh (165 firings)
' mesh', ' Mesh', ' meshes', 'Mesh',
# fry (160 firings)
' Fry',
# omaha (146 firings)
' Omaha',
# methane (139 firings)
' methane',
# bryce (135 firings)
' Bryce',
# nder (128 firings)
'nder',
# ethan (127 firings)
' Ethan',
# huffpost (125 firings)
' HuffPost',
# nitrogen (117 firings)
' nitrogen',
# fresno (116 firings)
' Fresno',
# mali (116 firings)
' Mali',
# mesa (115 firings)
' Mesa',
# norri (114 firings)
' Norris',
# kristen (112 firings)
' Kristen',
# swat (112 firings)
' SWAT',
# lsu (109 firings)
' LSU',
# purdue (107 firings)
' Purdue',
# goat (106 firings)
' goat',
# peanut (105 firings)
' peanut',
# gron (104 firings)
' Gron',
# usda (103 firings)
' USDA',
# heb (100 firings)
' Heb',
# betsy (100 firings)
' Betsy',
# northwestern (100 firings)
' Northwestern',
# lank (99 firings)
' Lank',
# hondura (93 firings)
' Honduras',
# mud (92 firings)
' Mud',
# sussex (91 firings)
' Sussex',
# cosmo (87 firings)
' Cosmos',
# neh (86 firings)
' Neh',
# woody (84 firings)
' Woody',
# morton (84 firings)
' Morton',
# hay (83 firings)
' hay',
# petraeu (80 firings)
' Petraeus',
# assy (75 firings)
' Assy',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ne-family', 1
def prog_L08F08069(input_tokens):
"""19 token families, 71 surface forms (train F1 of the uncapped optimum: 0.443).
Fires at 169,632 of 36,786,566 positions.
Fires at 0.46% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: technical terms or titles related to performances, services, and structured content"""
# --- end external annotation ---
fire = {
# test (836 firings)
' Test', ' Tests', 'Test', ' Testing', 'Testing', ' TEST',
# sci (811 firings)
' Science', 'Science', ' Sci',
# work (798 firings)
' Work', ' Working', 'Work', ' Works', ' Workers', ' Worker', 'Working', 'Works', ' WORK',
# pow (776 firings)
' Power', ' Powers', 'Power', ' POWER',
# life (743 firings)
' Life', 'Life', ' LIFE',
# commun (742 firings)
' Community', 'Community', ' Commun', ' Communities', ' COMMUN',
# mus (717 firings)
' Music', 'Music', ' Musical',
# law (716 firings)
' Law', ' Laws', ' LAW',
# sport (701 firings)
' Sports', ' Sport', ' Sporting', 'Sports', 'Sport',
# time (676 firings)
' Time', 'Time', ' TIME',
# home (661 firings)
' Home', 'Home', ' HOME', 'HOME',
# introduc (640 firings)
'Introduction', ' Introduction',
# play (604 firings)
' Play', ' Player', ' Players', ' Playing', ' Plays', ' PLAY', ' Played',
# project (554 firings)
' Project', ' Projects', 'Project',
# championship (495 firings)
' Championship', ' Championships',
# season (470 firings)
' Season', ' Seasons', 'Season',
# radio (456 firings)
' Radio', 'Radio',
# busi (432 firings)
' Business', 'Business',
# fami (430 firings)
' Family', 'Family',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'test-family', 1
def prog_L08F08070(input_tokens):
"""50 token families, 53 surface forms (train F1 of the uncapped optimum: 0.695).
Fires at 27,974 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to the cities Mumbai and Shanghai"""
# --- end external annotation ---
fire = {
# moscow (1,189 firings)
' Moscow', 'Moscow',
# melbourne (888 firings)
' Melbourne',
# mumbai (465 firings)
' Mumbai',
# milan (409 firings)
' Milan',
# munich (377 firings)
' Munich',
# shanghai (350 firings)
' Shanghai',
# istanbul (346 firings)
' Istanbul',
# barcelona (343 firings)
' Barcelona',
# mcgregor (252 firings)
' McGregor',
# kar (238 firings)
'kar',
# stockholm (222 firings)
' Stockholm',
# satan (217 firings)
' Satan', ' Satanic',
# vienna (215 firings)
' Vienna',
# petersburg (207 firings)
' Petersburg',
# vatican (204 firings)
' Vatican',
# windsor (188 firings)
' Windsor',
# jacksonville (188 firings)
' Jacksonville',
# salman (176 firings)
' Salman',
# frankfurt (173 firings)
' Frankfurt',
# base (171 firings)
' basement',
# monaco (164 firings)
' Monaco',
# parish (157 firings)
' parish',
# gotham (155 firings)
' Gotham',
# maharashtra (154 firings)
' Maharashtra',
# bundesliga (149 firings)
' Bundesliga',
# wichita (145 firings)
' Wichita',
# valencia (144 firings)
' Valencia',
# squadron (142 firings)
' Squadron', ' squadron',
# stamford (141 firings)
' Stamford',
# к (140 firings)
'к',
# sochi (140 firings)
' Sochi',
# serie (138 firings)
' Serie',
# chennai (136 firings)
' Chennai',
# phantom (131 firings)
' Phantom',
# shiv (127 firings)
' Shiv',
# zurich (122 firings)
' Zurich',
# meteor (122 firings)
' Meteor',
# karachi (120 firings)
' Karachi',
# daytona (115 firings)
' Daytona',
# chandl (114 firings)
' Chandler',
# budap (111 firings)
' Budapest',
# sunset (111 firings)
' Sunset',
# sharma (109 firings)
' Sharma',
# bledon (106 firings)
'bledon',
# stanton (104 firings)
' Stanton',
# deutsche (104 firings)
' Deutsche',
# ahar (103 firings)
'ahar',
# arkham (103 firings)
' Arkham',
# bangalore (102 firings)
' Bangalore',
# balcony (99 firings)
' balcony',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'moscow-family', 1
def prog_L08F08071(input_tokens):
"""15 token families, 32 surface forms (train F1 of the uncapped optimum: 0.261).
Fires at 342,382 of 36,786,566 positions.
Fires at 0.93% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases related to significant events or occurrences in society"""
# --- end external annotation ---
fire = {
# - (4,735 firings)
'-',
# to (2,130 firings)
' To', ' TO',
# follow (2,016 firings)
' Following', ' Follow',
# ' (1,906 firings)
"'", " '",
# new (1,806 firings)
' New', ' NEW', ' Newly',
# ed (1,497 firings)
'ed', 'ED',
# block (1,342 firings)
'block', ' Block', ' Blocks',
# in (1,302 firings)
' In', ' IN',
# you (1,131 firings)
' You', ' YOU',
# is (1,119 firings)
' Is', ' IS',
# of (1,103 firings)
' Of', ' OF',
# un (1,076 firings)
' Un',
# for (1,021 firings)
' For', ' FOR',
# ock (1,020 firings)
'ocked', 'ocking', 'ocks', 'OCK',
# your (991 firings)
' Your', ' YOUR',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '--family', 1
def prog_L08F08072(input_tokens):
"""41 token families, 52 surface forms (train F1 of the uncapped optimum: 0.654).
Fires at 12,157 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: mentions of specific hurricanes, particularly Hurricanes Sandy, Katrina, and Maria"""
# --- end external annotation ---
fire = {
# storm (1,031 firings)
' storm', ' storms',
# hurricane (886 firings)
' Hurricane', ' hurricane',
# disast (553 firings)
' disaster', ' disasters', ' Disaster',
# harvey (470 firings)
' Harvey',
# trop (357 firings)
' tropical', ' Tropical',
# hurrican (260 firings)
' Hurricanes', ' hurricanes',
# opioid (241 firings)
' opioid', ' opioids',
# lebron (229 firings)
' LeBron',
# destruct (167 firings)
' destructive',
# radeon (165 firings)
' Radeon',
# irma (159 firings)
' Irma',
# torrent (153 firings)
' torrent', ' Torrent',
# read (149 firings)
' Reader',
# whale (145 firings)
' whale',
# tornado (137 firings)
' tornado', ' Tornado',
# airbu (136 firings)
' Airbus',
# cycl (127 firings)
' cycl',
# hur (124 firings)
'Hur', 'hur',
# dictator (124 firings)
' dictator',
# katrina (121 firings)
' Katrina',
# batter (118 firings)
' battered',
# mosquito (113 firings)
' mosquito',
# meteor (111 firings)
' Meteor',
# apache (100 firings)
' Apache',
# wildfire (91 firings)
' wildfire',
# blist (88 firings)
' blister',
# typh (70 firings)
' typh', ' Typh',
# cop (69 firings)
' COP',
# landfall (65 firings)
' landfall',
# titan (65 firings)
' Titanic',
# igator (63 firings)
'igator',
# incendiary (62 firings)
' incendiary',
# typhoon (60 firings)
' Typhoon',
# ricane (54 firings)
'ricane',
# vernight (52 firings)
'vernight',
# fema (51 firings)
' FEMA',
# turbulent (51 firings)
' turbulent',
# hepatiti (50 firings)
' hepatitis',
# hms (43 firings)
' HMS',
# cata (37 firings)
' cataly',
# clone (36 firings)
'clone',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'storm-family', 1
def prog_L08F08073(input_tokens):
"""21 token families, 35 surface forms (train F1 of the uncapped optimum: 0.368).
Fires at 21,853 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "cross" and its variations, particularly in relation to boundaries or limits"""
# --- end external annotation ---
fire = {
# acros (1,731 firings)
' across',
# cros (1,571 firings)
' cross', 'cross', 'Cross',
# cross (1,335 firings)
' crossing', ' crossed', ' crosses', ' Crossing', ' crossings',
# jump (388 firings)
' jumped', ' jumping', ' jumps',
# walk (338 firings)
' walked',
# spill (184 firings)
' spilled', ' spilling',
# climb (184 firings)
' climb', ' climbing',
# board (159 firings)
' boarded', ' boarding',
# part (153 firings)
' parted', ' parting',
# transc (115 firings)
' transc',
# crossov (114 firings)
' crossover',
# touch (109 firings)
' touched',
# check (106 firings)
' checked',
# hopp (88 firings)
' hopped', ' hopping',
# breach (86 firings)
' breached', ' breaching',
# wander (83 firings)
' wandered',
# lin (78 firings)
' lined',
# span (73 firings)
' spans',
# tipp (68 firings)
' tipping',
# spann (63 firings)
' spanning',
# parallel (58 firings)
' parallels',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'acros-family', 1
def prog_L08F08074(input_tokens):
"""33 token families, 48 surface forms (train F1 of the uncapped optimum: 0.584).
Fires at 15,041 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: variations of the word "kick" in different contexts"""
# --- end external annotation ---
fire = {
# kick (3,393 firings)
' kick', ' kicked', ' kicking', ' kicks', ' Kick', 'kick', ' kicker', 'Kick',
# ick (572 firings)
'icks', 'icking', 'icked',
# boot (356 firings)
' boot', ' booted',
# knock (193 firings)
' knocked', ' knocks',
# rip (157 firings)
' rip', 'rip',
# jab (145 firings)
' Jab',
# kickoff (143 firings)
' kickoff',
# chas (135 firings)
' chased',
# expell (127 firings)
' expelled',
# whipp (109 firings)
' whipped', ' whipping',
# kab (103 firings)
' Kab',
# pave (97 firings)
' pave',
# stomp (90 firings)
' stomp',
# blist (87 firings)
' blister',
# piss (85 firings)
' pissed',
# punt (82 firings)
' punt',
# pepp (75 firings)
' pepp',
# whip (75 firings)
' whip',
# capp (75 firings)
' capped',
# pis (72 firings)
' piss',
# claw (72 firings)
' claw',
# kham (71 firings)
' Kham',
# cock (66 firings)
' Cock',
# slap (66 firings)
' slap',
# stab (63 firings)
' Stab', 'stab',
# kap (57 firings)
' Kap',
# expel (55 firings)
' expel',
# drib (54 firings)
' drib',
# escort (53 firings)
' escorted',
# pun (47 firings)
'pun',
# expul (40 firings)
' expulsion',
# lick (40 firings)
' licking', ' lick',
# kemp (40 firings)
' Kemp',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'kick-family', 1
def prog_L08F08075(input_tokens):
"""25 token families, 48 surface forms (train F1 of the uncapped optimum: 0.147).
Fires at 41,699 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: specific years or time periods related to historical events or positions"""
# --- end external annotation ---
fire = {
# 2015 (801 firings)
' 2015', '2015',
# 2016 (781 firings)
' 2016', '2016',
# 2012 (707 firings)
' 2012', '2012',
# 2014 (700 firings)
' 2014', '2014',
# 2013 (676 firings)
' 2013', '2013',
# 2011 (457 firings)
' 2011', '2011',
# 2010 (432 firings)
' 2010', '2010',
# 2009 (345 firings)
' 2009', '2009',
# 2018 (308 firings)
' 2018', '2018',
# 2008 (303 firings)
' 2008', '2008',
# 2007 (240 firings)
' 2007', '2007',
# pm (234 firings)
'pm', ' pm',
# season (232 firings)
' seasons',
# 2000 (218 firings)
' 2000', '2000',
# 2006 (204 firings)
' 2006', '2006',
# 2005 (165 firings)
' 2005', '2005',
# 2020 (164 firings)
' 2020', '2020',
# 2019 (148 firings)
' 2019', '2019',
# 19 (146 firings)
'19',
# 1990 (141 firings)
' 1990', '1990',
# 2002 (130 firings)
' 2002', '2002',
# 1980 (124 firings)
' 1980', '1980',
# 2003 (120 firings)
' 2003', '2003',
# 2001 (118 firings)
' 2001', '2001',
# 1970 (113 firings)
' 1970', '1970',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '2015-family', 1
def prog_L08F08076(input_tokens):
"""1 token families, 2 surface forms (train F1 of the uncapped optimum: 0.204).
Fires at 38,984 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases indicating the concept of 'room for' various aspects such as improvement or creativity"""
# --- end external annotation ---
fire = {
# for (29,626 firings)
' for', ' FOR',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'for-family', 1
def prog_L08F08077(input_tokens):
"""47 token families, 55 surface forms (train F1 of the uncapped optimum: 0.640).
Fires at 11,857 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to various types of spectra"""
# --- end external annotation ---
fire = {
# spectrum (609 firings)
' spectrum', ' Spectrum',
# montgomery (305 firings)
' Montgomery',
# syndrome (278 firings)
' syndrome',
# sand (210 firings)
' sands', ' Sands',
# receptor (195 firings)
' receptor', ' receptors',
# lte (169 firings)
' LTE',
# metadata (156 firings)
' metadata', 'metadata',
# mickey (152 firings)
' Mickey',
# wav (149 firings)
'waves',
# siriu (136 firings)
' Sirius',
# cryst (132 firings)
' crystal',
# slate (129 firings)
' slate',
# leslie (128 firings)
' Leslie',
# bandwidth (127 firings)
' bandwidth',
# mhz (127 firings)
' MHz', 'MHz',
# route (127 firings)
' Route', 'Route',
# seism (120 firings)
' seismic', ' seism',
# spect (118 firings)
' spect', 'Spect',
# airspace (111 firings)
' airspace',
# constella (104 firings)
' constellation',
# leigh (100 firings)
' Leigh',
# cassidy (98 firings)
' Cassidy',
# jazz (96 firings)
' Jazz',
# navajo (95 firings)
' Navajo',
# swath (94 firings)
' swath',
# quire (93 firings)
'quire',
# rock (87 firings)
' Rockies',
# arn (85 firings)
' Arn',
# schwartz (85 firings)
' Schwartz',
# sinclair (81 firings)
' Sinclair',
# frequenc (78 firings)
' frequencies',
# zinski (77 firings)
'zinski',
# ghz (69 firings)
' GHz',
# cortex (69 firings)
' cortex',
# rez (63 firings)
' Rez',
# trey (63 firings)
' Trey',
# myst (62 firings)
' myst',
# armac (62 firings)
'armac',
# palette (54 firings)
' palette',
# ing (53 firings)
'inges',
# nic (53 firings)
' nic',
# modem (48 firings)
' modem',
# nebula (47 firings)
' Nebula',
# metast (46 firings)
' metast',
# malone (45 firings)
' Malone',
# otom (44 firings)
'otom',
# ambigu (39 firings)
' ambiguity',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'spectrum-family', 1
def prog_L08F08078(input_tokens):
"""35 token families, 71 surface forms (train F1 of the uncapped optimum: 0.427).
Fires at 44,763 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases that emphasize approximation or near certainty"""
# --- end external annotation ---
fire = {
# almost (2,255 firings)
' almost', 'Almost', ' Almost', 'almost',
# near (1,765 firings)
' nearly', 'Nearly', ' Nearly', 'Near',
# quite (908 firings)
' quite', ' Quite', 'Quite', 'quite',
# many (851 firings)
'Many',
# yet (845 firings)
' Yet', 'Yet',
# much (685 firings)
'Much', ' Much', 'much',
# still (659 firings)
' Still', 'Still',
# perhap (652 firings)
'Perhaps', ' Perhaps',
# ever (568 firings)
'Ever', ' Ever',
# very (512 firings)
' Very', 'Very', 'very',
# just (451 firings)
'Just',
# un (443 firings)
'Un',
# rath (387 firings)
' Rather', 'Rather',
# les (369 firings)
'Less', ' Less', 'LESS',
# part (327 firings)
'Part', ' Partial',
# twenty (260 firings)
'Twenty', ' Twenty',
# often (257 firings)
' Often', 'Often',
# probab (222 firings)
' Probably', 'Probably', 'probably',
# sometim (216 firings)
'Sometimes',
# apparent (213 firings)
'Apparently', ' Apparently',
# too (211 firings)
'Too', ' Too',
# only (204 firings)
'Only',
# usual (195 firings)
'Usually', ' Usually',
# neith (193 firings)
' Neither', 'Neither',
# half (189 firings)
'Half', ' Half',
# known (188 firings)
'Known', ' Known',
# already (185 firings)
'Already', ' Already',
# nev (170 firings)
'Never',
# few (164 firings)
'Few',
# short (163 firings)
'Shortly', ' Shortly',
# basical (160 firings)
'Basically', ' Basically',
# pretty (153 firings)
' Pretty', 'Pretty',
# especial (149 firings)
' Especially', 'Especially',
# noth (142 firings)
'Nothing',
# maybe (131 firings)
'Maybe',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'almost-family', 1
def prog_L08F08079(input_tokens):
"""42 token families, 60 surface forms (train F1 of the uncapped optimum: 0.633).
Fires at 14,401 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: occurrences of the word "drag" in various contexts"""
# --- end external annotation ---
fire = {
# drag (712 firings)
' drag', ' Drag', 'Drag', ' DRAG',
# opera (371 firings)
' opera', ' Opera',
# rub (336 firings)
' Rub', 'Rub',
# dragg (266 firings)
' dragged', ' dragging',
# ru (233 firings)
' Ru', 'Ru',
# rac (229 firings)
' Rac', ' racer',
# redirect (225 firings)
' redirect',
# custom (218 firings)
' Customs',
# rack (205 firings)
' rack', ' Rack', ' racks',
# cos (168 firings)
'Cos', ' Cos',
# dr (167 firings)
'dr',
# ig (164 firings)
' Ig', ' IG',
# cub (138 firings)
' Cub',
# bra (134 firings)
' Bras', ' bras',
# gomez (124 firings)
' Gomez',
# hod (121 firings)
' Hod',
# rag (118 firings)
'rag', ' Rag', 'RAG',
# craw (115 firings)
' craw', ' Craw',
# barb (112 firings)
' Barber', ' Barb',
# hug (112 firings)
'Hug', ' Hug',
# dug (110 firings)
' Dug',
# lewd (109 firings)
' lewd',
# hip (104 firings)
' Hip',
# tug (102 firings)
' tug',
# acht (102 firings)
'acht',
# drum (101 firings)
' Drum',
# mot (99 firings)
' mot',
# bian (97 firings)
' Bian',
# armac (89 firings)
'armac',
# zin (86 firings)
' Zin',
# asp (79 firings)
' ASP',
# queen (78 firings)
' queens',
# igor (78 firings)
' Igor',
# borg (78 firings)
' Borg',
# salon (73 firings)
' salon',
# grimm (68 firings)
' Grimm',
# cadillac (65 firings)
' Cadillac',
# cor (63 firings)
' Cors',
# men (62 firings)
' Mens',
# ng (61 firings)
' ng',
# rez (53 firings)
' Rez',
# yg (49 firings)
'yg',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'drag-family', 1
def prog_L08F08080(input_tokens):
"""4 token families, 6 surface forms (train F1 of the uncapped optimum: 0.370).
Fires at 14,489 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: the repetition of the phrase "over the years." """
# --- end external annotation ---
fire = {
# year (8,701 firings)
' years', ' Years', 'Years',
# decad (905 firings)
' decades',
# generation (129 firings)
' generations',
# wiv (49 firings)
'wives',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'year-family', 1
def prog_L08F08081(input_tokens):
"""9 token families, 20 surface forms (train F1 of the uncapped optimum: 0.164).
Fires at 177,903 of 36,786,566 positions.
Fires at 0.48% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to books and their authors"""
# --- end external annotation ---
fire = {
# the (6,247 firings)
' The',
# " (3,576 firings)
' "',
# new (1,487 firings)
' New', ' Newly',
# why (990 firings)
'Why', ' Why', ' WHY', 'why',
# american (859 firings)
' American', 'American',
# & (707 firings)
' &',
# how (670 firings)
' How', 'How', 'HOW', ' HOW',
# black (512 firings)
' Black', 'Black', ' BLACK',
# is (503 firings)
' Is', 'Is',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'the-family', 1
def prog_L08F08082(input_tokens):
"""3 token families, 3 surface forms (train F1 of the uncapped optimum: 0.271).
Fires at 8,933 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: segments of text that involve the concept of "part" or participation in an action or event"""
# --- end external annotation ---
fire = {
# part (3,288 firings)
' part',
# entirety (34 firings)
' entirety',
# pain (23 firings)
' pains',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'part-family', 1
def prog_L08F08083(input_tokens):
"""46 token families, 56 surface forms (train F1 of the uncapped optimum: 0.631).
Fires at 30,646 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: terms related to awkwardness and embarrassment"""
# --- end external annotation ---
fire = {
# embarrass (664 firings)
' embarrassing', ' embarrassed', ' embarrassment',
# annoy (552 firings)
' annoying', ' annoyed', ' annoy', ' annoyance',
# ugly (423 firings)
' ugly',
# awkward (420 firings)
' awkward', ' awkwardly',
# pain (399 firings)
' painful',
# unfortunate (355 firings)
' unfortunate',
# disappoint (352 firings)
' disappointing',
# convenient (351 firings)
' convenient', ' conveniently',
# frustrat (328 firings)
' frustrating',
# setback (328 firings)
' setback', ' setbacks',
# uncomfort (305 firings)
' uncomfortable',
# awful (283 firings)
' awful',
# cost (254 firings)
' costly',
# nasty (247 firings)
' nasty',
# unpopular (224 firings)
' unpopular',
# hat (213 firings)
' hated', ' hates',
# hilari (208 firings)
' hilarious',
# unwant (184 firings)
' unwanted',
# contenti (183 firings)
' contentious',
# bor (179 firings)
' boring',
# invas (160 firings)
' invasive',
# humiliat (158 firings)
' humiliating', ' humiliated',
# unpleasant (143 firings)
' unpleasant',
# messy (141 firings)
' messy',
# boo (141 firings)
' boo',
# unnecessary (137 firings)
' unnecessary',
# inappropriate (134 firings)
' inappropriate',
# headache (129 firings)
' headache',
# unhappy (124 firings)
' unhappy',
# headach (123 firings)
' headaches',
# traumat (121 firings)
' traumatic',
# miser (121 firings)
' miserable',
# amus (115 firings)
' amusing',
# unaccept (114 firings)
' unacceptable',
# discomfort (112 firings)
' discomfort',
# infuri (109 firings)
' infuri',
# dislike (109 firings)
' dislike',
# distrac (109 firings)
' distraction',
# undo (105 firings)
' undo',
# tedi (104 firings)
' tedious',
# abrupt (104 firings)
' abrupt',
# stress (102 firings)
' stressful',
# painful (97 firings)
' painfully',
# unsettl (93 firings)
' unsettling',
# embarras (90 firings)
' embarrass',
# debilitat (87 firings)
' debilitating',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'embarrass-family', 1
def prog_L08F08084(input_tokens):
"""44 token families, 53 surface forms (train F1 of the uncapped optimum: 0.591).
Fires at 13,652 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "exploit" and its variations, indicating a focus on discussing abuse or taking advantage of something"""
# --- end external annotation ---
fire = {
# exploit (827 firings)
' exploit', ' exploited', ' exploits', ' exploiting',
# abus (309 firings)
' abused', ' abusing',
# bypas (234 firings)
' bypass',
# exploita (230 firings)
' exploitation',
# exot (217 firings)
' exotic',
# vulner (216 firings)
' vulnerability',
# expl (212 firings)
' Expl',
# har (208 firings)
' harness',
# leverage (208 firings)
' leverage',
# comprom (201 firings)
' compromised',
# vulnerabil (196 firings)
' vulnerabilities',
# util (178 firings)
' util', ' utilized', ' utilization',
# reproduc (140 firings)
' reproduced', ' reproduction',
# lev (134 firings)
' lever', ' Lever',
# equ (132 firings)
' Equ',
# utilize (119 firings)
' utilize',
# loophole (112 firings)
' loophole',
# lucrat (110 firings)
' lucrative',
# oit (105 firings)
'oit',
# zion (103 firings)
' Zion',
# circumvent (95 firings)
' circumvent',
# reproduce (91 firings)
' reproduce',
# capitalize (90 firings)
' capitalize',
# manipulat (84 firings)
' manipulated',
# exaggerat (83 firings)
' exaggerated',
# cultivat (83 firings)
' cultivated', ' cultivating',
# engineer (82 firings)
' engineered',
# manipula (79 firings)
' manipulation',
# loophol (77 firings)
' loopholes',
# leverag (66 firings)
' leveraging',
# penetra (66 firings)
' penetration',
# fetish (60 firings)
' fetish',
# inflat (58 firings)
' inflated',
# cv (54 firings)
' CV',
# opportun (54 firings)
' opportun',
# ploy (52 firings)
' ploy',
# exacerbat (47 firings)
' exacerbated',
# overflow (46 firings)
' overflow',
# reprodu (46 firings)
' reprodu',
# patch (43 firings)
' patched',
# emula (41 firings)
' emulation',
# explo (40 firings)
' explo',
# workaround (40 firings)
' workaround',
# reciproc (37 firings)
' reciproc',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'exploit-family', 1
def prog_L08F08085(input_tokens):
"""48 token families, 76 surface forms (train F1 of the uncapped optimum: 0.687).
Fires at 90,650 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: terms related to opposition, complaints, and historical context"""
# --- end external annotation ---
fire = {
# ad (1,098 firings)
'Ad',
# con (1,095 firings)
'Con', 'Cons', 'cons', ' Cons', 'CON',
# ex (828 firings)
'Ex',
# sen (803 firings)
'Sen',
# un (753 firings)
'Un',
# rep (753 firings)
'Rep', 'REP',
# re (695 firings)
'Re',
# pre (642 firings)
'Pre', 'Pres', ' Pres', 'pres', 'PRE', 'PRES',
# don (627 firings)
'Don',
# pro (599 firings)
'Pro',
# int (569 firings)
'Int', 'Inter', 'INTER',
# cl (558 firings)
'Cl',
# tran (502 firings)
'Trans',
# cont (484 firings)
'Cont', ' Cont', 'CONT',
# com (470 firings)
'Com', 'Comment', ' Comed',
# pl (446 firings)
'Pl',
# se (444 firings)
'Se',
# per (434 firings)
'Per', 'Pers',
# sp (420 firings)
'Sp',
# part (417 firings)
'Part',
# dr (402 firings)
'Dr',
# law (401 firings)
'Law',
# comp (397 firings)
'Comp', ' Comp', 'comp',
# co (396 firings)
'Co',
# activ (394 firings)
'Activ', ' Activ', 'activ',
# tr (389 firings)
'Tr',
# ann (372 firings)
'Ann', 'ANN',
# app (365 firings)
'App',
# dis (354 firings)
'Dis', 'DIS', ' Diss',
# te (348 firings)
'Te',
# def (336 firings)
'Def',
# gu (333 firings)
'Gu',
# sub (328 firings)
'Sub',
# sc (319 firings)
'Sc',
# col (311 firings)
'Col',
# spe (310 firings)
'Spe', ' Spe',
# rec (306 firings)
'Rec', 'REC',
# sur (305 firings)
'Sur',
# ind (305 firings)
'Ind',
# min (300 firings)
'Min',
# tim (300 firings)
'Tim',
# pat (294 firings)
'Pat',
# dec (288 firings)
'Dec', 'DEC',
# dep (281 firings)
'Dep',
# att (272 firings)
'Att',
# ill (271 firings)
'Ill',
# im (263 firings)
'Im',
# phot (248 firings)
'Phot', 'phot',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ad-family', 1
def prog_L08F08086(input_tokens):
"""42 token families, 59 surface forms (train F1 of the uncapped optimum: 0.652).
Fires at 11,856 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: technical terms or abbreviations related to a specific field of study"""
# --- end external annotation ---
fire = {
# ec (1,499 firings)
'EC', 'ec', ' EC', ' ec', ' Ec',
# ea (568 firings)
' EA', 'EA',
# em (351 firings)
'EM', ' EM',
# cloud (348 firings)
' Cloud', 'Cloud',
# ce (278 firings)
'CE', ' CE',
# aca (199 firings)
'ACA', ' ACA',
# aws (177 firings)
' AWS',
# fx (174 firings)
'FX', ' FX',
# ij (167 firings)
'IJ',
# electro (157 firings)
' electro', ' Electro', ' electroly',
# wc (150 firings)
'WC', ' WC',
# fi (144 firings)
' FI', 'FI',
# ecb (126 firings)
' ECB',
# petro (125 firings)
' Petro',
# ic (120 firings)
' IC',
# fia (118 firings)
' FIA',
# decree (110 firings)
' decree',
# crawford (103 firings)
' Crawford',
# form (103 firings)
'Form', 'FORM', ' FORM',
# ge (99 firings)
' GE',
# conc (98 firings)
' CONC',
# ab (96 firings)
' AB',
# ober (94 firings)
' Ober',
# ee (93 firings)
' EE',
# orche (88 firings)
' orche', ' Orche',
# ei (83 firings)
'ei',
# cri (79 firings)
' Cris',
# eva (75 firings)
'EVA',
# seaw (75 firings)
' seaw',
# ems (73 firings)
' EMS',
# dw (70 firings)
' DW',
# xia (69 firings)
' Xia',
# fec (65 firings)
' FEC',
# crom (61 firings)
' Crom',
# ant (60 firings)
' antic',
# tec (59 firings)
' Tec',
# ecake (54 firings)
'ecake',
# fisa (52 firings)
' FISA',
# nec (51 firings)
' NEC',
# scaff (49 firings)
' scaff',
# neoc (46 firings)
' neoc',
# alec (40 firings)
' ALEC',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'ec-family', 1
def prog_L08F08087(input_tokens):
"""17 token families, 27 surface forms (train F1 of the uncapped optimum: 0.101).
Fires at 125,759 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: adjectives that describe essential characteristics or qualities"""
# --- end external annotation ---
fire = {
# a (5,540 firings)
' a',
# up (903 firings)
' up',
# used (902 firings)
' used',
# known (646 firings)
' known', 'known',
# part (570 firings)
' part', ' partly',
# best (564 firings)
' best',
# avail (534 firings)
' available',
# direct (467 firings)
' direct', ' directly', 'direct',
# good (437 firings)
' good',
# very (420 firings)
' very', ' VERY',
# well (413 firings)
' well',
# place (412 firings)
' place',
# close (390 firings)
' close', ' closely', 'close',
# hard (378 firings)
' hard', ' harder', 'hard',
# little (358 firings)
' little', 'little',
# bett (344 firings)
' better',
# involv (343 firings)
' involved',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'a-family', 1
def prog_L08F08088(input_tokens):
"""2 token families, 3 surface forms (train F1 of the uncapped optimum: 0.065).
Fires at 31,361 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases indicating the necessity of time for various actions or processes"""
# --- end external annotation ---
fire = {
# to (23,212 firings)
' to', ' TO',
# furth (217 firings)
' further',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'to-family', 1
def prog_L08F08089(input_tokens):
"""41 token families, 46 surface forms (train F1 of the uncapped optimum: 0.496).
Fires at 13,801 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to phenomena or significant occurrences"""
# --- end external annotation ---
fire = {
# phenomenon (689 firings)
' phenomenon',
# syndrome (369 firings)
' syndrome', ' Syndrome',
# sensa (269 firings)
' sensation',
# epidem (255 firings)
' epidemic',
# paradox (207 firings)
' paradox', ' Paradox',
# spectacle (181 firings)
' spectacle',
# forma (153 firings)
' formation', ' Formation',
# phenomena (147 firings)
' phenomena',
# itu (125 firings)
'itus',
# occurr (124 firings)
' occurrence',
# characteristic (121 firings)
' characteristics',
# fam (120 firings)
' Fam',
# myster (112 firings)
' mysteries',
# sight (104 firings)
' sightings',
# phen (99 firings)
' Phen', ' phen',
# formation (96 firings)
' formations',
# wond (96 firings)
' wonders', ' Wonders',
# meme (92 firings)
' meme',
# commonplace (90 firings)
' commonplace',
# hoax (88 firings)
' hoax',
# anoma (84 firings)
' anomaly',
# novelty (82 firings)
' novelty',
# hysteria (75 firings)
' hysteria',
# mad (75 firings)
' madness',
# scourge (69 firings)
' scourge',
# menace (61 firings)
' menace',
# stimuli (58 firings)
' stimuli',
# paradigm (56 firings)
' paradigm',
# trope (52 firings)
' trope',
# rar (51 firings)
' rarity',
# ich (50 firings)
' Ich',
# hazard (48 firings)
' hazards',
# anomal (47 firings)
' anomalies',
# fiasco (45 firings)
' fiasco',
# sensation (43 firings)
' sensations',
# manifesta (43 firings)
' manifestation',
# iae (41 firings)
'iae',
# relic (40 firings)
' relics',
# sas (39 firings)
' Sas',
# fallacy (37 firings)
' fallacy',
# jah (36 firings)
' Jah',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'phenomenon-family', 1
def prog_L08F08090(input_tokens):
"""8 token families, 10 surface forms (train F1 of the uncapped optimum: 0.104).
Fires at 28,878 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases indicating causation or conditions leading to certain outcomes"""
# --- end external annotation ---
fire = {
# cit (716 firings)
' citing', ' cited',
# follow (546 firings)
' following',
# despite (381 firings)
' despite',
# due (285 firings)
' due',
# avoid (241 firings)
' avoid',
# iting (185 firings)
'iting',
# amid (161 firings)
' amid',
# flee (160 firings)
' fleeing', ' flee',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'cit-family', 1
def prog_L08F08091(input_tokens):
"""26 token families, 31 surface forms (train F1 of the uncapped optimum: 0.232).
Fires at 97,045 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: incidents involving gunfire or shooting"""
# --- end external annotation ---
fire = {
# fire (1,592 firings)
' fire', 'fire', ' FIRE',
# dead (824 firings)
' dead',
# action (822 firings)
' action', ' ACTION',
# money (796 firings)
' money',
# control (751 firings)
' control',
# war (748 firings)
' war',
# guilty (689 firings)
' guilty',
# sale (612 firings)
' sale',
# sex (556 firings)
' sex',
# fun (549 firings)
' fun',
# debut (521 firings)
' debut',
# vir (435 firings)
' viral',
# contract (434 firings)
' contract',
# effect (430 firings)
' effect',
# lawsuit (395 firings)
' lawsuit',
# street (363 firings)
' streets',
# tri (356 firings)
' trial',
# respons (348 firings)
' responsibility', 'responsive',
# attack (336 firings)
' attacked',
# contact (322 firings)
' contact',
# surgery (296 firings)
' surgery',
# door (288 firings)
' doors',
# weight (285 firings)
' weight', ' weights',
# birth (268 firings)
' birth',
# suit (266 firings)
' suit',
# suicide (258 firings)
' suicide',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'fire-family', 1
def prog_L08F08092(input_tokens):
"""5 token families, 9 surface forms (train F1 of the uncapped optimum: 0.745).
Fires at 5,869 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "enough." """
# --- end external annotation ---
fire = {
# enough (4,571 firings)
' enough', ' Enough', 'Enough', 'enough',
# sufficient (316 firings)
' sufficient', ' sufficiently',
# insufficient (62 firings)
' insufficient',
# leftov (25 firings)
' leftover',
# ufficient (20 firings)
'ufficient',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'enough-family', 1
def prog_L08F08093(input_tokens):
"""16 token families, 24 surface forms (train F1 of the uncapped optimum: 0.116).
Fires at 77,180 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: references to specific dates and times"""
# --- end external annotation ---
fire = {
# on (6,599 firings)
' on',
# time (1,508 firings)
' time',
# th (1,469 firings)
'th', ' th',
# p (1,388 firings)
' p',
# every (1,089 firings)
' every', ' Every',
# ear (1,059 firings)
' early',
# late (833 firings)
' late', 'late',
# each (799 firings)
' each', ' Each', 'Each',
# next (663 firings)
' next',
# morn (591 firings)
' morning', 'morning',
# hour (579 firings)
' hours',
# saturday (443 firings)
' Saturday', ' Saturdays',
# even (259 firings)
' evening', ' evenings',
# 30 (241 firings)
'30',
# christma (222 firings)
' Christmas',
# middle (221 firings)
' middle',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'on-family', 1
def prog_L08F08094(input_tokens):
"""1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.195).
Fires at 88,503 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: the indefinite article "a" used in various contexts"""
# --- end external annotation ---
fire = {
# a (71,509 firings)
' a',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'a-family', 1
def prog_L08F08095(input_tokens):
"""22 token families, 55 surface forms (train F1 of the uncapped optimum: 0.214).
Fires at 126,301 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: specific terms related to medical and health topics"""
# --- end external annotation ---
fire = {
# f (2,125 firings)
'F', ' F', 'f',
# p (2,034 firings)
'P', ' P', 'p',
# c (2,003 firings)
'C', 'c', ' c',
# b (1,611 firings)
'B', 'b',
# g (1,398 firings)
'G', 'g',
# l (1,144 firings)
'L', 'l',
# h (1,091 firings)
'H',
# t (1,010 firings)
'T',
# d (911 firings)
'D',
# r (833 firings)
'R',
# sp (603 firings)
' Sp', 'Sp', 'sp', ' SP',
# v (543 firings)
'V',
# sc (532 firings)
' Sc', 'sc', 'Sc', ' SC',
# te (525 firings)
'te', 'Te', ' Te', ' te', ' TE',
# w (514 firings)
'W',
# sh (504 firings)
'Sh', 'sh', ' SH',
# co (496 firings)
'co', 'Co', ' CO',
# u (442 firings)
'U',
# cl (426 firings)
'cl', 'Cl', ' Cl', ' CL',
# ch (388 firings)
'Ch', ' CH',
# dr (386 firings)
'Dr', 'dr', ' DR',
# con (371 firings)
'Cons', 'Con', 'cons', ' Cons', ' CON',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'f-family', 1
def prog_L08F08096(input_tokens):
"""8 token families, 10 surface forms (train F1 of the uncapped optimum: 0.649).
Fires at 10,365 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: numerical values, particularly those associated with thresholds or measurements"""
# --- end external annotation ---
fire = {
# 0 (6,344 firings)
' 0', '0',
# 00 (254 firings)
' 00',
# ft (118 firings)
'FT',
# enrique (80 firings)
' Enrique',
# nil (74 firings)
' Nil', ' nil',
# toby (52 firings)
' Toby',
# // (44 firings)
'//',
# sleepy (36 firings)
' sleepy',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return '0-family', 1
def prog_L08F08097(input_tokens):
"""11 token families, 19 surface forms (train F1 of the uncapped optimum: 0.100).
Fires at 240,929 of 36,786,566 positions.
Fires at 0.65% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: phrases that discuss situations or conditions that are recurring or typical"""
# --- end external annotation ---
fire = {
# in (7,169 firings)
' in',
# one (2,015 firings)
' one', ' ones',
# first (1,587 firings)
' first', ' FIRST',
# just (1,117 firings)
' just', ' JUST',
# last (1,013 firings)
' last', ' LAST',
# only (888 firings)
' only', 'only',
# many (826 firings)
' many',
# great (662 firings)
' great', ' greatest', ' GREAT',
# part (656 firings)
' part', ' partly',
# good (651 firings)
' good',
# among (606 firings)
' among',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'in-family', 1
def prog_L08F08098(input_tokens):
"""11 token families, 37 surface forms (train F1 of the uncapped optimum: 0.138).
Fires at 231,175 of 36,786,566 positions.
Fires at 0.63% of positions, and
token rules do not serve features this dense: median test F1 0.086
across the 37,173 features above 0.39% of positions, against 0.763 for
the 76,283 below 0.016%. Read this fire set as the best available
approximation, not as what the feature does."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: patterns related to dietary habits and food consumption"""
# --- end external annotation ---
fire = {
# to (8,098 firings)
' to',
# time (1,566 firings)
' time', 'time',
# day (1,507 firings)
' day', ' days', 'day', 'Day', ' Days',
# eat (1,197 firings)
' eat', ' eating', ' eats', ' Eat', ' Eating', 'Eat', 'eating', 'eat',
# your (1,072 firings)
' your', ' Your', ' YOUR',
# diet (797 firings)
' diet', ' diets', ' Diet',
# dur (787 firings)
' during',
# dai (741 firings)
' daily', 'daily', 'Daily',
# hour (685 firings)
' hours', ' hour', 'hour', 'Hour',
# every (636 firings)
' every', ' Every', 'every',
# food (603 firings)
' food', ' foods', 'food', 'Food',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'to-family', 1
def prog_L08F08099(input_tokens):
"""6 token families, 14 surface forms (train F1 of the uncapped optimum: 0.675).
Fires at 15,921 of 36,786,566 positions."""
# --- begin external annotation (ignored at runtime) ---
"""prev SAE description: instances of the word "share." """
# --- end external annotation ---
fire = {
# share (5,512 firings)
' share', 'Share', ' Share', 'share',
# shar (4,025 firings)
' shared', ' sharing', ' shares', 'sharing', ' Sharing', 'shared',
# publish (195 firings)
' publish',
# selfie (100 firings)
' selfie',
# self (54 firings)
' selfies',
# exchang (46 firings)
' exchanging',
}
token = input_tokens[-1]
if token not in fire:
return "no match", 0
return 'share-family', 1