Add synthetic numeracy pretraining corpus
Browse files- .gitattributes +1 -0
- README.md +16 -0
- generate_math_numeracy.py +121 -0
- metadata.json +29 -0
- train.jsonl +3 -0
.gitattributes
CHANGED
|
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 58 |
# Video files - compressed
|
| 59 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 58 |
# Video files - compressed
|
| 59 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
| 61 |
+
train.jsonl filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- synthetic
|
| 5 |
+
- math
|
| 6 |
+
- numeracy
|
| 7 |
+
- pretraining
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# AGILLM Math Numeracy Synthetic
|
| 11 |
+
|
| 12 |
+
Synthetic plain-text numeracy corpus for AGILLM pretraining. Each JSONL row has `text` and `kind`.
|
| 13 |
+
|
| 14 |
+
Coverage includes addition/subtraction/multiplication/division facts, multi-digit arithmetic, comparisons, order/parity, place value, missing-number equations, simple word problems, fractions, money, and two-step arithmetic.
|
| 15 |
+
|
| 16 |
+
Generated deterministically with `generate_math_numeracy.py`.
|
generate_math_numeracy.py
ADDED
|
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
import json, random, hashlib, time
|
| 3 |
+
from pathlib import Path
|
| 4 |
+
OUT = Path('/workspace/agillm_math_numeracy_synth/train.jsonl')
|
| 5 |
+
META = Path('/workspace/agillm_math_numeracy_synth/metadata.json')
|
| 6 |
+
README = Path('/workspace/agillm_math_numeracy_synth/README.md')
|
| 7 |
+
N = 2_000_000
|
| 8 |
+
SEED = 731095
|
| 9 |
+
rng = random.Random(SEED)
|
| 10 |
+
objects = ['apples','coins','books','marbles','pencils','blocks','tokens','stickers']
|
| 11 |
+
names = ['Ada','Ben','Cara','Dion','Eli','Fay','Gus','Hana']
|
| 12 |
+
counts = {}
|
| 13 |
+
|
| 14 |
+
def jwrite(f, text, kind):
|
| 15 |
+
counts[kind] = counts.get(kind, 0) + 1
|
| 16 |
+
f.write(json.dumps({'text': text.strip() + '\n', 'kind': kind}, ensure_ascii=True, separators=(',', ':')) + '\n')
|
| 17 |
+
|
| 18 |
+
def choose(xs): return xs[rng.randrange(len(xs))]
|
| 19 |
+
|
| 20 |
+
def gen_add(big=False):
|
| 21 |
+
m = 9999 if big else 100
|
| 22 |
+
a,b = rng.randint(0,m), rng.randint(0,m); c=a+b
|
| 23 |
+
return choose([f'{a} + {b} = {c}.', f'{a}+{b}={c}', f'Compute {a} + {b}. Answer: {c}.', f'What is {a} plus {b}? {c}.', f'The sum of {a} and {b} is {c}.']), 'multi_digit_addition' if big else 'addition'
|
| 24 |
+
|
| 25 |
+
def gen_sub():
|
| 26 |
+
b,c = rng.randint(0,1000), rng.randint(0,1000); a=b+c
|
| 27 |
+
return choose([f'{a} - {b} = {c}.', f'{a}-{b}={c}', f'Compute {a} minus {b}. Answer: {c}.', f'What is {a} - {b}? {c}.']), 'subtraction'
|
| 28 |
+
|
| 29 |
+
def gen_mul():
|
| 30 |
+
a,b = (rng.randint(0,20), rng.randint(0,20)) if rng.random()<0.75 else (rng.randint(0,99), rng.randint(0,12))
|
| 31 |
+
c=a*b
|
| 32 |
+
return choose([f'{a} * {b} = {c}.', f'{a} times {b} equals {c}.', f'Compute {a} x {b}. Answer: {c}.', f'The product of {a} and {b} is {c}.']), 'multiplication'
|
| 33 |
+
|
| 34 |
+
def gen_div():
|
| 35 |
+
b,c = rng.randint(1,20), rng.randint(0,100); a=b*c
|
| 36 |
+
return choose([f'{a} / {b} = {c}.', f'{a} divided by {b} equals {c}.', f'Compute {a} / {b}. Answer: {c}.']), 'exact_division'
|
| 37 |
+
|
| 38 |
+
def gen_compare():
|
| 39 |
+
a,b = rng.randint(-200,200), rng.randint(-200,200)
|
| 40 |
+
sym = '>' if a>b else '<' if a<b else '='
|
| 41 |
+
word = 'greater than' if a>b else 'less than' if a<b else 'equal to'
|
| 42 |
+
return choose([f'{a} {sym} {b}.', f'{a} is {word} {b}.', f'Compare {a} and {b}: {a} {sym} {b}.']), 'comparison'
|
| 43 |
+
|
| 44 |
+
def gen_order_parity():
|
| 45 |
+
n = rng.randint(-1000,1000)
|
| 46 |
+
if rng.random()<0.5:
|
| 47 |
+
return choose([f'The number after {n} is {n+1}.', f'The number before {n} is {n-1}.', f'{n-1}, {n}, {n+1}.']), 'number_order'
|
| 48 |
+
par = 'even' if n % 2 == 0 else 'odd'
|
| 49 |
+
return f'{n} is {par}.', 'parity'
|
| 50 |
+
|
| 51 |
+
def gen_place():
|
| 52 |
+
n = rng.randint(0,9999); ones=n%10; tens=(n//10)%10; hundreds=(n//100)%10; thousands=(n//1000)%10
|
| 53 |
+
return choose([f'In {n}, the ones digit is {ones}.', f'In {n}, the tens digit is {tens}.', f'In {n}, the hundreds digit is {hundreds}.', f'In {n}, the thousands digit is {thousands}.', f'{n} = {thousands} thousands + {hundreds} hundreds + {tens} tens + {ones} ones.']), 'place_value'
|
| 54 |
+
|
| 55 |
+
def gen_missing():
|
| 56 |
+
a = rng.randint(0,100); x = rng.randint(0,100); b = a+x
|
| 57 |
+
if rng.random()<0.5:
|
| 58 |
+
return f'Solve x + {a} = {b}. x = {x}. Check: {x} + {a} = {b}.', 'missing_number'
|
| 59 |
+
return f'Solve {b} - x = {a}. x = {x}. Check: {b} - {x} = {a}.', 'missing_number'
|
| 60 |
+
|
| 61 |
+
def gen_word():
|
| 62 |
+
name=choose(names); obj=choose(objects); a=rng.randint(0,30); b=rng.randint(0,30)
|
| 63 |
+
if rng.random()<0.55:
|
| 64 |
+
return f'{name} has {a} {obj} and gets {b} more {obj}. {name} has {a+b} {obj}.', 'word_addition'
|
| 65 |
+
if b>a: a,b=b,a
|
| 66 |
+
return f'{name} has {a} {obj} and gives away {b}. {name} has {a-b} {obj} left.', 'word_subtraction'
|
| 67 |
+
|
| 68 |
+
def gen_multistep():
|
| 69 |
+
a,b,c = rng.randint(0,50), rng.randint(0,50), rng.randint(0,50)
|
| 70 |
+
if rng.random()<0.5:
|
| 71 |
+
r=a+b-c
|
| 72 |
+
return f'{a} + {b} - {c} = {r}. First {a} + {b} = {a+b}; then {a+b} - {c} = {r}.', 'two_step'
|
| 73 |
+
r=a*b+c
|
| 74 |
+
return f'{a} * {b} + {c} = {r}. First {a} * {b} = {a*b}; then {a*b} + {c} = {r}.', 'two_step'
|
| 75 |
+
|
| 76 |
+
def gen_fraction():
|
| 77 |
+
d=rng.randint(2,12); n=rng.randint(1,d-1); k=rng.randint(2,10)
|
| 78 |
+
return f'{n}/{d} = {n*k}/{d*k}. Multiplying numerator and denominator by {k} makes an equivalent fraction.', 'fractions'
|
| 79 |
+
|
| 80 |
+
def gen_money():
|
| 81 |
+
cents=rng.randint(0,9999); dollars=cents//100; rem=cents%100
|
| 82 |
+
return f'{cents} cents = {dollars} dollars and {rem} cents.', 'money'
|
| 83 |
+
|
| 84 |
+
gens = [gen_add, lambda: gen_add(True), gen_sub, gen_mul, gen_div, gen_compare, gen_order_parity, gen_place, gen_missing, gen_word, gen_multistep, gen_fraction, gen_money]
|
| 85 |
+
weights = [0.18,0.07,0.14,0.14,0.08,0.08,0.08,0.06,0.07,0.07,0.05,0.04,0.04]
|
| 86 |
+
|
| 87 |
+
def weighted_gen():
|
| 88 |
+
total = sum(weights)
|
| 89 |
+
r=rng.random() * total; acc=0.0
|
| 90 |
+
for g,w in zip(gens,weights):
|
| 91 |
+
acc += w
|
| 92 |
+
if r <= acc: return g()
|
| 93 |
+
return gens[-1]()
|
| 94 |
+
|
| 95 |
+
start=time.time(); sha=hashlib.sha256(); rows=0
|
| 96 |
+
with OUT.open('w', encoding='utf-8') as f:
|
| 97 |
+
# Exhaustive core facts first, repeated in clean forms.
|
| 98 |
+
for a in range(0,101):
|
| 99 |
+
for b in range(0,101):
|
| 100 |
+
jwrite(f, f'{a} + {b} = {a+b}.', 'addition_table'); rows+=1
|
| 101 |
+
if a >= b:
|
| 102 |
+
jwrite(f, f'{a} - {b} = {a-b}.', 'subtraction_table'); rows+=1
|
| 103 |
+
for a in range(0,21):
|
| 104 |
+
for b in range(0,21):
|
| 105 |
+
jwrite(f, f'{a} * {b} = {a*b}.', 'multiplication_table'); rows+=1
|
| 106 |
+
if b != 0:
|
| 107 |
+
jwrite(f, f'{a*b} / {b} = {a}.', 'division_table'); rows+=1
|
| 108 |
+
while rows < N:
|
| 109 |
+
text, kind = weighted_gen()
|
| 110 |
+
jwrite(f, text, kind); rows += 1
|
| 111 |
+
if rows % 250000 == 0:
|
| 112 |
+
print(f'generated {rows:,}', flush=True)
|
| 113 |
+
# Hash after close.
|
| 114 |
+
with OUT.open('rb') as f:
|
| 115 |
+
for chunk in iter(lambda: f.read(1<<20), b''):
|
| 116 |
+
sha.update(chunk)
|
| 117 |
+
meta = {'name':'agillm-math-numeracy-synthetic','rows':rows,'seed':SEED,'sha256':sha.hexdigest(),'counts':counts,'created_unix':int(time.time()),'description':'Synthetic numeracy pretraining JSONL with exact arithmetic facts, comparisons, place value, word problems, and simple step-by-step arithmetic.'}
|
| 118 |
+
META.write_text(json.dumps(meta, indent=2, sort_keys=True) + '\n')
|
| 119 |
+
README.write_text('''---\nlicense: apache-2.0\ntags:\n- synthetic\n- math\n- numeracy\n- pretraining\n---\n\n# AGILLM Math Numeracy Synthetic\n\nSynthetic plain-text numeracy corpus for AGILLM pretraining. Each JSONL row has `text` and `kind`.\n\nCoverage includes addition/subtraction/multiplication/division facts, multi-digit arithmetic, comparisons, order/parity, place value, missing-number equations, simple word problems, fractions, money, and two-step arithmetic.\n\nGenerated deterministically with `generate_math_numeracy.py`.\n''')
|
| 120 |
+
print(json.dumps(meta, indent=2, sort_keys=True))
|
| 121 |
+
print(f'elapsed_sec={time.time()-start:.1f}')
|
metadata.json
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"counts": {
|
| 3 |
+
"addition": 324553,
|
| 4 |
+
"addition_table": 10201,
|
| 5 |
+
"comparison": 144775,
|
| 6 |
+
"division_table": 420,
|
| 7 |
+
"exact_division": 144311,
|
| 8 |
+
"fractions": 72327,
|
| 9 |
+
"missing_number": 126466,
|
| 10 |
+
"money": 71675,
|
| 11 |
+
"multi_digit_addition": 125494,
|
| 12 |
+
"multiplication": 253019,
|
| 13 |
+
"multiplication_table": 441,
|
| 14 |
+
"number_order": 72122,
|
| 15 |
+
"parity": 71839,
|
| 16 |
+
"place_value": 108699,
|
| 17 |
+
"subtraction": 251795,
|
| 18 |
+
"subtraction_table": 5151,
|
| 19 |
+
"two_step": 90161,
|
| 20 |
+
"word_addition": 69737,
|
| 21 |
+
"word_subtraction": 56814
|
| 22 |
+
},
|
| 23 |
+
"created_unix": 1780047216,
|
| 24 |
+
"description": "Synthetic numeracy pretraining JSONL with exact arithmetic facts, comparisons, place value, word problems, and simple step-by-step arithmetic.",
|
| 25 |
+
"name": "agillm-math-numeracy-synthetic",
|
| 26 |
+
"rows": 2000000,
|
| 27 |
+
"seed": 731095,
|
| 28 |
+
"sha256": "11e893285638c733c7068510f07011273ac856586f2aa1239a7d526533a24280"
|
| 29 |
+
}
|
train.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:11e893285638c733c7068510f07011273ac856586f2aa1239a7d526533a24280
|
| 3 |
+
size 132737866
|