3.57 MB

Ctrl+K

1 contributor

History: 6 commits

Atharva1232

final report + talk+math ratio-curve results (7 configs)

eded994 verified 11 days ago

.gitattributes

1.63 kB
final verdict: smallest arithmetic LLM = pythia-14m 14.1M (99.8%); smallest from-scratch = 805,632 (100%); corrected param counts; full stage A/B records 12 days ago
README.md

3.67 kB
tokenizer ablation: digit-split pre-tok = 0.2% -> 99.7% on TinyStories-1M; smallest arithmetic LLM now 3.75M 12 days ago
REPORT.md

8.69 kB
final report + talk+math ratio-curve results (7 configs) 11 days ago
STATUS.md

14.5 kB
tokenizer ablation: digit-split pre-tok = 0.2% -> 99.7% on TinyStories-1M; smallest arithmetic LLM now 3.75M 12 days ago
VERDICT.md

7.13 kB
tokenizer ablation: digit-split pre-tok = 0.2% -> 99.7% on TinyStories-1M; smallest arithmetic LLM now 3.75M 12 days ago
bakeoff_results.jsonl

2.99 kB
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
best.pt

3.24 MB
xet

1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
curve.csv

486 Bytes
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
extrapolation.jsonl

2.57 kB
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
phase_diagram.png

109 kB
xet

1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
stage_a_results.jsonl

3.91 kB
final verdict: smallest arithmetic LLM = pythia-14m 14.1M (99.8%); smallest from-scratch = 805,632 (100%); corrected param counts; full stage A/B records 12 days ago
stage_b_all.jsonl

7.09 kB
final verdict: smallest arithmetic LLM = pythia-14m 14.1M (99.8%); smallest from-scratch = 805,632 (100%); corrected param counts; full stage A/B records 12 days ago
stage_b_round7_tokablation.jsonl

1.22 kB
tokenizer ablation: digit-split pre-tok = 0.2% -> 99.7% on TinyStories-1M; smallest arithmetic LLM now 3.75M 12 days ago
summary.json

297 Bytes
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
sweep_results.jsonl

5.33 kB
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
talkmath_A2_3to1_B2_lora.jsonl

4.03 kB
final report + talk+math ratio-curve results (7 configs) 11 days ago
talkmath_A3_7to1.jsonl

2.21 kB
final report + talk+math ratio-curve results (7 configs) 11 days ago
talkmath_A4_anneal.jsonl

1.92 kB
final report + talk+math ratio-curve results (7 configs) 11 days ago
talkmath_runA_1to1_runB_lora.jsonl

2.16 kB
final report + talk+math ratio-curve results (7 configs) 11 days ago
tokenizer_ablation_check.py

3.47 kB
tokenizer ablation: digit-split pre-tok = 0.2% -> 99.7% on TinyStories-1M; smallest arithmetic LLM now 3.75M 12 days ago
train.py

13.4 kB
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off) 13 days ago
verdict.json

163 Bytes
final verdict: smallest arithmetic LLM = pythia-14m 14.1M (99.8%); smallest from-scratch = 805,632 (100%); corrected param counts; full stage A/B records 12 days ago
verdict_chart.png

137 kB
xet

final verdict: smallest arithmetic LLM = pythia-14m 14.1M (99.8%); smallest from-scratch = 805,632 (100%); corrected param counts; full stage A/B records 12 days ago