🤝 Open to Collab

Samuel Bartlett PRO

sbartlett97

8 19

webbrain-one-451498's profile picture

Siriusda's profile picture

21world's profile picture

https://sbartlett97.github.io/

sbartlett97
samuel-bartlett-ai

AI & ML interests

Transformers for text, audio, images.... pretty much anything really. I enjoy trying to push the limits of what I know and what a lot of people take for granted, tinkering with how models are put together and exploring different ideas/directions on top of what we already have.

Recent Activity

updated a collection 1 day ago

SOT phase1 — wikitext103

updated a collection 1 day ago

SOT phase1 — wikitext103

published a model 1 day ago

sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v9-zeroeffect

View all activity

Organizations

None yet

sbartlett97 's collections 6

SOT phase2 — wikitext103

Phase 2 birth/death, d768, attempts 4-9. 4-8: supernet ppl (8: 132 present-only). 9: zero-effect, 30.61 ppl at 0.425 dense FLOPs.

sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v4

Text Generation • 0.2B • Updated 1 day ago • 229
sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v5-imp

Text Generation • 0.2B • Updated 1 day ago • 230
sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v6-allfixes

Text Generation • 0.2B • Updated 1 day ago • 231
sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v7-lrfreeze

Text Generation • 0.2B • Updated 1 day ago • 231

SOT phase1 — wikitext103

Phase 1 gates on WikiText-103: d384 (2 budgets) and d768. SOT trails dense 3% at d384, 0.1% at d768. Retrained structure loses.

sbartlett97/sot-phase1-wikitext103-dense-seed42-tc017

Text Generation • 47.7M • Updated 1 day ago • 161
sbartlett97/sot-phase1-wikitext103-dense_8l-seed42-tc017

Text Generation • 38.3M • Updated 1 day ago • 193
sbartlett97/sot-phase1-wikitext103-dense_half_ffn-seed42-tc017

Text Generation • 37.1M • Updated 1 day ago • 201
sbartlett97/sot-phase1-wikitext103-sot-seed42-tc017

Text Generation • 47.7M • Updated 1 day ago • 219

Back to Basics

A collection of "Back to Basics" models/aretfacts

sbartlett97/test_eli5_clm-model

Text Generation • 81.9M • Updated Dec 14, 2025 • 42
sbartlett97/distilbert-imdb-finetuned

Text Classification • 67M • Updated Dec 20, 2025 • 15
sbartlett97/distilbert-ner-finetune

Token Classification • 66.4M • Updated Dec 20, 2025 • 8

SOT phase1 — tinystories

Phase 1 gates, d384, 3 seeds x 6 conditions. SOT trails dense ~2.6% ppl. Gate cost is a proxy, not compute. Start: dense/sot seed42.

sbartlett97/sot-phase1-tinystories-dense-seed42

Text Generation • 47.7M • Updated 1 day ago • 69
sbartlett97/sot-phase1-tinystories-dense_8l-seed42

Text Generation • 38.3M • Updated 1 day ago • 70
sbartlett97/sot-phase1-tinystories-dense_half_ffn-seed42

Text Generation • 37.1M • Updated 1 day ago • 83
sbartlett97/sot-phase1-tinystories-sot-seed42

Text Generation • 47.7M • Updated 1 day ago • 119

Experiments

A collection of (probably useless) model experiments, where I tinker with existing architectures or sometimes try my hand at something completely new.

sbartlett97/gqa-opus-mt-de-en

Text Generation • 77.4M • Updated Jan 6 • 233

Interesting Papers

Papers that I find interesting

Improving Rule-based Reasoning in LLMs using Neurosymbolic Representations

Paper • 2502.01657 • Published Jan 31, 2025 • 4
Stronger Normalization-Free Transformers

Paper • 2512.10938 • Published Dec 11, 2025 • 25
The FACTS Leaderboard: A Comprehensive Benchmark for Large Language Model Factuality

Paper • 2512.10791 • Published Dec 11, 2025 • 8
TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration

Paper • 2512.14550 • Published Dec 16, 2025 • 6

SOT phase2 — wikitext103

Phase 2 birth/death, d768, attempts 4-9. 4-8: supernet ppl (8: 132 present-only). 9: zero-effect, 30.61 ppl at 0.425 dense FLOPs.

sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v4

Text Generation • 0.2B • Updated 1 day ago • 229
sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v5-imp

Text Generation • 0.2B • Updated 1 day ago • 230
sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v6-allfixes

Text Generation • 0.2B • Updated 1 day ago • 231
sbartlett97/sot-phase2-wikitext103-heuristic-seed42-tc768d013-v7-lrfreeze

Text Generation • 0.2B • Updated 1 day ago • 231

SOT phase1 — tinystories

Phase 1 gates, d384, 3 seeds x 6 conditions. SOT trails dense ~2.6% ppl. Gate cost is a proxy, not compute. Start: dense/sot seed42.

sbartlett97/sot-phase1-tinystories-dense-seed42

Text Generation • 47.7M • Updated 1 day ago • 69
sbartlett97/sot-phase1-tinystories-dense_8l-seed42

Text Generation • 38.3M • Updated 1 day ago • 70
sbartlett97/sot-phase1-tinystories-dense_half_ffn-seed42

Text Generation • 37.1M • Updated 1 day ago • 83
sbartlett97/sot-phase1-tinystories-sot-seed42

Text Generation • 47.7M • Updated 1 day ago • 119

SOT phase1 — wikitext103

Phase 1 gates on WikiText-103: d384 (2 budgets) and d768. SOT trails dense 3% at d384, 0.1% at d768. Retrained structure loses.

sbartlett97/sot-phase1-wikitext103-dense-seed42-tc017

Text Generation • 47.7M • Updated 1 day ago • 161
sbartlett97/sot-phase1-wikitext103-dense_8l-seed42-tc017

Text Generation • 38.3M • Updated 1 day ago • 193
sbartlett97/sot-phase1-wikitext103-dense_half_ffn-seed42-tc017

Text Generation • 37.1M • Updated 1 day ago • 201
sbartlett97/sot-phase1-wikitext103-sot-seed42-tc017

Text Generation • 47.7M • Updated 1 day ago • 219

Experiments

A collection of (probably useless) model experiments, where I tinker with existing architectures or sometimes try my hand at something completely new.

sbartlett97/gqa-opus-mt-de-en

Text Generation • 77.4M • Updated Jan 6 • 233

Back to Basics

A collection of "Back to Basics" models/aretfacts

sbartlett97/test_eli5_clm-model

Text Generation • 81.9M • Updated Dec 14, 2025 • 42
sbartlett97/distilbert-imdb-finetuned

Text Classification • 67M • Updated Dec 20, 2025 • 15
sbartlett97/distilbert-ner-finetune

Token Classification • 66.4M • Updated Dec 20, 2025 • 8

Interesting Papers

Papers that I find interesting

Improving Rule-based Reasoning in LLMs using Neurosymbolic Representations

Paper • 2502.01657 • Published Jan 31, 2025 • 4
Stronger Normalization-Free Transformers

Paper • 2512.10938 • Published Dec 11, 2025 • 25
The FACTS Leaderboard: A Comprehensive Benchmark for Large Language Model Factuality

Paper • 2512.10791 • Published Dec 11, 2025 • 8
TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration

Paper • 2512.14550 • Published Dec 16, 2025 • 6