Open to Collab

John Broadway

john-broadway

https://john-broadway.github.io

AI & ML interests

Independent AI researcher. Sovereign models on bare metal. RYS layer duplication — train-free reasoning-circuit discovery across 21 model variants and 10 architecture families. Retired disabled U.S. veteran building from home. The work speaks for itself.

Recent Activity

new activity 4 days ago

TinyLlama/TinyLlama-1.1B-Chat-v1.0:Cross-architecture RYS sweep — TinyLlama-1.1B-Chat-v1.0 (EQ-amplifier extreme +47.85; refines hypothesis to weakest-dimension)

new activity 4 days ago

01-ai/Yi-1.5-6B-Chat:Cross-architecture RYS sweep — Yi-1.5-6B-Chat (rare three-way positive lift across math + EQ + reasoning)

new activity 4 days ago

google/gemma-2-9b-it:Cross-architecture RYS sweep — gemma-2-9b-it (early-layer reasoning circuit L14; highest baseline EQ in corpus)

View all activity

Organizations

None yet

New activity in TinyLlama/TinyLlama-1.1B-Chat-v1.0 4 days ago

Cross-architecture RYS sweep — TinyLlama-1.1B-Chat-v1.0 (EQ-amplifier extreme +47.85; refines hypothesis to weakest-dimension)

#64 opened 4 days ago by

john-broadway

New activity in 01-ai/Yi-1.5-6B-Chat 4 days ago

Cross-architecture RYS sweep — Yi-1.5-6B-Chat (rare three-way positive lift across math + EQ + reasoning)

#5 opened 4 days ago by

john-broadway

New activity in google/gemma-2-9b-it 4 days ago

Cross-architecture RYS sweep — gemma-2-9b-it (early-layer reasoning circuit L14; highest baseline EQ in corpus)

#69 opened 4 days ago by

john-broadway

New activity in mistralai/Mistral-7B-Instruct-v0.3 4 days ago

Cross-architecture RYS sweep — Mistral-7B-Instruct-v0.3 (same circuit-position as Llama-3.1-8B; weaker baseline → 28 boosters)

#207 opened 4 days ago by

john-broadway

New activity in Qwen/Qwen2.5-Coder-1.5B-Instruct 4 days ago

Cross-architecture RYS sweep — Qwen2.5-Coder-1.5B-Instruct (+35.29% reasoning, specialization unlock)

#7 opened 4 days ago by

john-broadway

New activity in ibm-granite/granite-3.1-1b-a400m-instruct 4 days ago

Cross-architecture RYS sweep — Granite-3.1-1B-A400M-Instruct, first MoE result (+52.94% reasoning)

#2 opened 4 days ago by

john-broadway

New activity in meta-llama/Llama-3.2-3B-Instruct 4 days ago

Cross-architecture RYS sweep — Llama-3.2-3B-Instruct (biggest math lift in corpus +14.32; family spans entire baseline curve)

#282 opened 4 days ago by

john-broadway

New activity in meta-llama/Llama-3.1-8B-Instruct 4 days ago

Cross-architecture RYS sweep — Llama-3.1-8B-Instruct (richer multi-circuit than Qwen2.5-7B at same scale; 15 of 66 boosters)

#370 opened 4 days ago by

john-broadway

New activity in meta-llama/Llama-3.2-1B-Instruct 4 days ago

Cross-architecture RYS sweep — Llama-3.2-1B-Instruct lifts reasoning 0%→76%

#174 opened 4 days ago by

john-broadway

New activity in HuggingFaceTB/SmolLM2-1.7B-Instruct 4 days ago

Cross-architecture RYS sweep — SmolLM2-1.7B-Instruct (negative result; 135M and 360M siblings respond normally)

#33 opened 4 days ago by

john-broadway

New activity in Qwen/Qwen2.5-Coder-1.5B-Instruct 4 days ago

Cross-architecture RYS sweep — Qwen2.5-Coder-1.5B-Instruct (+35.29% reasoning, specialization unlock)

#7 opened 4 days ago by

john-broadway

New activity in ibm-granite/granite-3.1-1b-a400m-instruct 4 days ago

Cross-architecture RYS sweep — Granite-3.1-1B-A400M-Instruct, first MoE result (+52.94% reasoning)

#2 opened 4 days ago by

john-broadway

New activity in meta-llama/Llama-3.2-1B-Instruct 4 days ago

Cross-architecture RYS sweep — Llama-3.2-1B-Instruct lifts reasoning 0%→76%

#174 opened 4 days ago by

john-broadway

John Broadway

AI & ML interests

Recent Activity

Organizations

john-broadway's activity

Cross-architecture RYS sweep — TinyLlama-1.1B-Chat-v1.0 (EQ-amplifier extreme +47.85; refines hypothesis to weakest-dimension)

Cross-architecture RYS sweep — Yi-1.5-6B-Chat (rare three-way positive lift across math + EQ + reasoning)

Cross-architecture RYS sweep — gemma-2-9b-it (early-layer reasoning circuit L14; highest baseline EQ in corpus)

Cross-architecture RYS sweep — Mistral-7B-Instruct-v0.3 (same circuit-position as Llama-3.1-8B; weaker baseline → 28 boosters)

Cross-architecture RYS sweep — Qwen2.5-Coder-1.5B-Instruct (+35.29% reasoning, specialization unlock)

Cross-architecture RYS sweep — Granite-3.1-1B-A400M-Instruct, first MoE result (+52.94% reasoning)

Cross-architecture RYS sweep — Llama-3.2-3B-Instruct (biggest math lift in corpus +14.32; family spans entire baseline curve)

Cross-architecture RYS sweep — Llama-3.1-8B-Instruct (richer multi-circuit than Qwen2.5-7B at same scale; 15 of 66 boosters)

Cross-architecture RYS sweep — Llama-3.2-1B-Instruct lifts reasoning 0%→76%

Cross-architecture RYS sweep — SmolLM2-1.7B-Instruct (negative result; 135M and 360M siblings respond normally)

Cross-architecture RYS sweep — Qwen2.5-Coder-1.5B-Instruct (+35.29% reasoning, specialization unlock)

Cross-architecture RYS sweep — Granite-3.1-1B-A400M-Instruct, first MoE result (+52.94% reasoning)

Cross-architecture RYS sweep — Llama-3.2-1B-Instruct lifts reasoning 0%→76%