john-broadway/rys-sovereign-collection-v2
Viewer • Updated • 1.16k • 159
RYS Sovereign Collection v2
Qwen3-family cohort (0.6B/1.7B/8B) + cross-architecture sweep corpus (21 variants, 10 families). v1 collection: Qwen2.5 cross-scale + Qwen3-32B.
Note v2 cross-architecture sweep corpus (21 model variants, 10 architecture families)
john-broadway/Qwen3-0.6B-RYS-10-13-GGUF
0.6B • Updated • 207Note v2 Qwen3-family cohort — 0.6B
john-broadway/Qwen3-1.7B-RYS-7-10-GGUF
2B • Updated • 211Note v2 Qwen3-family cohort — 1.7B
john-broadway/Qwen3-8B-RYS-16-19-GGUF
9B • Updated • 221Note v2 Qwen3-family cohort — 8B
john-broadway/SmolLM2-1.7B-RYS-eval
UpdatedNote Eval-only — first published negative result (0 of 30 configs boost reasoning)
john-broadway/Llama-3.2-1B-RYS-10-13-GGUF
1B • Updated • 198Note Deployable weights — 0%→64.71% reasoning (best-combined); +76.47% peak across full sweep
john-broadway/Granite-3.1-1B-A400M-RYS-12-15-GGUF
1B • Updated • 144Note Deployable weights — MoE +52.94% reasoning (EQ trade-off uniform)
john-broadway/Qwen2.5-Coder-1.5B-RYS-4-9-GGUF
2B • Updated • 212Note Deployable weights — code-specialization unlock +29.41% reasoning
john-broadway/TinyLlama-1.1B-RYS-10-14-GGUF
1B • Updated • 164Note Deployable weights — +47.85 EQ (corpus extreme; weakest-dimension recovery)
john-broadway/Llama-3.2-3B-RYS-21-24-GGUF
4B • Updated • 190Note Deployable weights — +14.32 math (biggest math lift in corpus)
john-broadway/Llama-3.1-8B-RYS-18-22-GGUF
9B • Updated • 193Note Deployable weights — +11.76 reasoning across 15 boosting configs (richer multi-circuit)
john-broadway/Mistral-7B-v0.3-RYS-18-23-GGUF
8B • Updated • 172Note Deployable weights — +17.65 reasoning (same circuit-position as Llama-3.1-8B, weaker baseline)
john-broadway/Granite-3.1-2B-RYS-19-24-GGUF
3B • Updated • 150Note Deployable weights — dense control +11.76 reasoning, EQ stable
john-broadway/Gemma-2-2B-RYS-19-23-GGUF
3B • Updated • 190Note Deployable weights — smaller-Gemma scaling control +5.88 reasoning
john-broadway/Gemma-2-9B-RYS-14-18-GGUF
10B • Updated • 194Note Deployable weights — early-layer reasoning L14 +23.53, EQ stable, highest baseline EQ
john-broadway/Yi-1.5-6B-RYS-20-23-GGUF
7B • Updated • 154Note Deployable weights — rare three-way positive lift (math + EQ + reasoning)
john-broadway/SmolLM2-135M-RYS-18-22-GGUF
0.1B • Updated • 173Note Deployable weights — tiniest in corpus +17.65 reasoning + +13.05 EQ
john-broadway/SmolLM2-360M-RYS-12-15-GGUF
0.4B • Updated • 165Note Deployable weights — +23.53 reasoning + +8.98 math (full block search)