Outlier Research
Collection
Ternary MoE overlays on Qwen2.5. 10B/40B/70B/150B scales. V3.3 active, V3.2/V2 archived. MMLU verified at n=14,042. • 8 items • Updated
⚠️ SUPERSEDED — use Outlier-Ai/Outlier-10B instead. These weights are retained live for reproducibility of earlier benchmark runs. All current research has moved to the successor.
Outlier 10B V3.2 (Superseded)
Earlier ternary MoE overlay on frozen Qwen2.5-7B-Instruct. Superseded by Outlier-10B (V3.3 alpha-fixed).
- Scale: 30.4B / 13.3B active
- Base (frozen): Qwen/Qwen2.5-7B-Instruct
- Status: archival — do not use for new work
What changed
V3.2 reported MMLU 76.19% was a smoke-test artifact. V3.3 canonical re-measurement (n=14,042, lm-eval-harness v0.4.9.1) is 70.87% ±0.37% with alpha-fix training applied.
Historical benchmark (reference only)
| Benchmark | Score | Notes |
|---|---|---|
| MMLU 5-shot | 76.19% (smoke-test, n<100) | Pre-V3.3 measurement |
Use the successor
- Current: Outlier-Ai/Outlier-10B
- Research collection: Outlier Research
Why this is still public
ML research norms: earlier checkpoints stay live so external benchmarks and papers that cite this URL remain reproducible. This is not dead weight — it's the historical record.
Related
- Outlier desktop app: outlier.host — v1.4 shipping
- Discord: discord.gg/Hapennmdn9
Patents
Architecture covered by US provisional patents 64/026,886, 64/030,368, 64/034,028 (Kerr & Company LLC, 2026).
- Downloads last month
- 2,204