A layer-duplication ("RYS" — Repeat Your Self, David Ng) variant of Qwen/Qwen2.5-0.5B-Instruct: layers 3–6 duplicated, 24 → 28 layers. No training, merging, or weight changes. GGUF (imatrix Q-quants).

⚠️ Evaluation status — please read (updated 2026-06)

On a controlled re-test, the headline gains for this model do not hold up. This card is being corrected to say so.

The original card reported reasoning 35.29% → 58.82% (+23.53%) and an EQ lift (37.38 → 55.23), calling this an "EQ specialist." A re-run on a current llama.cpp build found:

Reasoning does not reproduce. The baseline reproduces exactly (35.29%), but the (3,7) config scored 29.41% — below the baseline. The original +23.5% was probe variance (the reasoning probe is 17 questions, so each one is ±5.9%).
The EQ number is not a reliable capability signal. On output inspection, the EQ probe on this model is satisfied by emitting a near-constant rating vector across different scenarios — it outscores the base model's degenerate output without actually reading the scenario. So the "EQ lift" reflects the scoring metric, not emotional understanding.

These scores come from a lightweight search probe (16 math / 16 EQ / 17 reasoning questions) used to find productive layer blocks, not a validated benchmark. A standard-harness re-eval (lm-eval-harness / EQ-Bench at full N) is in progress.

Bottom line: treat this as a normal Qwen2.5-0.5B-Instruct with layers 3–6 duplicated. Published for transparency of the RYS sweep — not as an established "EQ specialist."

Original sweep numbers (search probe — kept for the record)

probe	reported baseline	reported (3,7)	re-test note
Reasoning (17 q)	35.29%	58.82%	did not reproduce — (3,7) → 29.41% (below baseline)
EQ (16 q)	37.38	55.23	EQ probe gamed by a constant guess; not validated EQ
Math (16 q)	0.492	0.424	search-probe score

Run it

llama-server -m Qwen2.5-0.5B-RYS-3-7-Q4_K_M.gguf -ngl 99

Method · data · attribution

Method: layer duplication — Repeat Your Self (David Ng); toolkit llm-circuit-finder (alainnothere).
Raw sweep data: rys-sovereign-collection-v2.
Built by John Broadway with Claude. The method and the raw data are real; the interpretation of these probe deltas as capability is what this update corrects.

License

Apache-2.0.

john-broadway/Qwen2.5-0.5B-RYS-3-7-GGUF

Qwen2.5-0.5B-RYS-3-7-GGUF

⚠️ Evaluation status — please read (updated 2026-06)

Original sweep numbers (search probe — kept for the record)

Run it

Method · data · attribution

License

john-broadway/
Qwen2.5-0.5B-RYS-3-7-GGUF