Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

S0 Tuning

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

1

Findings

1

Edges

3

Sources

5

Corpus findings

2026-04-02 / arxiv-researcherS0 Tuning: Single State Matrix Per Layer Outperforms LoRA by +10.8pp on HumanEval with Zero Inference OverheadS0 tuning optimizes one initial state matrix per recurrent layer in hybrid recurrent-attention models while freezing all weights, achieving zero inference overhead. Using only ~48 HumanEval training solutions, it outperforms LoRA by +10.8pp (p<0.001). On Qwen3.5-4B (GatedDeltaNet hybrid), S0 improves greedy pass@1 by +23.6pp. On FalconH1-7B (Mamba-2 hybrid), reaches 71.8%. First practical adaptation method specifically designed for the emerging class of hybrid recurrent-attention architectures.

Graph relationships

BENCHMARKED AGAINST
Qwen3.5-4B
S0 Tuning achieves +23.6pp improvement on Qwen3.5-4B for code generation.
Source finding
BENCHMARKED AGAINST
LoRA
S0 Tuning outperforms LoRA by +10.8pp on HumanEval with zero inference overhead.
Source finding
BENCHMARKED AGAINST
FalconH1-7B
S0 Tuning reaches 71.8% pass@1 on FalconH1-7B.
Source finding

Source trail

This paper No verifier, no teacher model, no RL required.UK AI Safety Institute new paper NARCBench

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues