CodeSOTA · Research question · RQ-0001Updated 2026-08-20
What we want to learn

Which evaluations provide useful signal for 25M–500M language models?

Which evaluation metrics provide non-saturated, reproducible learning signal for base language models between 25M and 500M parameters?

Status
active
Evidence
weak evidence
Graph
3 hypotheses · 2 experiments
Claims
2 public · 2 attachments
Reproductions
0 recorded

Training loss clearly moves at 25M parameters, but comparable downstream evidence across the 25M–500M range is still missing.

Why this question exists.

Large-model leaderboards often collapse small base models to chance. The Slayer small-model programme needs a ladder that exposes progress before MMLU-style accuracy becomes informative.

Why it matters. A scale-appropriate suite can decide which checkpoints deserve more compute and can prevent a genuine training improvement from being mistaken for no progress.

Keep the supported and refuted branches.

A negative result remains visible research memory. Each branch links its protocol and the claim the evidence supports or refutes.

H-0001testing

Byte-level loss is an early signal

Validation bits-per-byte changes smoothly before downstream exact-match benchmarks separate 25M-class models.

H-0002proposed

SciQ separates small models before MMLU

SciQ provides above-chance separation at smaller parameter scales than standard MMLU.

H-0003proposed

MMLU is floor-limited below 300M

Standard MMLU accuracy remains too close to chance to be a primary KPI for base models below roughly 300M parameters.

What the graph can currently say.

F-0001 · confirmed

Training signal precedes capability evidence

The 25M checkpoint demonstrates convergence, but it does not yet answer which downstream evaluations are informative at that scale.

Visible ownership.

Stable ID
RQ-0001
Created by
Kacper Wikiel
Contributors
Kacper Wikiel
Visibility
public