looped-51M-base

A 50,869,760-parameter English base model trained from random initialization in the Tiny llm ablation experiment.

Architecture and references

10 shared decoder layers applied 6 times (60 effective layers); width 512, SwiGLU 1792, GQA 8 query / 2 KV heads, RoPE, RMSNorm, tied embeddings and learned loop embeddings. Causal next-token cross-entropy. Parameters are shared across loops; compute is not matched to the other ablations. Context length: 2048. The architecture/tokenizer baseline is Q-50M-Base; these weights are not a fine-tune of that checkpoint. The unchanged tokenizer has 32,768 entries. Recurrent depth uses the general weight-sharing principle of Universal Transformers; this is a local fixed-six-loop implementation, not an exact reproduction of that architecture.

Training

Exactly 3,932,160,000 processed source tokens, 15,000 optimizer steps, FineWeb-Edu sample-10BT. This counts input blocks, not unique text or supervised target tokens. Effective batch: 4 × 32 accumulation × 2048 = 262,144 source tokens per step. Shuffle buffer 100,000; seed 2026.

BF16 compute / FP32 weights, one RTX 5070 Ti 16 GB. Fused AdamW, peak LR 0.001, 150-step linear warmup then cosine decay to 0.0001; betas (0.9, 0.95), weight decay 0.1 excluding bias/norm/1D parameters, gradient clipping 1.0. Exact configuration. Equal source-token budgets do not imply equal supervision or FLOPs.

Evaluation

Full splits, no added few-shot examples, lm-eval 0.4.12, BF16, RTX 5070 Ti, max context 2048 (ArithMark: 1024), no chat template. Protocol: autoregressive likelihood. Accuracy is percent; ± means one standard error; brackets are 95% Wilson intervals. LAMBADA requires every final-word token to match. acc_norm normalizes option likelihood by harness length. All metrics and provenance.

Dataset	Split	Examples	Metric	Score ± SE (%)	95% CI (%)
HellaSwag	validation	10,042	acc_norm	29.62 ± 0.46	[28.73, 30.52]
ARC-Easy	test	2,376	acc_norm	44.28 ± 1.02	[42.29, 46.28]
ARC-Challenge	test	1,172	acc_norm	22.10 ± 1.21	[19.82, 24.56]
PIQA	validation	1,838	acc_norm	60.28 ± 1.14	[58.03, 62.50]
WinoGrande	validation	1,267	acc	50.12 ± 1.41	[47.37, 52.87]
OpenBookQA	test	500	acc_norm	29.00 ± 2.03	[25.19, 33.13]
BoolQ	validation	3,270	acc	61.59 ± 0.85	[59.91, 63.24]
LAMBADA OpenAI	test	5,153	acc	20.90 ± 0.57	[19.81, 22.03]
ArithMark-3	train	1,000	acc_norm	35.70 ± 1.52	[32.79, 38.72]
Balanced COPA	train	1,000	acc	54.80 ± 1.57	[51.70, 57.86]
CommonsenseQA	validation	1,221	acc	19.66 ± 1.14	[17.52, 21.98]
SciQ (with support)	test	1,000	acc_norm	64.00 ± 1.52	[60.98, 66.92]
TruthfulQA MC2	validation	817	acc	42.43 ± 1.50	—
BananaMind Base 1.1	test	350	raw_accuracy	48.00 ± 2.67	[42.82, 53.23]
MMLU continuation	test	14,042	acc	25.67 ± 0.37	—
BLiMP	train	67,000	acc	75.84 ± 0.14	—

WikiText-2 raw test continuation: 291 nonoverlapping 1024-token blocks, 512 prefix + 512 scored suffix tokens; 148,992 scored tokens, 335 tail tokens excluded. GPU FP32 PPL 33.245 [31.978, 34.545], NLL 3.503892 [3.465063, 3.542263]. Percentile block bootstrap, 10,000 resamples, seed 2026. BF16 results are stored separately. This is conditional continuation PPL, not standard rolling or word PPL. Intervals do not capture training-seed variability or all within-document dependence.

model-index contains author-reported scores; no official leaderboard submission or verified badge is claimed.

Full selected splits; lm-eval 0.4.12; seed 1234; BF16 on RTX 5070 Ti; context cap 2048 (ArithMark 1024), TF32 disabled, no chat template and no added few-shot examples. TruthfulQA retains the harness's fixed six-QA preamble. ArithMark and BananaMind normalize by continuation token count; ordinary harness acc_norm uses its own length normalization. BananaMind is raw accuracy, not Elo. SciQ includes the support passage. Balanced COPA uses the mirrored 1000-item train-named evaluation split; cRia's split was inferred, not confirmed. MMLU scores full answer continuations across 57 subjects, weighted by item count; BLiMP averages 67 equal-sized minimal-pair subsets. Standard errors are retained from each evaluator. Wilson intervals are reported only where the runner logged binary item accuracy; MC2 is probability mass, not binary accuracy. These intervals do not model dependence between paired/templated examples or training seed variation. UL2, PrefixLM and experimental diffusion PLL use their documented conditional scoring protocols; PLL exposes the other answer tokens and is not autoregressive likelihood. cRia's published scores used a different precision and benchmark-adapted checkpoint; this completes our comparison coverage, not an independent reproduction of cRia or an official leaderboard submission.

Full results, provenance and group scores. Updated machine-readable results. TensorBoard events contain these new scores at step 15,000.

Usage

Install requirements.txt. Custom code requires trust_remote_code=True.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
repo = "d0rj/looped-51M-base"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, trust_remote_code=True).eval()
inputs = tokenizer("The purpose of science is", return_tensors="pt", add_special_tokens=False)

with torch.no_grad():
    output = model.generate(**inputs, max_new_tokens=32, do_sample=False)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Reproduce core scores from a downloaded repository after installing evaluation/requirements.txt:

python evaluation/run_core.py --device cuda:0 --dtype bfloat16 --batch-size 8 --output evaluation-rerun

To reproduce after downloading this model repository, accept the BananaMind dataset terms, authenticate with hf auth login, then run in a suitable CUDA environment:

pip install -r evaluation/comparison-20261001/repro/requirements.txt
python evaluation/comparison-20261001/repro/run.py --device cuda:0 --dtype bfloat16 --batch-size 8 --output comparison-rerun

The bundled runner uses the published model classes with the exact evaluation adapters and tokenizer. --limit produces smoke results only. Raw dataset examples are not included in this release.

TensorBoard and limitations

Training and evaluation event files contain 750 training-loss points (steps 20–15,000) and eval/ scalars at step 15,000, including CI bounds. Looped training resumed from a saved checkpoint; TensorBoard retains the valid history across the restart.

Small English base model, not instruction-tuned. Single training seed; benchmark contamination was not audited. Reference models with other tokenizers, prompts or data are not directly interchangeable. These checkpoints are research ablations.