Modelpublic

merileijona/quantumgpt-354m-finetuned

sourceHugging Facemitupdated 1mo agoView on Hugging Face

0likes241downloads

Model Card

QuantumGPT-354M (fine-tuned, format-anchored)

The 354M sibling of QuantumGPT-124M-v3: GPT-2 medium architecture (24 layers, 16 heads, 1024 embd, natively 512-token context) fine-tuned with the same format-anchor recipe that produced the v3 champion.

Corrected QSS-Core v0.1: 97.52 / 96.45 (two-seed replication) — final scores after correcting tokenizer stop-marker extraction and rerunning full robustness probes.

Benchmark results (QSemBench-dev-v0.1, n=120 stratified, greedy decode)

Scoring: qiskit qasm2 parse/compile + statevector fidelity ≥ 0.99 vs reference, entanglement checks, prompt-perturbation robustness probes. See TECHNICAL_REPORT.md in merileijona/quantumgpt-124m-v3 for full methodology.

Metric	Base 354M	Fine-tuned (seed 42)	Fine-tuned (seed 43)
qsyntax	100.00	100.00	100.00
qcompile	100.00	100.00	100.00
qsemantic	95.83	97.50	95.83
qexecute	95.83	97.50	95.83
qgeneralize	95.88	96.91	95.88
qrobust	95.65	94.87	95.65
qefficiency	100.00	100.00	100.00
QSS-Core v0.1	96.45	97.52	96.45

Final corrected replication: seed-42 fine-tuned model QSS 97.5235; seed-43 QSS 96.445. Spread: 1.079 QSS points. The evaluator strips tokenizer stop-marker drift before parsing.

Training

—Data: merileijona/quantum-circuits-21k (validated description→OpenQASM pairs), format-anchor oversampling ×2 (short perfectly-terminated circuits), long-circuit oversampling ×2, deterministic paraphrase augmentation — the recipe proven on the 124M lineage.
—Continuation training from the public base: lr 5e-6, cosine schedule, completion-only loss masking, effective batch 4 via gradient accumulation, one epoch per seed.
—Note: unlike the 124M v3, no architecture mutation was needed — this model already shipped with a 512-token context.

Family comparison (same benchmark)

Model	QSS-Core
quantumgpt-124m-v2 baseline	75.00
quantumgpt-354m base	96.45
quantumgpt-124m-v3 (ctx512)	97.70 / 97.28
quantumgpt-354m fine-tuned	97.52 / 96.45

The corrected seed-42 354M model scores 97.52, slightly below the 124M v3 seed-42 score of 97.70; both models are effectively near-ceiling on this development benchmark. The benchmark still shares its generator with training data.

Limitations

Same profile as the family: dev benchmark shares its generator with training data; negation constraints weakly followed; no refusal behavior; English only. Every generated circuit should be verified before execution.

Intended use

Research and education. Verify all outputs before running on real quantum hardware or simulators.