Model card · Pipeline and preflight stage — no trained model
quantum-1-echelon
—Parameter size
Final parameter size not yet defined
Configured architecture
Drawn from the committed preflight report. No Echelon model has been trained.
Summary
The current strategic Quantum model line. Public evidence covers an architecture preflight, a validated tokenizer and Garden data-pipeline smoke tests; no Echelon model has been trained or released.
- Status
- Pipeline and preflight stage — no trained model
- Model type
- In-development German-language model line
- Intended use
- Reproducible German-language base-model research · A later chat stage within the same Echelon model line
- Languages
- German-language focus; the production corpus is not complete
- Lineage
- Public path configuration treats quantum-1-echelon-base and quantum-1-echelon-chat as stages or variants within the same quantum-1-echelon model line and forbids reuse of the pilot models, tokenizers and data.
- Release status
- No trained Echelon weights, checkpoints, GGUF files or Hugging Face model repository are publicly available.
- License
- No model weights have been released. A license for any future quantum-1-echelon weights has not been documented.
Technical dossier
Documented model facts.
Information linked to public project sources. Unknown release dates, licensing terms and unsupported client compatibility are not inferred.
- Architecture status
- Configuration preflight only. The proposed Echelon Base configuration specifies 26 layers, hidden size 1,280, intermediate size 3,584, 20 attention heads and 5 KV heads. The committed preflight computes 506,333,440 trainable parameters; none of this is evidence of trained weights.
- Configured context
- 2,048 tokens in the architecture and Garden pipeline configurations; not validated on a trained Echelon model.
- Tokenizer
- SentencePiece BPE with a configured 32,768-token vocabulary. The committed validation report records 23 round-trip cases and 0 failures.
- Garden smoke test
- 5,001 documents seen, 1,559 accepted and 1,380,886 tokens produced. These are smoke-test results, not production-dataset totals.
- Production data run
- Not yet available. The committed Phase 3 report states that the full production run had not started.
- Model training
- Not yet available. No completed Echelon training is claimed.
- Evaluation
- Not yet available. No Echelon model outputs, loss curves or benchmark results are published.
Primary sources:Architecture configuration ·Architecture preflight report ·Tokenizer validation report ·Garden Phase 3 report ·Model-line path configuration
Research context
How this release fits the experiment.
The published work currently establishes pipeline readiness, artifact integrity and configuration boundaries. Echelon Base is the base-training stage; Echelon Chat is a later stage or variant in the same line. Neither is a separate released model family, and neither has public trained weights.
RELATED RESOURCES
Inspect the wider evidence.
Limitations
Read the limits first.
The proposed architecture has passed a configuration preflight only
The full Garden production-data run has not been published as completed
No Echelon model training logs, checkpoints or weights are public
No Echelon model outputs or benchmark results are available
Training hardware, runtime, cost and energy use are not published
A model license is not yet available because no model release exists
Not available for production use