REPRODUCIBILITY
Artifacts that can be inspected—and gaps that cannot be replayed.
Release integrity is public for the two pilot GGUF files. Full training reproducibility remains partial because final manifests, logs, environments and raw evaluations are missing.
PUBLIC F16 GGUF
Publishedquantum-1-pilot
- File
- quantum-1-base-v1.0.0-f16.gguf
- Size
- 98,990,560 bytes
- SHA-256
- aeab97e50a5789772b69cf1554ba74eb915b5621835d80d40785b473b62fd1a5
PUBLIC F16 GGUF
Publishedquantum-1.6-pilot
- File
- quantum-1.6-pilot-v1.6.0-f16.gguf
- Size
- 98,990,560 bytes
- SHA-256
- 6bda15fcd51286e55174d5876fe44aa9518fb18b75fb5aa4f7402ebd039bd994
LLAMA.CPP REFERENCE
llama-cli -m quantum-1.6-pilot-v1.6.0-f16.gguf -p "Berlin ist" -n 64 --seed 20260705 --temp 0 --top-p 1 --top-k 0 --no-cnvThis command mirrors the pinned diagnosis configuration. The website test suite does not execute model inference. A local llama.cpp build is required, and the exact external llama.cpp revision cannot be initialized from the reviewed repository because its Git link has no .gitmodules entry.
VERIFICATION WORKFLOW
A repeatable local record.
- 01
Download the GGUF from the pinned Hugging Face repository revision.
- 02
Compare the downloaded file against the published SHA-256 value.
- 03
Use completion mode rather than assuming a chat template.
- 04
Record the llama.cpp revision, hardware and full command before comparing output.
- 05
Preserve raw output alongside the prompt, seed and decoding settings.
| Area | Status | Available | Still missing |
|---|---|---|---|
| Pilot GGUF integrity | Published | Pinned files, manifests, sizes and SHA-256 checksums. | Artifact licenses remain unstated. |
| Pilot training method | Partial evidence | Code, documentation and YAML configurations. | Final run manifests, exact environments, completed-step records and complete logs. |
| Pilot evaluation | Partial evidence | Reported held-out metrics, fixed prompt data and a diagnosis method. | Versioned raw outputs, independently replayable metrics and standardized task reports. |
| Echelon tokenizer | Published | Configuration, checksums, quality summary and 23-case round-trip report. | No model-quality conclusion follows from tokenizer validation. |
| Echelon production and training | Not yet available | Production configuration and pipeline smoke evidence. | Production manifest, checkpoints, weights, outputs and evaluation. |
| Resource measurements | Not measured | No committed measurement report. | Hardware, runtime, memory, throughput, cost and energy measurements. |
DOCUMENTATION DIRECTORY
Continue through the evidence.
Every section keeps public evidence, configured targets and missing information visibly separate.