Verifiable research
Open source.
Auditable evidence.
GitHub preserves the public ASM snapshot with implementation, architecture, protocols, benchmarks and limitations. Subsequent development continues in private repositories.
A clear boundary
The public history remains. New development is private.
The snapshot lets anyone inspect what was implemented and verify results published up to that revision.
New architectures, checkpoints and experiments are not necessarily synchronized to GitHub. Access to the private line depends on diligence, a pilot, partnership or commercial agreement.
Official snapshot
ASM — Aletheion State Models
An experimental family of causal language models based on persistent state evolution, without claims of general superiority.
github.com/gnai-creator/ASMCode, configurations, scripts, documentation and results preserved as a public reference. Check the commit attached to each claim.
Open the public snapshot ↗Public materials
Start with the question you want to verify.
Architecture
Causal mechanisms, persistent state and experimental variants.
ASM family
Taxonomy, promotion criteria and relationships among ASM-CM, ASM-R, ASM-C, ASM-S, ASM-F and ASM-X.
Benchmarks
Protocols, artifacts and auditable results available in the public snapshot.
ASM-C streaming
Cache, VRAM, throughput, BF16 parity and MQAR validation up to 32K.
ASM-CM promotion
Frozen post-FP32 revalidation across three seeds: CE, throughput, VRAM, cache and parity.
ASM × Transformer
Paired 100M-token comparison with rescoring on the same validation set.
Limitations
Known constraints, methodological risks and open questions.
DRM theory
The mathematical formulation behind the geometric research line.
How to cite
Authorship and academic citation metadata for the project.
Reproduce the snapshot
From checkout to first experiment.
These commands reproduce the preserved public line. They do not grant access to subsequent private development.
git clone https://github.com/gnai-creator/ASM.git
cd ASM
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
python scripts/train_tiny.py --config configs/tiny.yaml --text data/tiny.txt- 01
Install the public snapshot in an isolated Python environment.
- 02
Start with the tiny configuration, which runs on CPU.
- 03
Use recorded seeds, configurations and protocols when comparing variants.
- 04
Keep frozen evaluation separate from training data.
- 05
Record the commit, environment, checkpoint and metrics for every run.
Licensing
The public snapshot's terms remain valid.
Copies distributed on GitHub remain under their accompanying terms, including AGPL-3.0-only. Making future development private does not revoke permissions already granted for previous public versions.
Access to private development, proprietary integration, support and alternative commercial terms requires a separate agreement.
Important boundary
A public snapshot does not mean continuous public development.
GitHub documents one verifiable revision of the project. Later results or capabilities should only be attributed to their corresponding reports, private commits or diligence materials.
Read the snapshot. Challenge the results. Discuss the next generation.
Public reproduction and private access serve different purposes in the current strategy.