Artificial Intelligence & Machine Learning.
Public AI/ML models, agent systems, interactive demos, and experimental architectures originating from Tonmoy Infrastructure & Vision and its research labs.
- M31 Architecture
- 3 Public Checkpoints
- 1 Interactive Demo
- System / Model / Demo
Hugging Face Profile
huggingface.co/eshanized
Authoritative host for public M31 checkpoints, tokenizer configs, agent runtimes, and interactive programmer spaces.
GitHub Repositories
github.com/eshanized
Source code repositories, agent harness architectures, tokenizer pipelines, and training specifications.
Research Checkpoint Integrity & Evaluation Policy
The M31 Ecosystem Architecture.
A clear distinction between agent systems, causal language models, runtime harnesses, and interactive programmer demos.
The M31 Engineering & Research Family
M31 encompasses autonomous developer systems, native causal language models, lightweight Python agent runtimes, and large-vocabulary pretraining checkpoints. Each project within the ecosystem maintains an independent lifecycle category, distinguishing stable operational software from experimental research checkpoints.
Relationships defined via explicit metadata, not nominal inference.
M31A
Rust-native autonomous software-engineering runtime
Single-crate, high-assurance Rust runtime with deterministic lifecycle control, 11-stage non-bypassable policy gates, evidence-based completion, and terminal cockpit.
M31Genesis
Native M31 causal language model for agentic coding research
Native M31 decoder-only transformer with 425M parameters, designed for software engineering, code synthesis, and agentic workflows.
M31Tesla
Standalone runtime and checkpoints for M31-Python-Agent-220M-v5
Custom 228.9M model with grouped-query attention and standalone Python runtime for code repair experiments.
M31Entropy
Large-vocabulary architecture exploration and pretraining checkpoints
Exploration of 128,000-token vocabulary representation density in a 314M parameter causal transformer.
M31 Q
Interactive demonstration Space for programmers on Hugging Face
Interactive demonstration Space hosted on Hugging Face allowing researchers and developers to test M31 capabilities directly in the browser.
Projects sharing the M31 prefix represent discrete engineering efforts with distinct lifecycle properties. M31A is an active stable release, whereas M31Genesis, M31Tesla, and M31Entropy are experimental research artifacts. Project names do not imply external entity affiliations (e.g. M31Tesla has no association with Tesla, Inc., nor does M31Entropy imply a thermodynamic architecture).
AI & Causal Language Models.
Open weights and checkpoint architectures published on Hugging Face for research and developer evaluation.
M31Genesis
AI ModelExperimentalORG:TIVM31-Genesis-AgenticCode-v1 — native M31 causal language model for agentic coding research. Checkpoints are experimental research checkpoints.
Undisclosed
Native M31 decoder-only transformer: 425M-class parameters, 65,536 ByteLevel BPE vocabulary, 24 layers, hidden size 1024, 16 attention heads / 8 KV heads (GQA), RMSNorm normalization, Rotary Position Embeddings (RoPE), and tied embeddings.
Standard
MIT
M31Tesla
AI ModelExperimentalORG:TIVM31Tesla — standalone runtime and experimental research checkpoints for the custom M31-Python-Agent-220M-v5 model.
Undisclosed
228,953,984 parameters; vocab 32,768; hidden 896; 20 layers; 14 query heads; 7 KV heads; intermediate 2,816; Grouped-Query Attention (GQA); RMSNorm; RoPE theta 10,000; tied embeddings. Defaults to 512-token working context, supporting positions up to 2,048.
Standard
MIT
M31Entropy
AI ModelExperimentalORG:TIVM31Entropy — experimental architecture exploration and pretraining checkpoints for large-vocabulary M31 variants (314M parameters).
Undisclosed
314,281,856 parameters; vocab 128,000; hidden 896; 20 layers; 14 attention heads; 7 KV heads; intermediate 2,816; RMSNorm eps 1e-6; RoPE theta 10,000; tied word embeddings.
Standard
MIT
Software & Agent Systems.
End-to-end applications and runtime engines integrating artificial intelligence into software workflows.
M31A
Software SystemStable ReleaseORG:TIVSingle-crate, high-assurance, Rust-native autonomous software engineering runtime. It provides deterministic lifecycle control, strict multi-layer security policies, resource-bounded execution, continuous verification, crash-resilient checkpoints, and local observability for autonomous coding agents.
Interactive Hugging Face Demos.
Web-based demonstrator spaces allowing developers to explore agent capabilities directly.
M31 Q // For programmers
Space / DemoExperimentalORG:TIVM31 Q // For programmers — Hugging Face Space interactive demonstration for exploring M31 programmer tooling and capabilities.
Public Datasets Disclosure
Governance & Data IntegrityNo public training corpora or benchmark datasets are currently published. TIV internalizes proprietary training datasets and partner splits under non-disclosure and strict copyright clearance. When benchmark datasets meet open-source reproducibility standards, they will be registered directly under authoritative organizational namespaces on Hugging Face and GitHub.
Related Publications & Papers.
Peer-reviewed papers, system whitepapers, and technical reports on AI systems and computing.