ModelTrainingAstraGPT18 Apr 2026
Training AstraGPT-7B on 2× RTX 4090: the actual recipe
The full configuration behind the 7B model: architecture, LoRA settings, batch math, hardware and what we would change next time.
Read post6 min read
BLOG
Notes from the lab.
Training recipes, inference plumbing, dataset design, weight forensics and open-source tooling. Written from the work, not around it.
InfrastructureGateway10 Jul 2026
One endpoint, many models: how the Astra Gateway routes
Why a gateway sits between applications and model providers, and the specific mechanics — channels, fallbacks, streaming, connection pooling — that make it useful.
Read post5 min read
DatasetReasoning20 Feb 2026
Why the reasoning trace is the dataset
AstraGPT Reasoning Dataset v1 treats chain-of-thought as the primary artifact. Here is what that changes for training, verification and uncertainty.
Read post6 min read
SecurityForensicsResearch2 Nov 2026
Numbat Forensics: how to tell whether a checkpoint was trained
A public 7.7 TB checkpoint claimed a mythos-class training run. A deterministic hash of tensor names rebuilt 633 tensors bit-for-bit. Here is the method and what it does not prove.
Read post7 min read
Open sourceModerationGo3 Nov 2026
Astra Moderation: a self-hosted moderation API in one Go binary
Vision-LLM moderation with 15 categories, tenant keys, webhooks and Prometheus — plus the benchmark numbers and the latency trick that matters most.
Read post7 min read