Tantra AI Labs — Blog https://tantra-ai-labs.com/blog/ Notes from the lab: training recipes, inference infrastructure, datasets and forensics. en Training AstraGPT-7B on 2× RTX 4090: the actual recipe https://tantra-ai-labs.com/blog/training-astragpt-7b-recipe/ https://tantra-ai-labs.com/blog/training-astragpt-7b-recipe/ The full configuration behind the 7B model: architecture, LoRA settings, batch math, hardware and what we would change next time. Sat, 18 Apr 2026 09:00:00 +0000 One endpoint, many models: how the Astra Gateway routes https://tantra-ai-labs.com/blog/one-endpoint-many-models/ https://tantra-ai-labs.com/blog/one-endpoint-many-models/ Channels, fallbacks, streaming and connection pooling — the mechanics that make a gateway useful. Fri, 10 Jul 2026 09:00:00 +0000 Why the reasoning trace is the dataset https://tantra-ai-labs.com/blog/reasoning-trace-is-the-dataset/ https://tantra-ai-labs.com/blog/reasoning-trace-is-the-dataset/ AstraGPT Reasoning Dataset v1 treats chain-of-thought as the primary artifact. What that changes for training and verification. Fri, 20 Feb 2026 09:00:00 +0000 Numbat Forensics: how to tell whether a checkpoint was trained https://tantra-ai-labs.com/blog/numbat-forensics-audit/ https://tantra-ai-labs.com/blog/numbat-forensics-audit/ A deterministic hash of tensor names rebuilt 633 tensors bit-for-bit. The method, the findings, and what it does not prove. Mon, 02 Nov 2026 09:00:00 +0000 Astra Moderation: a self-hosted moderation API in one Go binary https://tantra-ai-labs.com/blog/astra-moderation-self-hosted/ https://tantra-ai-labs.com/blog/astra-moderation-self-hosted/ Vision-LLM moderation with 15 categories, tenant keys, webhooks and benchmarks — plus the latency trick that matters. Tue, 03 Nov 2026 09:00:00 +0000