FIBO is a SOTA, first open-source, JSON-native text-to-image model built for controllable, predictable, and legally safe image generation.

FIBO-Edit brings the power of structured prompt generation to image editing

Running small but capable language models entirely offline

finetune gpt-oss and smollm3 on your data easily and cheaply

A small, readable Go runtime for SmolLM3-3B local inference, with tokenizer, int8 weight-only quantization, KV cache, and ARM64/AMD64 SIMD.

TMLR 2026 | Mechanistic interpretability: attention-head binding (EB*) as a marker of concept emergence. 7 models, 5 architectures (Pythia 160M–2.8B, OLMo-1B, CRFM GPT-2, SmolLM3-3B, Qwen2.5-1.5B), 41 terms.

Deploy AI models with an API through quantization and containerization.

Independent benchmark of HuggingFace SmolLM3-3B (base) — full 5-benchmark suite, two clean vendor validations, reproducible vLLM methodology.

Custom kernels for Smol-lm3 3B Inference.