In-depth articles, capability tours and field notes — written from actually running the models, tools and pipelines they're about.
A hands-on walkthrough of OpenWiki, the open-source CLI from the LangChain team: one command documents a repository with a docs-only sandboxed agent, another distills 100 real Gmail messages into a personal wiki you can chat with from the terminal. Guardrails, the SHA-256 snapshot gate, live demos with screenshots — and the full video walkthrough.
Read the article →More articles
An illustrated, jargon-light tour of FalkorDB — the graph database that stores your graph as sparse matrices and answers queries with matrix multiplication, with a native vector index for AI and RAG.
A visual field guide to twelve illustrator and artist styles, reproduced entirely local in Krea 2 Turbo on one 24 GB RTX 4090 — each with the style described, a short artist bio, and a same-seed triptych with its prompt recipe.
A 12-billion-parameter diffusion transformer, quantized to a Q6_K GGUF (~10.6 GB) and run entirely local on one 24 GB 4090 under Pop!_OS. Install path, a 16-style quality gallery with honest per-image verdicts, and where diffusion text still breaks.
Baidu's DeepSeek-OCR-based parsing VLM, served with SGLang on one
24 GB 4090. Near-perfect on clean English — including exact table extraction — and
much harder on 1890s Arabic. Install steps, full-page → rendered-Markdown comparisons,
and the runaway loop image_mode=base fixes.
Graph-RAG without GraphRAG's bill. LightRAG (HKUDS) turns your documents into a
knowledge graph and answers with a dual-level retrieval paradigm —
a single API call instead of community traversal, and incremental updates instead of rebuilds — run
end-to-end on OpenAI gpt-5-mini / gpt-5.5 and a Qdrant store.
GraphRAG without relation extraction. LinearRAG (ICLR 2026) builds its
Tri-Graph with zero LLM tokens — just NER and embeddings — then does
single-pass multi-hop retrieval via semantic bridging and Personalized
PageRank, run end-to-end on a 24 GB 4090 with gpt-5-mini.
The Apache-2.0 10B unified image model on a 24 GB 4090: instruction-based photo editing and a 4-step Turbo text-to-image — install with screenshots, prompt/config galleries, and the honest blur/OOM/group offload and steps/CFG/identity story.
Microsoft's 3.8B text-to-image model with a 4-bit GPT-OSS-20B encoder: why no-offload bf16 OOMs above 1024, two eviction dead-ends, and the FP8 DiT quantization that unlocks 1440 fully on the GPU — with a multi-category gallery and honest per-image quality notes.
The 9.3B open-weight design model: JSON captions with bounding-box layout,
LLM magic prompts, a 14-image quality-scored style gallery (Arabic included) and a 21-run
speed benchmark — measured locally in nf4 on a 24 GB 4090.
PrismML's 4B image generator quantized to ternary (1.58-bit) and distilled to
4 steps — ~5 GB VRAM, sub-second warm renders, ten --style presets,
and the measured throughput numbers.
ByteDance's Lance does image/video understanding, generation and editing in one 3B transformer — all seven modes, the official benchmarks, and the two code fixes needed to fit 24 GB.
~1.98× average speedup (peak 2.21× on math) via MTP on Jackrong/Qwopus3.6-27B-v2-MTP-GGUF —
with the failed Ollama detour and the production llama-server recipe.
A capability tour of the open 8B Pixel-DiT model: text-to-image, instruction editing, IP reference, bbox layout, openpose — end-to-end on an RTX 4090.