AI Blog

Insights, tutorials, and benchmarks from the world of AI

JUST DROPPED Research CUDA Flash Attention May 9, 2026

Fused TBQ4 Flash Attention: 82 tok/s with Lossless 4-bit KV at 200K

We fused TurboQuant dequant directly into the flash attention kernel — reading raw TBQ4 blocks inline via centroid lookup in the FWHT-rotated domain. 82+ tok/s with lossless 4.25 bpv KV at 200K context on RTX 4090. Nobody else has done this.

Fused FA TBQ4 llama.cpp CUDA Kernel MTP Qwen3.6
Read Article →
NEW Benchmarks Research May 5, 2026

Turbo4 KV Cache: Better Than Q8 at Half the VRAM

Trellis-Coded Quantization benchmark: Turbo4 scores 100/100 vs Q8_0's 91 on hardened agentic benchmark. Lossless FP16 quality, 40 t/s, 256K context on RTX 4090 24GB.

KV Cache TCQ TurboQuant RTX 4090 Qwen3.6
Read Article →
NEW Benchmarks Optimization February 20, 2026

Q4 KV Cache: Surprisingly Viable

Benchmarks show Q4 KV cache producing faster, higher-quality code than Q8. The conventional wisdom about Q4 being unusable may be wrong.

KV Cache Q4 Quantization llama.cpp RTX 4090
Read Article →
NEW Benchmarks February 15, 2026

IQ2 vs IQ3 Quantization: 2x Speed, Same Quality

Comprehensive RTX 4090 benchmark: IQ2_XS hits 86 t/s vs IQ3_XXS at 44 t/s. Quality testing across coding, debugging, and agentic tasks shows negligible difference.

Quantization RTX 4090 Benchmarks Qwen3-Coder-Next
Read Article →
Benchmarks Updated Feb 15, 2026

Maxing Out Qwen3-Coder-Next Abliterated: 94 t/s

The abliterated version hits 94 t/s at 168K context with Q8 KV cache — over 5x faster than the base model. Complete optimization guide with red teaming verification.

Abliterated Q8 KV Cache Local AI RTX 4090
Read Article →
Benchmarks February 9, 2026

Qwen3-Coder-Next: IQ2 vs IQ3 Benchmarks

IQ2_XXS achieves 22 t/s on RTX 4090 at 200K context — 85% faster than IQ3 with no measurable quality loss. Full benchmark data and configuration.

Local AI Benchmarks IQ2 Quantization RTX 4090
Read Article →
Tutorial February 2, 2026

Running Uncensored AI Locally: My PRISM Setup

How I set up GLM-4.7-Flash locally with web search and vision capabilities. No content filters, no API subscription, just my hardware doing what I tell it to.

Local AI GLM-4.7 Uncensored Claude Code
Read Article →

Get AI Tips in Your Inbox

Subscribe for tutorials on local AI, Stable Diffusion, LoRA training, and Claude Code workflows.

More Articles Coming Soon

We're preparing additional tutorials and guides:

  • Stable Diffusion Mastery — Advanced prompting and workflow optimization
  • LoRA Training from Scratch — Create custom models for any subject
  • LLM Fine-tuning Guide — Adapt open-source models for your domain
  • AI Automation Pipelines — Build end-to-end workflows with n8n
  • Claude Code Pro Tips — Advanced usage and MCP development

Have a topic you'd like us to cover? Let us know