Blog

Local AI research · 40 → 64 t/s · 4 experiments
16 May 2026 · Experiment 03
Experiment 03 LM Cache MTP llama.cpp
Reaching 64 t/s: LM Cache, KV Checkpointing, and MTP

The proxy showed 57 TPS, but llama-cli gave 36. Same hardware. This post answers why — then adds MTP to reach 64 TPS at 131k context.

64 t/s · 131k ctx · MTP
12 May 2026 · Experiment 02
Experiment 02 Claude Code Cline Proxy
Using Claude Code and Cline with a Local LLM

A 600-line proxy bridges Anthropic's API and llama-server — with lifetime metrics tracking and a live dashboard.

57 t/s · LM cache
9 May 2026 · Experiment 01
Experiment 01 MoE llama.cpp RTX 5060 Ti
Running Qwen3.6 35B at 40 TPS on Consumer Hardware

Ollama leaves 2.2× performance on the table for MoE models — here's how to recover it.

36 t/s → 40 t/s
25 May 2025
Explainer LLMs AI
What Are LLMs (Large Language Models)?

A clear breakdown of what large language models are, how they work, and why they matter for local inference.

25 May 2025
Setup Ollama Windows
Change Default Directory for Ollama

How to change the default directory for Ollama models on Windows using an environment variable.

No posts match your search.