Articles by tag: llm-optimization

Fine-tuning AI for technical writing: lessons after failure | IT Practice

How to personalize a language model for technical content. Practical steps for fine-tuning, hyperparameter tuning, and quality evaluation. Learn how to avoid failures.

Meta-Harness optimizes LLM better than engineers

The Meta-Harness system automates LLM harnesses with access to 10 million tokens. Results: +7.7 pp on classification, 1st place on TerminalBench-2. Explore the details for your projects.

TurboQuant: lossless KV-cache compression for AI

Learn how Google's TurboQuant compresses transformer memory down to 3 bits with PolarQuant and QJL. Benchmarks on Gemma, Mistral. Optimization for AI developers.

TurboQuant: 3-bit KV-cache for LLM without losses

Learn how Google TurboQuant compresses key-value cache of LLM to 3 bits, speeding up inference 8x. Benchmarks on Gemma/Mistral, application in RAG. For developers.

Toolc: proxy for MCP in VS Code — 60% savings

Optimize MCP servers for VS Code agents with toolc on Go. Token reduction up to 60%, model benchmarks. Install and test the savings.

Caveman LLM: token savings up to 87%

Learn how the caveman style reduces LLM tokens by 65% on average while preserving accuracy. Benchmarks, code examples, and caveman tool for developers. Optimize API requests right now.

SCALE framework for optimizing LLM costs

SCALE Framework reduces AI feature budget overrun by 3x without retention loss. Diagnosis of drivers, smart routing, cache, metrics. Instructions for PM and DevOps.

Reducing AI costs to $20/mo: MiniMax and routing

Learn how to optimize LLM costs from $200 to $20/month. MiniMax subscriptions, Kimi K2.5, cascade routing. Real benchmarks and setup for production agents. Test it yourself

Dual-Process Architecture for AI agents

How to split AI into «consciousness» and «reflexes» for 16 ms response. Technical implementation, benchmarks, integration into UE5/Unity.