# Antonio V. Franco

Antonio V. Franco is an AI engineer and independent researcher. This blog publishes first-hand technical work on LLMs, SLMs, open-source models, fine-tuning, distillation, agentic AI, pricing/cost analysis, and practical model stack decisions. Articles are written from real testing and real workloads, not from press releases.

Canonical: https://antoniovfranco.com/

## Scope

- Agentic AI and Small Language Models (SLMs)
- Open-source vs. closed model evaluation
- Model pricing and cost-per-token analysis
- LLM fine-tuning and distillation tooling
- Practical AI stacks for independent researchers and startups

## Articles
- Title: Quantizing an LLM Isn’t Enough: How to Prove That a 4-bit Model Still Deserves to Go Into Production
  URL: https://antoniovfranco.com/posts/2026/quantizing-an-llm-isnt-enough-how-to-prove-that-a-4-bit-model-still-deserves-to-go-into/
  Date: 2026-08-29
  Description: When I started designing how quantization should work in Huyawo, the first question was the most obvious one: how much could I reduce the size of a model wi.
  Topics: LLM quantization, GPTQ, AWQ, Model compression, Quantization evaluation, Production ML
- Title: The "DeepSeek V4 Flash + GLM-5.2" Combination Is Currently Enough for Me (and Will Probably Be Enough for You Too)
  URL: https://antoniovfranco.com/posts/2026/deepseek-v4-flash-glm-5-2-combination/
  Date: 2026-07-24
  Description: Why DeepSeek V4 Flash handles 70% of my workload and GLM-5.2 covers the rest — a practical AI stack for serious development work
  Topics: AI model stack, DeepSeek V4 Flash, GLM-5.2, Practical AI engineering
- Title: GLM-5.2: I thought Sonnet 4.5 and similar open models were enough, but…
  URL: https://antoniovfranco.com/posts/2026/glm-5-2-sonnet-4-5-and-open-models-were-enough/
  Date: 2026-06-25
  Description: A first-hand account of testing GLM-5.2 and realizing the open ecosystem has closed the gap on proprietary flagship models — in code quality, cost, and access.
  Topics: Open-source LLMs, GLM-5.2, Model comparison, Open AI advocacy
- Title: Agentic AI, SLMs, and Why Models Above US$0.50 Output per 1M Tokens Are Equivalent to Burning Money
  URL: https://antoniovfranco.com/posts/2026/agentic-ai-slms-why-models-above-50-cents-burning-money/
  Date: 2026-06-13
  Description: A first-hand look at why DeepSeek V4 Flash and Qwen3-Coder-Next make anything above US$0.50 per million output tokens a waste of money in agentic pipelines, and why SLMs are the future of industrial agents.
  Topics: Small Language Models (SLMs), Agentic AI, LLM API pricing, Model cost optimization, DeepSeek V4 Flash
- Title: QDoRA Explained: Why It Became the New PEFT Standard in 2025
  URL: https://antoniovfranco.com/posts/2025/qdora-explained-the-new-peft-standard-for-2025/
  Date: 2025-11-11
  Description: QDoRA combines DoRA’s weight decomposition with 4-bit quantization for memory-efficient LLM fine-tuning. Learn the mathematics, implementation, and trade-offs.
  Topics: QDoRA, PEFT, Parameter-efficient fine-tuning, LoRA, QLoRA, DoRA, LLM quantization

## Feeds

- Sitemap: https://antoniovfranco.com/sitemap.xml
- RSS: https://antoniovfranco.com/index.xml