AINewsnow

Full fine-tune vs LoRA vs QLoRA on one L40S: Qwen2.5-1.5B, memory, speed and held-out loss, measured

"Use LoRA, it needs far less memory" and "use QLoRA if LoRA doesn't fit" are the standard advice for fine-tuning on one GPU. I wanted to see what each one costs on a small model, where the weights are not the problem, so I rented an L40S (48 GB) on Runcrate today (2026-10-28) and fine-tuned Qwen2.5…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-10-28 15:15 · DEV Community — Machine Learning
    Full fine-tune vs LoRA vs QLoRA on one L40S: Qwen2.5-1.5B, memory, speed and held-out loss, measured

More stories

  1. NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt — Hugging Face Blog
  2. Anthropic reveals plans for blockbuster IPO — Sky News Technology
  3. OpenAI launching election war room to track interference efforts — Politico Technology
  4. 3 Questions: Why infrastructure is entering a defining era — MIT News AI
  5. GitHub says it's seeing 7.4B commits per month, up from 2.9B in August, and has added 4M CPU cores and 160PB of storage in 2026 so far to support this growth — Techmeme
  6. DeepMind unveils Genie 4 world model promising more realistic physics and humans — Ars Technica AI
  7. Elon Musk’s AI Odyssey review – the dullest, most pointless riposte possible to Christopher Nolan’s brilliant epic — The Guardian AI
  8. What is artificial general intelligence – and is it already here? — The Conversation AI (US)

Get the daily brief of stories like this at 6:30 every morning →