{"version":"1","article":{"id":"nvfp4-lora-consumer-gpu-study-2026","slug":"nvfp4-lora-consumer-gpu-study-2026","language":"en","url":"https://haberpaper.com/en/news/nvfp4-lora-consumer-gpu-study-2026","api_url":"https://haberpaper.com/api/v1/articles/nvfp4-lora-consumer-gpu-study-2026?lang=en","title":"PACT: NVFP4 Data-Flow LoRA Fine-Tuning on Consumer GPUs","summary":"Researchers introduce PACT, a system for LoRA fine-tuning on consumer GPUs that uses native NVFP4 matrix operations for frozen projections. The paper also addresses memory spent on activations and the cost of quantizing and dequantizing data.","category":"business","region":"world","topics":["ai","semiconductors"],"published_at":"2026-10-20T19:09:32+03:00","modified_at":"2026-10-20T19:09:32+03:00","publication_basis":"publication-record","event_period":"2026-10-18","event_period_basis":"event-or-reference-period","sources":[{"name":"arXiv","url":"https://arxiv.org/abs/2610.25048v1","publishedAt":"2026-10-18","checkedAt":"2026-10-20"}],"translations":[{"language":"tr","url":"https://haberpaper.com/tr/news/nvfp4-lora-consumer-gpu-study-2026","api_url":"https://haberpaper.com/api/v1/articles/nvfp4-lora-consumer-gpu-study-2026?lang=tr"},{"language":"en","url":"https://haberpaper.com/en/news/nvfp4-lora-consumer-gpu-study-2026","api_url":"https://haberpaper.com/api/v1/articles/nvfp4-lora-consumer-gpu-study-2026?lang=en"},{"language":"ar","url":"https://haberpaper.com/ar/news/nvfp4-lora-consumer-gpu-study-2026","api_url":"https://haberpaper.com/api/v1/articles/nvfp4-lora-consumer-gpu-study-2026?lang=ar"},{"language":"fr","url":"https://haberpaper.com/fr/news/nvfp4-lora-consumer-gpu-study-2026","api_url":"https://haberpaper.com/api/v1/articles/nvfp4-lora-consumer-gpu-study-2026?lang=fr"},{"language":"es","url":"https://haberpaper.com/es/news/nvfp4-lora-consumer-gpu-study-2026","api_url":"https://haberpaper.com/api/v1/articles/nvfp4-lora-consumer-gpu-study-2026?lang=es"}],"data_types":[],"data":null,"data_compilations":[],"data_downloads":[],"content":{"paragraphs":["Researchers introduce PACT, a system for LoRA fine-tuning on consumer GPUs that uses native NVFP4 matrix operations for frozen projections. The paper also addresses memory spent on activations and the cost of quantizing and dequantizing data.","The work targets computation and memory bottlenecks during fine-tuning. It is a preprint describing an evaluated system, not a new GPU or a guarantee that every consumer setup will achieve the same results."],"key_points":[],"sections":[]},"text":"Researchers introduce PACT, a system for LoRA fine-tuning on consumer GPUs that uses native NVFP4 matrix operations for frozen projections. The paper also addresses memory spent on activations and the cost of quantizing and dequantizing data.\n\nResearchers introduce PACT, a system for LoRA fine-tuning on consumer GPUs that uses native NVFP4 matrix operations for frozen projections. The paper also addresses memory spent on activations and the cost of quantizing and dequantizing data.\n\nThe work targets computation and memory bottlenecks during fine-tuning. It is a preprint describing an evaluated system, not a new GPU or a guarantee that every consumer setup will achieve the same results.","corrections":[],"editorial_illustration":{"kind":"symbolic-editorial-illustration","documentary":false,"scope":"section","asset_id":"business","url":"https://haberpaper.com/assets/illustrations/business-1536.webp","width":1536,"height":1024,"alt":"Symbolic industrial workshop with machinery, a warehouse and a crane.","language":"en"},"section_illustration":{"kind":"symbolic-editorial-illustration","documentary":false,"scope":"section","asset_id":"business","url":"https://haberpaper.com/assets/illustrations/business-1536.webp","width":1536,"height":1024,"alt":"Symbolic industrial workshop with machinery, a warehouse and a crane.","language":"en"}}}