Unverified60% confidenceFactExact time
QLoRA builds on LoRA by introducing the NF4 (Normal Float 4) 4-bit quantization format, and employs double quantization and a paged optimizer, enabling fine-tuning of a 7B model on a single consumer-grade 24GB RTX 4090 GPU.
2
Sources
60%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
VerifiedQLoRA introduces 4-bit quantization technology, making it possible to fine-tune models with 65 billion parameters on a single RTX 3090/4090 GPU with 24 GB of VRAM.80% similarPartially VerifiedThe QLoRA paper verified that NF4 quantization combined with LoRA fine-tuning can reduce memory requirements from 780GB to 48GB on a 65B parameter model while maintaining performance close to full-precision fine-tuning74% similarUnverifiedSingle-card fine-tuning usually relies on parameter-efficient fine-tuning techniques (such as LoRA); with LoRA fine-tuning of the 1.7B model on a 4-bit quantized base, a single consumer-grade GPU (such as one with 24 GB of VRAM) is sufficient.73% similarUnverified使用IQ4_XS配合-fa 1 -ctk q8_0 -ctv q8_0对KV缓存进行q8_0量化后，可加载完整32k上下文，常驻内存约16.6GB72% similarUnverifiedLlama 3 70B量化版（Q4精度）仍需约35GB内存71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/527823API
curl https://kongchang.com/api/v1/knowledge/claims/527823MCP
get_claim(id=527823)