Unverified50% confidenceFactExact time
llama.cpp uses mmap by default to load GGUF models, loading pages from the SSD on demand into the Linux page cache via major page faults.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
Unverifiedllama.cpp traditionally reads the entire model weights into memory at once and keeps them resident when the service starts, which is unfriendly to machines with limited memory.61% similarVerifiedllama.cpp通过GGUF格式对模型权重进行4-bit、8-bit等量化压缩，大幅降低显存和内存占用60% similarUnverifiedllama.cpp is the standard for partial offloading, capable of loading some model layers into VRAM with the remaining layers spilling over to system memory.58% similarUnverifiedA major page fault means the target page is not in physical memory and must be read from disk, with overhead at the millisecond level; a minor fault only requires establishing a mapping in the page table or page cache, with overhead at the microsecond level.57% similarUnverifiedThe deployment uses a llama.cpp server with 4-bit KV cache, context length set to 100,000, 32 layers offloaded to system memory, and lazy mode enabled to force the Ngram table to be read from SSD.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1014113API
curl https://kongchang.com/api/v1/knowledge/claims/1014113MCP
get_claim(id=1014113)