Re:Show HN: Open-source engine running Gemma 4 26B in 2 GB R...
by gitpusher42 on Wednesday July 29, 2026 @05:16PM
(#49103209)
Thank you!
afaik ollama relies on llama.cpp and mmap. mmap loads pages on demand and doesn't use the same explicit cache or parallel reads like my engine. Most likely ollama/llama.cpp will be way slower in this case
afaik ollama relies on llama.cpp and mmap. mmap loads pages on demand and doesn't use the same explicit cache or parallel reads like my engine. Most likely ollama/llama.cpp will be way slower in this case