--- a/envconfig/config.go
+++ b/envconfig/config.go
@@ -222,6 +222,8 @@
 	NoHistory = Bool("OLLAMA_NOHISTORY")
 	// NoPrune disables pruning of model blobs on startup.
 	NoPrune = Bool("OLLAMA_NOPRUNE")
+	// NoMMap disables memory mapping of the model file.
+	NoMMap = Bool("OLLAMA_NO_MMAP")
 	// SchedSpread allows scheduling models across all GPUs.
 	SchedSpread = Bool("OLLAMA_SCHED_SPREAD")
 	// MultiUserCache optimizes prompt caching for multi-user scenarios
@@ -326,6 +328,7 @@
 		"OLLAMA_NO_CLOUD":             {"OLLAMA_NO_CLOUD", NoCloud(), "Disable Ollama cloud features (remote inference and web search)"},
 		"OLLAMA_NOHISTORY":            {"OLLAMA_NOHISTORY", NoHistory(), "Do not preserve readline history"},
 		"OLLAMA_NOPRUNE":              {"OLLAMA_NOPRUNE", NoPrune(), "Do not prune model blobs on startup"},
+		"OLLAMA_NO_MMAP":              {"OLLAMA_NO_MMAP", NoMMap(), "Disable memory mapping of the model file"},
 		"OLLAMA_NUM_PARALLEL":         {"OLLAMA_NUM_PARALLEL", NumParallel(), "Maximum number of parallel requests"},
 		"OLLAMA_ORIGINS":              {"OLLAMA_ORIGINS", AllowedOrigins(), "A comma separated list of allowed origins"},
 		"OLLAMA_SCHED_SPREAD":         {"OLLAMA_SCHED_SPREAD", SchedSpread(), "Always schedule model across all GPUs"},

--- a/llm/llama_server.go
+++ b/llm/llama_server.go
@@ -628,7 +628,7 @@
 		}
 	}
 
-	if opts.UseMMap != nil && !*opts.UseMMap {
+	if envconfig.NoMMap() || (opts.UseMMap != nil && !*opts.UseMMap) {
 		return append(params, "--load-mode", "none")
 	}