ggml-qnn: benchmark of real LLM inference on a Snapdragon 8 Gen3 phone

zhouwg · zhouwg · commit eb47de0e5665 · 2025-02-26T19:20:07.000+08:00
diff --git a/scripts/build-run-android.sh b/scripts/build-run-android.sh
@@ -141,7 +141,7 @@ function run_llamacli()
 
     adb shell "cd ${REMOTE_PATH} \
                && export LD_LIBRARY_PATH=${REMOTE_PATH} \
-               && ${REMOTE_PATH}/llama-cli -mg ${qnnbackend} -m ${GGUF_MODEL_NAME} -p \"introduce the movie Once Upon a Time in America briefly.\n\""
+               && ${REMOTE_PATH}/llama-cli -mg ${qnnbackend} -no-cnv -m ${GGUF_MODEL_NAME} -p \"introduce the movie Once Upon a Time in America briefly.\n\""
 
 }
 

Original file line number	Diff line number	Diff line change
`@@ -141,7 +141,7 @@ function run_llamacli()`
`141`	`141`
`142`	`142`	`adb shell "cd ${REMOTE_PATH} \`
`143`	`143`	`&& export LD_LIBRARY_PATH=${REMOTE_PATH} \`
`144`		`- && ${REMOTE_PATH}/llama-cli -mg ${qnnbackend} -m ${GGUF_MODEL_NAME} -p \"introduce the movie Once Upon a Time in America briefly.\n\""`
	`144`	`+ && ${REMOTE_PATH}/llama-cli -mg ${qnnbackend} -no-cnv -m ${GGUF_MODEL_NAME} -p \"introduce the movie Once Upon a Time in America briefly.\n\""`
`145`	`145`
`146`	`146`	`}`
`147`	`147`