From 6058861b3f2542aed9c5dbd4e6a2285d2055cc6e Mon Sep 17 00:00:00 2001 From: Hans Date: Fri, 3 May 2024 13:41:08 +0800 Subject: [PATCH] feat: increase limit to capable Llama 7B --- src/ggml-qnn/ggml-qnn.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/ggml-qnn/ggml-qnn.cpp b/src/ggml-qnn/ggml-qnn.cpp index 42e9325..f88ebda 100644 --- a/src/ggml-qnn/ggml-qnn.cpp +++ b/src/ggml-qnn/ggml-qnn.cpp @@ -4282,7 +4282,8 @@ static size_t ggml_backend_qnn_buffer_type_get_max_size(ggml_backend_buffer_type GGML_UNUSED(buft); //works fine with ggml-tiny.en-q8_0.bin for whisper.cpp //return (38 * 1024 * 1024); - return (96 * 1024 * 1024); + // return (96 * 1024 * 1024); + return (128 * 1024 * 1024); }