From 9e730dd29a090a5552d9ed3448b82585377b07a0 Mon Sep 17 00:00:00 2001 From: DandinPower Date: Mon, 6 Oct 2025 12:22:42 +0800 Subject: [PATCH] fix: update the bug of the sparsity --- src/llama.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/llama.cpp b/src/llama.cpp index e332f855..04ed9638 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -18189,11 +18189,11 @@ static void llama_send_tensors(zmq::socket_t & socket, struct llama_ubatch * uba return; } - float sparse_ratio = comm_sparse_percentage / 100; + float sparse_ratio = (float) comm_sparse_percentage / 100; sparse_array_t *sparse_array = NULL; start_compute_time = get_iso8601_ms_timestamp(); - if (compress(ubatch->backend_embd, tensors->sub_gf_out->ne[0], tensors->sub_gf_out->ne[1], sparse_ratio, &sparse_array)) { + if (compress(ubatch->backend_embd, tensors->sub_gf_out->ne[1], tensors->sub_gf_out->ne[0], sparse_ratio, &sparse_array)) { fprintf(stderr, "compress failed for ratio %.2f\n", sparse_ratio); free_sparse_array(sparse_array); return;