model : print tensor size during load (ggml-org#12711)

ggerganov · slaren · web-flow · commit 833e2b740921 · 2025-04-02T16:38:54.000+03:00
* model : print tensor size during load

* cont : fix units MB -&gt; MiB

Co-authored-by: Diego Devesa &lt;slarengh@gmail.com&gt;

---------

Co-authored-by: Diego Devesa &lt;slarengh@gmail.com&gt;
diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
@@ -603,7 +603,9 @@ llama_model_loader::llama_model_loader(
 
             if (trace > 0) {
                 const uint16_t sid = w.idx;
-                LLAMA_LOG_INFO("%s: - tensor split %2d: %32s %-8s [ %s ]\n", __func__, sid, ggml_get_name(tensor), ggml_type_name(type), llama_format_tensor_shape(tensor).c_str());
+                LLAMA_LOG_INFO("%s: - tensor split %2d: %32s %-8s [ %s ] %8.2f MiB\n", __func__,
+                        sid, ggml_get_name(tensor), ggml_type_name(type), llama_format_tensor_shape(tensor).c_str(),
+                        ggml_nbytes(tensor)/1024.0f/1024.0f);
             }
         }
 

Original file line number	Diff line number	Diff line change
`@@ -603,7 +603,9 @@ llama_model_loader::llama_model_loader(`
`603`	`603`
`604`	`604`	`if (trace > 0) {`
`605`	`605`	`const uint16_t sid = w.idx;`
`606`		`- LLAMA_LOG_INFO("%s: - tensor split %2d: %32s %-8s [ %s ]\n", __func__, sid, ggml_get_name(tensor), ggml_type_name(type), llama_format_tensor_shape(tensor).c_str());`
	`606`	`+ LLAMA_LOG_INFO("%s: - tensor split %2d: %32s %-8s [ %s ] %8.2f MiB\n", __func__,`
	`607`	`+ sid, ggml_get_name(tensor), ggml_type_name(type), llama_format_tensor_shape(tensor).c_str(),`
	`608`	`+ ggml_nbytes(tensor)/1024.0f/1024.0f);`
`607`	`609`	`}`
`608`	`610`	`}`
`609`	`611`