Minor: remove unnecesssary calls to build_inp_out_ids (#935)

ikawrakow · Iwan Kawrakow · web-flow · commit 489554bc1126 · 2025-11-10T17:38:46.000+02:00
Co-authored-by: Iwan Kawrakow &lt;iwan.kawrakow@gmail.com&gt;
diff --git a/src/llama-build-context.cpp b/src/llama-build-context.cpp
@@ -6474,7 +6474,6 @@ ggml_cgraph * llm_build_context::build_glm4_moe() {
         // crop output on last layer
         if (il == n_transformer_layers - 1 && inp_out_ids) {
             // skip computing output for unused tokens
-            ggml_tensor * inp_out_ids = build_inp_out_ids();
             cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
             inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
         }
@@ -7776,7 +7775,6 @@ ggml_cgraph * llm_build_context::build_ernie4_5() {
         }
 
         if (il == n_layer - 1 && inp_out_ids) {
-            ggml_tensor * inp_out_ids = build_inp_out_ids();
             cur = ggml_get_rows(ctx0, cur, inp_out_ids);
             inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
         }

Original file line number	Diff line number	Diff line change
`@@ -6474,7 +6474,6 @@ ggml_cgraph * llm_build_context::build_glm4_moe() {`
`6474`	`6474`	`// crop output on last layer`
`6475`	`6475`	`if (il == n_transformer_layers - 1 && inp_out_ids) {`
`6476`	`6476`	`// skip computing output for unused tokens`
`6477`		`- ggml_tensor * inp_out_ids = build_inp_out_ids();`
`6478`	`6477`	`cur = ggml_get_rows(ctx0, cur, inp_out_ids);`
`6479`	`6478`	`inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);`
`6480`	`6479`	`}`
`@@ -7776,7 +7775,6 @@ ggml_cgraph * llm_build_context::build_ernie4_5() {`
`7776`	`7775`	`}`
`7777`	`7776`
`7778`	`7777`	`if (il == n_layer - 1 && inp_out_ids) {`
`7779`		`- ggml_tensor * inp_out_ids = build_inp_out_ids();`
`7780`	`7778`	`cur = ggml_get_rows(ctx0, cur, inp_out_ids);`
`7781`	`7779`	`inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);`
`7782`	`7780`	`}`