whisper : adjust to ggml changes (#0)

diff --git a/src/whisper.cpp b/src/whisper.cpp
index b6581f2b409c43fc11b618806ea89e8a66ced380..5b6e4b4be486de7759c2154afb4631f679f250b0 100644 (file)
--- a/src/whisper.cpp
+++ b/src/whisper.cpp
@@ -2505,7 +2505,7 @@ static struct ggml_cgraph * whisper_build_graph_decoder(
 
     const float KQscale = pow(float(n_state_head), -0.25);
 
-    struct ggml_tensor * KQ_mask = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD), 1);
+    struct ggml_tensor * KQ_mask = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_kv, n_tokens, 1);
     ggml_set_name(KQ_mask, "KQ_mask");
     ggml_set_input(KQ_mask);
 
@@ -2929,7 +2929,7 @@ static bool whisper_decode_internal(
                     }
                 }
 
-                for (int i = n_tokens; i < GGML_PAD(n_tokens, GGML_KQ_MASK_PAD); ++i) {
+                for (int i = n_tokens; i < n_tokens; ++i) {
                     for (int j = 0; j < n_kv; ++j) {
                         data[h*(n_kv*n_tokens) + i*n_kv + j] = -INFINITY;
                     }