whisper : alternative way to handle the external encoders

whisper.cpp

@@ -1659,24 +1659,15 @@ static struct ggml_cgraph * whisper_build_graph_conv(
         ggml_set_name(cur, "embd_conv");
         wstate.embd_conv = cur;
     } else {
-#ifdef WHISPER_USE_COREML
+        // transform the "mel" tensor to "embd_enc" via a sequence of ggml ops
-        cur = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_state, n_ctx);
+        // these are not actually executed when using external encoder
-        ggml_allocr_alloc(alloc, cur);
+        // necessary only to prepare tensors with the appropriate memory sizes
-
+        cur = ggml_conv_1d_ph(ctx0, model.e_conv_1_w, mel, 1, 1); // (conv)
-        if (!ggml_allocr_is_measure(alloc)) {
+        cur = ggml_conv_1d_ph(ctx0, model.e_conv_2_w, cur, 2, 1); // (conv)
-            whisper_coreml_encode(wstate.ctx_coreml, mel->ne[0], mel->ne[1], (float *) mel->data, (float *) cur->data);
+        cur = ggml_add(ctx0, model.e_pe, ggml_cont(ctx0, ggml_transpose(ctx0, cur))); // (cross)
-        }
-#endif
-#ifdef WHISPER_USE_OPENVINO
-        cur = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_state, n_ctx);
-        ggml_allocr_alloc(alloc, cur);
-
-        if (!ggml_allocr_is_measure(alloc)) {
-            whisper_openvino_encode(wstate.ctx_openvino, mel, cur);
-        }
-#endif
-
         ggml_set_name(cur, "embd_enc");
+        ggml_set_output(cur);
+
         wstate.embd_enc = cur;
     }
 
@@ -1708,14 +1699,6 @@ static struct ggml_cgraph * whisper_build_graph_encoder(
 
     ggml_cgraph * gf = ggml_new_graph_custom(ctx0, WHISPER_MAX_NODES, false);
 
-    //ggml_allocr * alloc = wstate.alloc_encode.alloc;
-
-    //struct ggml_tensor * cur = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_ctx, n_state);
-    //ggml_allocr_alloc(alloc, cur);
-
-    //if (!ggml_allocr_is_measure(alloc)) {
-    //    ggml_backend_tensor_copy(wstate.embd_conv, cur);
-    //}
     struct ggml_tensor * cur = ggml_view_tensor(ctx0, wstate.embd_conv);
 
     const float KQscale = 1.0f/sqrtf(float(n_state)/n_head);
@@ -1957,14 +1940,6 @@ static struct ggml_cgraph * whisper_build_graph_cross(
 
     ggml_cgraph * gf = ggml_new_graph(ctx0);
 
-    //ggml_allocr * alloc = wstate.alloc_cross.alloc;
-
-    //struct ggml_tensor * cur = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_state, n_ctx);
-    //ggml_allocr_alloc(alloc, cur);
-
-    //if (!ggml_allocr_is_measure(alloc)) {
-    //    ggml_backend_tensor_copy(wstate.embd_enc, cur);
-    //}
     struct ggml_tensor * cur = ggml_view_tensor(ctx0, wstate.embd_enc);
 
     const float  Kscale = pow(float(n_state) / n_head, -0.25);
@@ -2037,13 +2012,13 @@ static bool whisper_encode_internal(
             return false;
         }
 
+        struct ggml_tensor * mel = ggml_graph_get_tensor(gf, "mel");
+
         // set the input
         {
             const auto & mel_inp = wstate.mel;
             const int n_ctx      = wstate.exp_n_audio_ctx > 0 ? wstate.exp_n_audio_ctx : wctx.model.hparams.n_audio_ctx;
 
-            struct ggml_tensor * mel = ggml_graph_get_tensor(gf, "mel");
-
             assert(mel->type == GGML_TYPE_F32);
             assert(mel_inp.n_mel == wctx.model.hparams.n_mels);
 
@@ -2068,6 +2043,12 @@ static bool whisper_encode_internal(
             if (!ggml_graph_compute_helper(wstate.backend, gf, n_threads)) {
                 return false;
             }
+        } else {
+#if defined(WHISPER_USE_COREML)
+            whisper_coreml_encode(wstate.ctx_coreml, mel->ne[0], mel->ne[1], (float *) mel->data, (float *) wstate.embd_enc->data);
+#elif defined(WHISPER_USE_OPENVINO)
+            whisper_openvino_encode(wstate.ctx_openvino, mel, wstate.embd_enc);
+#endif
         }
     }