sync : whisper.cpp

diff --git a/examples/whisper/convert-pt-to-ggml.py b/examples/whisper/convert-pt-to-ggml.py
index 22bd12e5c84320c55f8cfa27cbe22ab7239c0de7..9e9b2dcebefb92b0e38a600d3c0fee78e6fe8717 100644 (file)
--- a/examples/whisper/convert-pt-to-ggml.py
+++ b/examples/whisper/convert-pt-to-ggml.py
@@ -271,7 +271,7 @@ byte_decoder = {v:k for k, v in byte_encoder.items()}
 fout.write(struct.pack("i", len(tokens)))
 
 for key in tokens:
-    text = bytearray([byte_decoder[c] for c in key]).decode('utf-8', errors='replace').encode('utf-8')
+    text = bytearray([byte_decoder[c] for c in key])
     fout.write(struct.pack("i", len(text)))
     fout.write(text)
 

diff --git a/examples/whisper/whisper.cpp b/examples/whisper/whisper.cpp
index 1826ce960c7a7206bd98b5c414d313fa09314b52..988527811c3cda0c253fc28908f95db7159f8f0e 100644 (file)
--- a/examples/whisper/whisper.cpp
+++ b/examples/whisper/whisper.cpp
@@ -156,11 +156,11 @@ static const std::map<e_model, size_t> MEM_REQ_ENCODE_LAYER = {
 };
 
 static const std::map<e_model, size_t> MEM_REQ_DECODE = {
-    { MODEL_TINY,     94ull*MB },
-    { MODEL_BASE,     96ull*MB },
-    { MODEL_SMALL,    98ull*MB },
-    { MODEL_MEDIUM,  100ull*MB },
-    { MODEL_LARGE,   102ull*MB },
+    { MODEL_TINY,    200ull*MB },
+    { MODEL_BASE,    202ull*MB },
+    { MODEL_SMALL,   204ull*MB },
+    { MODEL_MEDIUM,  206ull*MB },
+    { MODEL_LARGE,   208ull*MB },
 };
 
 static const std::map<e_model, size_t> MEM_REQ_DECODE_LAYER = {
@@ -2314,6 +2314,12 @@ int whisper_full(
         struct whisper_full_params params,
         const float * samples,
         int n_samples) {
+    // clear old results
+    auto & result_all = ctx->result_all;
+    auto & result_cur = ctx->result_cur;
+
+    result_all.clear();
+
     // compute log mel spectrogram
     if (whisper_pcm_to_mel(ctx, samples, n_samples, params.n_threads) != 0) {
         fprintf(stderr, "%s: failed to compute log mel spectrogram\n", __func__);
@@ -2344,11 +2350,6 @@ int whisper_full(
         }
     }
 
-    auto & result_all = ctx->result_all;
-    auto & result_cur = ctx->result_cur;
-
-    result_all.clear();
-
     int progress_prev = 0;
     int progress_step = 5;
 
@@ -2424,7 +2425,7 @@ int whisper_full(
                 whisper_token id  = 0;
                 whisper_token tid = whisper_token_beg(ctx);
 
-                id = whisper_sample_best(ctx, result_len == 0);
+                id = whisper_sample_best(ctx, result_len == 0 || i > 32);
                 if (i > 0) {
                     tid = whisper_sample_timestamp(ctx);
                 }
@@ -2444,7 +2445,9 @@ int whisper_full(
                 // end of text token
                 if (id == whisper_token_eot(ctx)) {
                     if (result_len == 0) {
-                        result_len = i + 1;
+                        // TODO: figure out how to resolve this
+                        fprintf(stderr, "\n%s: failed to generate timestamp token - this should not happen\n\n", __func__);
+                        //result_len = i + 1;
                     }
                     break;
                 }

diff --git a/src/ggml.c b/src/ggml.c
index 5d5c5fa4a231c8704ce26a22c96e63cf4ca81db3..7f11c96faa58eacd466719de7f22cd58df947d28 100644 (file)
--- a/src/ggml.c
+++ b/src/ggml.c
@@ -2,7 +2,7 @@
 
 #if defined(_MSC_VER) || defined(__MINGW32__)
 #include <malloc.h> // using malloc.h with MSC/MINGW
-#else
+#elif !defined(__FreeBSD__)
 #include <alloca.h>
 #endif