whisper : return ptr for whisper_get_timings

ggerganov · Nov 10, 2024 · 60238af · 60238af
1 parent b1cd3be
commit 60238af
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 19 deletions.
diff --git a/examples/whisper.swiftui/whisper.cpp.swift/LibWhisper.swift b/examples/whisper.swiftui/whisper.cpp.swift/LibWhisper.swift
@@ -119,7 +119,7 @@ actor WhisperContext {
         whisper_print_timings(context)
 
         let system_info = self.system_info()
-        let timings: whisper_timings = whisper_get_timings(context)
+        let timings: whisper_timings = whisper_get_timings(context).pointee
         let encode_ms = String(format: "%.2f", timings.encode_ms)
         let decode_ms = String(format: "%.2f", timings.decode_ms)
         let batchd_ms = String(format: "%.2f", timings.batchd_ms)

diff --git a/include/whisper.h b/include/whisper.h
@@ -430,7 +430,7 @@ extern "C" {
         float batchd_ms;
         float prompt_ms;
     };
-    WHISPER_API struct whisper_timings whisper_get_timings(struct whisper_context * ctx);
+    WHISPER_API struct whisper_timings * whisper_get_timings(struct whisper_context * ctx);
     WHISPER_API void whisper_print_timings(struct whisper_context * ctx);
     WHISPER_API void whisper_reset_timings(struct whisper_context * ctx);
 

diff --git a/scripts/bench-all.sh b/scripts/bench-all.sh
@@ -24,14 +24,7 @@ else
     fattn="-fa"
 fi
 
-models=(                                                                                                    \
-      "tiny"     "tiny-q4_0"     "tiny-q4_1"     "tiny-q5_0"     "tiny-q5_1"     "tiny-q8_0"                \
-      "base"     "base-q4_0"     "base-q4_1"     "base-q5_0"     "base-q5_1"     "base-q8_0"                \
-     "small"    "small-q4_0"    "small-q4_1"    "small-q5_0"    "small-q5_1"    "small-q8_0"                \
-    "medium"   "medium-q4_0"   "medium-q4_1"   "medium-q5_0"   "medium-q5_1"   "medium-q8_0"   "medium-dis" \
-  "large-v2" "large-v2-q4_0" "large-v2-q4_1" "large-v2-q5_0" "large-v2-q5_1" "large-v2-q8_0" "large-v2-dis" \
-  "large-v3-turbo"                           "large-v3-turbo-q5_0"           "large-v3-turbo-q8_0"          \
-)
+models=("tiny")
 
 if [ "$encoder_only" -eq 0 ]; then
     printf "\n"

diff --git a/src/whisper.cpp b/src/whisper.cpp
@@ -4186,22 +4186,16 @@ whisper_token whisper_token_transcribe(struct whisper_context * ctx) {
     return ctx->vocab.token_transcribe;
 }
 
-struct whisper_timings whisper_get_timings(struct whisper_context * ctx) {
+struct whisper_timings * whisper_get_timings(struct whisper_context * ctx) {
     if (ctx->state == nullptr) {
-        return whisper_timings {
-            .sample_ms = 0,
-            .encode_ms = 0,
-            .decode_ms = 0,
-            .batchd_ms = 0,
-            .prompt_ms = 0,
-        };
+        return nullptr;
     }
     const int32_t n_sample = std::max(1, ctx->state->n_sample);
     const int32_t n_encode = std::max(1, ctx->state->n_encode);
     const int32_t n_decode = std::max(1, ctx->state->n_decode);
     const int32_t n_batchd = std::max(1, ctx->state->n_batchd);
     const int32_t n_prompt = std::max(1, ctx->state->n_prompt);
-    return whisper_timings {
+    return new whisper_timings {
         .sample_ms = 1e-3f * ctx->state->t_sample_us / n_sample,
         .encode_ms = 1e-3f * ctx->state->t_encode_us / n_encode,
         .decode_ms = 1e-3f * ctx->state->t_decode_us / n_decode,