whisper : add memory sizes for Q8_0 (close #846)

diff --git a/examples/whisper.wasm/README.md b/examples/whisper.wasm/README.md
index e7429f35659a37148fbe2ca54ce754d4c6b05a56..f228a8b6e5f90f3d8f482fa2b537408ca8748f40 100644 (file)
--- a/examples/whisper.wasm/README.md
+++ b/examples/whisper.wasm/README.md
@@ -37,6 +37,6 @@ emcmake cmake ..
 make -j
 
 # copy the produced page to your HTTP path
-cp bin/whisper.wasm/*       /path/to/html/
+cp bin/whisper.wasm/*    /path/to/html/
 cp bin/libmain.worker.js /path/to/html/
 ```

diff --git a/whisper.cpp b/whisper.cpp
index b08bf0faa5841f316f9e9af3d3431f755607f04d..cca949fa89be54d7abd76191426ac011a2448241 100644 (file)
--- a/whisper.cpp
+++ b/whisper.cpp
@@ -284,11 +284,11 @@ static const std::map<ggml_type, std::map<e_model, size_t>> MEM_REQ_MODEL = {
     },
     { GGML_TYPE_Q4_1,
         {
-            { MODEL_TINY,     31ull*MB },
-            { MODEL_BASE,     57ull*MB },
-            { MODEL_SMALL,   181ull*MB },
-            { MODEL_MEDIUM,  559ull*MB },
-            { MODEL_LARGE,  1122ull*MB },
+            { MODEL_TINY,     32ull*MB },
+            { MODEL_BASE,     58ull*MB },
+            { MODEL_SMALL,   182ull*MB },
+            { MODEL_MEDIUM,  562ull*MB },
+            { MODEL_LARGE,  1124ull*MB },
         },
     },
     { GGML_TYPE_Q4_2,
@@ -300,22 +300,31 @@ static const std::map<ggml_type, std::map<e_model, size_t>> MEM_REQ_MODEL = {
             { MODEL_LARGE,   940ull*MB },
         },
     },
-    { GGML_TYPE_Q5_0, // TODO: fix
+    { GGML_TYPE_Q5_0,
         {
-            { MODEL_TINY,     31ull*MB },
-            { MODEL_BASE,     57ull*MB },
-            { MODEL_SMALL,   181ull*MB },
-            { MODEL_MEDIUM,  559ull*MB },
-            { MODEL_LARGE,  1122ull*MB },
+            { MODEL_TINY,     30ull*MB },
+            { MODEL_BASE,     54ull*MB },
+            { MODEL_SMALL,   170ull*MB },
+            { MODEL_MEDIUM,  516ull*MB },
+            { MODEL_LARGE,  1034ull*MB },
         },
     },
     { GGML_TYPE_Q5_1,
         {
-            { MODEL_TINY,     31ull*MB },
-            { MODEL_BASE,     57ull*MB },
-            { MODEL_SMALL,   181ull*MB },
-            { MODEL_MEDIUM,  559ull*MB },
-            { MODEL_LARGE,  1122ull*MB },
+            { MODEL_TINY,     32ull*MB },
+            { MODEL_BASE,     58ull*MB },
+            { MODEL_SMALL,   182ull*MB },
+            { MODEL_MEDIUM,  562ull*MB },
+            { MODEL_LARGE,  1124ull*MB },
+        },
+    },
+    { GGML_TYPE_Q8_0,
+        {
+            { MODEL_TINY,     45ull*MB },
+            { MODEL_BASE,     84ull*MB },
+            { MODEL_SMALL,   268ull*MB },
+            { MODEL_MEDIUM,  834ull*MB },
+            { MODEL_LARGE,  1674ull*MB },
         },
     },
 };