tests: Use tiny-rwkv-6v0-3m

Signed-off-by: Molly Sophia <[email protected]>
RWKV · Jun 25, 2024 · 29ca869 · 29ca869
1 parent 2c8daac
commit 29ca869
Show file tree

Hide file tree

Showing 13 changed files with 25 additions and 25 deletions.
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
@@ -39,11 +39,11 @@ file(COPY tiny-rwkv-5v2-730K-Q5_0.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
 file(COPY tiny-rwkv-5v2-730K-Q5_1.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
 file(COPY expected-logits-5v2-730K.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
 
-file(COPY tiny-rwkv-6v0-1m-FP32.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
-file(COPY tiny-rwkv-6v0-1m-FP16.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
-file(COPY tiny-rwkv-6v0-1m-Q5_0.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
-file(COPY tiny-rwkv-6v0-1m-Q5_1.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
-file(COPY expected-logits-6v0-1m.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
+file(COPY tiny-rwkv-6v0-3m-FP32.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
+file(COPY tiny-rwkv-6v0-3m-FP16.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
+file(COPY tiny-rwkv-6v0-3m-Q5_0.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
+file(COPY tiny-rwkv-6v0-3m-Q5_1.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
+file(COPY expected-logits-6v0-3m.bin DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
 
 rwkv_add_test(test_ggml_basics.c)
 rwkv_add_test(test_quantized_matmul_on_gpu.c)

diff --git a/tests/expected-logits-6v0-1m.bin b/tests/expected-logits-6v0-1m.bin
diff --git a/tests/expected-logits-6v0-3m.bin b/tests/expected-logits-6v0-3m.bin
diff --git a/tests/test_quantization_format_compatibility.c b/tests/test_quantization_format_compatibility.c
@@ -15,7 +15,7 @@ int main(void) {
         "4v0-660K",
         "5v1-730K",
         "5v2-730K",
-        "6v0-1m"
+        "6v0-3m"
     };
 
     // See the explanation of huge expected differences for v5 models in test_tiny_rwkv.c
@@ -30,8 +30,8 @@ int main(void) {
         +025.273308F,
         +048.068733F,
         // 6v0
-        -005.441267F,
-        -017.046452F
+        -019.400530F,
+        +003.576909F
     };
 
     for (int i = 0; i < VERSION_COUNT; i++) {

diff --git a/tests/test_tiny_rwkv.c b/tests/test_tiny_rwkv.c
@@ -21,7 +21,7 @@ int main(void) {
         "4v0-660K",
         "5v1-730K",
         "5v2-730K",
-        "6v0-1m"
+        "6v0-3m"
     };
 
     const char * formats[FORMAT_COUNT] = {
@@ -37,16 +37,16 @@ int main(void) {
     const float expected_difference_sum_full[VERSION_COUNT * 2] = {
         // 4v0
         +0.001000F, // FP32
-        -0.005320F, // FP16
+        -0.013652F, // FP16
         // 5v1
         +0.001000F, // FP32
         -0.289921F, // FP16
         // 5v2
         +0.001000F, // FP32
-        +0.206919F, // FP16
+        +0.455912F, // FP16
         // 6v0
-        +0.001000F, // FP32
-        -0.184410F  // FP16
+        +0.001566F, // FP32
+        -0.416620F  // FP16
     };
 
     // *** Why the hell the expected logit difference sum for v4 models is < 1, and for v5 models it can be as high as 160? ***
@@ -83,20 +83,20 @@ int main(void) {
         +048.068733F, // Q5_1
         -009.441034F, // Q8_0
         // 6v0
-        +039.715752F, // Q4_0
-        +049.779972F, // Q4_1
-        -005.838017F, // Q5_0
-        -017.046452F, // Q5_1
-        -000.220227F  // Q8_0
+        -003.824263F, // Q4_0
+        +021.939022F, // Q4_1
+        -021.191444F, // Q5_0
+        +003.576909F, // Q5_1
+        -009.539596F  // Q8_0
     };
 
     const float expected_difference_sum_quantized_FP16[VERSION_COUNT * (FORMAT_COUNT - 2)] = {
          // 4v0
         +000.154614F, // Q4_0
         -000.539827F, // Q4_1
-        -000.170043F, // Q5_0
+        -000.180142F, // Q5_0
         +000.294953F, // Q5_1
-        +000.070944F, // Q8_0
+        +000.077226F, // Q8_0
         // 5v1
         +119.471931F, // Q4_0
         -028.245888F, // Q4_1
@@ -110,11 +110,11 @@ int main(void) {
         +029.726818F, // Q5_1
         -007.242277F, // Q8_0
         // 6v0
-        +039.676075F, // Q4_0
-        +049.956646F, // Q4_1
-        -006.077929F, // Q5_0
-        -016.773785F, // Q5_1
-        -000.038582F  // Q8_0
+        -003.487368F, // Q4_0
+        +021.797060F, // Q4_1
+        -021.271053F, // Q5_0
+        +003.405264F, // Q5_1
+        -009.734720F  // Q8_0
     };
 
     for (int i_version = 0; i_version < VERSION_COUNT; i_version++) {

diff --git a/tests/tiny-rwkv-6v0-1m-FP16.bin b/tests/tiny-rwkv-6v0-1m-FP16.bin
diff --git a/tests/tiny-rwkv-6v0-1m-FP32.bin b/tests/tiny-rwkv-6v0-1m-FP32.bin
diff --git a/tests/tiny-rwkv-6v0-1m-Q5_0.bin b/tests/tiny-rwkv-6v0-1m-Q5_0.bin
diff --git a/tests/tiny-rwkv-6v0-1m-Q5_1.bin b/tests/tiny-rwkv-6v0-1m-Q5_1.bin
diff --git a/tests/tiny-rwkv-6v0-3m-FP16.bin b/tests/tiny-rwkv-6v0-3m-FP16.bin
diff --git a/tests/tiny-rwkv-6v0-3m-FP32.bin b/tests/tiny-rwkv-6v0-3m-FP32.bin
diff --git a/tests/tiny-rwkv-6v0-3m-Q5_0.bin b/tests/tiny-rwkv-6v0-3m-Q5_0.bin
diff --git a/tests/tiny-rwkv-6v0-3m-Q5_1.bin b/tests/tiny-rwkv-6v0-3m-Q5_1.bin