Autoparser - complete refactoring of parser architecture (llama/18675)

* Autoparser - full single commit squish

* Final pre-merge changes: minor fixes, Kimi 2.5 model parser

diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp
index c4b9540f4f40da49e135eadc876ddc8ed6b2c469..faa771e0869d85e025172d5c7fdddb52c1d332a8 100644 (file)
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -7820,6 +7820,8 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
         test_cases.emplace_back(new test_mul_mat(type_a, GGML_TYPE_F32, 1, 64, 256, {1,  1}, {1, 1}));
     }
 
+    test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q8_0, GGML_TYPE_F32, 6, 4096, 5120, {1, 1}, {1, 1}));
+
 #if 0
     // test the mat-mat path for Metal
     for (int k = 1; k < 512; ++k) {