test-llama-arch: ignore mtp heads

am17an · am17an · commit 17df5830e72b · 2026-05-06T00:17:07.000+08:00
diff --git a/tests/test-llama-archs.cpp b/tests/test-llama-archs.cpp
@@ -406,6 +406,9 @@ static bool arch_supported(const llm_arch arch) {
     if (arch == LLM_ARCH_DEEPSEEK2OCR) {
         return false;
     }
+    if (arch == LLM_ARCH_QWEN35_MTP || arch == LLM_ARCH_QWEN35MOE_MTP) {
+        return false; // MTP-only arch; requires a sibling trunk model and cannot run standalone.
+    }
 
     // FIXME some models are segfaulting with WebGPU:
 #ifdef GGML_USE_WEBGPU

Original file line number	Diff line number	Diff line change
`@@ -406,6 +406,9 @@ static bool arch_supported(const llm_arch arch) {`
`406`	`406`	`if (arch == LLM_ARCH_DEEPSEEK2OCR) {`
`407`	`407`	`return false;`
`408`	`408`	`}`
	`409`	`+ if (arch == LLM_ARCH_QWEN35_MTP \|\| arch == LLM_ARCH_QWEN35MOE_MTP) {`
	`410`	`+ return false; // MTP-only arch; requires a sibling trunk model and cannot run standalone.`
	`411`	`+ }`
`409`	`412`
`410`	`413`	`// FIXME some models are segfaulting with WebGPU:`
`411`	`414`	`#ifdef GGML_USE_WEBGPU`