Skip to content

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

test(cuda): group tier regression rules with related tests

ZacharyZcRf5ec92f478eb
diff --git a/c/Makefile b/c/Makefile
index f9dde799..af644695 100644
--- a/c/Makefile
+++ b/c/Makefile
@@ -1877,6 +1877,10 @@ tests/test_qwen36_tier_autoplace$(EXE): tests/test_qwen36_tier_autoplace.c tests
tests/test_qwen36_tier_fp8$(EXE): tests/test_qwen36_tier_fp8.c tests/qwen36_fake_cuda.h qwen36_tier.c qwen36_tier.h backend_cuda.h compat.h
$(CC) $(CFLAGS) -DCOLI_CUDA $< -o $@ $(LDFLAGS)
+# Failed gate/up/down uploads must release every unpublished CUDA tensor.
+tests/test_qwen36_tier_rollback$(EXE): tests/test_qwen36_tier_rollback.c tests/qwen36_fake_cuda.h qwen36_tier.c qwen36_tier.h backend_cuda.h compat.h
+ $(CC) $(CFLAGS) -DCOLI_CUDA $< -o $@ $(LDFLAGS)
+
# Generic resident dense matrices (the Qwen3.8 trunk) and per-offer placement.
tests/test_qwen36_tier_dense$(EXE): tests/test_qwen36_tier_dense.c tests/qwen36_fake_cuda.h qwen36_tier.c qwen36_tier.h backend_cuda.h
$(CC) $(CFLAGS) -DCOLI_CUDA $< -o $@ $(LDFLAGS)
@@ -2180,7 +2184,3 @@ tests/test_kv_prefix$(EXE): tests/test_kv_prefix.c kv_prefix.h
tests/test_kimi_request_state$(EXE): tests/test_kimi_request_state.c kimi_k3.c kv_prefix.h serve_codec.h st.h json.h tok.h tok_unicode.h tok_unicode_o200k.h compat.h quant.h idot.h omp_tune.h route_trace.h
$(CC) $(NOCUDA_CFLAGS) tests/test_kimi_request_state.c $(VK_OBJ) -o tests/test_kimi_request_state$(EXE) $(NOCUDA_LDFLAGS)
-
-# Failed gate/up/down uploads must release every unpublished CUDA tensor.
-tests/test_qwen36_tier_rollback$(EXE): tests/test_qwen36_tier_rollback.c tests/qwen36_fake_cuda.h qwen36_tier.c qwen36_tier.h backend_cuda.h compat.h
- $(CC) $(CFLAGS) -DCOLI_CUDA $< -o $@ $(LDFLAGS)