Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
Merge remote-tracking branch 'origin/dev' into HEAD
JustVugg046f604fbf97
diff --cc c/Makefileindex 6bda50b6,c3a55fa7..24eee2b1--- a/c/Makefile+++ b/c/Makefile@@@ -840,7 -865,7 +865,7 @@@ rans: $(RANSLIB$(RANSLIB): tools/rans_ctypes.c rans.h$(CC) $(CFLAGS) -fPIC -shared $< -o $@ $(LDFLAGS)- cuda-test: backend_cuda.cu backend_cuda.h backend_gpu_compat.h tests/test_backend_cuda.cu tests/test_ragged_attention.cu tests/test_absorb_determinism.cu tests/test_mxfp4_cuda.cu tests/mxfp4_ref.c tests/test_fp8_warp_cuda.cu tests/test_cuda_fmt_trap_cuda.cu-cuda-test: backend_cuda.cu backend_cuda.h backend_gpu_compat.h tests/test_backend_cuda.cu tests/test_ragged_attention.cu tests/test_absorb_determinism.cu tests/test_mxfp4_cuda.cu tests/mxfp4_ref.c tests/test_fp8_warp_cuda.cu tests/test_fp8_cuda.cu tests/test_weights_owned_cuda.cu++cuda-test: backend_cuda.cu backend_cuda.h backend_gpu_compat.h tests/test_backend_cuda.cu tests/test_ragged_attention.cu tests/test_absorb_determinism.cu tests/test_mxfp4_cuda.cu tests/mxfp4_ref.c tests/test_fp8_warp_cuda.cu tests/test_fp8_cuda.cu tests/test_weights_owned_cuda.cu tests/test_cuda_fmt_trap_cuda.cu@command -v "$(GPUCC)" >/dev/null 2>&1 || { echo "$(GPUCC_NAME) not found" >&2; exit 1; }"$(GPUCC)" $(GPUFLAGS) backend_cuda.cu tests/test_backend_cuda.cu -o backend_cuda_test$(EXE) $(ANS_NVCC_LIBS)./backend_cuda_test$(EXE)@@@ -908,7 -937,7 +944,8 @@@ gpu-compile: backend_cuda."$(GPUCC)" $(GPUFLAGS) backend_cuda.cu tests/test_ragged_attention.cu -o ragged_attention_test$(EXE)"$(GPUCC)" $(GPUFLAGS) backend_cuda.cu tests/test_absorb_determinism.cu -o absorb_determinism_test$(EXE)"$(GPUCC)" $(GPUFLAGS) tests/test_fp8_warp_cuda.cu -o fp8_warp_test$(EXE) $(ANS_NVCC_LIBS)+ "$(GPUCC)" $(GPUFLAGS) tests/test_weights_owned_cuda.cu -o weights_owned_test$(EXE) $(ANS_NVCC_LIBS)+ "$(GPUCC)" $(GPUFLAGS) tests/test_cuda_fmt_trap_cuda.cu -o cuda_fmt_trap_test$(EXE) $(ANS_NVCC_LIBS)cuda-bench: backend_cuda.cu backend_cuda.h backend_gpu_compat.h tests/bench_tensor_core.cu@command -v "$(GPUCC)" >/dev/null 2>&1 || { echo "$(GPUCC_NAME) not found" >&2; exit 1; }