Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
Merge branch 'dev' into sse41-tier-upstream
Jaapjan Tinbergenc3bf90daadd8
diff --cc c/Makefileindex 342c9cd7,2fe13efd..f9e92f8d--- a/c/Makefile+++ b/c/Makefile@@@ -1055,7 -1066,7 +1075,7 @@@ fp8-bench: fp8_bench$(EXE# the GPU sits idle. Same guard #783 put on kimi_k3, which since gaining an# MXFP4 expert path no longer needs it. tests/test_makefile_cuda_scope.py# asserts this shape.- olmoe$(EXE): olmoe.c cli_args.h st.h json.h compat.h sample.h tok.h tok_unicode.h tok_unicode_o200k.h omp_tune.h kv_prefix.h route_trace.h serve_codec.h edge_adapters.h edge_runtime.h edge_tok_internal.h fused_simd.h segment_adapter_internal.h segment_adapters.h segment_runtime.h sse41_kernels.h-olmoe$(EXE): olmoe.c cli_args.h st.h json.h compat.h sample.h tok.h tok_unicode.h tok_unicode_o200k.h omp_tune.h kv_prefix.h pin_pool.h route_trace.h serve_codec.h edge_adapters.h edge_runtime.h edge_tok_internal.h fused_simd.h segment_adapter_internal.h segment_adapters.h segment_runtime.h++olmoe$(EXE): olmoe.c cli_args.h st.h json.h compat.h sample.h tok.h tok_unicode.h tok_unicode_o200k.h omp_tune.h kv_prefix.h pin_pool.h route_trace.h serve_codec.h edge_adapters.h edge_runtime.h edge_tok_internal.h fused_simd.h segment_adapter_internal.h segment_adapters.h segment_runtime.h sse41_kernels.h$(CC) $(NOCUDA_CFLAGS) olmoe.c -o olmoe$(EXE) $(NOCUDA_LDFLAGS)# Qwen3.6-35B-A3B engine (hybrid Gated Attention + Gated DeltaNet + streaming@@@ -1081,7 -1092,7 +1101,7 @@@ QWEN36_TIER_SRCQWEN36_CFLAGS = $(NOCUDA_CFLAGS)QWEN36_LDFLAGS = $(NOCUDA_LDFLAGS)endif- qwen36$(EXE): qwen36.c serve_poll.h cli_args.h qwen36_tier.h expert_ffn.h st.h json.h compat.h omp_tune.h kv_prefix.h edge_adapter_internal.h edge_adapters.h edge_runtime.h segment_adapter_internal.h segment_adapters.h segment_runtime.h gsgemv.h qgemv.h $(QWEN36_TIER_SRC) $(CUDA_OBJ)-qwen36$(EXE): qwen36.c decode_batch.h serve_poll.h cli_args.h qwen36_tier.h expert_ffn.h st.h json.h compat.h omp_tune.h kv_prefix.h pin_pool.h edge_adapter_internal.h edge_adapters.h edge_runtime.h segment_adapter_internal.h segment_adapters.h segment_runtime.h $(QWEN36_TIER_SRC) $(CUDA_OBJ)++qwen36$(EXE): qwen36.c decode_batch.h serve_poll.h cli_args.h qwen36_tier.h expert_ffn.h st.h json.h compat.h omp_tune.h kv_prefix.h pin_pool.h edge_adapter_internal.h edge_adapters.h edge_runtime.h segment_adapter_internal.h segment_adapters.h segment_runtime.h gsgemv.h qgemv.h $(QWEN36_TIER_SRC) $(CUDA_OBJ)$(CC) $(QWEN36_CFLAGS) qwen36.c $(QWEN36_TIER_SRC) $(CUDA_OBJ) -o qwen36$(EXE) $(QWEN36_LDFLAGS)# DeepSeek V4.1 Flash: one file, like every other portable engine. The fp4 experts