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
NameLast commit date
..
clef_tiny
deepseek_v4_tiny
gliner_decide_tiny
kimi_k3_tiny
laya_tiny
mimo_tiny
profiles
qwen36_tiny_bad
scripts
shaders
tests
tools
.gitignore
abl.h
affine_quant.h
autotune.py
backend_cuda_dsv4.cu
backend_cuda_dsv4.h
backend_cuda_ink.cu
backend_cuda_ink.h
backend_cuda.cu
backend_cuda.h
backend_gpu_compat.h
backend_loader_dsv4.c
backend_loader.c
backend_metal.h
backend_metal.mm
backend_vulkan.c
backend_vulkan.h
backend_xdna_helper.cpp
backend_xdna.c
backend_xdna.h
build_cuda.bat
cfse_pack.c
chat_olmoe.sh
clef_head.h
cli_args.h
cluster.py
coli
coli.cmd
colibri.c
compat_prefetch_win.h
compat.h
decide_serve.h
decide_vk.h
decode_batch.h
deepseek_v4_bank_pair.h
deepseek_v4_chain.h
deepseek_v4_dspark.inc
deepseek_v4_hybrid.h
deepseek_v4_internal.h
deepseek_v4.c
deepseek_v4.h
deepseek_v41_chain.h
deepseek_v41.c
delta_attention.h
doctor.py
download_fp8.py
dsv4_mhc.h
dsv4_quant.h
dsv4.def
edge_adapter_internal.h
edge_adapters.h
edge_runtime.c
edge_runtime.h
edge_tok_internal.h
evidence_digest.h
exact_dot.h
experiment_manifest.py
expert_ffn.h
expert_store_registry.c
expert_store_registry.h
expert_store.h
family_registry.py
fp8_format.h
fse_coli.h
fused_simd.h
glibc_c23_math_compat.h
gliner_decide.c
gliner_unicode.h
glm_chain.h
glm53_chain.h
glm53.c
grammar.h
gsgemv.h
hybrid_split.h
hyper_connections.h
idot.h
image_engine.py
inkling_chain.h
inkling.c
iobench.c
json.h
kimi_k3_chain.h
kimi_k3.c
kv_fp8.h
kv_image_key.h
kv_persist.h
kv_prefix.h
kv_tq.h
laya.c
load_fail.h
Makefile
Makefile.deepseek-v4
Makefile.deepseek-v4.units
matmul_f32.h
mcp_server.py
mimo_chain.h
mimo_vision.h
mimo.c
native_quant_batch.h
native_quant_dual.h
native_quant_fp4_rows16.h
native_quant.h
olmoe_chain.h
olmoe.c
omp_tune.h
openai_server.py
oracle.h
own_file.h
patches-deepgemm-sm120-msvc.patch
pin_pool.h
qgemv.h
qi_gemm.h
qpack_inspect.c
qpack.c
qpack.h
quant.h
qwen36_chain.h
qwen36_qpack.c
qwen36_qpack.h
qwen36_tier.c
qwen36_tier.h
qwen36.c
qwen38_chain.h
qwen38_core.h
qwen38_nfc_tables.h
qwen38_nfc.h
qwen38_vision.h
qwen38.c
qwenimage_chain.h
qwenimage_preview.h
qwenimage_vae_vk.h
qwenimage_vae.h
qwenimage.c
rans.h
ref_glm.json
ref_olmoe_real.json
ref_qwen36.json
resource_plan.py
route_trace.h
sample.h
schema_gbnf.h
segment_adapter_internal.h
segment_adapters.h
segment_runtime.c
segment_runtime.h
serve_budget.h
serve_codec.h
serve_poll.h
setup_catalog.py
setup_download.py
setup_flow.py
setup_hw.py
setup.sh
sparse_attn.h
sparse_index.h
spec_draft.h
sse41_kernels.h
st.h
stop_ids.h
telemetry.h
tensor.h
term_image.py
test_expert_store_registry.c
tier.h
tok_unicode_deepseek.h
tok_unicode_o200k.h
tok_unicode.h
tok.h
uring.h
v4_dsml.py
v41_dsml.py
version.py
vision_tower.h
vk_alloc.h
vk_chain.c
vk_chain.h
vk_kvsplit.h
vk_load.h
vk_tier.c
vk_tier.h
warmup.ps1