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
..
expert_atlas
jev-sdk
__init__.py
analyze_split_entropy.py
bench_dsv4_deepgemm_real.py
bench_fp4_matmul.c
bench_fp4_matmul.sh
bench_ux.sh
benchmark_baseline.py
benchmark_continuous_batch.py
benchmark_cuda_fixture.py
benchmark_http_serving.py
benchmark_kv_prefix.py
benchmark_serving_interference.py
benchmark_split_entropy.cu
benchmark_split_matvec.cu
blocksweep.py
build_xdna_package.py
check_ablate_evidence.py
check_glm53_checkpoint.py
check_glm53_container.c
clean.py
compare_clef.py
compare_gliner_decide.py
compare_laya.py
convert_fmt4_to_fmt2.py
convert_fp8_to_int4.py
convert_gguf_to_olmoe.py
convert_glm53.py
convert_inkling_dense_int4.py
convert_inkling_int4.py
convert_olmoe_merged.py
convert_olmoe.py
convert_qwen36.py
convert_qwen38_experts_int4.py
datapoint.py
diag_harness.py
disaggregation_break_even.py
download_glm52.py
dsv4_attn_oracle.py
dsv4_attn2_oracle.py
dsv4_bf16_deepgemm_probe.py
dsv4_block_oracle.py
dsv4_compressor_oracle.py
dsv4_dense_deepgemm_probe.py
dsv4_fc1_deepgemm_oracle.py
dsv4_masked_moe_probe.py
dsv4_mhc_oracle.py
dsv4_vllm_benchmark.py
dsv4_vllm_oracle.py
dsv4_vllm_reference.json
dsv4_vllm_reference.py
dsv4_vllm_shape_probe.py
dsv4_wo_a_deepgemm_probe.py
dsv41_image.py
dsv41_ref.py
efficiency.py
engine_evidence.py
eval_glm.py
fetch_benchmarks.py
fetch_deepgemm.sh
fp8_collision_census.py
gen_gliner_unicode.py
gen_qwen38_nfc.py
gen_unicode_deepseek.py
gen_unicode.py
gguf_dequant.py
gguf_olmoe_profile.py
gguf_reader.py
glm_fp8_emit.py
glm53_image.py
iq3_encode.c
iq3_pack.py
iq3xxs_grid.json
k3_ref.py
k3_repack.py
k3_tokenizer.py
make_clef_ref.py
make_clef_tiny.py
make_deepseek_tok_fixture.py
make_deepseek_v4_oracle.py
make_deepseek_v4_tiny.py
make_dsv41_tiny.py
make_e8_fixture.py
make_edge_tiny_tokenizer.py
make_gliner_decide_ref.py
make_gliner_decide_tiny.py
make_glm_bench_model.py
make_glm_mtp_tiny.py
make_glm_oracle.py
make_glm53_multimodal_tiny.py
make_glm53_streaming_pair.py
make_glm53_tiny.py
make_glm53_tokenizer.py
make_glm53_vision_tiny.py
make_kimi_k3_ref.py
make_kimi_k3_tiny.py
make_laya_ref.py
make_laya_tiny.py
make_mimo_ref.py
make_mimo_tiny.py
make_olmoe_real_oracle.py
make_olmoe_tiny.py
make_qwen36_oracle.py
make_qwen36_qpack_snap.py
make_qwen36_tiny.py
make_qwen36_vl_tiny.py
make_qwen38_multimodal_tiny.py
make_qwen38_oracle.py
make_qwen38_tiny.py
make_qwen38_vision_tiny.py
make_qwenimage_tiny.py
make_qwenimage_vae_tiny.py
make_tiny_inkling_audio.py
make_tiny_inkling.py
mimo_real_check.py
mirror_plan.py
mw_ur3_raw_adapter.py
oracle-requirements.txt
pack_python.py
placement_balance.py
placement_crossval.py
prepare_dsv41.py
qpack_http_install.py
qpack_install_policy.py
qpack_mirror_install.py
quant_ablation.py
qwen36_tensor_kinds.py
qwen38_image.py
qwen38_mtp_ref.py
qwenimage_fit_preview.py
qwenimage_ref.py
qwenimage_stub.py
rans_ctypes.c
rans_format.py
rans_verify.py
README.md
repack_fp8_passthrough.py
repack_rans.py
repair_mtp_int8.py
requirements-deepseek-v4-tiny.txt
requirements-gguf.txt
requirements-glm53-tiny.txt
requirements-jev-sdk.txt
requirements-kimi-k3-ref.txt
requirements-kimi-k3-tiny.txt
requirements-mimo-tiny.txt
requirements-qwen38-tiny.txt
requirements-qwenimage-tiny.txt
requirements.txt
residency_sim.py
route_coupling_report.py
route_pairs.py
run_f8_bench.sh
run_tests.py
test_olmoe_real.py
try_tool_calling.py
verify_fp8_decode.c