aiter
AI Tensor Engine for ROCm
파일 탐색기
최종 버전 다운로드 (.zip)파일 수가 많아 일부만 표시됩니다. 전체 파일은 위 다운로드 버튼으로 확인해 주세요.
- SKILL.md
- SKILL.md
- SKILL.md
- SKILL.md
- SKILL.md
- SKILL.md
- SKILL.md
- install
- pre-commit
- aiter-ops-triton.instructions.md
- triton-test.txt
- aiter_gpu_visibility_check.sh
- aiter_prebuild_summary.py
- aiter_test.sh
- build_aiter_triton.sh
- check_deps.sh
- check_signal.sh
- check_tuned_op_regression.sh
- clean_up_rocm.sh
- collect_logs.py
- compare_benchmark.py
- docker_pull_with_retry.sh
- download_triton_wheel.sh
- generate_summary.py
- install_triton.sh
- kimi_sglang_accuracy.sh
- kimi_sglang_perf.sh
- kimi_vllm_accuracy.sh
- kimi_vllm_perf.sh
- list_jobs.py
- op_tune.sh
- prebuild_vllm_disagg_cktile.py
- query_job_status.py
- select_triton_tests.py
- sglang_downstream.py
- split_tests.sh
- update_split_test_times.py
- verify_triton_pin.py
- actionlint.yaml
- aiter-release.yaml
- aiter-test.yaml
- amd-ci-job-monitor.yml
- atom-di-ci-smoke-workflow.yaml
- atom-test.yaml
- ci-config.yaml
- docs.yml
- ffm-bringup.yaml
- flash_attention_integration.yaml
- kimi-downstream.yaml
- kimi-perf-downstream.yaml
- nightly.yaml
- operators-tuning.yaml
- opus-test.yaml
- pr-title-tags.yaml
- pr-welcome-comment.yaml
- pre-checks.yaml
- prepare-triton-wheel.yaml
- promote.yaml
- sglang_downstream.yaml
- test-network.yaml
- test-whl.yaml
- triton-test.yaml
- tuning-tests.yaml
- update-split-tests.yaml
- vllm-disagg-ci-smoke-workflow.yaml
- vllm_benchmark.yaml
- actionlint.yaml
- runner-config.yml
- config.h
- composable_kernel
- chunk_gdn_h.py
- common.py
- fhmoe.py
- gemm.py
- grouped_moe.py
- moe.py
- mxfp4_moe.py
- README.md
- matmul_fp16.py
- test.sh
- test_hsaco.py
- test_matmul.cpp
- decode_mla.py
- norm.py
- __init__.py
- asm_mla_decode_fwd.py
- pa.py
- pa_ragged.py
- pa_v1.py
- sampling.py
- a8w8_blockscale_bpreshuffle_tuned_gemm_dsv3.csv
- a8w8_blockscale_bpreshuffle_tuned_gemm_glm5.2.csv
- a8w8_blockscale_bpreshuffle_tuned_gemm_mm2.5.csv
- a8w8_blockscale_bpreshuffle_tuned_gemm_qwen3.5_397b.csv
- a8w8_blockscale_bpreshuffle_tuned_gemm_qwen3_235b.csv
- a8w8_blockscale_bpreshuffle_untuned_gemm_dsv3.csv
- a8w8_blockscale_bpreshuffle_untuned_gemm_mm2.5.csv
- a8w8_blockscale_bpreshuffle_untuned_gemm_qwen3.5_397b.csv
- a8w8_blockscale_bpreshuffle_untuned_gemm_qwen3_235b.csv
- a8w8_blockscale_tuned_fmoe_ds_v3.csv
- a8w8_blockscale_tuned_fmoe_glm5.csv
- a8w8_blockscale_tuned_fmoe_glm5_1.csv
- a8w8_blockscale_tuned_fmoe_minimax-m2_5.csv
- a8w8_blockscale_tuned_fmoe_qwen3_235b.csv
- a8w8_blockscale_tuned_fmoe_qwen3_5_397b.csv
- a8w8_blockscale_tuned_gemm_ds_v3.csv
- a8w8_blockscale_tuned_gemm_glm5_1.csv
- a8w8_blockscale_tuned_gemm_qwen36_27b.csv
- a8w8_blockscale_tuned_gemm_qwen3_235b.csv
- a8w8_blockscale_tuned_gemm_qwen3_5_397b_a13b.csv
- a8w8_blockscale_tuned_gemm_qwen3_next_80b_a3b.csv
- a8w8_blockscale_tuned_gemm_qwen3_vl_235b.csv
- a8w8_blockscale_tuned_gemm_qwen3_vl_32b.csv
- a8w8_blockscale_untuned_fmoe_ds_v3.csv
- a8w8_blockscale_untuned_fmoe_glm5.csv
- a8w8_blockscale_untuned_fmoe_minimax-m2_5.csv
- a8w8_blockscale_untuned_fmoe_qwen3_235b.csv
- a8w8_blockscale_untuned_fmoe_qwen3_5_397b.csv
- a8w8_blockscale_untuned_gemm_ds_v3.csv
- a8w8_blockscale_untuned_gemm_qwen3_235b.csv
- a8w8_blockscale_untuned_gemm_qwen3_next_80b_a3b.csv
- a8w8_bpreshuffle_tuned_gemm_glm47.csv
- a8w8_bpreshuffle_tuned_gemm_glm5.2.csv
- a8w8_bpreshuffle_tuned_gemm_kimik3.csv
- a8w8_bpreshuffle_tuned_gemm_minimax_m3.csv
- a8w8_bpreshuffle_tuned_gemm_qwen3_5_397b_a17b_mxfp4_attnfp8.csv
- a8w8_bpreshuffle_untuned_gemm_glm47.csv
- a8w8_bpreshuffle_untuned_gemm_glm5.2.csv
- a8w8_bpreshuffle_untuned_gemm_kimik3.csv
- a8w8_bpreshuffle_untuned_gemm_minimax_m3.csv
- dsv3_a4w4_blockscale_tuned_gemm.csv
- dsv3_a4w4_blockscale_untuned_gemm.csv
- dsv3_a8w8_bpreshuffle_tuned_gemm.csv
- dsv3_bf16_tuned_gemm.csv
- dsv3_bf16_untuned_gemm.csv
- dsv3_fp4_tuned_fmoe.csv
- dsv3_fp4_untuned_fmoe.csv
- dsv4_a8w8_blockscale_bpreshuffle_tuned_gemm.csv
- dsv4_a8w8_blockscale_tuned_gemm.csv
- dsv4_a8w8_blockscale_untuned_gemm.csv
- dsv4_batched_gemm_a8w8_blockscale_mxscale_tuned.csv
- dsv4_bf16_tuned_gemm.csv
- dsv4_bf16_untuned_gemm.csv
- dsv4_fp8fp4_tuned_fmoe.csv
- dsv4_fp8fp4_untuned_fmoe.csv
- dsv4pro_bf16_tuned_gemm.csv
- flux2_bf16_tuned_gemm.csv
- glm47_bf16_tuned_gemm.csv
- glm47_bf16_untuned_gemm.csv
- glm47_fp8_tuned_fmoe.csv
- glm47_fp8_untuned_fmoe.csv
- glm5_a8w8_blockscale_bpreshuffle_tuned_gemm.csv
- glm5_a8w8_blockscale_bpreshuffle_untuned_gemm.csv
- glm5_a8w8_bpreshuffle_tuned_gemm.csv
- glm5_a8w8_bpreshuffle_untuned_gemm.csv
- glm5_bf16_tuned_gemm.csv
- glm5_bf16_untuned_gemm.csv
- glm5_fp4_tuned_fmoe.csv
- glm5_fp4_untuned_fmoe.csv
- glm5_mxfp8_tuned_fmoe.csv
- glm5_mxfp8_untuned_fmoe.csv
- glm5_ptpc_tuned_fmoe.csv
- glm5_ptpc_untuned_fmoe.csv
- glm5_untuned_gemm_bf16.csv
- gptoss_a16w4_tuned_fmoe.csv
- gptoss_a16w4_untuned_fmoe.csv
- gptoss_bf16_tuned_gemm.csv
- gptoss_bf16_untuned_gemm.csv
- gptoss_fp4_tuned_fmoe.csv
- gptoss_fp4_untuned_fmoe.csv
- gptoss_fp8fp4_tuned_fmoe.csv
- gptoss_fp8fp4_untuned_fmoe.csv
- kimi_bf16_tuned_gemm.csv
- kimi_bf16_untuned_gemm.csv
- kimik2_bf16_tuned_gemm.csv
- kimik2_bf16_untuned_gemm.csv
- kimik2_fp4_tuned_fmoe.csv
- kimik2_fp4_untuned_fmoe.csv
- kimik2_fp8fp4_tuned_fmoe.csv
- kimik2_fp8fp4_untuned_fmoe.csv
- kimik2_i4_tuned_fmoe.csv
- kimik2_i4_untuned_fmoe.csv
- kimik3_a4w4_blockscale_tuned_gemm.csv
- kimik3_a4w4_blockscale_untuned_gemm.csv
- kimik3_a4w4_tuned_fmoe.csv
- kimik3_a4w4_untuned_fmoe.csv
- kimik3_a8w4_tuned_fmoe.csv
- kimik3_a8w4_untuned_fmoe.csv
- kimik3_bf16_tuned_gemm.csv
- kimik3_bf16_untuned_gemm.csv
- kimik3_fp4_tuned_fmoe.csv
- kimik3_fp4_untuned_fmoe.csv
- llama405B_bf16_tuned_gemm.csv
- llama405B_untuned_gemm.csv
- llama405B_untuned_gemm_bf16.csv
- llama70B_bf16_tuned_gemm.csv
- llama70B_untuned_gemm.csv
- llama70B_untuned_gemm_bf16.csv
- mimo_v2_5_pro_tuned_fmoe.csv
- mimo_v2_5_pro_untuned_fmoe.csv
- minimax_m25_fp4_tuned_fmoe.csv
- minimax_m25_fp4_untuned_fmoe.csv
- minimax_m3_eagle_bf16_tuned_gemm.csv
- minimax_m3_eagle_bf16_untuned_gemm.csv
- minimax_m3_fp4_tuned_fmoe.csv
- minimax_m3_fp4_untuned_fmoe.csv
- minimax_m3_mxfp8_tuned_fmoe.csv
- minimax_m3_mxfp8_untuned_fmoe.csv
- qwen32B_bf16_tuned_gemm.csv
- qwen32B_untuned_gemm.csv
- qwen32B_untuned_gemm_bf16.csv
- qwen3_235b_bf16_tuned_fmoe.csv
- qwen3_235b_bf16_untuned_fmoe.csv
- qwen3_5_35b_chunk_gdn_h_opt_tuned.csv
- qwen3_5_35b_chunk_gdn_h_opt_untuned.csv
- qwen3_5_397b_bf16_tuned_gemm.csv
- qwen3_5_397b_chunk_gdn_h_opt_tuned.csv
- qwen3_5_397b_chunk_gdn_h_opt_untuned.csv
- qwen3_5_397b_fp4_tuned_fmoe.csv
- qwen3_5_397b_fp4_untuned_fmoe.csv
- qwen3_5_397b_untuned_gemm.csv
- qwen3_8_2400b_fp4_tuned_fmoe.csv
- qwen3_8_2400b_fp4_untuned_fmoe.csv
- qwen3_8b_bf16_tuned_gemm.csv
- qwen3_vl_fp4_tuned_fmoe.csv
- qwen3_vl_fp4_untuned_fmoe.csv
- qwen3next_80b_fp8_tuned_fmoe.csv
- qwen3next_80b_fp8_untuned_fmoe.csv
- README.md
- __init__.py
- a4w4_blockscale_tuned_gemm.csv
- a4w4_blockscale_untuned_gemm.csv
- a8w8_blockscale_bpreshuffle_tuned_gemm.csv
- a8w8_blockscale_bpreshuffle_untuned_gemm.csv
- a8w8_blockscale_tuned_gemm.csv
- a8w8_blockscale_untuned_gemm.csv
- a8w8_bpreshuffle_tuned_gemm.csv
- a8w8_bpreshuffle_untuned_gemm.csv
- a8w8_tuned_batched_gemm.csv
- a8w8_tuned_gemm.csv
- a8w8_untuned_batched_gemm.csv
- a8w8_untuned_gemm.csv
- asm_a8w8_gemm.csv
- bf16_tuned_batched_gemm.csv
- bf16_tuned_gemm.csv
- bf16_untuned_batched_gemm.csv
- bf16_untuned_gemm.csv
- chunk_gdn_h_opt_tuned.csv
- chunk_gdn_h_opt_untuned.csv
- tuned_fhmoe.csv
- tuned_fmoe.csv
- tuned_grouped_fmoe.csv
- untuned_fhmoe.csv
- untuned_fmoe.csv
- untuned_grouped_fmoe.csv
- all2all.py
- base_device_communicator.py
- communicator_cuda.py
- communicator_pynccl.py
- custom_all_reduce.py
- pynccl_wrapper.py
- quick_all_reduce.py
- README.md
- rocm_version.py
- vmm_allocator.py
- __init__.py
- communication_op.py
- cuda_wrapper.py
- parallel_state.py
- shm_broadcast.py
- utils.py
- __init__.py
- constants.py
- cuda_to_hip_mappings.py
- hipify_python.py
- __init__.py
- _cpp_extension_versioner.py
- build_targets.py
- chip_info.py
- cpp_extension.py
- file_baton.py
- mha_recipes.py
- moe_recipes.py
- torch_guard.py
- __init__.py
- core.py
- optCompilerConfig.json
- flydsl_gemm_a8w8_bpreshuffle_common.py
- flydsl_gemm_a8w8_bpreshuffle_wmma_common.py
- flydsl_gemm_mxfp8_128_bpreshuffle_wmma_common.py
- __init__.py
- fmha_core_loop.py
- fmha_kernel.py
- fmha_prologue.py
- fmha_schedule.py
- __init__.py
- dispatch.py
- gemm1.py
- gemm2.py
- gemm_util.py
- mega_moe_config.py
- mega_moe_stage1.py
- mega_moe_stage2.py
- mega_moe_v2.py
- quant.py
- __init__.py
- combine.py
- config.py
- dispatch.py
- mega_moe.py
- tdm_gather_shim.py
- types.py
- flydsl_gfx950_mi355x_IntraNode_ep8.json
- __init__.py
- gemm1.py
- gemm2.py
- utils.py
- __init__.py
- fp8_mqa_logits.py
- pa_mqa_logits_fp4.py
- pa_mqa_logits_fp4_common.py
- pa_mqa_logits_fp4_prefill.py
- __init__.py
- act.py
- buffer_ops.py
- chunk_gated_delta_h.py
- chunk_gated_delta_h_opt.py
- communication_ops_utils.py
- dpp_utils.py
- fhmoe.py
- flash_attn_func_gfx1201.py
- flydsl_dispatch_combine_intranode_kernel.py
- flydsl_dispatch_combine_intranode_op.py
- fused_compress_attn.py
- fused_compress_attn_common.py
- fused_compress_attn_gfx1250.py
- fused_compress_attn_hca.py
- fused_compress_attn_hca_gfx1250.py
- gdn_prepare.py
- gdr_decode.py
- gemm_a8w8_256x256_gfx1250.py
- gemm_a8w8_8wave.py
- gemm_a8w8_gfx1250.py
- gemm_a8w8_splitk_reduce_gfx1250.py
- gemm_common_gfx1250.py
- gfx1250_cluster.py
- hgemm_dispatch.py
- kernels_common.py
- layout_utils.py
- mfma_epilogues.py
- mfma_preshuffle_pipeline.py
- mixed_moe_gemm_2stage.py
- mixed_moe_gemm_2stage_common.py
- mla_reduce.py
- moe_contiguous_psum.py
- moe_fused_route_quant_scatter.py
- moe_g2l_lut.py
- moe_gather_reduce.py
- moe_m_tile_map.py
- moe_reduce.py
- moe_route_maps.py
- moe_scatter_copy_preshuffle_scale.py
- moe_scatter_copy_token.py
- moe_sorting_kernel.py
- mxfp4_gemm1.py
- mxfp4_gemm2.py
- mxfp4_gemm_common.py
- mxfp4_preshuffle_batched_gemm_gfx1250_tdm.py
- mxfp4_preshuffle_gfx1250_tdm.py
- mxmoe_dispatcher.py
- mxmoe_gemm_v2.py
- preshuffle_gemm.py
- qk_norm_rope_quant.py
- qk_norm_rope_quant_gfx1250.py
- quant_utils.py
- silu_and_mul_fq.py
- small_m_hgemm.py
- splitk_hgemm.py
- swiglu_and_mul.py
- tdm_ops_gfx1250.py
- tensor_shim.py
- vector.py
- __init__.py
- batched_gemm_mxfp4.py
- bpreshuffle_gemm_gfx1250.py
- causal_conv1d_flydsl.py
- chunk_gdn_h_tuned.csv
- fhmoe.py
- fmha_kernels.py
- gdr_decode_tuned.csv
- gemm_a8w8_bpreshuffle_8wave.py
- gemm_kernels.py
- grouped_gemm_mxfp4.py
- grouped_moe_gfx1250.py
- linear_attention_kernels.py
- linear_attention_prefill_kernels.py
- mla_reduce_kernels.py
- moe_common.py
- moe_kernels.py
- moe_sorting.py
- mxfp4_gemm1_kernels.py
- mxfp4_gemm2_kernels.py
- mxfp4_kname.py
- mxfp4_v2_tune_utils.py
- mxfp8_128_bpreshuffle_gemm_gfx1250.py
- test_flydsl_linear_attention.py
- test_flydsl_moe_a4w4.py
- test_flydsl_splitk_hgemm.py
- utils.py
- __init__.py
- _arch.py
- bmm_op.py
- common.py
- gemm_op_a16w16.py
- gemm_op_a8w8.py
- moe_stage1_a8w4.py
- moe_stage2_a8w4.py
- README.md
- __init__.py
- fused_compress_attn.py
- fp8_mqa_logits.py
- mla.py
- pa_decode_sparse.py
- pa_prefill_sparse.py
- unified_attention_2d.py
- unified_attention_3d.py
- unified_attention_reduce.py
- fused_kv_cache.py
- gemm_a16w16.py
- gemm_a8w8_blockscale.py
- gemm_mxfp4.py
- batched_gemm_bf16.py
- moe_op_gemm_a16w4.py
- moe_op_gemm_a4w4.py
- moe_op_gemm_a8w4.py
- reduce.py
- fused_rmsnorm_add.py
- __init__.py
- fused_mxfp4_quant.py
- __init__.py
- moe_op_gemm_int8_smoothquant.py
- __init__.py
- fp8_mqa_logits.py
- pa_decode_sparse.py
- __init__.py
- __init__.py
- block_lut.py
- chunked_pa_prefill.py
- extend_attention.py
- fav3_sage_attention.py
- fav3_sage_attention_mxfp4.py
- fp8_mqa_logits.py
- hstu_attention.py
- lean_atten.py
- lean_atten_paged.py
- mha.py
- mha_fused_bwd.py
- mha_onekernel_bwd.py
- mla.py
- mla_decode_rope.py
- pa_decode.py
- pa_decode_sparse.py
- pa_mqa_logits.py
- pa_prefill.py
- pod_attention.py
- prefill_attention.py
- sparse_attention_dsv4.py
- unified_attention.py
- unified_attention_sparse_mla.py
- __init__.py
- cumsum.py
- index.py
- l2norm.py
- __init__.py
- chunk_delta_attn_utils.py
- chunk_fwd.py
- flash_kda.py
- gate.py
- gla_output.py
- intra_attn.py
- wy_fast.py
- __init__.py
- splitk_reduce.py
- __init__.py
- causal_conv1d.py
- causal_conv1d_update_single_token.py
- conv_1x1.py
- conv_3x3.py
- conv_3x3_winograd_f4x3.py
- conv_general.py
- helpers.py
- __init__.py
- bwd.py
- common.py
- fwd_decode.py
- fwd_prefill.py
- interface_v2.py
- interface_v3.py
- utils.py
- __init__.py
- attn_res.py
- fused_bmm_rope_kv_cache.py
- fused_clamp_act_mul.py
- fused_kv_cache.py
- fused_mul_add.py
- fused_qk_concat.py
- fused_reduce_qk_norm_rope_swa_write.py
- fused_routing_from_topk.py
- mhc.py
- __init__.py
- causal_conv1d_update_split_qkv.py
- fused_rearrange_sigmoid_gdr.py
- fused_recurrent.py
- fused_sigmoid_gating_recurrent.py
- __init__.py
- causal_conv1d_fwd_split_qkv.py
- chunk.py
- chunk_delta_h.py
- chunk_o.py
- fused_cumsum_kkt.py
- fused_gdn_gating_prefill.py
- fused_solve_tril_recompute.py
- __init__.py
- cumsum.py
- index.py
- l2norm.py
- op.py
- prefill_metadata.py
- solve_tril.py
- wy_representation.py
- __init__.py
- fused_qkvzba_split.py
- gated_delta_rule_utils.py
- __init__.py
- gemm_a16w16.py
- gemm_a16w16_atomic.py
- gemm_a16w16_gated.py
- gemm_a16w8_blockscale.py
- gemm_a16wfp4.py
- gemm_a8w8.py
- gemm_a8w8_blockscale.py
- gemm_a8w8_per_token_scale.py
- gemm_a8wfp4.py
- gemm_afp4wfp4.py
- gemm_afp8wfp8.py
- __init__.py
- batched_gemm_a16wfp4.py
- batched_gemm_a8w8.py
- batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant.py
- batched_gemm_afp4wfp4.py
- batched_gemm_bf16.py
- __init__.py
- ff_a16w16_fused_gated.py
- ff_a16w16_fused_ungated.py
- __init__.py
- fused_gemm_a16w16_quant_x.py
- fused_gemm_a8w8_blockscale_a16w16.py
- fused_gemm_a8w8_blockscale_mul_add.py
- fused_gemm_a8w8_blockscale_split_cat.py
- fused_gemm_afp4wfp4_a16w16.py
- fused_gemm_afp4wfp4_mul_add.py
- fused_gemm_afp4wfp4_split_cat.py
- __init__.py
- bitmatrix.py
- expt_data.py
- minunique.py
- routing.py
- topk.py
- __init__.py
- activations.py
- moe_align_block_size.py
- moe_op.py
- moe_op_e2e.py
- moe_op_gelu.py
- moe_op_gemm_a16w4.py
- moe_op_gemm_a4w4.py
- moe_op_gemm_a8w4.py
- moe_op_gemm_a8w8.py
- moe_op_gemm_a8w8_blockscale.py
- moe_op_gemm_int8_smoothquant.py
- moe_op_mxfp4.py
- moe_op_mxfp4_silu_fused.py
- moe_op_silu_fused.py
- moe_routing_sigmoid_top1_fused.py
- quant_moe.py
- reduce.py
- __init__.py
- fused_add_rmsnorm_pad.py
- fused_rmsnorm_add.py
- norm.py
- rmsnorm.py
- __init__.py
- fused_fp8_quant.py
- fused_mxfp4_quant.py
- fused_mxfp8_quant.py
- quant.py
- sage_attention_quant.py
- __init__.py
- fused_qkv_split_qk_norm_rope_cache.py
- fused_qkv_split_qk_rope.py
- rope.py
- activation.py
- gather_kv_b_proj.py
- gmm.py
- kv_cache.py
- softmax.py
- topk.py
- __init__.py
- chunked_pa_prefill.py
- extend_attention.py
- fav3_sage.py
- fav3_sage_attention_mxfp4_wrapper.py
- fp8_mqa_logits.py
- hstu_attention.py
- lean_atten.py
- lean_atten_paged.py
- mha.py
- mha_fused_bwd.py
- mha_onekernel_bwd.py
- mha_v3.py
- mla.py
- mla_decode.py
- mla_decode_rope.py
- pa_decode.py
- pa_decode_sparse.py
- pa_mqa_logits.py
- pa_prefill.py
- pa_prefill_sparse.py
- pod_attention.py
- prefill_attention.py
- unified_attention.py
- unified_attention_sparse_mla.py
- utils.py
- __init__.py
- reduce_scatter_rmsnorm_quant_all_gather.py
- __init__.py
- all_gather.py
- iris.py
- reduce_scatter.py
- gfx1100-CONV-1X1.json
- gfx1100-CONV-3X3-CBLOCKED.json
- gfx1100-CONV-3X3-NHWC.json
- gfx1100-CONV-GENERAL.json
- gfx1100-CONV-WINO-F4X3-GEMM.json
- gfx1100-CONV-WINO-F4X3-INPUT.json
- gfx1100-CONV-WINO-F4X3-OUTPUT.json
- gfx1151-CONV-1X1.json
- gfx1151-CONV-3X3-CBLOCKED.json
- gfx1151-CONV-3X3-NHWC.json
- gfx1151-CONV-GENERAL.json
- gfx1151-CONV-WINO-F4X3-GEMM.json
- gfx1151-CONV-WINO-F4X3-INPUT.json
- gfx1151-CONV-WINO-F4X3-OUTPUT.json
- gfx1200-CONV-1X1.json
- gfx1200-CONV-3X3-CBLOCKED.json
- gfx1200-CONV-3X3-NHWC.json
- gfx1200-CONV-GENERAL.json
- gfx1200-CONV-WINO-F4X3-GEMM.json
- gfx1200-CONV-WINO-F4X3-INPUT.json
- gfx1200-CONV-WINO-F4X3-OUTPUT.json
- gfx1201-CONV-1X1.json
- gfx1201-CONV-3X3-CBLOCKED.json
- gfx1201-CONV-3X3-NHWC.json
- gfx1201-CONV-GENERAL.json
- gfx1201-CONV-WINO-F4X3-GEMM.json
- gfx1201-CONV-WINO-F4X3-INPUT.json
- gfx1201-CONV-WINO-F4X3-OUTPUT.json
- gfx1250-CONV-1X1.json
- gfx1250-CONV-3X3-CBLOCKED.json
- gfx1250-CONV-3X3-NHWC.json
- gfx1250-CONV-GENERAL.json
- gfx1250-CONV-WINO-F4X3-GEMM.json
- gfx1250-CONV-WINO-F4X3-INPUT.json
- gfx1250-CONV-WINO-F4X3-OUTPUT.json
- gfx942-CONV-1X1.json
- gfx942-CONV-3X3-CBLOCKED.json
- gfx942-CONV-3X3-NHWC.json
- gfx942-CONV-GENERAL.json
- gfx942-CONV-WINO-F4X3-GEMM.json
- gfx942-CONV-WINO-F4X3-INPUT.json
- gfx942-CONV-WINO-F4X3-OUTPUT.json
- gfx950-CONV-1X1.json
- gfx950-CONV-3X3-CBLOCKED.json
- gfx950-CONV-3X3-NHWC.json
- gfx950-CONV-GENERAL.json
- gfx950-CONV-WINO-F4X3-GEMM.json
- gfx950-CONV-WINO-F4X3-INPUT.json
- gfx950-CONV-WINO-F4X3-OUTPUT.json
- gfx1250-BATCHED_GEMM-A16W16-B=16-N=1024-K=4096.json
- gfx1250-BATCHED_GEMM-A16W16-B=4-N=1024-K=4096.json
- gfx1250-BATCHED_GEMM-A16W16-B=8-N=1024-K=4096.json
- gfx1250-BATCHED_GEMM-A16W16.json
- gfx1250-GEMM-A8W8_BLOCKSCALE.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=1536-K=4096.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=1536-K=7168.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=16384-K=1536.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=2048-K=7168.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=32768-K=1024.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=2048.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=4096.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=8192.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=6144-K=7168.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=65536-K=1536.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=16384.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=3072.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=4096.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=768.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=8192-K=1024.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=8192-K=1536.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED.json
- gfx950-GEMM-A8W8.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=2112-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=3072-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=4608-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=512-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=7168-K=2048.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=7168-K=256.json
- gfx950-GEMM-A8W8_BLOCKSCALE.json
- gfx1151-GEMM-A8W8.json
- gfx1201-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=128-K=512.json
- gfx1201-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=512-K=128.json
- gfx1201-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT.json
- gfx1201-GEMM-A8W8-N=28672-K=4096.json
- gfx1201-GEMM-A8W8-N=4096-K=14336.json
- gfx1201-GEMM-A8W8-N=4096-K=4096.json
- gfx1201-GEMM-A8W8-N=6144-K=4096.json
- gfx1201-GEMM-A8W8.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=1024-K=4096.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=1024-K=8192.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=1536-K=4096.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=16384-K=1536.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=2048-K=2048.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=2112-K=7168.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=24576-K=1536.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=2624-K=6144.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=3072-K=1536.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=3072-K=6144.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=32768-K=512.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=32768-K=8192.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=3584-K=512.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=36864-K=7168.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=4096-K=1024.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=4096-K=256.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=4096-K=512.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=4096-K=7168.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=4608-K=7168.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=512-K=4096.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=512-K=7168.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=6144-K=1536.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=6144-K=2048.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=7168-K=16384.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=7168-K=18432.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=7168-K=2048.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=7168-K=2304.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=7168-K=256.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=8192-K=1024.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=8192-K=32768.json
- gfx1201-GEMM-A8W8_BLOCKSCALE-N=8192-K=8192.json
- gfx1201-GEMM-A8W8_BLOCKSCALE.json
- gfx1201-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=24576-K=4096.json
- gfx1201-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=12288.json
- gfx1201-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=4096.json
- gfx1201-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=6144-K=4096.json
- gfx1201-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED.json
- gfx1250-BATCHED_GEMM-A16W16-B=4-N=1024-K=4096.json
- gfx1250-BATCHED_GEMM-A16W16-B=8-N=1024-K=4096.json
- gfx1250-BATCHED_GEMM-A16W16-N=1024-K=4096.json
- gfx1250-BATCHED_GEMM-A16W16-N=1024-K=7168.json
- gfx1250-BATCHED_GEMM-A16W16.json
- gfx1250-BATCHED_GEMM-A16WFP4.json
- gfx1250-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT.json
- gfx1250-BATCHED_GEMM-A8W8.json
- gfx1250-BATCHED_GEMM-AFP4WFP4.json
- gfx1250-FF-A16W16-fused.json
- gfx1250-FUSED-GEMM-A8W8_BLOCKSCALE-A16W16.json
- gfx1250-FUSED-GEMM-AFP4WFP4-A16W16.json
- gfx1250-FUSED-GEMM-AFP4WFP4_PRESHUFFLED-A16W16.json
- gfx1250-GEMM-A16W16-ATOMIC.json
- gfx1250-GEMM-A16W16-gated.json
- gfx1250-GEMM-A16W16-N=1024-K=4096.json
- gfx1250-GEMM-A16W16-N=1024-K=7168.json
- gfx1250-GEMM-A16W16-N=10248-K=2880.json
- gfx1250-GEMM-A16W16-N=128-K=2880.json
- gfx1250-GEMM-A16W16-N=128-K=3072.json
- gfx1250-GEMM-A16W16-N=129280-K=4096.json
- gfx1250-GEMM-A16W16-N=2048-K=4096.json
- gfx1250-GEMM-A16W16-N=2048-K=7168.json
- gfx1250-GEMM-A16W16-N=256-K=4096.json
- gfx1250-GEMM-A16W16-N=2880-K=4096.json
- gfx1250-GEMM-A16W16-N=32320-K=7168.json
- gfx1250-GEMM-A16W16-N=36864-K=6144.json
- gfx1250-GEMM-A16W16-N=384-K=7168.json
- gfx1250-GEMM-A16W16-N=512-K=4096.json
- gfx1250-GEMM-A16W16-N=512-K=7168.json
- gfx1250-GEMM-A16W16-N=5120-K=2880.json
- gfx1250-GEMM-A16W16-N=5120-K=3072.json
- gfx1250-GEMM-A16W16-N=55296-K=6144.json
- gfx1250-GEMM-A16W16-N=6144-K=18432.json
- gfx1250-GEMM-A16W16-N=6144-K=24576.json
- gfx1250-GEMM-A16W16-N=6144-K=6144.json
- gfx1250-GEMM-A16W16-N=64-K=4096.json
- gfx1250-GEMM-A16W16-N=64-K=7168.json
- gfx1250-GEMM-A16W16-N=7168-K=16384.json
- gfx1250-GEMM-A16W16.json
- gfx1250-GEMM-A16W8_BLOCKSCALE.json
- gfx1250-GEMM-A16W8_BLOCKSCALE_PRESHUFFLED.json
- gfx1250-GEMM-A16WFP4.json
- gfx1250-GEMM-A16WFP4_PRESHUFFLED.json
- gfx1250-GEMM-A8W8.json
- gfx1250-GEMM-A8W8_BLOCKSCALE.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=1536-K=4096.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=32768-K=1024.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=2048.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=4096.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=8192.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=8192-K=1024.json
- gfx1250-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED.json
- gfx1250-GEMM-A8W8_PER_TOKEN_SCALE.json
- gfx1250-GEMM-A8WFP4.json
- gfx942-BATCHED_GEMM-A16W16-N=1024-K=4096.json
- gfx942-BATCHED_GEMM-A16W16.json
- gfx942-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=128-K=512.json
- gfx942-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=512-K=128.json
- gfx942-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT.json
- gfx942-BATCHED_GEMM-A8W8.json
- gfx942-FF-A16W16-fused.json
- gfx942-FUSED-GEMM-A8W8_BLOCKSCALE-A16W16.json
- gfx942-GEMM-A16W16-ATOMIC.json
- gfx942-GEMM-A16W16-gated.json
- gfx942-GEMM-A16W16.json
- gfx942-GEMM-A16W8_BLOCKSCALE.json
- gfx942-GEMM-A16W8_BLOCKSCALE_PRESHUFFLED.json
- gfx942-GEMM-A8W8-N=128-K=2048.json
- gfx942-GEMM-A8W8-N=128-K=4096.json
- gfx942-GEMM-A8W8-N=128-K=6144.json
- gfx942-GEMM-A8W8-N=256-K=2048.json
- gfx942-GEMM-A8W8-N=256-K=4096.json
- gfx942-GEMM-A8W8-N=256-K=6144.json
- gfx942-GEMM-A8W8.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=2048-K=2048.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=2112-K=7168.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=2624-K=6144.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=3072-K=1536.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=3072-K=6144.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=3584-K=512.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=6144-K=1536.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=6144-K=2048.json
- gfx942-GEMM-A8W8_BLOCKSCALE-N=7168-K=2048.json
- gfx942-GEMM-A8W8_BLOCKSCALE.json
- gfx942-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=1536-K=4096.json
- gfx942-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=1024.json
- gfx942-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=8192-K=1024.json
- gfx942-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED.json
- gfx942-GEMM-A8W8_PER_TOKEN_SCALE.json
- gfx942-GEMM-AFP8WFP8.json
- gfx942-GEMM-AFP8WFP8_PRESHUFFLED.json
- gfx950-BATCHED_GEMM-A16W16-N=1024-K=4096.json
- gfx950-BATCHED_GEMM-A16W16-N=1024-K=7168.json
- gfx950-BATCHED_GEMM-A16W16.json
- gfx950-BATCHED_GEMM-A16WFP4-N=128-K=512.json
- gfx950-BATCHED_GEMM-A16WFP4-N=8192-K=8192.json
- gfx950-BATCHED_GEMM-A16WFP4.json
- gfx950-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=128-K=512.json
- gfx950-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=512-K=128.json
- gfx950-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT-N=8192-K=8192.json
- gfx950-BATCHED_GEMM-A8W8-A_PER_TOKEN_GROUP_PREQUANT_W_PER_BATCHED_TENSOR_QUANT.json
- gfx950-BATCHED_GEMM-A8W8.json
- gfx950-BATCHED_GEMM-AFP4WFP4-N=128-K=512.json
- gfx950-BATCHED_GEMM-AFP4WFP4-N=512-K=128.json
- gfx950-BATCHED_GEMM-AFP4WFP4.json
- gfx950-FF-A16W16-fused.json
- gfx950-FUSED-GEMM-A8W8_BLOCKSCALE-A16W16-N8=512-N16=256-K=7168.json
- gfx950-FUSED-GEMM-A8W8_BLOCKSCALE-A16W16.json
- gfx950-FUSED-GEMM-A8W8_BLOCKSCALE-MUL_ADD-N=7168-K=256.json
- gfx950-FUSED-GEMM-A8W8_BLOCKSCALE-MUL_ADD.json
- gfx950-FUSED-GEMM-AFP4WFP4-A16W16-N4=512-N16=256-K=7168.json
- gfx950-FUSED-GEMM-AFP4WFP4-A16W16.json
- gfx950-FUSED-GEMM-AFP4WFP4-MUL_ADD-N=7168-K=256.json
- gfx950-FUSED-GEMM-AFP4WFP4-MUL_ADD.json
- gfx950-FUSED-GEMM-AFP4WFP4_PRESHUFFLED-A16W16.json
- gfx950-GEMM-A16W16-ATOMIC-N=256-K=6144.json
- gfx950-GEMM-A16W16-ATOMIC-N=256-K=7168.json
- gfx950-GEMM-A16W16-ATOMIC.json
- gfx950-GEMM-A16W16-gated.json
- gfx950-GEMM-A16W16-N=128-K=2880.json
- gfx950-GEMM-A16W16-N=128-K=4096.json
- gfx950-GEMM-A16W16-N=128-K=5120.json
- gfx950-GEMM-A16W16-N=256-K=7168.json
- gfx950-GEMM-A16W16-N=2880-K=4096.json
- gfx950-GEMM-A16W16-N=2880-K=512.json
- gfx950-GEMM-A16W16-N=384-K=7168.json
- gfx950-GEMM-A16W16-N=5120-K=2880.json
- gfx950-GEMM-A16W16-N=640-K=2880.json
- gfx950-GEMM-A16W16.json
- gfx950-GEMM-A16W8_BLOCKSCALE-N=7168-K=2048.json
- gfx950-GEMM-A16W8_BLOCKSCALE.json
- gfx950-GEMM-A16W8_BLOCKSCALE_PRESHUFFLED-N=2112-K=7168.json
- gfx950-GEMM-A16W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=2048.json
- gfx950-GEMM-A16W8_BLOCKSCALE_PRESHUFFLED.json
- gfx950-GEMM-A16WFP4-N=512-K=7168.json
- gfx950-GEMM-A16WFP4-N=7168-K=2048.json
- gfx950-GEMM-A16WFP4.json
- gfx950-GEMM-A16WFP4_PRESHUFFLED-N=2112-K=7168.json
- gfx950-GEMM-A16WFP4_PRESHUFFLED.json
- gfx950-GEMM-A8W8-N=1280-K=8192.json
- gfx950-GEMM-A8W8-N=2048-K=7168.json
- gfx950-GEMM-A8W8-N=2112-K=7168.json
- gfx950-GEMM-A8W8.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=1024-K=8192.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=16384-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=2112-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=24576-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=3072-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=32768-K=512.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=32768-K=8192.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=36864-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=4096-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=4608-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=512-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=7168-K=16384.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=7168-K=18432.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=7168-K=2048.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=7168-K=256.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=8192-K=1024.json
- gfx950-GEMM-A8W8_BLOCKSCALE-N=8192-K=32768.json
- gfx950-GEMM-A8W8_BLOCKSCALE.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=1536-K=4096.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=16384-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=2048-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=2112-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=24576-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=3072-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=32768-K=512.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=36864-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=512.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4096-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=4608-K=7168.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=65536-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=16384.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=18432.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=2048.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=2304.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=7168-K=4096.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=8192-K=1536.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED-N=8192-K=8192.json
- gfx950-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED.json
- gfx950-GEMM-A8W8_PER_TOKEN_SCALE-N=1024-K=8192.json
- gfx950-GEMM-A8W8_PER_TOKEN_SCALE-N=32768-K=8192.json
- gfx950-GEMM-A8W8_PER_TOKEN_SCALE-N=8192-K=1024.json
- gfx950-GEMM-A8W8_PER_TOKEN_SCALE-N=8192-K=32768.json
- gfx950-GEMM-A8W8_PER_TOKEN_SCALE.json
- gfx950-GEMM-A8WFP4-N=4096-K=4160.json
- gfx950-GEMM-A8WFP4.json
- gfx950-GEMM-AFP8WFP8.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=1536-K=4096.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=2048-K=7168.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=4096-K=1024.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=4096-K=256.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=512-K=4096.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=7168-K=2048.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=7168-K=384.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=768-K=7168.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=8192-K=1024.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED-N=8192-K=1536.json
- gfx950-GEMM-AFP8WFP8_PRESHUFFLED.json
- DEFAULT.json
- .gitkeep
- DEFAULT.json
- .gitkeep
- .gitkeep
- DEFAULT.json
- DEFAULT.json
- .gitkeep
- DEFAULT.json
- DEFAULT.json
- GEMM-AFP4WFP4-N=106496-K=16384.json
- GEMM-AFP4WFP4-N=1280-K=8192.json
- GEMM-AFP4WFP4-N=13312-K=16384.json
- GEMM-AFP4WFP4-N=16384-K=13312.json
- GEMM-AFP4WFP4-N=16384-K=16384.json
- GEMM-AFP4WFP4-N=16384-K=2048.json
- GEMM-AFP4WFP4-N=16384-K=26624.json
- GEMM-AFP4WFP4-N=16384-K=4096.json
- GEMM-AFP4WFP4-N=16384-K=53248.json
- GEMM-AFP4WFP4-N=16384-K=6656.json
- GEMM-AFP4WFP4-N=16384-K=8192.json
- GEMM-AFP4WFP4-N=18432-K=16384.json
- GEMM-AFP4WFP4-N=2112-K=7168.json
- GEMM-AFP4WFP4-N=2304-K=16384.json
- GEMM-AFP4WFP4-N=24576-K=1536.json
- GEMM-AFP4WFP4-N=26624-K=16384.json
- GEMM-AFP4WFP4-N=28672-K=4096.json
- GEMM-AFP4WFP4-N=3072-K=1536.json
- GEMM-AFP4WFP4-N=32768-K=512.json
- GEMM-AFP4WFP4-N=36864-K=7168.json
- GEMM-AFP4WFP4-N=4096-K=14336.json
- GEMM-AFP4WFP4-N=4096-K=7168.json
- GEMM-AFP4WFP4-N=4608-K=16384.json
- GEMM-AFP4WFP4-N=4608-K=7168.json
- GEMM-AFP4WFP4-N=512-K=128.json
- GEMM-AFP4WFP4-N=512-K=7168.json
- GEMM-AFP4WFP4-N=53248-K=16384.json
- GEMM-AFP4WFP4-N=57344-K=8192.json
- GEMM-AFP4WFP4-N=7168-K=16384.json
- GEMM-AFP4WFP4-N=7168-K=2048.json
- GEMM-AFP4WFP4-N=7168-K=2304.json
- GEMM-AFP4WFP4-N=7168-K=256.json
- GEMM-AFP4WFP4-N=8192-K=28672.json
- GEMM-AFP4WFP4-N=8192-K=8192.json
- GEMM-AFP4WFP4-N=9216-K=16384.json
- DEFAULT.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=10240-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=106496-K=16384.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=1280-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=14336-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=16384-K=16384.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=16384-K=53248.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=18432-K=16384.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=2112-K=7168.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=2560-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=28672-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=3072-K=1536.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=36864-K=7168.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=4096-K=14336.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=4096-K=512.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=5120-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=57344-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=7168-K=8192.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=1024.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=14336.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=2048.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=28672.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=3584.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=4096.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=7168.json
- GEMM-AFP4WFP4_PRESHUFFLED-N=8192-K=8192.json
- .gitkeep
- gfx942-HSTU_ATTN_BWD.json
- gfx942-HSTU_ATTN_FWD.json
- gfx950-HSTU_ATTN_BWD.json
- gfx950-HSTU_ATTN_FWD.json
- gfx1250-A4W4.json
- gfx1250-A8W4.json
- gfx942-MOE-DEFAULT.json
- gfx942-MOE-FP8_W8A8.json
- gfx942-MOE-INT4_W4A16.json
- gfx942-MOE-INT8_W8A16.json
- gfx942-MOE-INT8_W8A8.json
- gfx942-MOE_ROUTING_SIGMOID_TOPK1.json
- gfx950-A8W4.json
- gfx950-MOE-DEFAULT.json
- gfx950-MOE-FP8_W8A8.json
- gfx950-MOE-INT4_W4A16.json
- gfx950-MOE-INT8_W8A16.json
- gfx950-MOE-INT8_W8A8.json
- gfx950-MOE-MX_FP4.json
- gfx950-MOE_ROUTING_SIGMOID_TOPK1.json
- CLAUDE.md
- gfx1151-MHA-DEFAULT.json
- gfx1250-EXTEND_ATTENTION.json
- gfx1250-GMM.json
- gfx1250-MHA-DEFAULT.json
- gfx1250-MLA_DECODE_ROPE-DEFAULT.json
- gfx942-EXTEND_ATTENTION.json
- gfx942-GMM.json
- gfx942-LEANATTN-DEFAULT.json
- gfx942-MHA-DEFAULT.json
- gfx942-MHC_FUSED_SINKHORN-C=1024.json
- gfx942-MHC_FUSED_SINKHORN-C=128.json
- gfx942-MHC_FUSED_SINKHORN-C=32768.json
- gfx942-MHC_FUSED_SINKHORN-C=4096.json
- gfx942-MHC_FUSED_SINKHORN-C=512.json
- gfx942-MHC_FUSED_SINKHORN.json
- gfx942-MHC_POST.json
- gfx942-MLA_DECODE_ROPE-DEFAULT.json
- gfx950-EXTEND_ATTENTION.json
- gfx950-GMM.json
- gfx950-MHA-DEFAULT.json
- gfx950-MHC_FUSED_SINKHORN-C=1024.json
- gfx950-MHC_FUSED_SINKHORN-C=128.json
- gfx950-MHC_FUSED_SINKHORN-C=32768.json
- gfx950-MHC_FUSED_SINKHORN-C=4096.json
- gfx950-MHC_FUSED_SINKHORN-C=512.json
- gfx950-MHC_FUSED_SINKHORN-C=7168.json
- gfx950-MHC_FUSED_SINKHORN.json
- gfx950-MHC_POST.json
- gfx950-MLA_DECODE_ROPE-DEFAULT.json
- __init__.py
- _launch.py
- _prepack.py
- _utils.py
- causal_conv1d.py
- causal_conv1d_update_single_token.py
- conv2d.py
- DESIGN.md
- README.md
- __init__.py
- attn_res.py
- fused_bmm_rope_kv_cache.py
- fused_clamp_act_mul.py
- fused_kv_cache.py
- fused_mul_add.py
- fused_qk_concat.py
- fused_reduce_qk_norm_rope_swa_write.py
- fused_routing_from_topk.py
- mhc.py
- __init__.py
- causal_conv1d_decode.py
- causal_conv1d_prefill.py
- fused_rearrange_sigmoid_gdr.py
- gated_delta_rule.py
- __init__.py
- gemm_a16w16.py
- gemm_a16w16_atomic.py
- gemm_a16w16_gated.py
- gemm_a16w8_blockscale.py
- gemm_a16wfp4.py
- gemm_a8w8.py
- gemm_a8w8_blockscale.py
- gemm_a8w8_per_token_scale.py
- gemm_a8wfp4.py
- gemm_afp4wfp4.py
- gemm_afp4wfp4_pre_quant_atomic.py
- gemm_afp8wfp8.py
- __init__.py
- batched_gemm_a16wfp4.py
- batched_gemm_a8w8.py
- batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant.py
- batched_gemm_afp4wfp4.py
- batched_gemm_afp4wfp4_pre_quant.py
- batched_gemm_bf16.py
- __init__.py
- ff_a16w16.py
- ff_a16w16_fused_gated.py
- ff_a16w16_fused_ungated.py
- __init__.py
- fused_gemm_a16w16_quant_x.py
- fused_gemm_a8w8_blockscale_a16w16.py
- fused_gemm_a8w8_blockscale_mul_add.py
- fused_gemm_a8w8_blockscale_split_cat.py
- fused_gemm_afp4wfp4_a16w16.py
- fused_gemm_afp4wfp4_mul_add.py
- fused_gemm_afp4wfp4_split_cat.py
- __init__.py
- __init__.py
- gemm_a8w8.py
- gemm_a8w8_blockscale.py
- gemm_afp4wfp4.py
- mla_gluon.py
- pa_decode_gluon.py
- pa_mqa_logits.py
- README.md
- __init__.py
- chunk_delta_attn.py
- bitmatrix.py
- minunique.py
- routing.py
- topk.py
- __init__.py
- moe_align_block_size.py
- moe_op.py
- moe_op_e2e.py
- moe_op_gelu.py
- moe_op_gemm_a16w4.py
- moe_op_gemm_a4w4.py
- moe_op_gemm_a8w4.py
- moe_op_gemm_a8w8.py
- moe_op_gemm_a8w8_blockscale.py
- moe_op_gemm_int8_smoothquant.py
- moe_op_mxfp4.py
- moe_op_mxfp4_silu_fused.py
- moe_op_silu_fused.py
- moe_routing_sigmoid_top1_fused.py
- quant_moe.py
- reduce.py
- __init__.py
- fused_add_rmsnorm_pad.py
- fused_rmsnorm_add.py
- norm.py
- rmsnorm.py
- __init__.py
- fused_fp8_quant.py
- fused_mxfp4_quant.py
- fused_mxfp8_quant.py
- mxfp6_fmha_pack.py
- quant.py
- sage_attention_quant_wrappers.py
- __init__.py
- fused_qkv_split_qk_norm_rope_cache.py
- fused_qkv_split_qk_rope.py
- rope.py
- .gitignore
- _utils.py
- README.md
- rprof.py
- screen.py
- ut_a16w16_gemm.py
- ut_a16w16_gemm_atomic.py
- ut_a16w16_gemm_gated.py
- ut_a16w8_gemm_blockscale.py
- ut_a16w8_gemm_blockscale_preshuffle.py
- ut_a16wfp4_gemm.py
- ut_a8w8_gemm.py
- ut_a8w8_gemm_blockscale.py
- ut_a8w8_gemm_blockscale_preshuffle.py
- ut_a8w8_gemm_per_token_scale.py
- ut_a8wfp4_gemm.py
- ut_afp4wfp4_gemm.py
- ut_afp4wfp4_gemm_pre_quant_atomic.py
- ut_afp4wfp4_gemm_preshuffle.py
- ut_afp8wfp8_gemm_preshuffle.py
- ut_batched_gemm_a16w16.py
- ut_template.py
- verify-perf.py
- view-screen.py
- arch_info.py
- gemm_tune_check.py
- kernel_repr.py
- mha_kernel_utils.py
- moe_common.py
- pid_preprocessing.py
- __init__.py
- common_utils.py
- conv_config_utils.py
- core.py
- device_info.py
- gemm_config_utils.py
- gmm_common.py
- la_kernel_utils.py
- logger.py
- mha_kernel_utils.py
- mhc_config_utils.py
- moe_common.py
- moe_config_utils.py
- shuffle.py
- tuned_config_utils.py
- types.py
- __init__.py
- activation.py
- gather_kv_b_proj.py
- gmm.py
- kv_cache.py
- README.md
- softmax.py
- topk.py
- __init__.py
- activation.py
- aiter_operator.py
- attention.py
- batched_gemm_op_a8w8.py
- batched_gemm_op_bf16.py
- cache.py
- causal_conv1d_fwd_split_qkv.py
- causal_conv1d_update.py
- chunk_gated_delta_rule_fwd_h.py
- communication.py
- custom.py
- custom_all_reduce.py
- deepgemm.py
- enum.py
- fused_qk_norm_mrope_cache_quant.py
- fused_qk_norm_rope_cache_quant.py
- fused_qk_rmsnorm_group_quant.py
- fused_qknorm_idxrqknorm.py
- fused_split_gdr_update.py
- gated_rmsnorm_fp8_group_quant.py
- gated_rmsnorm_fp8_per_token_quant.py
- gdr_decode_packed_bf16.py
- gemm_op_a16w16.py
- gemm_op_a4w4.py
- gemm_op_a8w4.py
- gemm_op_a8w8.py
- gemm_op_common.py
- gradlib.py
- groupnorm.py
- inverse_rope_group_quant.py
- mha.py
- mha_v4.md
- mha_v4.py
- mhc.py
- moe_mxfp4_aux.py
- moe_op.py
- moe_sorting.py
- moe_sorting_opus.py
- norm.py
- pa_sparse_prefill_opus.py
- pos_encoding.py
- prefill_batch_metadata.py
- quant.py
- quick_all_reduce.py
- rmsnorm.py
- rope.py
- sample.py
- sampling.py
- shuffle.py
- topk.py
- topk_plain.py
- trans_ragged_layout.py
- vsa_sparse_attention.py
- README.md
- triton_metadata_redirect.py
- aiter_types.py
- base_tuner.py
- dtypes.py
- fp4_utils.py
- mp_tuner.py
- mx_types.py
- pretune.py
- __init__.py
- bert_padding.py
- fhmoe.py
- fused_moe.py
- fused_moe_bf16_asm.py
- fused_moe_dp_shared_expert.py
- int4_utils.py
- mla.py
- paged_attn.py
- rotary_embedding.py
- test_common.py
- test_mha_common.py
- tuned_gemm.py
- readme.md
- run.py
- batched_gemm_a8w8.h
- batched_gemm_a8w8_common.cuh
- batched_gemm_a8w8.cu
- batched_gemm_a8w8_common.py
- batched_gemm_a8w8_tune.cu
- batched_gemm_a8w8_tune.py
- gen_instances.py
- README.md
- batched_gemm_bf16.h
- batched_gemm_bf16_common.cuh
- batched_gemm_bf16.cu
- batched_gemm_bf16_common.py
- batched_gemm_bf16_tune.cu
- batched_gemm_bf16_tune.py
- gen_instances.py
- README.md
- deepgemm.h
- deepgemm_common.cuh
- deepgemm.cu
- deepgemm_common.py
- gen_instances.py
- gemm_a4w4_blockscale.h
- gemm_a4w4_blockscale_common.cuh
- gemm_a4w4_blockscale.cu
- gemm_a4w4_blockscale_common.py
- gemm_a4w4_blockscale_tune.cu
- gemm_a4w4_blockscale_tune.py
- gen_instances.py
- README.md
- validate_tuned_csv.py
- gemm_a8w8.h
- gemm_a8w8_common.cuh
- gemm_a8w8.cu
- gemm_a8w8_common.py
- gemm_a8w8_tune.cu
- gemm_a8w8_tune.py
- gen_instances.py
- README.md
- gemm_a8w8_blockscale.h
- gemm_a8w8_blockscale_bpreshuffle_cktile.h
- gemm_a8w8_blockscale_cktile.h
- gemm_a8w8_blockscale_cktile_common.cuh
- gemm_a8w8_blockscale_common.cuh
- gemm_a8w8_blockscale.cu
- gemm_a8w8_blockscale_cktile.cu
- gemm_a8w8_blockscale_cktile_instance.py
- gemm_a8w8_blockscale_cktile_tune.cu
- gemm_a8w8_blockscale_instance.py
- gemm_a8w8_blockscale_lookup_bf16.cu
- gemm_a8w8_blockscale_lookup_fp16.cu
- gemm_a8w8_blockscale_tune.cu
- gemm_a8w8_blockscale_tune.py
- gen_instances.py
- gen_instances_cktile.py
- README.md
- gemm_a8w8_blockscale_bpreshuffle.h
- gemm_a8w8_blockscale_bpreshuffle_common.cuh
- gemm_a8w8_blockscale_bpreshuffle.cu
- gemm_a8w8_blockscale_bpreshuffle_common.py
- gemm_a8w8_blockscale_bpreshuffle_tune.cu
- gen_instances.py
- README.md
- gemm_a8w8_bpreshuffle.h
- gemm_a8w8_bpreshuffle_common.cuh
- gemm_a8w8_bpreshuffle.cu
- gemm_a8w8_bpreshuffle_common.py
- gemm_a8w8_bpreshuffle_tune.cu
- gemm_a8w8_bpreshuffle_tune.py
- gen_instances.py
- README.md
- gemm_moe_ck2stages.cu
- gemm_moe_ck2stages.h
- gemm_moe_ck2stages_common.cuh
- gemm_moe_ck2stages_common.py
- gemm_moe_ck2stages_common_blockscale.cuh
- gemm_moe_ck2stages_common_mxfp4.cuh
- gemm_moe_ck2stages_common_mxfp4_bns.cuh
- gemm_moe_tune.py
- gen_instances.py
- README.md
- moe_cktile2stages.h
- moe_cktile2stages_common.cuh
- gen_instances.py
- moe_cktile2stages.cu
- moe_cktile2stages_common.py
- gemm_a8w8_bpreshuffle_cktile.h
- gemm_a8w8_bpreshuffle_cktile_common.cuh
- gemm_a8w8_bpreshuffle_cktile.cu
- gemm_a8w8_bpreshuffle_cktile_common.py
- gemm_a8w8_bpreshuffle_cktile_tune.cu
- gen_instances.py
- README.md
- compile.cpp
- compile.h
- compile.py
- compile_gluon.py
- asm_mla_decode_fwd.cpp
- asm_mla_decode_fwd.cpp.jinja
- asm_mla_decode_fwd.h
- asm_mla_decode_fwd.py
- asm_mla_decode_fwd_test.cpp
- asm_mla_decode_fwd_test.py
- Makefile
- asm_moe.cpp.jinja
- asm_moe.py
- test_asm_moe.py
- __init__.py
- Makefile
- pa.cpp.jinja
- pa.cuh
- pa.py
- pa_common.cuh
- pa_kernels.cuh
- pa_ps.cpp.jinja
- pa_ps.cuh
- pa_ps.py
- pa_ragged.cpp
- pa_ragged.cpp.jinja
- pa_ragged.cuh
- pa_ragged.h
- pa_ragged.py
- pa_ragged_test.cpp
- pa_ragged_test.py
- pa_test.py
- pa_v1.cpp.jinja
- pa_v1.cuh
- pa_v1.py
- Makefile
- pa_attention_kernel_test.py
- pa_decode_attention_kernel.cpp.jinja
- pa_decode_attention_reduce_kernel.cpp.jinja
- pa_decode_gluon_aot.cpp
- pa_decode_gluon_aot.h
- pa_decode_gluon_aot.py
- pa_decode_gluon_aot_prebuild.py
- pa_decode_gluon_aot_warmup.py
- pa_decode_reduce_kernel.cpp.jinja
- pa_reduce_kernel_test.py
- test_pa_gluon.cpp
- test_pa_gluon.py
- sampling.cuh
- top_k_renorm_probs.cpp.jinja
- top_k_renorm_probs.py
- top_k_top_p_sampling_from_probs.cpp.jinja
- top_k_top_p_sampling_from_probs.py
- top_p_sampling_from_probs.cpp.jinja
- top_p_sampling_from_probs.py
- util.py
- vec_dtypes.cuh
- __init__.py
- hsaco_launcher.py
- lru_cache.h
- mha_bwd.cu
- mha_fwd.cu
- mha_fwd_batch_prefill.cu
- mha_fwd_split.cu
- README.MD
- torch_utils.py
- utils.h
- utils.py
- chunk_gdn_h_opt_tune.py
- README.md
- gemm_a16w16_tune.py
- gemm_tuner.py
- README.md
- vec_convert.h
- hip_minimal.hpp
- logo.png
- opus.hpp
- README.md
- rmsnorm_opus_kernel.hpp
- fmha_fwd_bf16_opus.h
- mha_batch_prefill.h
- mha_bwd.h
- mha_fwd.h
- mha_v3_bwd.h
- mha_v3_fwd.h
- mha_v3_varlen_bwd.h
- mha_v3_varlen_fwd.h
- mha_v4_fwd.h
- mha_v4_quant.h
- mha_varlen_bwd.h
- mha_varlen_fwd.h
- activation.h
- aiter_ctypes_error.h
- aiter_dispatch.h
- aiter_enum.h
- aiter_hip_common.h
- aiter_logger.h
- aiter_operator.h
- aiter_opus_plus.h
- aiter_stream.h
- aiter_tensor.h
- aiter_unary.h
- attention_ck.h
- attention_common.cuh
- attention_dtypes.h
- attention_generic.cuh
- binary_operator.cuh
- cache.h
- causal_conv1d_fwd_split_qkv.h
- causal_conv1d_update.h
- chunk_gated_delta_rule_fwd_h.h
- ck_tile_shim.h
- custom.h
- custom_all_reduce.cuh
- custom_all_reduce.h
- custom_all_reduce_gfx1250.cuh
- custom_all_reduce_gfx1250.h
- dispatch_utils.h
- dsv4_rotate_quant.h
- dtype_bfloat16.cuh
- dtype_float16.cuh
- dtype_float32.cuh
- dtype_fp8.cuh
- fmha_fwd_hd128_bf16_opus.h
- fmha_fwd_hd128_bf16_opus_defs.h
- fmha_fwd_hd128_bf16_opus_kernel.hpp
- fmha_fwd_hd192_v128_bf16_opus.h
- fmha_fwd_hd192_v128_bf16_opus_defs.h
- fmha_fwd_hd192_v128_bf16_opus_kernel.hpp
- fused_ar_mhc_post.h
- fused_qk_norm_mrope_cache_quant.h
- fused_qk_norm_rope_cache_quant.h
- fused_qk_rmsnorm_group_quant.h
- fused_qknorm_idxrqknorm.h
- fused_split_gdr_update.h
- gated_rmsnorm_quant.h
- gdr_decode_packed_bf16.h
- gemm_common.h
- gemm_dispatch_utils.h
- groupnorm.hpp
- hip_float8.h
- hip_float8_impl.h
- hip_reduce.h
- inverse_rope_group_quant.h
- mha_bwd.h
- mha_common.h
- mha_fwd.h
- mhc.h
- mla.h
- mla_hk.h
- mla_metadata.h
- mla_opus.h
- mla_reduce.h
- moe_ck.h
- moe_mxfp4_aux.h
- moe_op.h
- moe_sorting.h
- moe_sorting_opus.h
- moe_topk_op.h
- mx_quant_utils.h
- norm.h
- pa.h
- pa_sparse_prefill_opus.h
- pos_encoding.h
- ps.h
- py_itfs_common.h
- quant.h
- quant_utils.cuh
- quick_all_reduce.cuh
- quick_all_reduce.h
- quick_all_reduce_base.h
- rmsnorm.h
- rmsnorm_quant.h
- rocm_ops.hpp
- rope.h
- sample.h
- smoothquant.h
- topk_gating_kernels.cuh
- topk_per_row.h
- topk_plain.h
- vectorization.cuh
- vsa_sparse_attention.h
- warp_sort.h
- op_combine.hpp
- op_epilog.hpp
- op_gemm.hpp
- op_lds.hpp
- op_softmax.hpp
- pipeline.hpp
- params.hpp
- fmha_fwd_d64_bf16_combine.cu
- fmha_fwd_d64_bf16_msk0_split.cu
- fmha_fwd_d64_bf16_msk1_split.cu
- mha_fwd_native_splitkv.cu
- mha_native.h
- mha_native_launch.h
- hk_mla_v40_gen1_spec.md
- hk_mla_buffer_managers.cuh
- hk_mla_softmax.cuh
- hk_mla_utils.cuh
- hk_mla_v40_buffer_managers_gen1.cuh
- hk_mla_v40_fwd_decode_gen1_common.cuh
- mi35x_v32_fwd_decode_m16x4_fp8_fp8.cuh
- mi35x_v32_fwd_decode_m16x8_fp8_fp8.cuh
- mi35x_v40_fwd_decode_m16x4_fp8bf16_fp8bf16_gen1.cuh
- mi35x_v40_fwd_decode_m16x8_fp8bf16_fp8bf16_gen1.cuh
- mi3xx_v32_fwd_decode_m16x8_fp8_fp8.cuh
- v1_0_device.cuh
- v1_1_device.cuh
- v1_2_device.cuh
- v1_2_host.cuh
- v1_2_pa_device.cuh
- v1_comm.cuh
- defs.h
- dsa_v32_splitkv.hpp
- mla_decode_fwd_16mx8_32nx1_fp8fp8_ps_opus.hpp
- mla_fp8fp8_def.h
- mla_global_load.hpp
- ds32_decode_fwd.cu
- hk_v32_decode_fwd.cu
- hk_v40_decode_fwd.cu
- metadata.cu
- mla_decode_fwd_opus.cu
- reduce.cu
- arithmetic.hpp
- gen_instances.py
- mxfp4_moe_aux_dispatch.h
- moe_3stage_sort.cuh
- moe_scatter_reduce.cuh
- moe_sort_quant.cuh
- moe_sort_scales.cuh
- moe_mxfp4_aux.cu
- rmsnorm_opus_norm.hpp
- rmsnorm_opus_norm_bf16.cu
- rmsnorm_opus_norm_entry.cu
- rmsnorm_opus_norm_fp16.cu
- rmsnorm_opus_norm_fp32.cu
- rmsnorm_opus_quant.cu
- general_1c_2d_bwd_kernels.cu
- general_1c_2d_fwd_kernels.cu
- general_1c_cached_bwd_kernels.cu
- general_1c_cached_fwd_kernels.cu
- general_1c_cached_positions_fwd_kernels.cu
- general_1c_cached_positions_offsets_fwd_kernels.cu
- general_1c_thd_bwd_kernels.cu
- general_1c_thd_fwd_kernels.cu
- general_1c_uncached_bwd_kernels.cu
- general_1c_uncached_fwd_kernels.cu
- general_2c_cached_bwd_kernels.cu
- general_2c_cached_fwd_kernels.cu
- general_2c_cached_positions_fwd_kernels.cu
- general_2c_cached_positions_offsets_fwd_kernels.cu
- general_2c_uncached_bwd_kernels.cu
- general_2c_uncached_fwd_kernels.cu
- rope_common.h
- lapack_sytrd.py
- Makefile
- README.md
- sytrd_benchmark.cu
- sytrd_kernels.cu
- topk_sigmoid.cu
- topk_softmax.cu
- topk_sqrtsoftplus.cu
- activation_kernels.cu
- binary_operator.cu
- cache_kernels.cu
- causal_conv1d_fwd_split_qkv.cu
- causal_conv1d_update.cu
- chunk_gated_delta_rule_fwd_h.cu
- custom_all_reduce.cu
- custom_all_reduce_gfx1250.cu
- custom_kernels.cu
- dsv4_rotate_quant.cu
- fused_ar_mhc_post.cu
- fused_kernels.cu
- fused_qk_norm.cu
- fused_qk_norm_mrope_cache_quant.cu
- fused_qk_norm_rope_cache_quant.cu
- fused_qk_rmsnorm_group_quant.cu
- fused_qknorm_idxrqknorm.cu
- fused_split_gdr_update.cu
- gated_rmsnorm_quant_kernels.cu
- gdr_decode_packed_bf16.cu
- generate_binaryop.py
- groupnorm.cu
- inverse_rope_group_quant.cu
- mha_common.cu
- mha_v4_quant.cu
- mhc_kernels.cu
- moe_align_block_size_kernels.cu
- moe_fused_gate.cu
- pos_encoding_kernels.cu
- quant_kernels.cu
- quant_mxfp4.cu
- quick_all_reduce.cu
- rmsnorm_quant_kernels.cu
- sample_kernels.cu
- topk_gating_kernels.cu
- topk_per_row_kernels.cu
- topk_plain_kernels.cu
- topk_softmax_kernels.cu
- topk_softmax_kernels_group.cu
- unary_operator.cu
- __init__.py
- common.py
- gen_instances_gfx1250.py
- gen_instances_gfx942.py
- gen_instances_gfx950.py
- opus_gemm_arch_gfx1250.cuh
- opus_gemm_heuristic_dispatch_gfx1250.cuh
- opus_gemm_pipeline_a16w16_cluster_tdm_splitk_ws_gfx1250.cuh
- opus_gemm_pipeline_a16w16_clusterlaunch_tdm_splitk_fuse_gfx1250.cuh
- opus_gemm_pipeline_a16w16_clusterlaunch_tdm_splitk_ws_gfx1250.cuh
- opus_gemm_traits_a16w16_gfx1250.cuh
- splitk_reduce_gfx1250.cuh
- splitk_reduce_launch_gfx1250.cuh
- opus_gemm_helpers_a16w16.cuh
- opus_gemm_mfma16x16x16_gfx942.cuh
- opus_gemm_pipeline_a16w16_em3en4_lds1_pgr2_sk.cuh
- opus_gemm_pipeline_a16w16_kbuf1.cuh
- opus_gemm_pipeline_a16w16_kbuf1_large_tile.cuh
- opus_gemm_pipeline_a16w16_kbuf2v.cuh
- opus_gemm_pipeline_a16w16_kbuf2v_bk128.cuh
- opus_gemm_pipeline_a16w16_quad_mfma32_kbuf1.cuh
- opus_gemm_pipeline_a16w16_wave_k_coop.cuh
- opus_gemm_traits_a16w16.cuh
- splitk_reduce_gfx942.cuh
- opus_gemm_pipeline_a8w8_blockscale_bpreshuffle.cuh
- opus_gemm_traits_a8w8_blockscale_bpreshuffle.cuh
- opus_gemm_arch_gfx942.cuh
- opus_gemm_heuristic_dispatch_gfx942.cuh
- opus_bmm_launchers_a8w8_mxscale_gfx950.cuh
- opus_bmm_pipeline_a8w8_mxscale_gfx950.cuh
- opus_gemm_arch_gfx950.cuh
- opus_gemm_heuristic_dispatch_gfx950.cuh
- opus_gemm_pipeline_a16w16_4g_safe_gfx950.cuh
- opus_gemm_pipeline_a16w16_flatmm_gfx950.cuh
- opus_gemm_pipeline_a16w16_flatmm_splitk_gfx950.cuh
- opus_gemm_pipeline_a16w16_gfx950.cuh
- opus_gemm_pipeline_a16w16_mono_tile_4g_safe_gfx950.cuh
- opus_gemm_pipeline_a16w16_mono_tile_gfx950.cuh
- opus_gemm_pipeline_a16w16_persistent_4g_safe_gfx950.cuh
- opus_gemm_pipeline_a16w16_persistent_gfx950.cuh
- opus_gemm_pipeline_a8w8_mxscale_flatmm_splitk_gfx950.cuh
- opus_gemm_pipeline_a8w8_noscale_gfx950.cuh
- opus_gemm_pipeline_a8w8_scale_gfx950.cuh
- opus_gemm_traits_a16w16_gfx950.cuh
- opus_gemm_traits_a8w8_noscale_gfx950.cuh
- opus_gemm_traits_a8w8_scale_gfx950.cuh
- splitk_reduce_gfx950.cuh
- opus_bmm.h
- opus_gemm.h
- opus_gemm_arch.cuh
- opus_gemm_common.cuh
- opus_gemm_utils.cuh
- gen_instances.py
- opus_bmm.cu
- opus_bmm_mxscale_tune.py
- opus_gemm.cu
- opus_gemm_common.py
- opus_gemm_tune.py
- README.md
- opus_moe_pipeline_stage2_gemmstyle_gfx950.cuh
- opus_moe_traits_stage2_gfx950.cuh
- opus_moe_stage1_a8w4_dispatch_gfx950.cuh
- opus_moe_stage1_a8w4_pipeline_common_gfx950.cuh
- opus_moe_stage1_a8w4_pipeline_group_split_gfx950.cuh
- opus_moe_stage1_a8w4_pipeline_pair_kwave_gfx950.cuh
- opus_moe_stage1_a8w4_traits_gfx950.cuh
- opus_moe_pipeline_stage2_a8w4_decode_main_gfx950.cuh
- opus_moe_pipeline_stage2_a8w4_decode_policy_gfx950.cuh
- opus_moe_stage2_a8w4_decode_dispatch_gfx950.cuh
- opus_moe_traits_stage2_a8w4_decode_gfx950.cuh
- opus_moe_arch_gfx950.cuh
- opus_moe_stage2_route_output_reduce_gfx950.cuh
- opus_moe_stage2_utils_gfx950.cuh
- opus_moe.h
- opus_moe_arch.cuh
- opus_moe_common.cuh
- opus_moe_host_impl.cuh
- gen_instances.py
- opus_moe.cu
- opus_moe_common.py
- README.md
- moe_ck_gemm.hpp
- moe_ck_gemm1_instance_pertensor_b16.cu
- moe_ck_gemm1_instance_pertensor_b16_f8.cu
- moe_ck_gemm1_instance_pertensor_b16_f8_wint4.cu
- moe_ck_gemm1_instance_pertensor_b16_i8.cu
- moe_ck_gemm1_instance_pertensor_f16.cu
- moe_ck_gemm1_instance_pertensor_f16_f8.cu
- moe_ck_gemm1_instance_pertensor_f16_f8_win4.cu
- moe_ck_gemm1_instance_pertensor_f16_i8.cu
- moe_ck_gemm1_instance_pertensor_mulweight_b16.cu
- moe_ck_gemm1_instance_pertensor_mulweight_f16.cu
- moe_ck_gemm1_instance_pertoken_b16.cu
- moe_ck_gemm1_instance_pertoken_b16_f8.cu
- moe_ck_gemm1_instance_pertoken_b16_f8_wint4.cu
- moe_ck_gemm1_instance_pertoken_b16_i8.cu
- moe_ck_gemm1_instance_pertoken_f16.cu
- moe_ck_gemm1_instance_pertoken_f16_f8.cu
- moe_ck_gemm1_instance_pertoken_f16_f8_win4.cu
- moe_ck_gemm1_instance_pertoken_f16_i8.cu
- moe_ck_gemm1_instance_pertoken_mulweight_b16.cu
- moe_ck_gemm1_instance_pertoken_mulweight_f16.cu
- moe_ck_gemm2_instance_pertensor_b16..cu
- moe_ck_gemm2_instance_pertensor_b16_f8.cu
- moe_ck_gemm2_instance_pertensor_b16_f8_wint4.cu
- moe_ck_gemm2_instance_pertensor_b16_i8.cu
- moe_ck_gemm2_instance_pertensor_f16.cu
- moe_ck_gemm2_instance_pertensor_f16_f8.cu
- moe_ck_gemm2_instance_pertensor_f16_f8_wint4.cu
- moe_ck_gemm2_instance_pertensor_f16_i8.cu
- moe_ck_gemm2_instance_pertensor_mulweight_b16.cu
- moe_ck_gemm2_instance_pertensor_mulweight_f16.cu
- moe_ck_gemm2_instance_pertoken_b16..cu
- moe_ck_gemm2_instance_pertoken_b16_f8.cu
- moe_ck_gemm2_instance_pertoken_b16_f8_wint4.cu
- moe_ck_gemm2_instance_pertoken_b16_i8.cu
- moe_ck_gemm2_instance_pertoken_f16.cu
- moe_ck_gemm2_instance_pertoken_f16_f8.cu
- moe_ck_gemm2_instance_pertoken_f16_f8_wint4.cu
- moe_ck_gemm2_instance_pertoken_f16_i8.cu
- moe_ck_gemm2_instance_pertoken_mulweight_b16.cu
- moe_ck_gemm2_instance_pertoken_mulweight_f16.cu
- moe_ck_gemm_common.cuh
- attention_kernels.cu
- mha_batch_prefill_kernels.cu
- mha_bwd_kernels.cu
- mha_fwd_kernels.cu
- mha_varlen_bwd_kernels.cu
- mha_varlen_fwd_kernels.cu
- moe_ck_2stages_kernel.cu
- moe_sorting_kernels.cu
- norm_kernels.cu
- smoothquant_kernels.cu
- topk_sigmoid_kernels.cu
- vsa_sparse_attention_kernels.cu
- asm_a8w8_blockscale_bpreshuffle.cu
- asm_f4gemm.cu
- asm_flatmm_a8w8_blockscale.cu
- asm_fmha_fwd_mxfp8.cu
- asm_fmha_fwd_with_sink.cu
- asm_fmha_fwd_with_sink_varlen.cu
- asm_fmoe.cu
- asm_gemm_a16w16.cu
- asm_gemm_a4w4.cu
- asm_gemm_a8w8.cu
- asm_mha_bwd.cu
- asm_mha_fwd.cu
- asm_mha_v4_fwd.cu
- asm_mha_varlen_bwd.cu
- asm_mha_varlen_fwd.cu
- asm_mla.cu
- asm_mla_v4.cu
- asm_moe_2stage.cu
- asm_mxfp8fp4gemm.cu
- asm_pa.cu
- asm_pa_decode_bf16.cu
- asm_topk_per_row_decode.cu
- asm_topk_per_row_prefill.cu
- asm_topksoftmax.cu
- custom.cu
- fmha_bwd_pre_post_kernel_generate.py
- fmha_fwd_bf16_opus_kernels.cu
- gemm_common.cu
- moe_sorting_opus_kernels.cu
- pa_sparse_prefill_opus_kernels.cu
- activation_pybind.cu
- aiter_core_pybind.cu
- aiter_operator_pybind.cu
- aiter_unary_pybind.cu
- attention_ck_pybind.cu
- batched_gemm_a8w8_pybind.cu
- batched_gemm_a8w8_tune_pybind.cu
- batched_gemm_bf16_pybind.cu
- batched_gemm_bf16_tune_pybind.cu
- cache_pybind.cu
- causal_conv1d_fwd_split_qkv_pybind.cu
- causal_conv1d_update_pybind.cu
- chunk_gated_delta_rule_fwd_h_pybind.cu
- custom_all_reduce_gfx1250_pybind.cu
- custom_all_reduce_pybind.cu
- custom_pybind.cu
- deepgemm_pybind.cu
- dsv4_rotate_quant_pybind.cu
- fmha_fwd_bf16_opus_pybind.cu
- fused_ar_mhc_post_pybind.cu
- fused_qk_norm_mrope_cache_quant_pybind.cu
- fused_qk_norm_rope_cache_quant_pybind.cu
- fused_qk_rmsnorm_group_quant_pybind.cu
- fused_qknorm_idxrqknorm_pybind.cu
- fused_split_gdr_update_pybind.cu
- gated_rmsnorm_quant_pybind.cu
- gdr_decode_packed_bf16_pybind.cu
- gemm_a16w16_asm_pybind.cu
- gemm_a4w4_blockscale_pybind.cu
- gemm_a4w4_blockscale_tune_pybind.cu
- gemm_a8w8_blockscale_bpreshuffle_cktile_pybind.cu
- gemm_a8w8_blockscale_bpreshuffle_cktile_tune_pybind.cu
- gemm_a8w8_blockscale_bpreshuffle_pybind.cu
- gemm_a8w8_blockscale_bpreshuffle_tune_pybind.cu
- gemm_a8w8_blockscale_cktile_pybind.cu
- gemm_a8w8_blockscale_cktile_tune_pybind.cu
- gemm_a8w8_blockscale_pybind.cu
- gemm_a8w8_blockscale_tune_pybind.cu
- gemm_a8w8_bpreshuffle_cktile_pybind.cu
- gemm_a8w8_bpreshuffle_cktile_tune_pybind.cu
- gemm_a8w8_bpreshuffle_pybind.cu
- gemm_a8w8_bpreshuffle_tune_pybind.cu
- gemm_a8w8_pybind.cu
- gemm_a8w8_tune_pybind.cu
- groupnorm_pybind.cu
- inverse_rope_group_quant_pybind.cu
- mha_batch_prefill_pybind.cu
- mha_bwd_asm_pybind.cu
- mha_bwd_pybind.cu
- mha_fwd_asm_pybind.cu
- mha_fwd_native_splitkv_pybind.cu
- mha_fwd_pybind.cu
- mha_v4_fwd_pybind.cu
- mha_varlen_bwd_asm_pybind.cu
- mha_varlen_bwd_pybind.cu
- mha_varlen_fwd_asm_pybind.cu
- mha_varlen_fwd_pybind.cu
- mhc_pybind.cu
- mla_ds32_pybind.cu
- mla_hk_v32_pybind.cu
- mla_hk_v40_pybind.cu
- mla_metadata_pybind.cu
- mla_reduce_pybind.cu
- moe_ck_2stages_pybind.cu
- moe_ck_pybind.cu
- moe_cktile_2stages_pybind.cu
- moe_mxfp4_aux_pybind.cu
- moe_op_pybind.cu
- moe_sorting_opus_pybind.cu
- moe_sorting_pybind.cu
- moe_topk_ck_pybind.cu
- moe_topk_pybind.cu
- norm_pybind.cu
- opus_gemm_pybind.cu
- opus_moe_pybind.cu
- pa_metadata_pybind.cu
- pa_sparse_prefill_opus_pybind.cu
- pos_encoding_pybind.cu
- ps_metadata_pybind.cu
- quant_pybind.cu
- quick_all_reduce_pybind.cu
- rmsnorm_quant_pybind.cu
- rope_1c_cached_positions_fwd_pybind.cu
- rope_1c_cached_positions_offsets_fwd_pybind.cu
- rope_2c_cached_positions_fwd_pybind.cu
- rope_2c_cached_positions_offsets_fwd_pybind.cu
- rope_general_1c_2d_bwd_pybind.cu
- rope_general_1c_2d_fwd_pybind.cu
- rope_general_1c_cached_bwd_pybind.cu
- rope_general_1c_cached_fwd_pybind.cu
- rope_general_1c_thd_bwd_pybind.cu
- rope_general_1c_thd_fwd_pybind.cu
- rope_general_1c_uncached_bwd_pybind.cu
- rope_general_1c_uncached_fwd_pybind.cu
- rope_general_2c_cached_bwd_pybind.cu
- rope_general_2c_cached_fwd_pybind.cu
- rope_general_2c_uncached_bwd_pybind.cu
- rope_general_2c_uncached_fwd_pybind.cu
- sample_pybind.cu
- smoothquant_pybind.cu
- topk_per_row_pybind.cu
- topk_plain_pybind.cu
- vsa_sparse_attention_pybind.cu
- .gitkeep
- .gitkeep
- attention.rst
- gemm.rst
- operators.rst
- aiter_logo.png
- aiter_logo_mini.png
- analyze_kernel.py
- Dockerfile
- extract_asm.py
- README.md
- roundtrip.sh
- autotuning_ci_pipeline_1.jpeg
- autotuning_ci_pipeline_2.jpeg
- 2026-05.md
- add_new_op.rst
- basic_usage.rst
- index.rst
- aiter_container_nonroot_setup.md
- autotuning_pipeline.md
- conf.py
- DEPLOYMENT.md
- DOCUMENTATION_AUDIT_REPORT.md
- index.rst
- installation.rst
- isa_kernel_optimization.md
- Makefile
- quickstart.rst
- README.md
- requirements.txt
- triton_comms.md
- grad_funcs.cu
- hipbsolgemm.cu
- rocsolgemm.cu
- gemm_tuner.py
- GemmTuner.py
- hipbsolgemm.cuh
- rocsolgemm.cuh
- README.md
- setup.py
- f4gemm.csv
- f4gemm_bf16_mxfp4_ABpreShuffle_256x256_4x4_ps.co
- f4gemm_bf16_nvfp4_ABpreShuffle_256x256_4x4_ps.co
- f4gemm_fp4_mxfp4_ABpreShuffle_256x256_4x4_ps_noscale.co
- f4gemm_fp8_mxfp4_ABpreShuffle_256x256_4x4_ps.co
- f4gemm_fp8_nvfp4_ABpreShuffle_256x256_4x4_ps.co
- fmha_bf16_pertokenBf16_hd128_128x256.co
- fmha_bf16_pertokenBf16_hd128_128x256_mask.co
- fmha_bf16_pertokenBf16_hd64_128x256.co
- fmha_bf16_pertokenBf16_hd64_128x256_mask.co
- fmha_fwd_bf16.csv
- fmha_bf16_pertokenBf16_hd128_128x256_mask_varlen.co
- fmha_bf16_pertokenBf16_hd128_128x256_varlen.co
- fmha_bf16_pertokenBf16_hd64_128x256_mask_varlen.co
- fmha_bf16_pertokenBf16_hd64_128x256_varlen.co
- fmha_fwd_bf16_varlen.csv
- fmha_fwd_mxfp8.csv
- fmha_fwd_mxfp8_d128.co
- fmha_fwd_mxfp8_d128_causal.co
- bwd_hd128_bf16_a32_pssk.co
- bwd_hd128_bf16_a32_pssk_perf.co
- bwd_hd128_bf16_causal_br_a32_pssk.co
- bwd_hd128_bf16_causal_br_a32_pssk_perf.co
- bwd_hd128_dq_convert_bf16.co
- bwd_hd128_odo_bf16.co
- fmha_bwd_dq_convert.csv
- fmha_bwd_dqdkdv.csv
- fmha_bwd_odo.csv
- mla_a8w8_qh128_1tg_16mx4_64nx1_np.co
- mla_a8w8_qh16_1tg_16mx1_32nx4_np_3p.co
- mla_a8w8_qh16_1tg_16mx2_32nx4_np_3p.co
- mla_a8w8_qh16_1tg_16mx4_64nx1_np.co
- mla_a8w8_qh32_1tg_32mx1_32nx4_np_3p.co
- mla_a8w8_qh64_1tg_16mx4_64nx1_np.co
- mla_a8w8_qh8_1tg_8mx1_32nx4_np_3p.co
- mla_a8w8_qh8_1tg_8mx2_32nx4_np_3p.co
- mla_a8w8_qh8_1tg_8mx3_32nx4_np_3p.co
- mla_a8w8_qh8_1tg_8mx4_32nx4_np_3p.co
- mla_asm.csv
- mla_a8w8_qh16_1tg_16mx4_64nx1_sparse.co
- mla_a8w8_qh64_1tg_16mx4_64nx1_sparse.co
- mla_v4_asm.csv
- f8gemm_bf16_mxfp8fp4_ABpreShuffle_256x256_4x4_ps.co
- f8gemm_bf16_mxfp8fp4_ABpreShuffle_64x512_4x1_ps.co
- f8gemm_bf16_mxfp8fp8_ABpreShuffle_256x256_4x4_ps.co
- f8gemm_bf16_mxfp8fp8_ABpreShuffle_64x512_4x1_ps.co
- mxfp8fp4gemm.csv
- pa_decode_bf16.csv
- pa_decode_bf16_d64_page256_gqa8_tq16.co
- bf16gemm_fp32bf16.csv
- bf16gemm_fp32bf16_tn_128x64_bshuffle_splitk.co
- bf16gemm_fp32bf16_tn_128x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_160x64_bshuffle_splitk.co
- bf16gemm_fp32bf16_tn_160x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_32x64_bshuffle_splitk.co
- bf16gemm_fp32bf16_tn_32x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_32x64_pf3_splitk.co
- bf16gemm_fp32bf16_tn_32x64_splitk_clean.co
- bf16gemm_fp32bf16_tn_48x64_bshuffle_splitk.co
- bf16gemm_fp32bf16_tn_48x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_48x64_pf3_splitk.co
- bf16gemm_fp32bf16_tn_48x64_splitk_clean.co
- bf16gemm_fp32bf16_tn_64x64_bshuffle_splitk.co
- bf16gemm_fp32bf16_tn_64x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_64x64_pf3_splitk.co
- bf16gemm_fp32bf16_tn_64x64_splitk_clean.co
- bf16gemm_fp32bf16_tn_80x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_80x64_splitk_clean.co
- bf16gemm_fp32bf16_tn_96x64_bshuffle_splitk.co
- bf16gemm_fp32bf16_tn_96x64_bshuffle_splitk_clean.co
- bf16gemm_fp32bf16_tn_96x64_pf3_splitk.co
- bf16gemm_fp32bf16_tn_96x64_splitk_clean.co
- f4gemm_bf16_per1x32Fp4.csv
- bwd_hd128_bf16_a16_rtna.co
- bwd_hd128_bf16_a16_rtna_pddv.co
- bwd_hd128_bf16_a16_rtne.co
- bwd_hd128_bf16_a16_rtne_pddv.co
- bwd_hd128_bf16_a16_rtz.co
- bwd_hd128_bf16_a16_rtz_pddv.co
- bwd_hd128_bf16_a32_rtna.co
- bwd_hd128_bf16_a32_rtna_pssk_group.co
- bwd_hd128_bf16_a32_rtna_psskddv.co
- bwd_hd128_bf16_a32_rtna_psskddv_group.co
- bwd_hd128_bf16_a32_rtne.co
- bwd_hd128_bf16_a32_rtne_pssk_group.co
- bwd_hd128_bf16_a32_rtne_psskddv.co
- bwd_hd128_bf16_a32_rtne_psskddv_group.co
- bwd_hd128_bf16_a32_rtz.co
- bwd_hd128_bf16_a32_rtz_pssk_group.co
- bwd_hd128_bf16_a32_rtz_psskddv.co
- bwd_hd128_bf16_a32_rtz_psskddv_group.co
- bwd_hd128_bf16_causal_a16_rtna.co
- bwd_hd128_bf16_causal_a16_rtna_pddv.co
- bwd_hd128_bf16_causal_a16_rtne.co
- bwd_hd128_bf16_causal_a16_rtne_pddv.co
- bwd_hd128_bf16_causal_a16_rtz.co
- bwd_hd128_bf16_causal_a16_rtz_pddv.co
- bwd_hd128_bf16_causal_a32_rtna.co
- bwd_hd128_bf16_causal_a32_rtna_pssk_group.co
- bwd_hd128_bf16_causal_a32_rtna_psskddv.co
- bwd_hd128_bf16_causal_a32_rtna_psskddv_group.co
- bwd_hd128_bf16_causal_a32_rtne.co
- bwd_hd128_bf16_causal_a32_rtne_pssk_group.co
- bwd_hd128_bf16_causal_a32_rtne_psskddv.co
- bwd_hd128_bf16_causal_a32_rtne_psskddv_group.co
- bwd_hd128_bf16_causal_a32_rtz.co
- bwd_hd128_bf16_causal_a32_rtz_pssk_group.co
- bwd_hd128_bf16_causal_a32_rtz_psskddv.co
- bwd_hd128_bf16_causal_a32_rtz_psskddv_group.co
- bwd_hd128_bf16_causal_br_a32_rtna_pssk_group.co
- bwd_hd128_bf16_causal_br_a32_rtna_psskddv.co
- bwd_hd128_bf16_causal_br_a32_rtna_psskddv_group.co
- bwd_hd128_bf16_causal_br_a32_rtne_pssk_group.co
- bwd_hd128_bf16_causal_br_a32_rtne_psskddv.co
- bwd_hd128_bf16_causal_br_a32_rtne_psskddv_group.co
- bwd_hd128_bf16_causal_br_a32_rtz_pssk_group.co
- bwd_hd128_bf16_causal_br_a32_rtz_psskddv.co
- bwd_hd128_bf16_causal_br_a32_rtz_psskddv_group.co
- bwd_hd128_bf16_swa_a32_rtna_psskddv.co
- bwd_hd128_bf16_swa_a32_rtne_psskddv.co
- bwd_hd128_bf16_swa_a32_rtz_psskddv.co
- bwd_hd128_dq_convert_bf16_rtna.co
- bwd_hd128_dq_convert_bf16_rtna_group.co
- bwd_hd128_dq_convert_bf16_rtne.co
- bwd_hd128_dq_convert_bf16_rtne_group.co
- bwd_hd128_dq_convert_bf16_rtz.co
- bwd_hd128_dq_convert_bf16_rtz_group.co
- bwd_hd128_dq_convert_fp16.co
- bwd_hd128_dq_convert_fp16_group.co
- bwd_hd128_fp16_a16.co
- bwd_hd128_fp16_a16_pddv.co
- bwd_hd128_fp16_a32.co
- bwd_hd128_fp16_a32_pssk_group.co
- bwd_hd128_fp16_a32_psskddv.co
- bwd_hd128_fp16_a32_psskddv_group.co
- bwd_hd128_fp16_causal_a16.co
- bwd_hd128_fp16_causal_a16_pddv.co
- bwd_hd128_fp16_causal_a32.co
- bwd_hd128_fp16_causal_a32_pssk_group.co
- bwd_hd128_fp16_causal_a32_psskddv.co
- bwd_hd128_fp16_causal_a32_psskddv_group.co
- bwd_hd128_fp16_causal_br_a32_pssk_group.co
- bwd_hd128_fp16_causal_br_a32_psskddv.co
- bwd_hd128_fp16_causal_br_a32_psskddv_group.co
- bwd_hd128_fp16_swa_a32_psskddv.co
- bwd_hd128_odo_bf16.co
- bwd_hd128_odo_bf16_group.co
- bwd_hd128_odo_fp16.co
- bwd_hd128_odo_fp16_group.co
- bwd_hd192_bf16_a32_rtna_psskddv.co
- bwd_hd192_bf16_a32_rtna_psskddv_group.co
- bwd_hd192_bf16_a32_rtne_psskddv.co
- bwd_hd192_bf16_a32_rtne_psskddv_group.co
- bwd_hd192_bf16_a32_rtz_psskddv.co
- bwd_hd192_bf16_a32_rtz_psskddv_group.co
- bwd_hd192_bf16_causal_a32_rtna_psskddv.co
- bwd_hd192_bf16_causal_a32_rtna_psskddv_group.co
- bwd_hd192_bf16_causal_a32_rtne_psskddv.co
- bwd_hd192_bf16_causal_a32_rtne_psskddv_group.co
- bwd_hd192_bf16_causal_a32_rtz_psskddv.co
- bwd_hd192_bf16_causal_a32_rtz_psskddv_group.co
- bwd_hd192_bf16_causal_br_a32_rtna_psskddv.co
- bwd_hd192_bf16_causal_br_a32_rtna_psskddv_group.co
- bwd_hd192_bf16_causal_br_a32_rtne_psskddv.co
- bwd_hd192_bf16_causal_br_a32_rtne_psskddv_group.co
- bwd_hd192_bf16_causal_br_a32_rtz_psskddv.co
- bwd_hd192_bf16_causal_br_a32_rtz_psskddv_group.co
- bwd_hd192_dq_convert_bf16_rtna.co
- bwd_hd192_dq_convert_bf16_rtna_group.co
- bwd_hd192_dq_convert_bf16_rtne.co
- bwd_hd192_dq_convert_bf16_rtne_group.co
- bwd_hd192_dq_convert_bf16_rtz.co
- bwd_hd192_dq_convert_bf16_rtz_group.co
- bwd_hd192_dq_convert_fp16.co
- bwd_hd192_dq_convert_fp16_group.co
- bwd_hd192_fp16_a32_psskddv.co
- bwd_hd192_fp16_a32_psskddv_group.co
- bwd_hd192_fp16_causal_a32_psskddv.co
- bwd_hd192_fp16_causal_a32_psskddv_group.co
- bwd_hd192_fp16_causal_br_a32_psskddv.co
- bwd_hd192_fp16_causal_br_a32_psskddv_group.co
- bwd_hd192_odo_bf16.co
- bwd_hd192_odo_bf16_group.co
- bwd_hd192_odo_fp16.co
- bwd_hd192_odo_fp16_group.co
- bwd_hd64_bf16_a16_rtna.co
- bwd_hd64_bf16_a16_rtne.co
- bwd_hd64_bf16_a16_rtz.co
- bwd_hd64_bf16_a32_rtna_pssk.co
- bwd_hd64_bf16_a32_rtna_pssk_group.co
- bwd_hd64_bf16_a32_rtne_pssk.co
- bwd_hd64_bf16_a32_rtne_pssk_group.co
- bwd_hd64_bf16_a32_rtz_pssk.co
- bwd_hd64_bf16_a32_rtz_pssk_group.co
- bwd_hd64_bf16_causal_a16_rtna.co
- bwd_hd64_bf16_causal_a16_rtne.co
- bwd_hd64_bf16_causal_a16_rtz.co
- bwd_hd64_bf16_causal_a32_rtna_pssk.co
- bwd_hd64_bf16_causal_a32_rtna_pssk_group.co
- bwd_hd64_bf16_causal_a32_rtne_pssk.co
- bwd_hd64_bf16_causal_a32_rtne_pssk_group.co
- bwd_hd64_bf16_causal_a32_rtz_pssk.co
- bwd_hd64_bf16_causal_a32_rtz_pssk_group.co
- bwd_hd64_bf16_causal_br_a32_rtna_pssk.co
- bwd_hd64_bf16_causal_br_a32_rtna_pssk_group.co
- bwd_hd64_bf16_causal_br_a32_rtne_pssk.co
- bwd_hd64_bf16_causal_br_a32_rtne_pssk_group.co
- bwd_hd64_bf16_causal_br_a32_rtz_pssk.co
- bwd_hd64_bf16_causal_br_a32_rtz_pssk_group.co
- bwd_hd64_dq_convert_bf16_rtna.co
- bwd_hd64_dq_convert_bf16_rtna_group.co
- bwd_hd64_dq_convert_bf16_rtne.co
- bwd_hd64_dq_convert_bf16_rtne_group.co
- bwd_hd64_dq_convert_bf16_rtz.co
- bwd_hd64_dq_convert_bf16_rtz_group.co
- bwd_hd64_dq_convert_fp16.co
- bwd_hd64_dq_convert_fp16_group.co
- bwd_hd64_fp16_a16.co
- bwd_hd64_fp16_a32_pssk.co
- bwd_hd64_fp16_a32_pssk_group.co
- bwd_hd64_fp16_causal_a16.co
- bwd_hd64_fp16_causal_a32_pssk.co
- bwd_hd64_fp16_causal_a32_pssk_group.co
- bwd_hd64_fp16_causal_br_a32_pssk.co
- bwd_hd64_fp16_causal_br_a32_pssk_group.co
- bwd_hd64_odo_bf16.co
- bwd_hd64_odo_bf16_group.co
- bwd_hd64_odo_fp16.co
- bwd_hd64_odo_fp16_group.co
- fmha_bwd_dq_convert.csv
- fmha_bwd_dq_shuffle.csv
- fmha_bwd_dqdkdv.csv
- fmha_bwd_odo.csv
- fwd_hd128_bf16_causal_rtna.co
- fwd_hd128_bf16_causal_rtna_group.co
- fwd_hd128_bf16_causal_rtne.co
- fwd_hd128_bf16_causal_rtne_group.co
- fwd_hd128_bf16_causal_rtz.co
- fwd_hd128_bf16_causal_rtz_group.co
- fwd_hd128_bf16_rtna.co
- fwd_hd128_bf16_rtna_group.co
- fwd_hd128_bf16_rtne.co
- fwd_hd128_bf16_rtne_group.co
- fwd_hd128_bf16_rtz.co
- fwd_hd128_bf16_rtz_group.co
- fwd_hd128_fp8.co
- fwd_hd128_fp8_causal.co
- fwd_hd128_fp8_causal_group.co
- fwd_hd128_fp8_group.co
- fwd_hd192x128_bf16_causal_rtna.co
- fwd_hd192x128_bf16_causal_rtna_group.co
- fwd_hd192x128_bf16_causal_rtne.co
- fwd_hd192x128_bf16_causal_rtne_group.co
- fwd_hd192x128_bf16_causal_rtz.co
- fwd_hd192x128_bf16_causal_rtz_group.co
- fwd_hd192x128_bf16_rtna.co
- fwd_hd192x128_bf16_rtna_group.co
- fwd_hd192x128_bf16_rtne.co
- fwd_hd192x128_bf16_rtne_group.co
- fwd_hd192x128_bf16_rtz.co
- fwd_hd192x128_bf16_rtz_group.co
- fwd_hd128_bf16_causal_rtna.co
- fwd_hd128_bf16_causal_rtna_group.co
- fwd_hd128_bf16_causal_rtne.co
- fwd_hd128_bf16_causal_rtne_group.co
- fwd_hd128_bf16_causal_rtz.co
- fwd_hd128_bf16_causal_rtz_group.co
- fwd_hd128_bf16_rtna.co
- fwd_hd128_bf16_rtna_group.co
- fwd_hd128_bf16_rtne.co
- fwd_hd128_bf16_rtne_group.co
- fwd_hd128_bf16_rtz.co
- fwd_hd128_bf16_rtz_group.co
- fwd_hd128_fp8.co
- fwd_hd128_fp8_causal.co
- fwd_hd128_fp8_causal_group.co
- fwd_hd128_fp8_group.co
- fwd_hd192x128_bf16_causal_rtna.co
- fwd_hd192x128_bf16_causal_rtna_group.co
- fwd_hd192x128_bf16_causal_rtne.co
- fwd_hd192x128_bf16_causal_rtne_group.co
- fwd_hd192x128_bf16_causal_rtz.co
- fwd_hd192x128_bf16_causal_rtz_group.co
- fwd_hd192x128_bf16_rtna.co
- fwd_hd192x128_bf16_rtna_group.co
- fwd_hd192x128_bf16_rtne.co
- fwd_hd192x128_bf16_rtne_group.co
- fwd_hd192x128_bf16_rtz.co
- fwd_hd192x128_bf16_rtz_group.co
- fmha_fwd.csv
- fwd_hd128_i8fp8.co
- fmha_v4_fwd.csv
- fmoe_bf16_blockscaleFp8_g1u1_gelu.csv
- fmoe_bf16_blockscaleFp8_g1u1_gelu_1tg_32x256.co
- fmoe_bf16_blockscaleFp8_g1u1_gelu_1tg_ps_32x256.co
- fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_32x128.co
- fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_32x256.co
- fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_ps_32x128.co
- fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_ps_32x256.co
- fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_32x128.co
- fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_32x256.co
- fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_64x256.co
- fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_ps_32x128.co
- fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_ps_32x256.co
- fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_psx_64x256.co
- fmoe_bf16_noquant_g1u0_gelu.csv
- fmoe_bf16_noquantBf16_g1u0_atm_inlv_gelu_1tg_32x512.co
- fmoe_bf16_noquantBf16_g1u0_atm_inlv_gelu_1tg_ps_32x512.co
- fmoe_bf16_noquantBf16_g1u0_vs_atm_inlv_gelu_1tg_32x512.co
- fmoe_bf16_noquantBf16_g1u0_vs_atm_inlv_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_gelu.csv
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_gelu_tkw1.csv
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_smf_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_smf_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_smf_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_smf_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x512.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_atm_opt_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_atm_opt_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_gelu.csv
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u0_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_smf_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_vs_atm_opt_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_vs_atm_opt_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u0_vs_smf_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_gelu.csv
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_gelu_tkw1.csv
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_smf_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_smf_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x128.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x192.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x320.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x384.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x448.co
- fmoe_bf16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x512.co
- fmoe_bf16_pertokenInt8_g1u1_vs_smf_gelu_1tg_1tg_32x256.co
- fmoe_bf16_pertokenInt8_g1u1_vs_smf_gelu_1tg_32x320.co
- fmoe_fp16_blockscaleFp8_g1u1_gelu.csv
- fmoe_fp16_blockscaleFp8_g1u1_gelu_1tg_32x256.co
- fmoe_fp16_blockscaleFp8_g1u1_gelu_1tg_ps_32x256.co
- fmoe_fp16_blockscaleFp8_g1u1_novs_gelu_1tg_32x256.co
- fmoe_fp16_blockscaleFp8_g1u1_novs_gelu_1tg_ps_32x256.co
- fmoe_fp16_blockscaleFp8_g1u1_vs_gelu_1tg_32x256.co
- fmoe_fp16_blockscaleFp8_g1u1_vs_gelu_1tg_ps_32x256.co
- fmoe_fp16_noquant_g1u0_gelu.csv
- fmoe_fp16_noquantBf16_g1u0_vs_atm_inlv_gelu_1tg_ps_32x512.co
- fmoe_fp16_noquantFp16_g1u0_atm_inlv_gelu_1tg_32x512.co
- fmoe_fp16_noquantFp16_g1u0_atm_inlv_gelu_1tg_ps_32x512.co
- fmoe_fp16_noquantFp16_g1u0_vs_atm_inlv_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_gelu.csv
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_gelu_tkw1.csv
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_multix_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_smf_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_smf_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_vs_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_vs_multix_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_smf_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_smf_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_32x512.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenFp8_g1u1_vs_tkw1_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_atm_opt_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_atm_opt_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_gelu.csv
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u0_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_smf_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_vs_atm_opt_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_vs_atm_opt_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u0_vs_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u0_vs_smf_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_gelu.csv
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_gelu_tkw1.csv
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_multix_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_smf_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_smf_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_tkw1_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_vs_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x128.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x192.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x320.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x384.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x448.co
- fmoe_fp16_pertokenInt8_g1u1_vs_multix_gelu_1tg_ps_32x512.co
- fmoe_fp16_pertokenInt8_g1u1_vs_smf_gelu_1tg_32x256.co
- fmoe_fp16_pertokenInt8_g1u1_vs_smf_gelu_1tg_32x320.co
- fmoe_fp8_g1u1_subGU_128_gelu.co
- fmoe_fp8_g1u1_subGU_128_gelu_tkw1.co
- fmoe_fp8_g1u1_subGU_192_gelu.co
- fmoe_fp8_g1u1_subGU_192_gelu_tkw1.co
- fmoe_fp8_g1u1_subGU_256_gelu.co
- fmoe_fp8_g1u1_subGU_256_gelu_tkw1.co
- fmoe_fp8_g1u1_subGU_320_gelu.co
- fmoe_fp8_g1u1_subGU_320_gelu_tkw1.co
- fmoe_fp8_g1u1_subGU_384_gelu.co
- fmoe_fp8_g1u1_subGU_384_gelu_tkw1.co
- fmoe_fp8_g1u1_subGU_448_gelu.co
- fmoe_fp8_g1u1_subGU_448_gelu_tkw1.co
- fmoe_fp8_g1u1_subGU_512_gelu.co
- fmoe_fp8_g1u1_subGU_512_gelu_tkw1.co
- fmoe_int8_g1u0_subGU_128_gelu.co
- fmoe_int8_g1u0_subGU_192_gelu.co
- fmoe_int8_g1u0_subGU_256_gelu.co
- fmoe_fp8_blockscale_g1u1_novs_subGU_256.co
- fmoe_fp8_blockscale_g1u1_novs_subGU_256_ps.co
- fmoe_fp8_blockscale_g1u1_subGU_256.co
- all_reduce.co
- allreduce_layernorm_N8192.co
- allreduce_rmsnorm_N8192.co
- allreduce_rmsnorm_qnt_N8192.co
- flatmm_uk_gfx9_f16f8_128x256x128_1x4x1_16x16x32.co
- codegen.py
- .clang-format
- .clang-tidy
- .gitattributes
- .gitignore
- .gitmodules
- CONTRIBUTE.md
- LICENSE
- MANIFEST.in
- README.md
- REPO_CLEANUP_PLAN.md
# 설치 가이드
1. 코드 내려받기
git clone https://github.com/ROCm/aiter
깃허브에서 프로젝트 코드 전체를 내 컴퓨터로 내려받습니다.
cd aiter
방금 내려받은 프로젝트 폴더 안으로 이동합니다.
2. Python
쉬움 추천사전 준비물
pip install -r requirements.txt
requirements.txt 등에 명시된 파이썬 라이브러리를 설치합니다.
pip install -e .
requirements.txt 등에 명시된 파이썬 라이브러리를 설치합니다.
pip install -r requirements-triton-comms.txt
requirements.txt 등에 명시된 파이썬 라이브러리를 설치합니다.
에러 메시지 없이 실행되고 터미널에 안내 문구가 출력되면 정상입니다.
이 레포의 README에 적힌 실제 명령어를 그대로 가져왔습니다.
3. Make
보통사전 준비물
- Git GitHub에서 프로젝트 코드를 내려받으려면 필요합니다.
- Make Linux/macOS는 보통 기본 설치되어 있습니다. Windows는 별도 설치(예: MSYS2, WSL)가 필요합니다.
⚠️ 이 프로젝트는 규모가 큰 저장소라, 이 방법이 실제 핵심 제품이 아니라 내부 하위 패키지를 가리키는 것일 수 있습니다. README 전체를 함께 확인해보세요.
cd csrc/cpp_itfs/mla
이 프로젝트의 관련 파일이 하위 폴더 안에 있어서, 먼저 그 폴더로 이동합니다.
make
생성된 빌드 설정을 바탕으로 실제 컴파일을 진행해 실행 파일을 만듭니다.
에러 없이 끝나면 성공입니다. 생성된 실행 파일을 직접 실행해보세요.
// repository documentation
Was this content helpful?
(0 ratings)