LICENSE
MANIFEST.in
NOTICE
README.md
THIRD_PARTY_NOTICES.md
TRADEMARKS.md
pyproject.toml
setup.py
LICENSES/Apache-2.0.txt
LICENSES/BSD-3-Clause-causal-conv1d.txt
LICENSES/MIT-flash-linear-attention.txt
eagle_harness/__init__.py
eagle_harness/cli.py
eagle_harness/compat.py
eagle_harness.egg-info/PKG-INFO
eagle_harness.egg-info/SOURCES.txt
eagle_harness.egg-info/dependency_links.txt
eagle_harness.egg-info/entry_points.txt
eagle_harness.egg-info/requires.txt
eagle_harness.egg-info/top_level.txt
eagle_harness/core/__init__.py
eagle_harness/core/aiter_gemm_bf16.csv
eagle_harness/core/attention_sweeps.py
eagle_harness/core/calibration.py
eagle_harness/core/candidate_calibration.py
eagle_harness/core/chain_width.py
eagle_harness/core/config.py
eagle_harness/core/cost_model.py
eagle_harness/core/cost_profiles.py
eagle_harness/core/cost_table.py
eagle_harness/core/draft_tree.py
eagle_harness/core/gemm_solutions.py
eagle_harness/core/limits.py
eagle_harness/core/live_costs.py
eagle_harness/core/models.py
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/README.md
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/candidate_topk.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/draft_config.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/gemm_solutions.csv
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/target_config.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_1.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_2.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_3.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_4.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_5.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_llm/cache_interval_6.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_1.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_2.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_3.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_4.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_5.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/cache_slm/cache_interval_6.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_1.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_2.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_3.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_4.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_5.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_llm/model_interval_6.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_1.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_2.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_3.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_4.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_5.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/model_slm/model_interval_6.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_llm/query_1.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_llm/query_16.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_llm/query_32.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_llm/query_8.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_slm/query_1.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_slm/query_2.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_slm/query_4.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_1/query_slm/query_8.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_32/model_llm/model_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_32/model_slm/model_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_8/model_llm/model_interval_0.json
eagle_harness/core/profiles/mi355x-qwen3.8-27b-tp1/cost/bs_8/model_slm/model_interval_0.json
eagle_harness/core/profiles/vllm-qwen38-tp4/README.md
eagle_harness/core/profiles/vllm-qwen38-tp4/rounds/tree.json
eagle_harness/kernels/__init__.py
eagle_harness/kernels/accept_tail.py
eagle_harness/kernels/arch.py
eagle_harness/kernels/build.py
eagle_harness/kernels/candidate.py
eagle_harness/kernels/candidate_ops.py
eagle_harness/kernels/candidate_triton.py
eagle_harness/kernels/gather_dot.py
eagle_harness/kernels/gdn_tree.py
eagle_harness/kernels/gdn_tree_chunk_flydsl.py
eagle_harness/kernels/gdn_tree_conv.py
eagle_harness/kernels/gdn_tree_gluon.py
eagle_harness/kernels/gdn_tree_records.py
eagle_harness/kernels/gdn_tree_recurrence.py
eagle_harness/kernels/l2_prefetch.py
eagle_harness/kernels/mamba2_tree.py
eagle_harness/kernels/native.py
eagle_harness/kernels/online_softmax.py
eagle_harness/kernels/row_attention.py
eagle_harness/kernels/tree_step.py
eagle_harness/kernels/verify_prep.py
eagle_harness/kernels/csrc/candidate.cu
eagle_harness/models/__init__.py
eagle_harness/models/eagle4_config.py
eagle_harness/models/eagle4_structure.py
eagle_harness/models/low_rank_head.py
eagle_harness/sglang/__init__.py
eagle_harness/sglang/algorithm.py
eagle_harness/sglang/calibrate.py
eagle_harness/sglang/candidate_logits.py
eagle_harness/sglang/chain_rounds.py
eagle_harness/sglang/draft_extend.py
eagle_harness/sglang/draft_head.py
eagle_harness/sglang/draft_steps.py
eagle_harness/sglang/drafter.py
eagle_harness/sglang/gdn_tree_verify.py
eagle_harness/sglang/graph_pool.py
eagle_harness/sglang/hooks.py
eagle_harness/sglang/l2_warm.py
eagle_harness/sglang/mamba2_tree_verify.py
eagle_harness/sglang/plain_decode.py
eagle_harness/sglang/plugin.py
eagle_harness/sglang/row_attention.py
eagle_harness/sglang/tail_mask.py
eagle_harness/sglang/verify_graphs.py
eagle_harness/sglang/verify_input.py
eagle_harness/sglang/verify_plan.py
eagle_harness/sglang/verify_step.py
eagle_harness/sglang/verify_tail.py
eagle_harness/sglang/worker.py
eagle_harness/sglang/models/__init__.py
eagle_harness/sglang/models/llama_eagle3.py
eagle_harness/sglang/models/llama_eagle4.py
eagle_harness/tokenspeed/__init__.py
eagle_harness/tokenspeed/config.py
eagle_harness/tokenspeed/costs.py
eagle_harness/tokenspeed/drafter.py
eagle_harness/tokenspeed/hooks.py
eagle_harness/tokenspeed/models.py
eagle_harness/tokenspeed/plugin.py
eagle_harness/tokenspeed/tree_lanes.py
eagle_harness/vllm/__init__.py
eagle_harness/vllm/adapter.py
eagle_harness/vllm/calibrate.py
eagle_harness/vllm/candidate_gather.py
eagle_harness/vllm/config.py
eagle_harness/vllm/depth_stop.py
eagle_harness/vllm/draft_attention.py
eagle_harness/vllm/draft_rows.py
eagle_harness/vllm/eagle3.py
eagle_harness/vllm/gdn_tree_verify.py
eagle_harness/vllm/model_rows.py
eagle_harness/vllm/models.py
eagle_harness/vllm/plugin.py
eagle_harness/vllm/round_commit.py
eagle_harness/vllm/round_recorder.py
eagle_harness/vllm/speculator.py
eagle_harness/vllm/tree_accept.py
eagle_harness/vllm/tree_attention.py
eagle_harness/vllm/tree_drafter.py
eagle_harness/vllm/verify_attention.py
eagle_harness/vllm/verify_layout.py
tests/__init__.py
tests/synthetic.py
tests/test_accept_tail_gpu.py
tests/test_accept_tail_graphs.py
tests/test_attention_sweeps.py
tests/test_calibration_step_timings.py
tests/test_candidate_calibration.py
tests/test_candidate_gather.py
tests/test_candidate_head.py
tests/test_candidate_ops_gpu.py
tests/test_chain_no_spec.py
tests/test_chain_policy_costs.py
tests/test_chain_rounds.py
tests/test_chain_width.py
tests/test_check_headers.py
tests/test_cli.py
tests/test_compat.py
tests/test_config.py
tests/test_config_cleanup.py
tests/test_cost_golden.py
tests/test_cost_policy_tables.py
tests/test_cost_profiles.py
tests/test_dense_target_commit.py
tests/test_depth_stop.py
tests/test_draft_extend.py
tests/test_draft_input_gpu.py
tests/test_draft_layer_input.py
tests/test_draft_rows.py
tests/test_draft_tree.py
tests/test_draft_vocab.py
tests/test_drafter.py
tests/test_eagle3_attention.py
tests/test_eagle3_config.py
tests/test_eagle3_sglang.py
tests/test_eagle3_vllm.py
tests/test_eagle4_model.py
tests/test_eagle4_port.py
tests/test_eagle4_structure.py
tests/test_engine_release.py
tests/test_fused_accept_step.py
tests/test_gdn_rounded_records_gpu.py
tests/test_gdn_tree_conv.py
tests/test_gdn_tree_gluon_gpu.py
tests/test_gdn_tree_kernels.py
tests/test_gdn_tree_records_gpu.py
tests/test_gemm_solutions.py
tests/test_gpu_kernels.py
tests/test_image_tag.py
tests/test_kernels.py
tests/test_l2_warm_gpu.py
tests/test_limits.py
tests/test_live_costs.py
tests/test_mamba2_tree_kernels.py
tests/test_mamba2_tree_verify_gpu.py
tests/test_models.py
tests/test_plain_decode.py
tests/test_prelaunch_gpu.py
tests/test_rocm_dispatch.py
tests/test_rocm_prelaunch.py
tests/test_row_attention_gpu.py
tests/test_serving_choice.py
tests/test_sglang_hooks.py
tests/test_sglang_logprobs.py
tests/test_sglang_overlap.py
tests/test_sglang_plugin.py
tests/test_target_models.py
tests/test_third_party_headers.py
tests/test_tokenspeed_adapter.py
tests/test_tokenspeed_nemotron.py
tests/test_tokenspeed_ownership.py
tests/test_tokenspeed_plugin.py
tests/test_tokenspeed_policy_cost.py
tests/test_tokenspeed_policy_golden.py
tests/test_tree_graph_memory_gpu.py
tests/test_tree_load_policy.py
tests/test_tree_rejection_sampler.py
tests/test_tree_round_kernels.py
tests/test_tree_sampling.py
tests/test_tree_step_attention.py
tests/test_tree_step_gpu.py
tests/test_verify_attention.py
tests/test_verify_attn_args_gpu.py
tests/test_verify_input.py
tests/test_verify_plan.py
tests/test_verify_prep_gpu.py
tests/test_verify_rows.py
tests/test_verify_tail_gpu.py
tests/test_vllm_calibrate.py
tests/test_vllm_model_rows.py
tests/test_vllm_plugin.py
tests/test_vllm_priced_serve.py
tests/test_vllm_round_recorder.py
tests/test_wheel_index.py
tests/test_worker_rounds.py
tests/triton_probe.py
tests/golden/fork_golden.json
tests/golden/nemotron35_super_eagle4_draft_config.json