TokenSpeed is a speed-of-light LLM inference engine.
This report summarizes scanner evidence for lightseekorg/tokenspeed, a Python project with 2,211 GitHub stars. SynthScan examined 957,777 lines of code across 2349 source files, recording 14463 pattern matches distributed across 23 syntactic categories. The legacy adjusted pattern signal is 19.1 (Elevated pattern signal); it is not an authorship estimate.
Use the findings to ask whether maintainers can explain intent, verify dependencies, simplify unnecessary scaffolding, and focus review effort. Risk Payload v1 adds ownership, supply-chain, over-engineering, and review-effort vectors when available. These are prompts for human verification, not probabilities, security findings, or measured maintenance costs.
Use each vector to guide manual verification of the underlying findings.
Can maintainers explain intent, assumptions, and trade-offs? 10434 finding(s).
Are imports declared and package/API names verified? 1964 finding(s).
Are wrappers, guards, and repeated structures necessary? 9933 finding(s).
Which changes need focused human review? 14463 finding(s).
This chart tracks the legacy adjusted pattern score across scan runs. Changes can reflect source changes, scan coverage, or detector/scoring versions; they do not measure AI adoption or prove that a team refactored code. Compare runs only when their scoring versions are compatible. The dashed line tracks the raw pattern-hit count.
Groups findings by configured severity weight. Severity is not a confidence estimate, security rating, or proof that a package is missing. Verify imports against the project's manifests and inspect each finding in its local design context.
This horizontal bar chart shows where configured pattern scores accumulate across directories. Treat the bars as locations to inspect, not as evidence of authorship or code quality. Use the file-level findings to decide whether a directory needs a focused ownership, dependency, complexity, or review check.
The scanner identified 14463 distinct pattern matches across 23 syntactic categories. Each entry below represents a discrete location in the source code where the engine recorded a detector match to inspect in context. Expand any category row to inspect the individual file paths, line numbers, code snippets, and the lexical context (CODE, COMMENT, or STRING) in which each match was detected.
Reading the findings table: The Severity column indicates the configured finding weight (CRITICAL / HIGH / MEDIUM / LOW). The Context column identifies whether the match occurred inside executable code, an inline comment, or a string literal — legacy comment-context matches receive a ×1.5 score weight. The ⚡ bolt icon marks clustered matches: three or more patterns within a 10-line window, each receiving an additional ×1.5 density multiplier under the legacy scoring model.
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tokenspeed-kernel-npu/test/test_layernorm.py | 53 | def test_add_rmsnorm_updates_residual(dtype: torch.dtype) -> None: | CODE |
| LOW | tokenspeed-kernel-npu/test/test_mha.py | 83 | def test_mha_extend_with_paged_cache() -> None: | CODE |
| LOW | tokenspeed-kernel-npu/test/test_mha.py | 123 | def test_mha_decode_with_paged_cache() -> None: | CODE |
| LOW | tokenspeed-kernel-npu/test/test_mha.py | 149 | def test_mha_decode_npugraph_updates_sequence_lengths() -> None: | CODE |
| LOW | test/test_utils.py | 377 | def write_github_step_summary(content): | CODE |
| LOW | test/test_host_cache_storage.py | 65 | def test_block_offsets_follow_group_packing_without_overlap(): | CODE |
| LOW | test/test_host_cache_storage.py | 77 | def test_null_and_out_of_range_blocks_are_rejected(group_index, block_id): | CODE |
| LOW | test/test_host_cache_storage.py | 85 | def test_host_fields_are_packed_without_padding(): | CODE |
| LOW | test/test_l3_mooncake_master.py | 145 | def test_wait_tcp_requires_timeout(self) -> None: | CODE |
| LOW | test/test_l3_mooncake_master.py | 151 | def test_host_requires_nbytes_and_page(self) -> None: | CODE |
| LOW | test/runners.py | 560 | def batch_forward_generation_raw( | CODE |
| LOW | test/runners.py | 594 | def check_close_model_outputs( | CODE |
| LOW | test/test_l2_layerwise_load.py | 44 | def test_layerwise_load_waits_for_the_selected_event_set(monkeypatch): | CODE |
| LOW | test/test_l2_layerwise_load.py | 69 | def test_layerwise_load_waits_for_bound_flags(monkeypatch): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 59 | def test_cache_field_layer_id_parses_layer_owned_field(): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 67 | def test_cache_field_layer_id_rejects_invalid_field(field_id): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 72 | def test_layout_rejects_duplicate_group_ids(): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 88 | def test_layout_rejects_duplicate_field_ids_across_groups(): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 98 | def test_layout_rejects_payload_larger_than_device_stride(): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 103 | def test_layout_rejects_unknown_consumer_field(): | CODE |
| LOW⚡ | test/test_cache_transfer_layout.py | 110 | def test_layout_requires_a_positive_num_lcm_blocks(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 122 | def test_lcm_layout_is_derived_from_planned_field_offsets(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 169 | def test_lcm_layout_uses_scheduler_group_order(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 224 | def test_merged_plan_views_filter_and_remap_fields_before_combining(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 309 | def test_target_and_draft_layouts_must_share_one_arena(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 330 | def test_aliased_target_and_draft_layout_is_transferred_once(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 342 | def test_merged_owner_keeps_aliased_field_on_target(): | CODE |
| LOW | test/test_cache_transfer_layout.py | 361 | def test_invalid_target_draft_field_overlap_is_rejected(draft_fields, error): | CODE |
| LOW | test/test_cache_transfer_layout.py | 370 | def test_draft_layout_must_use_target_block_geometry(): | CODE |
| LOW⚡ | test/test_l3_kvstore.py | 1935 | def test_namespace_clear_uses_anchored_escaped_regex(self): | CODE |
| LOW⚡ | test/test_l3_kvstore.py | 1945 | def test_namespace_clear_without_remove_by_regex_is_failure(self): | CODE |
| LOW⚡ | test/test_l3_kvstore.py | 1950 | def test_namespace_clear_failure_is_not_reported_as_success(self): | CODE |
| LOW⚡ | test/test_l3_kvstore.py | 1956 | def test_segment_is_divided_across_tp_and_pp_ranks(self): | CODE |
| LOW | test/test_l3_kvstore.py | 80 | def test_fake_host_requires_buffer_size(self): | CODE |
| LOW | test/test_l3_kvstore.py | 86 | def test_object_key_includes_group_offset_and_rank(self): | CODE |
| LOW | test/test_l3_kvstore.py | 102 | def test_attention_build_exports_resolved_backend_names(self): | CODE |
| LOW | test/test_l3_kvstore.py | 191 | def test_weight_version_factory_preserves_resolved_backend_identity(self): | CODE |
| LOW | test/test_l3_kvstore.py | 233 | def test_prefix_is_stable_and_separates_incompatible_cache_objects(self): | CODE |
| LOW | test/test_l3_kvstore.py | 348 | def test_checkpoint_id_prefers_loaded_commit_over_moving_branch(self): | CODE |
| LOW | test/test_l3_kvstore.py | 389 | def test_checkpoint_id_uses_snapshot_directory_commit(self): | CODE |
| LOW | test/test_l3_kvstore.py | 408 | def test_checkpoint_id_does_not_trust_hub_snapshots_without_refs(self): | CODE |
| LOW | test/test_l3_kvstore.py | 427 | def test_checkpoint_id_does_not_trust_snapshots_dir_outside_hf_hub_cache(self): | CODE |
| LOW | test/test_l3_kvstore.py | 456 | def test_checkpoint_id_fingerprints_local_weight_bytes(self): | CODE |
| LOW | test/test_l3_kvstore.py | 492 | def test_checkpoint_id_does_not_trust_hex_basename_outside_hf_snapshots(self): | CODE |
| LOW | test/test_l3_kvstore.py | 521 | def test_checkpoint_id_ignores_inherited_commit_on_local_dir(self): | CODE |
| LOW | test/test_l3_kvstore.py | 545 | def test_checkpoint_id_fingerprints_local_hf_quant_config(self): | CODE |
| LOW | test/test_l3_kvstore.py | 574 | def test_checkpoint_id_fingerprints_local_config_code(self): | CODE |
| LOW | test/test_l3_kvstore.py | 603 | def test_checkpoint_id_fingerprints_imported_custom_code_subdirectories(self): | CODE |
| LOW | test/test_l3_kvstore.py | 639 | def test_checkpoint_id_fingerprints_symlinked_custom_code_packages(self): | CODE |
| LOW | test/test_l3_kvstore.py | 699 | def test_checkpoint_id_uses_selected_load_format_weight_files(self): | CODE |
| LOW | test/test_l3_kvstore.py | 740 | def test_checkpoint_id_ignores_unselected_weight_files(self): | CODE |
| LOW | test/test_l3_kvstore.py | 778 | def test_checkpoint_id_fingerprints_sharded_state_weight_files(self): | CODE |
| LOW | test/test_l3_kvstore.py | 803 | def test_checkpoint_id_fingerprints_sharded_state_pattern_files(self): | CODE |
| LOW | test/test_l3_kvstore.py | 864 | def test_checkpoint_id_fingerprints_npcache_numpy_files(self): | CODE |
| LOW | test/test_l3_kvstore.py | 896 | def test_checkpoint_id_fingerprints_mistral_shard_index(self): | CODE |
| LOW | test/test_l3_kvstore.py | 934 | def test_checkpoint_id_rejects_unsupported_load_format(self): | CODE |
| LOW | test/test_l3_kvstore.py | 946 | def test_ext_def_file_parses_loader_yaml_forms(self): | CODE |
| LOW | test/test_l3_kvstore.py | 982 | def test_checkpoint_id_hashes_extensible_yaml_and_extension_code(self): | CODE |
| LOW | test/test_l3_kvstore.py | 1038 | def test_checkpoint_id_hashes_extension_code_for_quoted_key_and_flow_yaml(self): | CODE |
| LOW | test/test_l3_kvstore.py | 1084 | def test_checkpoint_id_fingerprints_extensible_local_weights_and_yaml(self): | CODE |
| 7605 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM⚡ | test/runtime/test_dspark_proposal.py | 87 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_dspark_proposal.py | 89 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_dspark_proposal.py | 216 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_dspark_proposal.py | 218 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_dspark_proposal.py | 292 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_dspark_proposal.py | 294 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 290 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 292 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 338 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 342 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 611 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 613 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 644 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 646 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 735 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 737 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 782 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_device_handle.py | 784 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_device_handle.py | 677 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_device_handle.py | 679 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_device_handle.py | 882 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_device_handle.py | 884 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_l3_storage_hits.py | 271 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_l3_storage_hits.py | 273 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_qwen4_exp.py | 1908 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_qwen4_exp.py | 1910 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 326 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 328 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 444 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 446 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 492 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 494 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 1040 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_retraction_image_executor.py | 1042 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_retraction_image_executor.py | 89 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_retraction_image_executor.py | 91 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_drafter_accept_indexing.py | 690 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_drafter_accept_indexing.py | 694 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_expert_location.py | 426 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_expert_location.py | 429 | # ---------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 72 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 74 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 1133 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 1135 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 1314 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 1316 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 298 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 300 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 946 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 948 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 1013 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 1015 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 1095 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | test/runtime/test_detokenizer_parity.py | 1097 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_input_logprobs.py | 73 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_input_logprobs.py | 75 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_input_logprobs.py | 342 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_input_logprobs.py | 344 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_input_logprobs.py | 783 | # -------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | test/runtime/test_input_logprobs.py | 785 | # -------------------------------------------------------------------------- | COMMENT |
| 1093 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tokenspeed-kernel-npu/test/test_rotary_embedding.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | tokenspeed-kernel-npu/test/test_layernorm.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | tokenspeed-kernel-npu/test/test_mha.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …ed-kernel-npu/python/tokenspeed_kernel_npu/__init__.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/_triton.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …rnel-npu/python/tokenspeed_kernel_npu/ops/layernorm.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …ernel-npu/python/tokenspeed_kernel_npu/ops/__init__.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …u/python/tokenspeed_kernel_npu/ops/rotary_embedding.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/ops/mha.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/test_host_cache_storage.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/test_l3_mooncake_master.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runners.py | 1 | # SPDX-License-Identifier: MIT AND Apache-2.0 | COMMENT |
| LOW | test/test_l2_layerwise_load.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/test_cache_transfer_layout.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/test_l3_kvstore.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | …ci/eval/kimi-k3-mxfp4-tp8ep1-evalscope-aime26-amd.yaml | 1 | api_version: ci.tokenspeed.io/v1 | COMMENT |
| LOW | …kimi-k3-mxfp4-tp8ep1-evalscope-random-4k-1k-mi35x.yaml | 1 | api_version: ci.tokenspeed.io/v1 | COMMENT |
| LOW | …kimi-k3-mxfp4-tp8ep1-evalscope-random-4k-1k-mi35x.yaml | 121 | # Values are [Latency (tps/user), Throughput (tps/gpu)]; gate fails any conc whose actual falls below ref * perf_thresho | COMMENT |
| LOW | test/runtime/test_unified_decode_path.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_cutedsl_mla_dcp_metadata.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_device_handle.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_cudagraph_probe_ladder.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_multimodal_warmup.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_kimi_k3_comm_arming.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_kda_state_eviction.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_executor_capture_step.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_l3_storage_hits.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_gpu_memory_summary.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_moe_ispp_padding.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_dummy_weights.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_unit_kv_scales.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_moe_global_padded_tp_loader.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_forward_routing.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_sharded_state_loader.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_input_pad_tail_tape.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_batch_log.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_cu_extend_seq_lens_cpu_metadata.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_qwen4_exp_cache_without_gdn.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_qwen4_exp.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_logging.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_tree_attention_cascade.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_generation_output_processor.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_kda_state_eviction_spec.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_kvp_cache_contract.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_deepseek_v4_slot_mappings.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_retraction_image_executor.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_runtime_states.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_flashmla_dcp.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_expert_location.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_cupti_warmup_vendor_guard.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_multimodal_encoded_offload.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_jit_compile_check.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_tree_spec.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_gluon_petit_backend.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_input_logprobs.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_deepseek_v41_inputs.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_qwen4_exp_ple_backend.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_cudagraph_support_resolution.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_executor_rebind.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| LOW | test/runtime/test_linear_method_construction.py | 1 | # Copyright (c) 2026 LightSeek Foundation | COMMENT |
| 1833 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/_triton.py | 25 | CODE | |
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/_triton.py | 26 | CODE | |
| LOW | …rnel-npu/python/tokenspeed_kernel_npu/ops/layernorm.py | 23 | CODE | |
| LOW | …u/python/tokenspeed_kernel_npu/ops/rotary_embedding.py | 23 | CODE | |
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/ops/mha.py | 23 | CODE | |
| LOW | test/test_host_cache_storage.py | 21 | CODE | |
| LOW | test/test_l3_mooncake_master.py | 28 | CODE | |
| LOW | test/test_l3_mooncake_master.py | 169 | CODE | |
| LOW | test/test_l2_layerwise_load.py | 21 | CODE | |
| LOW | test/test_cache_transfer_layout.py | 21 | CODE | |
| LOW | test/test_l3_kvstore.py | 21 | CODE | |
| LOW | test/ci/deepswe/test_kimi_code.py | 3 | CODE | |
| LOW | test/ci/deepswe/kimi_code_environment.py | 3 | CODE | |
| LOW | test/ci/deepswe/kimi_code_pier_agent.py | 3 | CODE | |
| LOW | test/ci/deepswe/summarize.py | 4 | CODE | |
| LOW | test/runtime/test_unified_decode_path.py | 36 | CODE | |
| LOW | test/runtime/test_dspark_proposal.py | 11 | CODE | |
| LOW | test/runtime/test_kimi_k3_eagle3.py | 3 | CODE | |
| LOW | test/runtime/test_block_tables_bridge.py | 1 | CODE | |
| LOW | test/runtime/test_device_handle.py | 40 | CODE | |
| LOW | test/runtime/test_inkling_reference_parity.py | 23 | CODE | |
| LOW | test/runtime/test_host_executor.py | 3 | CODE | |
| LOW | test/runtime/test_glm53_flash_cache_spec.py | 3 | CODE | |
| LOW | test/runtime/test_cudagraph_probe_ladder.py | 23 | CODE | |
| LOW | test/runtime/test_kimi_mxfp4_dense_linear.py | 3 | CODE | |
| LOW | test/runtime/test_multimodal_warmup.py | 21 | CODE | |
| LOW | test/runtime/test_kimi_k3_comm_arming.py | 23 | CODE | |
| LOW | test/runtime/test_kda_state_eviction.py | 30 | CODE | |
| LOW | test/runtime/test_executor_capture_step.py | 27 | CODE | |
| LOW | test/runtime/test_l3_storage_hits.py | 29 | CODE | |
| LOW | test/runtime/test_gpu_memory_summary.py | 23 | CODE | |
| LOW | test/runtime/test_trtllm_spec_seqlen_clamp.py | 27 | CODE | |
| LOW | test/runtime/test_moe_ispp_padding.py | 21 | CODE | |
| LOW | test/runtime/conftest.py | 3 | CODE | |
| LOW | test/runtime/test_dummy_weights.py | 24 | CODE | |
| LOW | test/runtime/test_cache_memory_plan.py | 1 | CODE | |
| LOW | test/runtime/test_moe_global_padded_tp_loader.py | 21 | CODE | |
| LOW | test/runtime/test_forward_routing.py | 27 | CODE | |
| LOW | test/runtime/test_input_pad_tail_tape.py | 29 | CODE | |
| LOW | test/runtime/test_batch_log.py | 23 | CODE | |
| LOW | test/runtime/test_fp8_linear_blockscale_flashinfer.py | 3 | CODE | |
| LOW | test/runtime/test_cu_extend_seq_lens_cpu_metadata.py | 31 | CODE | |
| LOW | test/runtime/test_qwen4_exp.py | 21 | CODE | |
| LOW | test/runtime/test_fp8_linear_mxfp8.py | 3 | CODE | |
| LOW | test/runtime/cache_pool_test_utils.py | 1 | CODE | |
| LOW | test/runtime/test_generation_output_processor.py | 21 | CODE | |
| LOW | test/runtime/test_glm53_flash_config.py | 3 | CODE | |
| LOW | test/runtime/test_kda_state_eviction_spec.py | 28 | CODE | |
| LOW | test/runtime/test_scheduler_config_guard.py | 3 | CODE | |
| LOW | test/runtime/test_deepseek_v4_slot_mappings.py | 28 | CODE | |
| LOW | test/runtime/test_retraction_image_executor.py | 33 | CODE | |
| LOW | test/runtime/test_kimi_k3_cache_pool.py | 1 | CODE | |
| LOW | test/runtime/test_unified_kv_slab_pool.py | 3 | CODE | |
| LOW | test/runtime/test_unified_kv_slab_pool.py | 469 | CODE | |
| LOW | test/runtime/test_unified_kv_slab_pool.py | 471 | CODE | |
| LOW | test/runtime/test_expert_location.py | 23 | CODE | |
| LOW | test/runtime/test_cache_group_router.py | 8 | CODE | |
| LOW | test/runtime/test_cupti_warmup_vendor_guard.py | 41 | CODE | |
| LOW | test/runtime/test_trtllm_mla_block_decode.py | 1 | CODE | |
| LOW | test/runtime/test_multimodal_encoded_offload.py | 22 | CODE | |
| 1834 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | …tic_benchmark/minimax_m3/tokenspeed/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …gentic_benchmark/kimi_k3/tokenspeed/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …gentic_benchmark/inkling/tokenspeed/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …chmark/deepseek_v4_flash/tokenspeed/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …st/agentic_benchmark/kimi_k2.5/vllm/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …ntic_benchmark/kimi_k2.5/tokenspeed/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …/agentic_benchmark/kimi_k2.5/sglang/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …/agentic_benchmark/kimi_k2.5/trtllm/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …agentic_benchmark/glm5.2/tokenspeed/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | test/agentic_benchmark/glm5.2/trtllm/collect_outputs.py | 0 | collect a sweep's per-config benchmark_summary.json files into one csv. | STRING |
| HIGH | …tic_benchmark/minimax_m3/tokenspeed/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …gentic_benchmark/kimi_k3/tokenspeed/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …gentic_benchmark/inkling/tokenspeed/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …chmark/deepseek_v4_flash/tokenspeed/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …st/agentic_benchmark/kimi_k2.5/vllm/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …ntic_benchmark/kimi_k2.5/tokenspeed/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …/agentic_benchmark/kimi_k2.5/sglang/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …/agentic_benchmark/kimi_k2.5/trtllm/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …agentic_benchmark/glm5.2/tokenspeed/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | test/agentic_benchmark/glm5.2/trtllm/collect_outputs.py | 0 | iteration-weighted accept length: sum(completion-1) / sum(n_chunks-1). the "decoded tok/iter" in benchmark_summary.json | STRING |
| HIGH | …time/layers/attention/backends/paged/tokenspeed_mla.py | 0 | per-bs decode metadata views over the persistent buffers. one builder for capture and refresh; cached per bs — pointer-s | STRING |
| HIGH | …/runtime/layers/attention/backends/paged/trtllm_mla.py | 0 | per-bs decode metadata views over the persistent buffers. one builder for capture and refresh; cached per bs — pointer-s | STRING |
| HIGH | …enspeed/runtime/layers/attention/backends/paged/mla.py | 0 | per-bs decode metadata views over the persistent buffers. one builder for capture and refresh; cached per bs — pointer-s | STRING |
| HIGH | …st/nvidia/ops/moe/test_mxfp4_situ_flashinfer_trtllm.py | 0 | minimal module carrying what the preprocessor and apply consume. | STRING |
| HIGH | …st/nvidia/ops/moe/test_nvfp4_situ_flashinfer_trtllm.py | 0 | minimal module carrying what the preprocessor and apply consume. | STRING |
| HIGH | …st/nvidia/ops/moe/test_bf16_relu2_flashinfer_trtllm.py | 0 | minimal module carrying what the preprocessor and apply consume. | STRING |
| HIGH | …t/nvidia/ops/moe/test_nvfp4_relu2_flashinfer_trtllm.py | 0 | minimal module carrying what the preprocessor and apply consume. | STRING |
| HIGH | …d-kernel/test/nvidia/thirdparty/bench_mnnvl_vs_nccl.py | 0 | true when the world spans hosts. cross-node cuda-ipc workspace creation fails and poisons the cuda context ('invalid res | STRING |
| HIGH | …ernel/test/nvidia/thirdparty/test_trtllm_mnnvl_comm.py | 0 | true when the world spans hosts. cross-node cuda-ipc workspace creation fails and poisons the cuda context ('invalid res | STRING |
| HIGH | …el/test/nvidia/thirdparty/test_trtllm_mnnvl_twoshot.py | 0 | true when the world spans hosts. cross-node cuda-ipc workspace creation fails and poisons the cuda context ('invalid res | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/kda/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/mha/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/qsa/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/msa/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …ython/tokenspeed_kernel/ops/attention/rmha/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/dsa/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/mla/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …ython/tokenspeed_kernel/ops/attention/dsv4/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/gdn/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …thon/tokenspeed_kernel/ops/attention/kpool/__init__.py | 0 | pick dense vs mxfp8 signature and build the scale kwargs splat. q_scale selects the block-scaled path; k_scale/v_scale m | STRING |
| HIGH | …l/ops/attention/rmha/_cute_dsl/flash_fwd_sm100_bias.py | 0 | set up configurations and parameters for the mla kernel operation. this method initializes and configures various attrib | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/fmha.py | 0 | set up configurations and parameters for the mla kernel operation. this method initializes and configures various attrib | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | set up configurations and parameters for the mla kernel operation. this method initializes and configures various attrib | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp8.py | 0 | the device split_kv kernel implementation of the multi-head latent attention. this kernel coordinates multiple specializ | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | the device split_kv kernel implementation of the multi-head latent attention. this kernel coordinates multiple specializ | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp16.py | 0 | the device split_kv kernel implementation of the multi-head latent attention. this kernel coordinates multiple specializ | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp8.py | 0 | get the current k_index, k_tile_count, and local split_kv value for the mla kernel. :param split_kv: split_kv value :typ | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | get the current k_index, k_tile_count, and local split_kv value for the mla kernel. :param split_kv: split_kv value :typ | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp16.py | 0 | get the current k_index, k_tile_count, and local split_kv value for the mla kernel. :param split_kv: split_kv value :typ | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp8.py | 0 | compute one k-tile of mma for q*k^t. updates the tiled mma qk and pipeline states. :param qk_params: the qk parameters : | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | compute one k-tile of mma for q*k^t. updates the tiled mma qk and pipeline states. :param qk_params: the qk parameters : | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp16.py | 0 | compute one k-tile of mma for q*k^t. updates the tiled mma qk and pipeline states. :param qk_params: the qk parameters : | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp8.py | 0 | compute one k-tile of mma for p*v. updates the tiled mma pv and pipeline states. :param common_params: the common parame | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | compute one k-tile of mma for p*v. updates the tiled mma pv and pipeline states. :param common_params: the common parame | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp16.py | 0 | compute one k-tile of mma for p*v. updates the tiled mma pv and pipeline states. :param common_params: the common parame | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp8.py | 0 | compute warp to compute the result of softmax, rescale, and epilogue. updates the related pipeline states. :param common | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | compute warp to compute the result of softmax, rescale, and epilogue. updates the related pipeline states. :param common | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp16.py | 0 | compute warp to compute the result of softmax, rescale, and epilogue. updates the related pipeline states. :param common | STRING |
| HIGH | tokenspeed-mla/python/tokenspeed_mla/mla_decode_fp8.py | 0 | compute warp to compute the result of softmax, rescale, and epilogue. updates the related pipeline states. :param common | STRING |
| HIGH | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 0 | compute warp to compute the result of softmax, rescale, and epilogue. updates the related pipeline states. :param common | STRING |
| 41 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/_triton.py | 41 | __all__ = ["libdevice", "proton", "tl", "triton"] | CODE |
| LOW | …rnel-npu/python/tokenspeed_kernel_npu/ops/layernorm.py | 79 | __all__ = ["qk_rmsnorm", "rmsnorm"] | CODE |
| LOW | …u/python/tokenspeed_kernel_npu/ops/rotary_embedding.py | 54 | __all__ = ["apply_rope"] | CODE |
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/ops/mha.py | 201 | __all__ = [ | CODE |
| LOW | test/runtime/cache_pool_test_utils.py | 111 | def set_kv_buffer(self, layer, loc, cache_k, cache_v) -> None: | CODE |
| LOW | test/runtime/test_event_loop_pause_hooks.py | 57 | def set_finish_with_abort(self, message: str, notify_client: bool = False) -> None: | CODE |
| LOW | test/runtime/test_epd_prefill_hooks.py | 51 | def set_finish_with_abort(self, message: str) -> None: | CODE |
| LOW | test/runtime/test_logits_processor.py | 199 | def _set_fabric(monkeypatch, supported: bool) -> None: | CODE |
| LOW | test/runtime/distributed/test_comm_ops.py | 49 | def _set_fabric(monkeypatch, supported: bool) -> None: | CODE |
| LOW | test/ci_system/ci_utils.py | 20 | logger = logging.getLogger(__name__) | CODE |
| LOW | test/ci_system/ci_register.py | 23 | __all__ = [ | CODE |
| LOW | python/tokenspeed/runtime/pd/kv_events.py | 43 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/pd/cache_protocol.py | 646 | __all__ = [ | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/pack.py | 41 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1293 | __all__ = ["MooncakeKVManagerPrefill", "MooncakeKVSender"] | CODE |
| LOW | python/tokenspeed/runtime/pd/base/mooncake_engine.py | 25 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/epd/encode_executor.py | 39 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/epd/prefill_admission.py | 58 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/epd/prefill_hooks.py | 38 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/layers/hyperconnection.py | 308 | __all__ = [ | CODE |
| LOW | python/tokenspeed/runtime/layers/rotary_embedding.py | 36 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/layers/parameter.py | 35 | __all__ = [ | CODE |
| LOW | python/tokenspeed/runtime/layers/qwen4_exp_ple.py | 1079 | __all__ = [ | CODE |
| LOW | …speed/runtime/layers/attention/mm_encoder_attention.py | 55 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/layers/attention/registry.py | 81 | logger = logging.getLogger(__name__) | CODE |
| LOW | …on/tokenspeed/runtime/layers/attention/qsa/__init__.py | 34 | __all__ = [ | CODE |
| LOW | …okenspeed/runtime/layers/attention/qsa/verify_state.py | 218 | __all__ = ["QSAVerifyState"] | CODE |
| LOW | …hon/tokenspeed/runtime/layers/attention/qsa/indexer.py | 525 | __all__ = ["QSAIndexer"] | CODE |
| LOW⚡ | …n/tokenspeed/runtime/layers/attention/backends/base.py | 473 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …n/tokenspeed/runtime/layers/attention/backends/base.py | 133 | def set_cache_pool(self, cache_pool: CachePool) -> None: | CODE |
| LOW⚡ | …/runtime/layers/attention/backends/specific/inkling.py | 717 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …/runtime/layers/attention/backends/specific/inkling.py | 81 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | …time/layers/attention/backends/specific/qsa_indexer.py | 324 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW⚡ | …untime/layers/attention/backends/specific/qwen4_exp.py | 257 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …untime/layers/attention/backends/specific/qwen4_exp.py | 285 | __all__ = ["Qwen4ExpBackend", "qwen4_exp_backend"] | CODE |
| LOW | …enspeed/runtime/layers/attention/backends/paged/msa.py | 65 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | …enspeed/runtime/layers/attention/backends/paged/msa.py | 500 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …runtime/layers/attention/backends/paged/tree_verify.py | 36 | __all__ = ["TreeCascadeRows", "TreeDraftInputs", "TreeVerifyInputs"] | CODE |
| LOW | …runtime/layers/attention/backends/paged/tree_verify.py | 84 | def set_frontier(self, bs: int, frontier: torch.Tensor) -> None: | CODE |
| LOW | …enspeed/runtime/layers/attention/backends/paged/dsa.py | 345 | def set_request_slots(self, req_pool_indices: torch.Tensor) -> None: | CODE |
| LOW | …enspeed/runtime/layers/attention/backends/paged/dsa.py | 532 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …time/layers/attention/backends/paged/tokenspeed_mla.py | 88 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | …peed/runtime/layers/attention/backends/paged/router.py | 872 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW⚡ | …nspeed/runtime/layers/attention/backends/paged/base.py | 405 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …nspeed/runtime/layers/attention/backends/paged/base.py | 444 | def set_request_slots(self, req_pool_indices: torch.Tensor) -> None: | CODE |
| LOW | …speed/runtime/layers/attention/backends/state/mamba.py | 97 | logger = logging.getLogger(__name__) | CODE |
| LOW | …speed/runtime/layers/attention/backends/state/mamba.py | 505 | def set_kv_pool(self, kv_pool: CachePool) -> None: | CODE |
| LOW⚡ | …eed/runtime/layers/attention/backends/hybrid/linear.py | 107 | def update_draft_forward_metadata(self, frontier: torch.Tensor) -> None: | CODE |
| LOW | …enspeed/runtime/layers/attention/kv_cache/qwen4_exp.py | 66 | __all__ = [ | CODE |
| LOW | …untime/layers/attention/kv_cache/hybrid_deepseek_v4.py | 616 | def set_kv_buffer(self, *args, **kwargs) -> None: | CODE |
| LOW | …/layers/attention/kv_cache/recipes/scheduler_bridge.py | 197 | __all__ = [ | CODE |
| LOW | …peed/runtime/layers/attention/kv_cache/recipes/spec.py | 663 | __all__ = [ | CODE |
| LOW | …/runtime/layers/attention/kv_cache/recipes/transfer.py | 228 | __all__ = [ | CODE |
| LOW | …runtime/layers/attention/kv_cache/recipes/qwen4_exp.py | 314 | __all__ = ["Qwen4ExpRecipe"] | CODE |
| LOW | python/tokenspeed/runtime/layers/dense/__init__.py | 29 | __all__ = [ | CODE |
| LOW | python/tokenspeed/runtime/layers/dense/nvfp4.py | 36 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/layers/dense/mxfp4.py | 151 | __all__ = [ | CODE |
| LOW | …kenspeed/runtime/layers/quantization/modelopt_mixed.py | 64 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/layers/quantization/fp8.py | 37 | logger = logging.getLogger(__name__) | CODE |
| LOW | python/tokenspeed/runtime/layers/quantization/nvfp4.py | 30 | logger = logging.getLogger(__name__) | CODE |
| 435 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/test_utils.py | 102 | CODE | |
| LOW | test/test_utils.py | 199 | CODE | |
| LOW | test/test_utils.py | 299 | CODE | |
| LOW | …t/long_context_benchmark/tokenspeed/collect_outputs.py | 39 | CODE | |
| LOW | test/runtime/test_unified_decode_path.py | 563 | CODE | |
| LOW | test/runtime/test_weight_loader_prefetch.py | 198 | CODE | |
| LOW | test/runtime/test_weight_loader_prefetch.py | 299 | CODE | |
| LOW | test/runtime/test_instanttensor_loader.py | 130 | CODE | |
| LOW | test/runtime/test_cudagraph_probe_ladder.py | 313 | CODE | |
| LOW | test/runtime/test_deepseek_v4_config.py | 6985 | CODE | |
| LOW | test/runtime/test_kda_state_eviction.py | 199 | CODE | |
| LOW | test/runtime/test_cli_config_compat.py | 415 | CODE | |
| LOW | test/runtime/test_tree_attention_cascade.py | 110 | CODE | |
| LOW | test/runtime/cache_pool_test_utils.py | 361 | CODE | |
| LOW | test/runtime/test_kda_state_eviction_spec.py | 251 | CODE | |
| LOW | test/runtime/test_flashmla_dcp.py | 59 | CODE | |
| LOW | test/runtime/test_drafter_accept_indexing.py | 593 | CODE | |
| LOW | test/runtime/test_expert_location.py | 92 | CODE | |
| LOW | test/runtime/test_inkling_real_checkpoint_load.py | 60 | CODE | |
| LOW | test/runtime/test_tree_spec.py | 93 | CODE | |
| LOW | test/runtime/test_tree_spec.py | 138 | CODE | |
| LOW | test/runtime/test_tree_spec.py | 180 | CODE | |
| LOW | test/runtime/test_kimi_k3_dspark_model.py | 321 | CODE | |
| LOW | test/runtime/test_dp_rank_pinning.py | 164 | CODE | |
| LOW | test/runtime/test_kimi_k3_config.py | 834 | CODE | |
| LOW | test/runtime/test_gdn_state_paging.py | 397 | CODE | |
| LOW | test/runtime/test_gdn_state_paging.py | 566 | CODE | |
| LOW | test/runtime/test_gdn_state_paging.py | 1134 | CODE | |
| LOW | test/runtime/test_gdn_state_paging.py | 1812 | CODE | |
| LOW | test/runtime/test_deepseek_v41_model.py | 209 | CODE | |
| LOW | test/runtime/test_deepseek_v41_model.py | 405 | CODE | |
| LOW | test/runtime/test_deepseek_v41_model.py | 2381 | CODE | |
| LOW | test/runtime/test_v4_sliding_window_groups_smoke.py | 121 | CODE | |
| LOW | test/runtime/test_v4_sliding_window_groups_smoke.py | 183 | CODE | |
| LOW | test/runtime/test_v4_sliding_window_groups_smoke.py | 240 | CODE | |
| LOW | test/runtime/test_inkling_load_weights.py | 84 | CODE | |
| LOW | test/runtime/test_slot_state.py | 690 | CODE | |
| LOW | test/runtime/test_dflash2.py | 336 | CODE | |
| LOW | test/runtime/run_deepseek_v41_eval.py | 54 | CODE | |
| LOW | test/runtime/run_deepseek_v41_eval.py | 130 | CODE | |
| LOW | test/runtime/test_pp_mapping.py | 266 | CODE | |
| LOW | test/runtime/test_kimi_k3_attn_res.py | 741 | CODE | |
| LOW | test/runtime/test_dispatch_autotune.py | 50 | CODE | |
| LOW | test/runtime/test_retraction_restore_driver.py | 158 | CODE | |
| LOW | test/runtime/test_prefill_graph.py | 97 | CODE | |
| LOW | test/runtime/test_prefill_graph.py | 226 | CODE | |
| LOW | test/runtime/test_paged_attention_prologue.py | 524 | CODE | |
| LOW | test/runtime/test_qsa_backend.py | 365 | CODE | |
| LOW | test/runtime/test_control_server.py | 180 | CODE | |
| LOW | test/runtime/test_logits_processor.py | 1073 | CODE | |
| LOW | test/runtime/test_draft_tree.py | 39 | CODE | |
| LOW | test/runtime/test_deepseek_v41_engram.py | 154 | CODE | |
| LOW | test/runtime/distributed/test_snapshot_op_ownership.py | 193 | CODE | |
| LOW | …ntime/distributed/validate_kimi_k3_shared_expert_tp.py | 50 | CODE | |
| LOW | test/runtime/distributed/test_kimi_k3_projection_tp.py | 79 | CODE | |
| LOW | test/runtime/distributed/test_kimi_k3_projection_tp.py | 203 | CODE | |
| LOW | test/runtime/distributed/test_comm_ops.py | 521 | CODE | |
| LOW | test/runtime/models/test_moe_routing_config.py | 17 | CODE | |
| LOW | test/runtime/models/test_kimi_k3_eagle3_e2e.py | 55 | CODE | |
| LOW | test/ci_system/slurm_submit.py | 876 | CODE | |
| 405 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | test/test_utils.py | 127 | except requests.RequestException: | CODE |
| HIGH | …t/long_context_benchmark/tokenspeed/collect_outputs.py | 63 | except (TypeError, ValueError): | CODE |
| HIGH | …t/long_context_benchmark/tokenspeed/collect_outputs.py | 53 | except (TypeError, ValueError): | CODE |
| HIGH | test/runtime/run_deepseek_v41_eval.py | 223 | except (URLError, TimeoutError): | CODE |
| HIGH | test/runtime/run_deepseek_v41_eval.py | 327 | except psutil.NoSuchProcess: | CODE |
| HIGH | test/runtime/run_deepseek_v41_eval.py | 333 | except psutil.NoSuchProcess: | CODE |
| HIGH | test/runtime/distributed/test_deepseek_v41_pd_1p1d.py | 90 | except Exception: | CODE |
| HIGH | test/runtime/distributed/test_qwen35_epd_1e1p2d.py | 142 | except Exception: | CODE |
| HIGH | test/runtime/distributed/test_qwen35_epd_1e1p2d.py | 77 | except Exception: | CODE |
| HIGH | test/runtime/distributed/test_qwen35_pd_1p1d.py | 77 | except Exception: | CODE |
| HIGH | test/runtime/models/test_kimi_models.py | 103 | except Exception: | CODE |
| HIGH | test/runtime/models/test_qwen35_vlm_e2e.py | 116 | except Exception: | CODE |
| HIGH | test/runtime/models/test_qwen35_vlm_e2e.py | 191 | except subprocess.TimeoutExpired: | CODE |
| HIGH | test/runtime/models/test_kimi_k3_eagle3_e2e.py | 50 | except requests.RequestException: | CODE |
| HIGH | test/runtime/models/test_llama_models.py | 97 | except Exception: | CODE |
| HIGH | test/runtime/models/test_gpt_oss.py | 82 | except Exception: | CODE |
| HIGH | test/ci_system/process_group_manager.py | 87 | except (ProcessLookupError, PermissionError): | CODE |
| HIGH | test/ci_system/pipeline.py | 1703 | except subprocess.TimeoutExpired: | CODE |
| HIGH | test/ci_system/pipeline.py | 1656 | except (URLError, ConnectionError, TimeoutError): | CODE |
| HIGH | test/cli/test_serve_smg_unit.py | 1100 | except SystemExit: | CODE |
| HIGH | test/cli/test_serve_smg_integration.py | 55 | except (urllib.error.URLError, ConnectionError, TimeoutError): | CODE |
| HIGH | python/tokenspeed/_logging.py | 109 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1048 | except (IndexError, UnicodeError, ValueError): | CODE |
| HIGH | python/tokenspeed/runtime/epd/prefill_admission.py | 111 | except Exception: # noqa: BLE001 -- best-effort; worst case the MR leaks | CODE |
| HIGH | python/tokenspeed/runtime/cache/l2/executor.py | 1483 | except BaseException: | CODE |
| HIGH | …tokenspeed/runtime/utils/torch_memory_saver_adapter.py | 28 | except ImportError: | CODE |
| HIGH | python/tokenspeed/runtime/utils/custom_ops.py | 49 | except (AttributeError, RuntimeError): | CODE |
| HIGH | python/tokenspeed/runtime/utils/startup_timing.py | 56 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/common.py | 1189 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/network.py | 76 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/network.py | 84 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/network.py | 107 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/network.py | 114 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/network.py | 121 | except Exception: | CODE |
| HIGH | python/tokenspeed/runtime/utils/process.py | 66 | except psutil.NoSuchProcess: | CODE |
| HIGH | python/tokenspeed/runtime/utils/process.py | 77 | except psutil.NoSuchProcess: | CODE |
| HIGH | python/tokenspeed/runtime/models/dflash.py | 565 | except TypeError: | CODE |
| HIGH | python/tokenspeed/runtime/models/deepseek_v4_next.py | 97 | except ValueError: | CODE |
| HIGH | python/tokenspeed/runtime/models/deepseek_v41_engram.py | 441 | except FileNotFoundError: | CODE |
| HIGH | python/tokenspeed/runtime/models/deepseek_v41_engram.py | 445 | except ValueError: | CODE |
| HIGH | python/tokenspeed/runtime/models/deepseek_v41_engram.py | 405 | except FileNotFoundError: | CODE |
| HIGH | python/tokenspeed/runtime/multimodal/shm_transport.py | 273 | except FileNotFoundError: | CODE |
| HIGH | python/tokenspeed/runtime/multimodal/shm_transport.py | 270 | except FileNotFoundError: | CODE |
| HIGH | python/tokenspeed/runtime/execution/drafter/__init__.py | 127 | except ValueError: | CODE |
| HIGH | python/tokenspeed/runtime/engine/async_llm.py | 506 | except StopAsyncIteration: | CODE |
| HIGH | python/tokenspeed/cli/serve_smg.py | 814 | except ImportError: | CODE |
| HIGH | python/tokenspeed/cli/serve_smg.py | 837 | except ImportError: | CODE |
| HIGH | python/tokenspeed/cli/serve_smg.py | 680 | except NotImplementedError: | CODE |
| HIGH | python/tokenspeed/cli/serve_smg.py | 798 | except Exception: # noqa: BLE001 | CODE |
| HIGH | tokenspeed-kernel/test/test_kimi_prefill_ops.py | 94 | except Exception: | CODE |
| HIGH | tokenspeed-kernel/python/setup.py | 197 | except AttributeError: | CODE |
| HIGH | tokenspeed-kernel/python/setup.py | 184 | except OSError: | CODE |
| HIGH | tokenspeed-kernel/python/setup.py | 572 | except Exception: | CODE |
| HIGH | tokenspeed-kernel/python/setup.py | 702 | except ImportError: | CODE |
| HIGH | tokenspeed-kernel/python/setup.py | 723 | except ImportError: | CODE |
| HIGH | tokenspeed-kernel/python/tokenspeed_kernel/platform.py | 55 | except Exception: | CODE |
| HIGH | tokenspeed-kernel/python/tokenspeed_kernel/platform.py | 659 | except Exception: | CODE |
| HIGH | tokenspeed-kernel/python/tokenspeed_kernel/_triton.py | 148 | except ValueError: | CODE |
| HIGH | …kenspeed-kernel/python/tokenspeed_kernel/ops/tuning.py | 361 | except FileNotFoundError: | CODE |
| HIGH | …thon/tokenspeed_kernel/ops/attention/dsa/flashinfer.py | 57 | except ImportError: | CODE |
| 18 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | test/ci/deepswe/kimi_code_pier_agent.py | 125 | "/logs/agent/kimi-code.log 2>/dev/null || true' EXIT; " | CODE |
| HIGH | test/runtime/test_host_executor.py | 1708 | executor._completions.push( | CODE |
| HIGH | test/runtime/test_host_executor.py | 1760 | executor._completions.push( | CODE |
| HIGH | test/runtime/test_host_executor.py | 1831 | executor._completions.push( | CODE |
| HIGH | test/runtime/test_inkling_mtp_conv_state.py | 204 | over [committed history || accepted chunk prefix].""" | STRING |
| HIGH | test/runtime/test_inkling_mtp_conv_state.py | 349 | """Per-request reference: causal conv over [prefix || x].""" | STRING |
| HIGH | test/runtime/test_inkling_mtp_conv_state.py | 361 | the chunk): rows of [prefix || x[:upto]].""" | STRING |
| HIGH | test/runtime/test_inkling_mtp_conv_state.py | 380 | conv over [ring history || x_t]; the kernel persists the token's own | STRING |
| HIGH⚡ | test/runtime/test_unit_kv_scales.py | 137 | '{"kv_cache": {"scaling_factor": {"0": {"0": null}}}}', | CODE |
| HIGH⚡ | test/runtime/test_unit_kv_scales.py | 139 | '{"kv_cache": null}', | CODE |
| HIGH⚡ | test/runtime/test_cli_config_compat.py | 377 | ["--model", "test/model", "--hf-overrides", '{"rope_scaling": null}'] | CODE |
| HIGH⚡ | test/runtime/test_cli_config_compat.py | 379 | self.assertEqual(args.hf_overrides, '{"rope_scaling": null}') | CODE |
| HIGH⚡ | test/runtime/test_retraction_image_executor.py | 1029 | queue.push(first, "first") | CODE |
| HIGH⚡ | test/runtime/test_retraction_image_executor.py | 1030 | queue.push(second, "second") | CODE |
| HIGH⚡ | test/runtime/test_retraction_image_executor.py | 1035 | queue.push(Mock(), "stale") | CODE |
| HIGH⚡ | test/runtime/test_draft_page_table_scrub.py | 103 | # Rows 1..3 are padding: their pages are null, so slots are 0. | COMMENT |
| HIGH | test/runtime/test_gdn_state_paging.py | 1571 | # Page 0 is null; pages 1..N fill as the sequence grows. | COMMENT |
| HIGH | test/runtime/test_kimi_k3_kda.py | 216 | conv/recurrent component slabs indexed by page id (page 0 = null).""" | STRING |
| HIGH | test/runtime/distributed/test_decode_tp_layouts.py | 558 | ``fused_qkv_a_proj_with_mqa`` takes ``[embeds || hidden]``).""" | STRING |
| HIGH | test/ci_system/test_dispatch_workflows.py | 457 | "${{ inputs.commit || inputs.pr || 'main' }}" | CODE |
| HIGH | test/ci_system/test_dispatch_workflows.py | 1060 | " && 30 || inputs.timeout_minutes }}" | CODE |
| HIGH | test/ci_system/test_dispatch_workflows.py | 1182 | "${{ (inputs.pr || inputs.commit) && '1' || '0' }}" | CODE |
| HIGH | test/ci_system/slurm_submit.py | 511 | nvidia_smi="$(command -v nvidia-smi 2>/dev/null || true)" | CODE |
| HIGH⚡ | test/ci_system/slurm_submit.py | 636 | if [ -n "$pid" ] && kill -0 "$pid" 2>/dev/null; then | CODE |
| HIGH⚡ | test/ci_system/slurm_submit.py | 637 | kill -TERM "$pid" 2>/dev/null || true | CODE |
| HIGH⚡ | test/ci_system/slurm_submit.py | 640 | [ -z "$client_step_pid" ] || wait "$client_step_pid" 2>/dev/null || true | CODE |
| HIGH⚡ | test/ci_system/slurm_submit.py | 641 | [ -z "$server_step_pid" ] || wait "$server_step_pid" 2>/dev/null || true | CODE |
| HIGH | test/ci_system/slurm_submit.py | 674 | kill -0 "$client_step_pid" 2>/dev/null; do | CODE |
| HIGH | test/ci_system/slurm_submit.py | 679 | if ! kill -0 "$server_step_pid" 2>/dev/null; then | CODE |
| HIGH | test/ci_system/test_package_cache.py | 53 | 'configure_package_cache >/dev/null; printf "%s|%s|%s" "${PIP_CACHE_DIR}" "${CI_WHEEL_CACHE_DIR}" "${CI_CCACHE_D | CODE |
| HIGH | test/ci_system/test_package_cache.py | 76 | 'configure_package_cache >/dev/null; printf "%s|%s|%s" "${PIP_CACHE_DIR}" "${CI_WHEEL_CACHE_DIR}" "${CI_CCACHE_D | CODE |
| HIGH | test/ci_system/test_package_cache.py | 121 | "configure_nvcc_cache >/dev/null; " | CODE |
| HIGH | test/ci_system/test_package_cache.py | 172 | "configure_nvcc_cache >/dev/null 2>/dev/null; " | CODE |
| HIGH⚡ | test/ci_system/pipeline.py | 684 | "sudo apt-get update -q && sudo apt-get install -y lsof; " | CODE |
| HIGH⚡ | test/ci_system/pipeline.py | 686 | f"pids=$(lsof -tiTCP:{port} -sTCP:LISTEN 2>/dev/null || true); " | CODE |
| HIGH⚡ | test/ci_system/pipeline.py | 689 | "kill -TERM $pids 2>/dev/null || true; " | CODE |
| HIGH⚡ | test/ci_system/pipeline.py | 691 | f"pids=$(lsof -tiTCP:{port} -sTCP:LISTEN 2>/dev/null || true); " | CODE |
| HIGH⚡ | test/ci_system/pipeline.py | 692 | 'if [ -n "$pids" ]; then kill -KILL $pids 2>/dev/null || true; fi; ' | CODE |
| HIGH | test/ci_system/pipeline.py | 642 | f'pkill -{signal_name} -f "{pattern}" 2>/dev/null || true', | CODE |
| HIGH | test/ci_system/test_worker_cleanup.py | 14 | if kill -0 "$worker" 2>/dev/null; then | CODE |
| HIGH | python/tokenspeed/runtime/cache/l2/executor.py | 1131 | self._completions.push( | CODE |
| HIGH | python/tokenspeed/runtime/cache/l2/executor.py | 1199 | self._completions.push( | CODE |
| HIGH | python/tokenspeed/runtime/cache/l2/executor.py | 1302 | self._completions.push( | CODE |
| HIGH | python/tokenspeed/runtime/cache/l2/executor.py | 1447 | self._completions.push( | CODE |
| HIGH | python/tokenspeed/runtime/utils/server_args.py | 3424 | help="Specify tp size for MoE part, default equals nprocs-per-node, if non dp_attn && combine_dense mode, th | CODE |
| HIGH⚡ | python/tokenspeed/runtime/models/llama_eagle3.py | 75 | the [embed || hidden] concat) and ``forward`` (= qkv_proj + o_proj | STRING |
| HIGH | python/tokenspeed/runtime/models/deepseek_v3.py | 2665 | concatenated [embeds || hidden_states] as input. | STRING |
| HIGH | python/tokenspeed/runtime/execution/drafter/mtp.py | 169 | prefix). Deeper depths are rolls of [window || drafts], built by the | STRING |
| HIGH | python/tokenspeed/runtime/execution/drafter/mtp.py | 197 | rows the stash. Over [stash || fresh] both cases are one gather at | STRING |
| HIGH | python/tokenspeed/runtime/execution/drafter/eagle.py | 290 | ``[prefill_part || decode_part]`` under MIXED, full prefill chunks | STRING |
| HIGH | …n/tokenspeed/runtime/entrypoints/sglang_compat_http.py | 258 | "weight_version": str | null}``. The ``flush_cache`` wire default mirrors | STRING |
| HIGH | …peed-kernel/test/nvidia/ops/moe/test_multicast_view.py | 72 | """The deleter must be null, not a no-op callback. | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/kda/cute_dsl.py | 143 | # full-tile loads. Conv output padding is undefined, so scrub all scan | COMMENT |
| HIGH | …l/python/tokenspeed_kernel/ops/attention/gdn/triton.py | 205 | float32 legacy decode path; output for that batch entry is undefined). | STRING |
| HIGH | …thon/tokenspeed_kernel/ops/attention/dsv41/cute_dsl.py | 202 | null, past ``visible``, or on an unmapped page. | STRING |
| HIGH | …l/thirdparty/cute_dsl/nvfp4_gemm_swiglu_nvfp4_quant.py | 2558 | :param c_sf_ptr: Pointer to scale factor tensor for C (can be null) | STRING |
| HIGH | …l/thirdparty/cute_dsl/nvfp4_gemm_swiglu_nvfp4_quant.py | 2562 | :param norm_const_ptr: Pointer to normalization constant for SFC generation (can be null) | STRING |
| HIGH | …peed_kernel/thirdparty/cute_dsl/mega_moe/flag_batch.py | 24 | flag_addr: Int64 # per-lane counter-slot address (0 == null) | CODE |
| HIGH⚡ | …lib/moe/rocm/ops/mega_moe/token_shuffle_direct_push.py | 96 | tokens = MakeBufferResource(null, 0) | CODE |
| HIGH⚡ | …lib/moe/rocm/ops/mega_moe/token_shuffle_direct_push.py | 97 | ids = MakeBufferResource(null, 0) | CODE |
| 1 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/test_utils.py | 191 | except Exception as e: | CODE |
| MEDIUM | test/test_utils.py | 192 | print(f"Error killing process after launch failure: {e}") | CODE |
| LOW | test/runtime/test_inkling_load_weights.py | 554 | except Exception as e: # pragma: no cover - environment specific | CODE |
| MEDIUM | test/runtime/test_zmq_msgpack.py | 757 | def _engine(): | CODE |
| LOW | test/runtime/test_zmq_msgpack.py | 772 | except Exception as exc: # surface into the test thread | CODE |
| LOW | test/runtime/test_zmq_msgpack.py | 856 | except Exception as exc: | CODE |
| LOW | test/runtime/test_control_server.py | 107 | except Exception: | CODE |
| LOW | test/runtime/distributed/test_deepseek_v41_pd_1p1d.py | 90 | except Exception: | CODE |
| LOW | test/runtime/distributed/test_qwen35_epd_1e1p2d.py | 77 | except Exception: | CODE |
| LOW | test/runtime/distributed/test_qwen35_epd_1e1p2d.py | 142 | except Exception: | CODE |
| LOW | test/runtime/distributed/test_dp_sampling_swap.py | 41 | except Exception: | CODE |
| MEDIUM | test/runtime/distributed/test_dp_sampling_swap.py | 23 | def _worker_main(rank, world_size, port, test_fn, error_dict, args): | CODE |
| LOW | …runtime/distributed/test_batch_invariant_all_reduce.py | 45 | except Exception: | CODE |
| MEDIUM | …runtime/distributed/test_batch_invariant_all_reduce.py | 42 | def worker_fn(rank, world_size, port, attn_tp, error_dict): | CODE |
| LOW | …t/runtime/distributed/test_vocab_parallel_topk_nccl.py | 201 | except Exception: | CODE |
| MEDIUM | …t/runtime/distributed/test_vocab_parallel_topk_nccl.py | 183 | def _worker(rank, world_size, port, test_fn, error_dict): | CODE |
| MEDIUM | test/runtime/distributed/test_dp_sampling_comm.py | 29 | def _worker_main(rank, world_size, port, test_fn, error_dict, args): | CODE |
| MEDIUM | test/runtime/distributed/test_dp_sampling_comm.py | 66 | def _onesided_available_for_test(group) -> bool: | CODE |
| LOW⚡ | test/runtime/distributed/test_dp_sampling_comm.py | 69 | except Exception: | CODE |
| LOW | test/runtime/distributed/test_dp_sampling_comm.py | 47 | except Exception: | CODE |
| LOW⚡ | test/runtime/distributed/test_comm_manager.py | 81 | except Exception as e: | CODE |
| LOW | test/runtime/distributed/test_comm_manager.py | 310 | except Exception: | CODE |
| MEDIUM | test/runtime/distributed/test_comm_manager.py | 76 | def worker_fn( | CODE |
| MEDIUM | test/runtime/distributed/test_comm_manager.py | 307 | def _qcp_worker(rank, port, error_dict): | CODE |
| LOW | test/runtime/distributed/test_qcp_prompt_logprobs.py | 60 | except Exception: # pragma: no cover - reported to the parent | CODE |
| LOW | test/runtime/distributed/test_qwen35_pd_1p1d.py | 77 | except Exception: | CODE |
| LOW | test/runtime/distributed/test_qcp_history_gather.py | 133 | except Exception: # pragma: no cover - reported to the parent | CODE |
| LOW | test/runtime/distributed/test_qcp_history_gather.py | 413 | except Exception: # pragma: no cover - reported to the parent | CODE |
| MEDIUM | test/runtime/distributed/test_comm_ops.py | 337 | def worker_fn(rank, world_size, port, test_fn, error_dict): | CODE |
| LOW⚡ | test/runtime/distributed/test_comm_ops.py | 340 | except Exception: | CODE |
| LOW | test/runtime/sampling/test_flashinfer_verify_dp.py | 67 | except Exception: | CODE |
| MEDIUM | test/runtime/sampling/test_flashinfer_verify_dp.py | 49 | def _worker_main(rank, world_size, port, test_fn, error_dict, args): | CODE |
| LOW | test/runtime/sampling/test_dp_sampling_logits_verify.py | 80 | except Exception: | CODE |
| MEDIUM | test/runtime/sampling/test_dp_sampling_logits_verify.py | 62 | def _worker_main(rank, world_size, port, test_fn, error_dict, args): | CODE |
| LOW | test/runtime/models/test_kimi_models.py | 103 | except Exception: | CODE |
| LOW | test/runtime/models/test_qwen35_vlm_e2e.py | 116 | except Exception: | CODE |
| LOW | test/runtime/models/test_qwen35_vlm_e2e.py | 127 | except Exception as exc: # noqa: BLE001 | CODE |
| MEDIUM | test/runtime/models/test_qwen35_vlm_e2e.py | 124 | def setUpClass(cls): | CODE |
| LOW | test/runtime/models/test_llama_models.py | 97 | except Exception: | CODE |
| LOW | test/runtime/models/test_gpt_oss.py | 82 | except Exception: | CODE |
| LOW | test/ci_system/pipeline.py | 2048 | except Exception as exc: | CODE |
| LOW | test/ci_system/pipeline.py | 2078 | except Exception as exc: | CODE |
| LOW | test/ci_system/kernel_benchmark_ci.py | 1108 | except Exception as exc: # noqa: BLE001 - failures must become CI artifacts | CODE |
| LOW | …entic_benchmark/kimi_k3/tokenspeed_disagg/pd_client.py | 113 | except Exception as e: | CODE |
| LOW | python/tokenspeed/_logging.py | 77 | except Exception: | CODE |
| LOW | python/tokenspeed/_logging.py | 109 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/kv_events.py | 313 | except Exception as exc: | CODE |
| LOW | python/tokenspeed/runtime/pd/kv_events.py | 334 | except Exception as exc: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/pack.py | 172 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/receiver.py | 86 | except Exception as exc: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/receiver.py | 105 | except Exception as exc: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 909 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1021 | except Exception as exc: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1117 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1143 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1177 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/prefill.py | 1285 | except Exception as exc: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/decode.py | 148 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/mooncake/decode.py | 190 | except Exception: | CODE |
| LOW | python/tokenspeed/runtime/pd/base/mooncake_engine.py | 73 | except Exception: | CODE |
| 178 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | python/tokenspeed/runtime/layers/logits_processor.py | 762 | Logprob of each named prompt row's next token, position-chunked. The rows (``logits_metadata.input_logprob_rows | STRING |
| HIGH | …n/tokenspeed/runtime/layers/attention/dcp/placement.py | 100 | Count, on the host, how many history rows of each request every rank owns. The query-context-parallel history gathe | STRING |
| HIGH | …untime/layers/attention/backends/specific/qwen4_exp.py | 79 | Build the consumers whose fields belong to this cache view. Args: config: This side's attention con | STRING |
| HIGH | …speed/runtime/layers/attention/backends/state/mamba.py | 240 | Host prefix sum of the scheduler's extend lengths, as an int64 tensor. The contents must equal ``query_start_loc`` | STRING |
| HIGH | …speed/runtime/layers/attention/backends/state/utils.py | 27 | Return the physical state-row stride in int32 words for batched copies. Args: tensor: State rows with a non | STRING |
| HIGH | …ed/runtime/layers/attention/kv_cache/virtual_blocks.py | 48 | Translate scheduler blocks to owned local pages, keeping the owner mask. Args: virtual_blocks: Scheduler bl | STRING |
| HIGH | …n/tokenspeed/runtime/layers/attention/kv_cache/base.py | 65 | Map each recurrent layer to the state-family group holding its fields. The memory plan is the single record of whic | STRING |
| HIGH | …n/tokenspeed/runtime/layers/attention/kv_cache/base.py | 117 | Map each attention layer to the history-family group holding its KV. The counterpart of :func:`derive_state_groups_ | STRING |
| HIGH | …/tokenspeed/runtime/layers/attention/kv_cache/arena.py | 214 | Return one planned field's view into the arena. The view is shaped the way the field is addressed. A field whos | STRING |
| HIGH | …peed/runtime/layers/attention/kv_cache/recipes/plan.py | 64 | Return the plan-facing name of the dtype the arena views bytes as. Args: dtype: A ``torch.dtype`` or an alr | STRING |
| HIGH | …peed/runtime/layers/attention/kv_cache/recipes/plan.py | 99 | Return the byte width of one element of a plan dtype name. Args: dtype_name: A name produced by :func:`cach | STRING |
| HIGH | …peed/runtime/layers/attention/kv_cache/recipes/plan.py | 713 | Pack declared cache groups into one capacity-independent LCM block. The second stage of the pipeline: ``group`` say | STRING |
| HIGH | …peed/runtime/layers/attention/kv_cache/recipes/spec.py | 316 | Normalize either window spelling to one entry per layer. Args: sliding_window_tokens: A scalar window to br | STRING |
| HIGH | …peed/runtime/layers/attention/kv_cache/recipes/spec.py | 498 | Walk the layers once, building each group whole. A cache group has two halves -- the scheduler-facing spec and the | STRING |
| HIGH | …ime/layers/attention/kv_cache/recipes/cache_runtime.py | 39 | Validate that ``value`` is a positive, non-boolean integer. Args: name: Field name used in the error messag | STRING |
| HIGH | python/tokenspeed/runtime/cache/transfer/lanes.py | 104 | The static field geometry one executor's transfers share. One row per field in consumer (layer) order, each naming | STRING |
| HIGH | python/tokenspeed/runtime/distributed/pp_stage.py | 46 | All stages' [start, end) layer windows. The single source of the stage-split arithmetic: the model build, the KV | STRING |
| HIGH | python/tokenspeed/runtime/distributed/pp_stage.py | 101 | Map stage execution windows to the target cache-layer namespace. Stage windows partition execution blocks (decoder | STRING |
| HIGH | python/tokenspeed/runtime/distributed/comm_manager.py | 191 | Per-rank rows of the head group for one leg of the head-TP exchange. Head TP shards the head projections over ranks | STRING |
| HIGH | …enspeed/runtime/distributed/comm_backend/self_check.py | 112 | Verify the in-switch all-reduce on every group the route sends to it. Args: backend: The communication back | STRING |
| HIGH | python/tokenspeed/runtime/utils/spec_block_geometry.py | 87 | Derive the launch widths a checkpoint block size implies. Args: algorithm: ``"DSPARK"`` or ``"DFLASH"`` (DF | STRING |
| HIGH | python/tokenspeed/runtime/utils/launcher.py | 67 | Return the first hostname of a Slurm-compressed hostlist. Args: hostlist: A hostlist such as ``cn10``, ``cn | STRING |
| HIGH | python/tokenspeed/runtime/utils/launcher.py | 244 | Derive the multi-node topology from the launcher environment. Args: env: Environment mapping to read; defau | STRING |
| HIGH | python/tokenspeed/runtime/execution/device.py | 880 | Apply one in-place RL weight-sync request, ordered against forwards. Type-dispatched on the request — join the | STRING |
| HIGH | python/tokenspeed/runtime/execution/model_executor.py | 222 | Return the stage's DSpark context producer, or None when nothing is produced. A DSpark draft reads target taps that | STRING |
| HIGH | …thon/tokenspeed/runtime/execution/accept_simulation.py | 40 | Validate the simulated tokens per verify step. Args: value: Raw ``TOKENSPEED_SPEC_SIMULATED_ACCEPT_LEN``; e | STRING |
| HIGH | python/tokenspeed/runtime/execution/query_shard.py | 61 | How many of the sorted batch-global ``rows`` fall in each rank's shard. The one row split of a query shard: the sam | STRING |
| HIGH | python/tokenspeed/runtime/execution/query_shard.py | 240 | How many of ``rows`` fall in each rank's shard. The split a collective over those rows needs (the prompt-logpro | STRING |
| HIGH | python/tokenspeed/runtime/execution/query_shard.py | 269 | Per-rank row counts of the rows a collective moves. A model that reports a collective sizing (``report_collecti | STRING |
| HIGH | python/tokenspeed/runtime/execution/drafter/dflash.py | 72 | Resolve (verify_width, draft_block_size) for a block drafter. Args: cfg: The draft model's config. | STRING |
| HIGH | python/tokenspeed/runtime/engine/logprobs.py | 48 | Resolve an SGLang ``logprob_start_len`` against a prompt of ``input_len``. SGLang semantics: ``-1`` (or omitted) me | STRING |
| HIGH | python/tokenspeed/runtime/engine/scheduler_utils.py | 290 | Floor ``max_scheduled_tokens`` to the state-snapshot grain, if any. Recurrent-state groups (family=State, the C++ ` | STRING |
| HIGH | python/tokenspeed/runtime/engine/scheduler_utils.py | 959 | Bridge the per-group block tables to GPU int32 tensors: absolute page indices, null hole = 0 preserved, ragged-row p | STRING |
| HIGH | python/tokenspeed/runtime/engine/zmq_wire.py | 120 | Map a tokenspeed/torch dtype onto SMG's wire dtype enum string. Args: dtype: A torch dtype or its string fo | STRING |
| HIGH | python/tokenspeed/cli/trace_merge.py | 247 | Merge a Proton chrome trace into a VizTracer report. Shifts every Proton event by the difference between the two fi | STRING |
| HIGH | python/tokenspeed/cli/trace_merge.py | 283 | Merge and align Proton/VizTracer trace pairs from several TP ranks. Every input pair is first aligned using its own | STRING |
| HIGH | …ernel/python/tokenspeed_kernel/benchmark/validation.py | 92 | Return the output validator registered under ``name``. Args: name: Validator name from an output validation | STRING |
| HIGH | …peed_kernel/ops/attention/rmha/_cute_dsl/score_mods.py | 54 | Build a memoized FA4 score_mod adding a learned relative-position bias. The returned closure adds ``rel_logits[glob | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/dsa/cute_dsl.py | 95 | Length-aware per-row top-k over a dense indexer-logits matrix. Args: logits: 2-D row-major contiguous CUDA | STRING |
| HIGH | …python/tokenspeed_kernel/ops/attention/dsa/__init__.py | 293 | Select, without running it, the ``dsa_prefill_topk`` leaf for index keys handed as rows in workspace-row order of `` | STRING |
| HIGH | …l/python/tokenspeed_kernel/ops/attention/gdn/triton.py | 1022 | Replay all GDN layers and write their final accepted states. This portable Triton implementation is shared by every | STRING |
| HIGH | …n/tokenspeed_kernel/ops/attention/gdn/_triton/chunk.py | 145 | Args: q (torch.Tensor): queries of shape `[B, T, H, K]` if `head_first=False` else `[B, H, T, K | STRING |
| HIGH | …rnel/python/tokenspeed_kernel/ops/sampling/cute_dsl.py | 1006 | Build the cross-rank argmax state, requiring NVLS multicast. Args: group: The process group the argmax redu | STRING |
| HIGH | …/tokenspeed_kernel/ops/sampling/triton/dspark_block.py | 239 | Score this rank's vocabulary shard for one block step. Args: base_logits: ``[rows, block, local_vocab]`` FP | STRING |
| HIGH | …/tokenspeed_kernel/ops/sampling/triton/dspark_block.py | 352 | Write the winners of the gathered ``candidates`` into ``output[:, step]``. Args: candidates: ``[tp, rows, n | STRING |
| HIGH | …hon/tokenspeed_kernel/ops/sampling/triton/spec_path.py | 73 | Walk the lattice one locally best edge at a time. Args: candidate_ids: ``[batch, num_steps, top_k]`` candid | STRING |
| HIGH | …el/python/tokenspeed_kernel/ops/conv/triton/grouped.py | 91 | Convolve packed blocks with per-row, per-group dynamic coefficients. Args: x: Input rows ``[T, C]`` packed | STRING |
| HIGH | …-kernel/python/tokenspeed_kernel/ops/kvcache/triton.py | 724 | Build batched verify-commit row ids for :func:`copy_state_rows`. Sinks the ``arange``/``clamp``/``where`` chain tha | STRING |
| HIGH | …-kernel/python/tokenspeed_kernel/ops/kvcache/triton.py | 1323 | Normalize, rotate and scatter every layer's latent KV in one launch. Each ``[kv_lora_rank + rope_dim]`` row is RMSN | STRING |
| HIGH | …speed-kernel/python/tokenspeed_kernel/ops/copy/cuda.py | 38 | Copy a 2D device rectangle asynchronously. Args: dst: Destination device pointer. dst_pitch: Destin | STRING |
| HIGH | …python/tokenspeed_kernel/ops/metadata/ngram_history.py | 226 | Seed the per-slot prefixes and write every input row's history. Args: tokens: ``[batch, context + 1]`` int6 | STRING |
| HIGH | …on/tokenspeed_kernel/ops/metadata/accepted_frontier.py | 159 | Add each request's accepted tokens to its cache length and history. Args: req_pool_indices: ``[batch]`` int | STRING |
| HIGH | …speed_kernel_amd/ops/gfx950/moe/mxfp4/decode_common.py | 94 | Return the compact e8m0 scale tile for a CDNA4 scaled-upcast result. ``gl.amd.cdna4.scaled_upcast`` takes a scale t | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/ops/mha.py | 101 | CODE | |
| LOW | …eed-kernel-npu/python/tokenspeed_kernel_npu/ops/mha.py | 151 | CODE | |
| LOW | test/runners.py | 361 | CODE | |
| LOW | test/runtime/conftest.py | 34 | CODE | |
| LOW | test/runtime/test_drafter_accept_indexing.py | 84 | CODE | |
| LOW | test/runtime/test_dp_sampling_routing_metadata.py | 49 | CODE | |
| LOW | test/runtime/cache_pd_test_utils.py | 84 | CODE | |
| LOW | test/runtime/layers/test_mla_verify_metadata.py | 20 | CODE | |
| LOW | test/ci_system/pipeline.py | 1786 | CODE | |
| LOW | test/ci_system/test_kernel_benchmark_comment.py | 109 | CODE | |
| LOW | python/tokenspeed/runtime/layers/linear.py | 375 | CODE | |
| LOW | python/tokenspeed/runtime/layers/linear.py | 556 | CODE | |
| LOW | python/tokenspeed/runtime/layers/linear.py | 807 | CODE | |
| LOW | python/tokenspeed/runtime/layers/linear.py | 1160 | CODE | |
| LOW | …/tokenspeed/runtime/layers/vocab_parallel_embedding.py | 253 | CODE | |
| LOW | …/tokenspeed/runtime/layers/vocab_parallel_embedding.py | 663 | CODE | |
| LOW | …speed/runtime/layers/attention/mm_encoder_attention.py | 352 | CODE | |
| LOW | …speed/runtime/layers/attention/linear/causal_conv1d.py | 1052 | CODE | |
| LOW | python/tokenspeed/runtime/layers/moe/expert.py | 69 | CODE | |
| LOW | python/tokenspeed/runtime/layers/moe/latent.py | 232 | CODE | |
| LOW | python/tokenspeed/runtime/layers/moe/latent.py | 523 | CODE | |
| LOW | python/tokenspeed/runtime/layers/moe/loader.py | 563 | CODE | |
| LOW | python/tokenspeed/runtime/layers/moe/loader.py | 338 | CODE | |
| LOW | python/tokenspeed/runtime/layers/moe/topk.py | 492 | CODE | |
| LOW | python/tokenspeed/runtime/distributed/mapping.py | 600 | CODE | |
| LOW | python/tokenspeed/runtime/models/glm5.py | 260 | CODE | |
| LOW | python/tokenspeed/runtime/models/kimi_k3.py | 368 | CODE | |
| LOW | python/tokenspeed/runtime/models/kimi_k3.py | 2244 | CODE | |
| LOW | python/tokenspeed/runtime/models/gpt_oss.py | 116 | CODE | |
| LOW | python/tokenspeed/runtime/models/qwen3.py | 106 | CODE | |
| LOW | python/tokenspeed/runtime/models/deepseek_v3.py | 563 | CODE | |
| LOW | python/tokenspeed/runtime/models/moonvit.py | 293 | CODE | |
| LOW | python/tokenspeed/runtime/models/moonvit.py | 501 | CODE | |
| LOW | python/tokenspeed/runtime/models/moonvit.py | 598 | CODE | |
| LOW | python/tokenspeed/runtime/execution/forward_step.py | 195 | CODE | |
| LOW | python/tokenspeed/runtime/configs/inkling_config.py | 156 | CODE | |
| LOW | python/tokenspeed/runtime/configs/inkling_config.py | 393 | CODE | |
| LOW | python/tokenspeed/runtime/configs/inkling_config.py | 460 | CODE | |
| LOW | python/tokenspeed/runtime/configs/model_config.py | 467 | CODE | |
| LOW | …thon/tokenspeed/runtime/configs/qwen3_vision_config.py | 28 | CODE | |
| LOW | …on/tokenspeed/runtime/configs/kimi_k3_dspark_config.py | 53 | CODE | |
| LOW | …tokenspeed/runtime/configs/qwen3_5_text_base_config.py | 191 | CODE | |
| LOW | python/tokenspeed/runtime/configs/qwen3_asr_config.py | 50 | CODE | |
| LOW | python/tokenspeed/runtime/configs/qwen2_config.py | 57 | CODE | |
| LOW | python/tokenspeed/runtime/configs/kimi_k3_config.py | 60 | CODE | |
| LOW | python/tokenspeed/runtime/configs/kimi_k3_config.py | 140 | CODE | |
| LOW | python/tokenspeed/runtime/configs/qwen3_moe_config.py | 31 | CODE | |
| LOW | python/tokenspeed/runtime/configs/qwen4_exp_config.py | 52 | CODE | |
| LOW | python/tokenspeed/runtime/configs/qwen4_exp_config.py | 156 | CODE | |
| LOW | python/tokenspeed/runtime/configs/qwen3_config.py | 173 | CODE | |
| LOW | python/tokenspeed/runtime/configs/kimi_k25_config.py | 60 | CODE | |
| LOW | python/tokenspeed/runtime/configs/glm53_flash_config.py | 41 | CODE | |
| LOW | python/tokenspeed/runtime/configs/glm53_flash_config.py | 92 | CODE | |
| LOW | python/tokenspeed/runtime/configs/glm53_flash_config.py | 341 | CODE | |
| LOW | python/tokenspeed/runtime/configs/minimax_m3_config.py | 62 | CODE | |
| LOW | python/tokenspeed/runtime/configs/minimax_m3_config.py | 165 | CODE | |
| LOW | python/tokenspeed/runtime/entrypoints/engine.py | 155 | CODE | |
| LOW | python/tokenspeed/runtime/entrypoints/engine.py | 206 | CODE | |
| LOW | python/tokenspeed/runtime/entrypoints/engine_base.py | 42 | CODE | |
| LOW | python/tokenspeed/runtime/engine/request.py | 50 | CODE | |
| 143 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/runtime/test_draft_advance_seqlens.py | 350 | # Step 0: the drafter publishes the accepted frontier (vc + N -> vc + a); the | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 124 | # Step 1: Determine CUDA index and FlashInfer architecture | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 152 | # Step 2: Upgrade base tools | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 164 | # Step 3: Sync FlashInfer JIT cache on Blackwell runners | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 170 | # Step 4: Install tokenspeed-kernel | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 196 | # Step 5: Install TokenSpeed Scheduler (C++) | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 210 | # Step 6: Install TokenSpeed | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 220 | # Step 7: Optionally override tokenspeed-mla with in-tree source | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 234 | # Step 8: Pin critical kernel deps to exact versions | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 304 | # Step 9: Fix Triton ptxas (CUDA 13+ only) | COMMENT |
| LOW | …hon/tokenspeed/runtime/moe/eplb_algorithms/deepseek.py | 165 | # Step 1: pack groups to nodes | COMMENT |
| LOW⚡ | …hon/tokenspeed/runtime/moe/eplb_algorithms/deepseek.py | 176 | # Step 2: construct redundant experts within nodes | COMMENT |
| LOW⚡ | …hon/tokenspeed/runtime/moe/eplb_algorithms/deepseek.py | 185 | # Step 3: pack physical_experts to GPUs | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 436 | # Step 1: Load LSE_partial from gmem to shared memory | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 477 | # Step 2: Load O_partial for pipeline stages | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 534 | # Step 3: Load and transpose LSE from smem to registers | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 552 | # Step 4: Compute final LSE along split dimension | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 619 | # Step 5: Store final LSE to gmem | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 645 | # Step 6: Read O_partial and accumulate final O | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 698 | # Step 7: Write final O to gmem | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 597 | # Step 1: Load LSE_partial from gmem to shared memory | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 736 | # Step 2: Load O_partial for pipeline stages | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 805 | # Step 3: Load and transpose LSE from smem to registers | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 831 | # Step 4: Compute final LSE along split dimension | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 953 | # Step 5: Store final LSE to gmem | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 990 | # Step 6: Read O_partial and accumulate final O | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 1048 | # Step 7: Write final O to gmem (fake→real via SMEM) | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1564 | # Step 0: load tmem | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1618 | # Step 1: perform swiglu on the first part, interleave with the second's 32x32 tmem transpose. | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1650 | # Step 2: Quant | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1661 | # Step 3: TMASTG | COMMENT |
| LOW⚡ | …eed_kernel/thirdparty/cute_dsl/mega_moe/kernel_fc12.py | 1988 | # Step 1: coarse-grain spin on ``fc1_done_counter[slot]`` | COMMENT |
| LOW | …eed_kernel/thirdparty/cute_dsl/mega_moe/kernel_fc12.py | 2083 | # Step 3: K-loop with 2x cute.copy per tile (B + | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 277 | # Step 1: Load chunk → smem (convert to ordered) | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 836 | # Step 3: Collect output indices and values | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 1215 | # Step 1: Load chunk to smem as ordered | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 1220 | # Step 2: Multi-round radix select | COMMENT |
| LOW | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 491 | # Step 1: Inclusive prefix sum directly on the histogram | COMMENT |
| LOW | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 510 | # Step 2: Find threshold bucket | COMMENT |
| LOW | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 1367 | # Step 3: Collect output | COMMENT |
| LOW⚡ | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 284 | # Step 3: Collect output indices and values (cluster variant) | COMMENT |
| LOW⚡ | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 491 | # Step 1: Load chunk to smem as ordered | COMMENT |
| LOW⚡ | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 496 | # Step 2: Multi-round radix select | COMMENT |
| LOW | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 548 | # Step 3: Collect output (cluster barriers) | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 87 | # Step 1: Warp-level prefix sum using shuffle | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 90 | # Step 2: Store warp prefix sums | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 95 | # Step 3: Prefix sum across warps | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 106 | # Step 4: Add warp-level prefix | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 110 | # Step 5: Get total sum if need_total_sum is True | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2231 | # Step 1: validate inputs and reject unsupported modes. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2263 | # Step 2: shuffle w1 to the MFMA-tile layout the kernel expects. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2397 | # Step 3: materialise scalar / None args as device tensors so the | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2479 | # Step 4: launch the GEMM. One CTA per (M-tile, N-tile); the per-CTA | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1758 | # Step 1: validate inputs and reject unsupported modes. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1777 | # Step 2: shuffle w2 to MFMA-tile layout if the caller didn't. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1789 | # Step 3: flatten the (token, slot) axes so each row is one routed | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1927 | # Step 5: GEMM. Writes either to `partials` (reduce mode) or directly | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1972 | # Step 6: reduce. Sum partials[token, :, n] over the topk dim | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/runtime/test_draft_advance_seqlens.py | 350 | # Step 0: the drafter publishes the accepted frontier (vc + N -> vc + a); the | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 124 | # Step 1: Determine CUDA index and FlashInfer architecture | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 152 | # Step 2: Upgrade base tools | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 164 | # Step 3: Sync FlashInfer JIT cache on Blackwell runners | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 170 | # Step 4: Install tokenspeed-kernel | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 196 | # Step 5: Install TokenSpeed Scheduler (C++) | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 210 | # Step 6: Install TokenSpeed | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 220 | # Step 7: Optionally override tokenspeed-mla with in-tree source | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 234 | # Step 8: Pin critical kernel deps to exact versions | COMMENT |
| LOW⚡ | test/ci_system/install_deps.sh | 304 | # Step 9: Fix Triton ptxas (CUDA 13+ only) | COMMENT |
| LOW | …hon/tokenspeed/runtime/moe/eplb_algorithms/deepseek.py | 165 | # Step 1: pack groups to nodes | COMMENT |
| LOW⚡ | …hon/tokenspeed/runtime/moe/eplb_algorithms/deepseek.py | 176 | # Step 2: construct redundant experts within nodes | COMMENT |
| LOW⚡ | …hon/tokenspeed/runtime/moe/eplb_algorithms/deepseek.py | 185 | # Step 3: pack physical_experts to GPUs | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 436 | # Step 1: Load LSE_partial from gmem to shared memory | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 477 | # Step 2: Load O_partial for pipeline stages | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 534 | # Step 3: Load and transpose LSE from smem to registers | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 552 | # Step 4: Compute final LSE along split dimension | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 619 | # Step 5: Store final LSE to gmem | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 645 | # Step 6: Read O_partial and accumulate final O | COMMENT |
| LOW⚡ | …rnel/ops/attention/rmha/_cute_dsl/flash_fwd_combine.py | 698 | # Step 7: Write final O to gmem | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 597 | # Step 1: Load LSE_partial from gmem to shared memory | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 736 | # Step 2: Load O_partial for pipeline stages | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 805 | # Step 3: Load and transpose LSE from smem to registers | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 831 | # Step 4: Compute final LSE along split dimension | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 953 | # Step 5: Store final LSE to gmem | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 990 | # Step 6: Read O_partial and accumulate final O | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/msa/cute/src/sm100/fwd/combine.py | 1048 | # Step 7: Write final O to gmem (fake→real via SMEM) | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1564 | # Step 0: load tmem | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1618 | # Step 1: perform swiglu on the first part, interleave with the second's 32x32 tmem transpose. | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1650 | # Step 2: Quant | COMMENT |
| LOW | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1661 | # Step 3: TMASTG | COMMENT |
| LOW⚡ | …eed_kernel/thirdparty/cute_dsl/mega_moe/kernel_fc12.py | 1988 | # Step 1: coarse-grain spin on ``fc1_done_counter[slot]`` | COMMENT |
| LOW | …eed_kernel/thirdparty/cute_dsl/mega_moe/kernel_fc12.py | 2083 | # Step 3: K-loop with 2x cute.copy per tile (B + | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 277 | # Step 1: Load chunk → smem (convert to ordered) | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 836 | # Step 3: Collect output indices and values | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 1215 | # Step 1: Load chunk to smem as ordered | COMMENT |
| LOW⚡ | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 1220 | # Step 2: Multi-round radix select | COMMENT |
| LOW | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 491 | # Step 1: Inclusive prefix sum directly on the histogram | COMMENT |
| LOW | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 510 | # Step 2: Find threshold bucket | COMMENT |
| LOW | …ute_dsl/topk/top_k/single_pass_multi_cta_radix_topk.py | 1367 | # Step 3: Collect output | COMMENT |
| LOW⚡ | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 284 | # Step 3: Collect output indices and values (cluster variant) | COMMENT |
| LOW⚡ | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 491 | # Step 1: Load chunk to smem as ordered | COMMENT |
| LOW⚡ | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 496 | # Step 2: Multi-round radix select | COMMENT |
| LOW | …topk/top_k/single_pass_multi_cta_radix_topk_cluster.py | 548 | # Step 3: Collect output (cluster barriers) | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 87 | # Step 1: Warp-level prefix sum using shuffle | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 90 | # Step 2: Store warp prefix sums | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 95 | # Step 3: Prefix sum across warps | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 106 | # Step 4: Add warp-level prefix | COMMENT |
| LOW⚡ | …ed_kernel/thirdparty/cute_dsl/topk/top_k/block_scan.py | 110 | # Step 5: Get total sum if need_total_sum is True | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2231 | # Step 1: validate inputs and reject unsupported modes. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2263 | # Step 2: shuffle w1 to the MFMA-tile layout the kernel expects. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2397 | # Step 3: materialise scalar / None args as device tensors so the | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage1.py | 2479 | # Step 4: launch the GEMM. One CTA per (M-tile, N-tile); the per-CTA | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1758 | # Step 1: validate inputs and reject unsupported modes. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1777 | # Step 2: shuffle w2 to MFMA-tile layout if the caller didn't. | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1789 | # Step 3: flatten the (token, slot) axes so each row is one routed | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1927 | # Step 5: GEMM. Writes either to `partials` (reduce mode) or directly | COMMENT |
| LOW | …peed_kernel_amd/ops/gfx950/moe/mxfp4/prefill_stage2.py | 1972 | # Step 6: reduce. Sum partials[token, :, n] over the topk dim | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| CRITICAL | test/runtime/test_dspark_proposal.py | 348 | drafter.markov_head.markov_w2.weight.data.zero_() | CODE |
| CRITICAL | test/runtime/test_dspark_proposal.py | 378 | drafter.markov_head.markov_w1.weight.data.normal_(0.0, 3.0) | CODE |
| CRITICAL | test/runtime/test_dspark_proposal.py | 379 | drafter.markov_head.markov_w2.weight.data.normal_(0.0, 3.0) | CODE |
| CRITICAL⚡ | test/runtime/test_qwen4_exp.py | 3242 | device.lookup.ngram_embedding.weight.data.copy_(payload) | CODE |
| CRITICAL⚡ | test/runtime/test_qwen4_exp.py | 3243 | host.lookup.ngram_embedding.weight.data.copy_(payload.cpu()) | CODE |
| CRITICAL | …n/tokenspeed_kernel/ops/attention/gdn/_triton/utils.py | 147 | triton.runtime.driver.active.utils.get_device_properties(i)[ | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | test/runtime/test_kda_state_eviction_spec.py | 124 | # harness, with the complete unmodified scheduler tables supplied here. | COMMENT |
| MEDIUM⚡ | test/runtime/test_detokenizer_parity.py | 73 | # Shared harness | COMMENT |
| MEDIUM | test/runtime/test_inkling_activation_parity.py | 336 | # (bind_cache_groups); this harness plans a plain pool and hand-binds | COMMENT |
| MEDIUM | test/runtime/test_cache_pd_shutdown.py | 54 | # The harness plans no device work and no batch; trace anything that | COMMENT |
| LOW | …speed/runtime/layers/attention/linear/causal_conv1d.py | 181 | # first chunk and does not have prior-token, so just set to 0 | COMMENT |
| MEDIUM | tokenspeed-kernel/test/test_benchmark_gemm_generator.py | 172 | # The harness routes quant="none" through prepare_mm to the dense generator, | COMMENT |
| MEDIUM | tokenspeed-kernel/python/tokenspeed_kernel/selection.py | 856 | # comprehensive warmup should pass explicit op signatures. | COMMENT |
| MEDIUM | …l/ops/attention/rmha/_cute_dsl/flash_fwd_sm100_bias.py | 1560 | # CLC buffers placed here to utilize padding before sO's 1024-byte alignment. | COMMENT |
| MEDIUM⚡ | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 998 | _EpilogueTokenTileSize = 64 # Fundamentally the epi_tile_n | CODE |
| MEDIUM⚡ | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 999 | _EpilogueFc1IntermediateGateUpTileSize = 128 # Fundamentally epi_tile_m | CODE |
| MEDIUM⚡ | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1000 | _EpilogueFc1IntermediateDownTileSize = 64 # Fundamentally epi_tile_m // 2 | CODE |
| MEDIUM⚡ | …rnel/thirdparty/cute_dsl/mega_moe/epilogue_refactor.py | 1001 | _EpilogueFc2HiddenTileSize = 128 # Fundamentally epi_tile_m | CODE |
| MEDIUM | …peed-mla/python/tokenspeed_mla/mla_decode_fp8_sm107.py | 8450 | # H and Q are contiguous in the harness's (B,Q,H,D) allocation. | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | …/tokenspeed/runtime/engine/data_parallel_controller.py | 330 | # Create a thread for each worker | COMMENT |
| MEDIUM | …nel/python/tokenspeed_kernel/ops/communication/nccl.py | 22 | # This file is a pure Python wrapper for the NCCL library. | COMMENT |
| MEDIUM | …kenspeed_kernel/thirdparty/trtllm_blockwise/_kernel.py | 44 | # This file is copied and modified from cutlass example https://github.com/NVIDIA/cutlass/blob/main/examples/python/CuTe | COMMENT |
| MEDIUM | …kenspeed_kernel/thirdparty/trtllm_blockwise/_kernel.py | 124 | """This class implements batched matrix multiplication (C = (SFA * A) * (SFB * B)) with support for fp8 (e4m3fn, | STRING |
| MEDIUM | …l/thirdparty/cute_dsl/nvfp4_gemm_swiglu_nvfp4_quant.py | 115 | """This class implements batched matrix multiplication (C = A x SFA x B x SFB) with support for various data types | STRING |
| MEDIUM | …python/tokenspeed_kernel/thirdparty/cute_dsl/argmax.py | 18 | # This file contains code derived from the Quack library: | COMMENT |
| MEDIUM | …speed_kernel/thirdparty/cute_dsl/mega_moe/moe_utils.py | 189 | # Create a new cute.Pointer with generic address space, preserving alignment | COMMENT |
| MEDIUM | …speed_kernel/thirdparty/cute_dsl/mega_moe/moe_utils.py | 217 | # Create a new cute.Pointer with gmem address space, preserving alignment | COMMENT |
| MEDIUM | …on/tokenspeed_kernel/thirdparty/cute_dsl/topk/utils.py | 48 | # This file is copied and modified from cutlass https://github.com/NVIDIA/cutlass/blob/main/python/CuTeDSL/cutlass/cute/ | COMMENT |
| MEDIUM | …on/tokenspeed_kernel/thirdparty/cute_dsl/topk/utils.py | 177 | # Create a numpy array | STRING |
| MEDIUM | tokenspeed-mla/python/tokenspeed_mla/utils.py | 243 | # Create a numpy array | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/runners.py | 328 | # Check if the text is empty or only whitespace. | COMMENT |
| LOW | test/runners.py | 536 | # Check if the text is empty or only whitespace. | COMMENT |
| LOW | test/ci_system/ci_utils.py | 225 | # Check if we should retry | COMMENT |
| LOW | python/tokenspeed/runtime/engine/async_llm.py | 394 | # Check if this was an abort/error created by scheduler | COMMENT |
| LOW | …kenspeed_kernel/thirdparty/trtllm_blockwise/_kernel.py | 442 | # Check if input data types are compatible with MMA instruction | COMMENT |
| LOW | …l/thirdparty/cute_dsl/nvfp4_gemm_swiglu_nvfp4_quant.py | 490 | # Check if input data types are compatible with MMA instruction | COMMENT |
| LOW | …arty/cute_dsl/topk/top_k/filtered_top_k_varlen_util.py | 1236 | # Check if the number of valid indices matches | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | README.md | 32 | - [2026/08] [Qwen3.8 Flash Next](https://developer.nvidia.com/blog/experiment-with-qwen3-8-flash-next-176b-model-on-nvid | CODE |
| MEDIUM | README.md | 38 | - [2026/05] 🚀 TokenSpeed hits 580 TPS on Qwen3.5-397B-A17B for agentic workloads. [[PyTorch blog](https://pytorch.org/bl | CODE |
| MEDIUM | test/cli/test_serve_smg_integration.py | 90 | # test exercises orchestration/teardown, not tokenizer loading. | COMMENT |
| MEDIUM | python/tokenspeed/runtime/models/kimi_k3_nextn.py | 422 | # NextN scaffolding around the decoder layer. | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | …thon/tokenspeed/runtime/utils/hf_transformers_utils.py | 940 | # Special handling for stop token <|eom_id|> generated by llama 3 tool use. | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/runtime/test_event_loop_in_flight.py | 181 | output_tokens="placeholder", | CODE |
| LOW | test/runtime/test_event_loop_in_flight.py | 182 | output_lengths="placeholder", | CODE |
| LOW | python/tokenspeed/runtime/execution/prefill_graph.py | 1369 | """Smallest captured request capacity, including native dummy-token room. | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/runners.py | 319 | # make sure to disable compile | COMMENT |
| LOW | …/python/tokenspeed_kernel/ops/communication/deep_ep.py | 687 | # Please make sure to change DeepEP code in internode_ll.cu dispatch / combine as below first and then reinstall | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | test/ci_system/pipeline.py | 1786 | def execute_task( | CODE |