verl/HybridFlow: A Flexible and Efficient RL Post-Training Framework
This report summarizes scanner evidence for verl-project/verl, a Python project with 23,822 GitHub stars. SynthScan examined 245,575 lines of code across 1096 source files, recording 4915 pattern matches distributed across 23 syntactic categories. The legacy adjusted pattern signal is 27.4 (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? 3775 finding(s).
Are imports declared and package/API names verified? 265 finding(s).
Are wrappers, guards, and repeated structures necessary? 3217 finding(s).
Which changes need focused human review? 4915 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 4915 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 | tests/test_protocol_v2_on_cpu.py | 67 | def test_tensor_dict_constructor(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 92 | def test_index_select_tensor_dict(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 152 | def test_tensordict_with_images(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 180 | def test_tensordict_with_packing(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 269 | def test_tensor_dict_make_iterator(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 509 | def test_torch_save_data_proto(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 654 | def test_dataproto_chunk_after_index(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 697 | def test_concat_nested_tensor(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 923 | def test_assign_non_tensor_stack_with_nested_lists(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 938 | def test_assign_non_tensor_stack_with_nested_dicts(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 953 | def test_assign_non_tensor_stack_with_complex_nested(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 972 | def test_assign_non_tensor_handles_wrappers(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 987 | def test_assign_non_tensor_stack_batch_size_check(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 995 | def test_assign_non_tensor_with_auto_detection(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 1018 | def test_get_tensordict_with_nested_lists(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 1033 | def test_get_tensordict_with_nested_dicts(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 1045 | def test_get_tensordict_with_complex_nested_structures(): | CODE |
| LOW | tests/test_protocol_v2_on_cpu.py | 1060 | def test_get_tensordict_agent_loop_scenario(): | CODE |
| LOW | tests/test_base_config_on_cpu.py | 33 | def test_getitem_nonexistent_attribute(base_config_mock): | CODE |
| LOW | tests/test_base_config_on_cpu.py | 39 | def test_getitem_invalid_key_type(base_config_mock): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 141 | def test_tensor_dict_constructor(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 155 | def test_tensor_dict_make_iterator(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 219 | def test_concat_metrics_from_multiple_workers(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 249 | def test_concat_with_empty_and_non_list_meta_info(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 272 | def test_concat_first_worker_missing_metrics(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 295 | def test_concat_non_list_metrics(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 315 | def test_concat_merge_different_non_metric_keys(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 339 | def test_concat_conflicting_non_metric_keys(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 447 | def test_dataproto_fold_unfold(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 470 | def test_torch_save_data_proto(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 570 | def test_old_vs_new_from_single_dict(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 642 | def test_dataproto_unfold_column_chunks(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 708 | def test_dataproto_chunk_after_index(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 785 | def test_to_tensordict_with_nested_lists(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 810 | def test_to_tensordict_with_nested_dicts(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 834 | def test_to_tensordict_with_complex_nested_structures(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 862 | def test_to_tensordict_and_back_with_nested_data(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 926 | def test_to_tensordict_agent_loop_scenario(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 993 | def test_serialize_deserialize_single_tensor(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1010 | def test_serialize_deserialize_tensordict_regular_tensors(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1039 | def test_serialize_deserialize_tensordict_nested_tensors(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1092 | def test_serialize_deserialize_tensordict_mixed_types(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1177 | def test_serialize_deserialize_tensordict_with_device(): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1208 | def test_numpy_dataproto_serialization_skips_tensordict_consolidation(monkeypatch): | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1232 | def test_serialize_dataproto_with_empty_tensordict(): | CODE |
| LOW | …l/fully_async_policy/test_async_genrm_config_on_cpu.py | 76 | def _validate_async_rm_config(config): | CODE |
| LOW | …l/fully_async_policy/test_async_genrm_config_on_cpu.py | 92 | def test_rm_enabled_standalone_passes(self): | CODE |
| LOW | …l/fully_async_policy/test_async_genrm_config_on_cpu.py | 98 | def test_rm_enabled_colocate_fails(self): | CODE |
| LOW | tests/experimental/vla/test_sim_envs.py | 37 | def test_sim_env_creation_and_step(simulator_type): | CODE |
| LOW | …xperimental/reward_loop/run_determinism_e2e_with_rm.py | 226 | def parse_reward_curve_from_stdout(stdout: str): | CODE |
| LOW | …ts/experimental/reward_loop/test_reward_model_disrm.py | 107 | def test_reward_model_manager(): | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 71 | def mock_sync_reward_function( | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 83 | async def mock_async_reward_function( | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 106 | async def mock_failing_api_function( | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 114 | async def mock_dict_result_function( | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 169 | async def test_basic_reward_computation(self, tokenizer): | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 258 | async def test_concurrency_limiting(self, tokenizer): | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 351 | async def test_sync_reward_function(self, tokenizer): | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 366 | async def test_combined_rate_limits(self, tokenizer): | CODE |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 402 | async def test_correct_vs_incorrect_answers(self, tokenizer): | CODE |
| 2228 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | tests/experimental/agent_loop/test_basic_agent_loop.py | 413 | # ────────────────────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tests/experimental/agent_loop/test_basic_agent_loop.py | 415 | # ────────────────────────────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 52 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 54 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 331 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 333 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 491 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 493 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 586 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/tools/test_function_tool_on_cpu.py | 588 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/utils/test_bucketed_weight_transfer.py | 122 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/utils/test_bucketed_weight_transfer.py | 124 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/utils/test_bucketed_weight_transfer.py | 158 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/utils/test_bucketed_weight_transfer.py | 160 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/utils/test_bucketed_weight_transfer.py | 208 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/utils/test_bucketed_weight_transfer.py | 210 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/utils/test_bucketed_weight_transfer.py | 244 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/utils/test_bucketed_weight_transfer.py | 246 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/utils/test_continuous_token_on_cpu.py | 1742 | # ============================================================================= | COMMENT |
| MEDIUM⚡ | tests/utils/test_continuous_token_on_cpu.py | 1745 | # ============================================================================= | COMMENT |
| MEDIUM⚡ | tests/utils/test_continuous_token_on_cpu.py | 1888 | # ============================================================================= | COMMENT |
| MEDIUM⚡ | tests/utils/test_continuous_token_on_cpu.py | 1890 | # ============================================================================= | COMMENT |
| MEDIUM | tests/utils/test_continuous_token_on_cpu.py | 1842 | # ============================================================================= | COMMENT |
| MEDIUM | tests/utils/test_continuous_token_on_cpu.py | 1844 | # ============================================================================= | COMMENT |
| MEDIUM | tests/utils/test_continuous_token_on_cpu.py | 2080 | # ============================================================================= | COMMENT |
| MEDIUM | tests/utils/test_continuous_token_on_cpu.py | 2082 | # ============================================================================= | COMMENT |
| MEDIUM | tests/utils/test_continuous_token_on_cpu.py | 2423 | # ============================================================================= | COMMENT |
| MEDIUM | tests/utils/test_continuous_token_on_cpu.py | 2426 | # ============================================================================= | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 317 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 320 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 337 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 341 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 407 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 409 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 474 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 476 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 524 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 530 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 605 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 607 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 639 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 641 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 877 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 882 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 691 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 694 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 217 | # =========================================================================== | COMMENT |
| MEDIUM⚡ | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 219 | # =========================================================================== | COMMENT |
| MEDIUM⚡ | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 267 | # =========================================================================== | COMMENT |
| MEDIUM⚡ | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 269 | # =========================================================================== | COMMENT |
| MEDIUM⚡ | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 602 | # =========================================================================== | COMMENT |
| MEDIUM⚡ | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 604 | # =========================================================================== | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 36 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 38 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 66 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 68 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 156 | # =========================================================================== | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 158 | # =========================================================================== | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 407 | # =========================================================================== | COMMENT |
| MEDIUM | …/utils/ckpt/test_megatron_checkpoint_manager_on_cpu.py | 409 | # =========================================================================== | COMMENT |
| 302 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | manage_envs.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | manage_envs.py | 161 | import tempfile | COMMENT |
| LOW | manage_envs.py | 181 | # no torch, so it rides on whichever cu130 backend the job synced) and | COMMENT |
| LOW | manage_envs.py | 221 | "cu130": INFERENCE_BACKENDS + TRAINING_BACKENDS, # torch 2.13 GPU backends | COMMENT |
| LOW | manage_envs.py | 781 | # does pin the per-combo cache-warming syncs below. | COMMENT |
| LOW | pyproject.toml | 81 | # on the training side: `ray.util.collective`'s NCCL backend is cupy-based, so | COMMENT |
| LOW | pyproject.toml | 121 | COMMENT | |
| LOW | pyproject.toml | 181 | COMMENT | |
| LOW | pyproject.toml | 321 | # combos and every CI invocation are identical on the two arches. | COMMENT |
| LOW | pyproject.toml | 341 | # convention flash-attn-cu130torch213 / cupy-cu130 use to share a pin): | COMMENT |
| LOW | pyproject.toml | 361 | ] | COMMENT |
| LOW | pyproject.toml | 381 | # same `flash_attn/cute/` files as flash-attn 2.8.3 (the fsdp / megatron pin), | COMMENT |
| LOW | pyproject.toml | 401 | # already searches ("nvidia/cuXX/lib since CUDA 13.0"). SONAME stays .so.0. | COMMENT |
| LOW | pyproject.toml | 441 | version = "0.3.0" | COMMENT |
| LOW | setup.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/test_protocol_v2_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/test_protocol_v2_on_cpu.py | 581 | # assert result_np_bool.keys() == data.keys() | COMMENT |
| LOW | tests/test_base_config_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/__init__.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/test_protocol_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/__init__.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/fully_async_policy/__init__.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …l/fully_async_policy/test_async_genrm_config_on_cpu.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/vla/test_sim_envs.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …xperimental/reward_loop/run_determinism_e2e_with_rm.py | 1 | #!/usr/bin/env python3 | COMMENT |
| LOW | …ts/experimental/reward_loop/test_reward_model_disrm.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …reward_loop/test_rate_limited_reward_manager_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …ental/reward_loop/test_agent_reward_loop_standalone.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/reward_loop/test_math_verify.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …imental/reward_loop/test_agent_reward_loop_colocate.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …ts/experimental/reward_loop/test_reward_model_genrm.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …rimental/reward_loop/test_async_token_bucket_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/reward_loop/reward_fn.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/agent_loop/test_multi_modal.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/agent_loop/test_call_tool_on_cpu.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/agent_loop/test_basic_agent_loop.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/agent_loop/agent_utils.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …sts/experimental/agent_loop/test_standalone_rollout.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/agent_loop/__init__.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …mental/agent_loop/test_audio_server_contract_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/experimental/agent_loop/function_tool_examples.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …ent_loop/test_agent_loop_extra_fields_schema_on_cpu.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …ts/experimental/agent_loop/test_tool_call_id_on_cpu.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …perimental/agent_loop/test_qwen3_tool_parser_on_cpu.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …ts/experimental/agent_loop/test_gpt_oss_tool_parser.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | …ntal/agent_loop/test_multimodal_position_ids_on_cpu.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/tools/test_mixed_tools_on_cpu.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/tools/__init__.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/tools/test_function_tool_on_cpu.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/tools/_stub_search_tools.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_docstrings.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_device_api_usage.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_dataproto_usage.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/test_check_example_naming.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_license.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_api_docs.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_pr_description.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/validate_imported_docs.py | 1 | # Copyright 2025 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_uv_gpu_only.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/special_sanity/check_example_naming.py | 1 | # Copyright 2026 Bytedance Ltd. and/or its affiliates | COMMENT |
| 818 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | manage_envs.py | 316 | except OSError: | CODE |
| HIGH | tests/special_sanity/type_coverage_check.py | 50 | except Exception: | CODE |
| HIGH | tests/utils/test_rollout_trace_on_cpu.py | 332 | except RuntimeError: | CODE |
| HIGH | tests/utils/test_timeout_decorator_cpu.py | 92 | except RuntimeError: | CODE |
| HIGH | …ts/utils/test_vllm_weight_name_normalization_on_cpu.py | 49 | except ImportError: | CODE |
| HIGH | …orkers/rollout/test_vllm_weight_update_utils_on_cpu.py | 35 | except ImportError: | CODE |
| HIGH | …kers/rollout/rollout_sglang/test_http_server_engine.py | 902 | except Exception: | CODE |
| HIGH | verl/__init__.py | 106 | except AttributeError: | CODE |
| HIGH | verl/experimental/fully_async_policy/detach_utils.py | 444 | except asyncio.CancelledError: | CODE |
| HIGH | verl/utils/transferqueue_utils.py | 100 | except RuntimeError: | CODE |
| HIGH | verl/utils/megatron_utils.py | 252 | except AttributeError: | CODE |
| HIGH | verl/utils/megatron_utils.py | 717 | except ImportError: | CODE |
| HIGH | verl/utils/groupwise.py | 146 | except Exception: | CODE |
| HIGH | verl/utils/fs.py | 221 | except ImportError: | CODE |
| HIGH | verl/utils/import_utils.py | 138 | except (ImportError, AttributeError): | CODE |
| HIGH | verl/utils/tracking.py | 278 | except (AttributeError, KeyError, TypeError): | CODE |
| HIGH | verl/utils/checkpoint/megatron_checkpoint_manager.py | 870 | except Exception: | CODE |
| HIGH | verl/utils/checkpoint/megatron_checkpoint_manager.py | 1011 | except Exception: | CODE |
| HIGH | verl/utils/checkpoint/fsdp_checkpoint_manager.py | 475 | except Exception: | CODE |
| HIGH | verl/utils/vllm/patch.py | 24 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 31 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 38 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 45 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 52 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 59 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 66 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 73 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/patch.py | 110 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/vllm_quant_utils.py | 117 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/vllm_quant_utils.py | 123 | except ImportError: | CODE |
| HIGH | verl/utils/vllm/vllm_fp8_utils.py | 87 | except AttributeError: | CODE |
| HIGH | verl/utils/vllm/utils.py | 148 | except ImportError: | CODE |
| HIGH | verl/utils/veomni/router_replay.py | 168 | except ImportError: | CODE |
| HIGH | verl/utils/profiler/torch_memory_profile.py | 170 | except Exception: | CODE |
| HIGH | verl/utils/profiler/torch_memory_profile.py | 75 | except Exception: | CODE |
| HIGH | verl/utils/profiler/torch_profile.py | 48 | except Exception: | CODE |
| HIGH | verl/utils/profiler/torch_profile.py | 62 | except Exception: | CODE |
| HIGH | verl/utils/profiler/torch_profile.py | 60 | except Exception: | CODE |
| HIGH | verl/utils/kernel/kernels.py | 96 | except (ImportError, AttributeError): | CODE |
| HIGH | verl/utils/megatron/dist_checkpointing.py | 69 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/__init__.py | 38 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 608 | except SystemExit: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 250 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 255 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 260 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 164 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 290 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 471 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_code/testing_util.py | 489 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_math/grader.py | 214 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_math/grader.py | 347 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_math/grader.py | 357 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_math/grader.py | 296 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_math/grader.py | 318 | except Exception: | CODE |
| HIGH | verl/utils/reward_score/prime_math/__init__.py | 222 | except Exception: | CODE |
| HIGH | verl/models/mcore/patch.py | 73 | except Exception: | CODE |
| HIGH | verl/workers/reward_manager/prime.py | 75 | except Exception: | CODE |
| HIGH | verl/workers/reward_manager/__init__.py | 36 | except ImportError: | CODE |
| HIGH | verl/workers/rollout/utils.py | 233 | except Exception: | CODE |
| HIGH | verl/workers/rollout/trtllm_rollout/trtllm_rollout.py | 73 | except pynvml.NVMLError: | CODE |
| 2 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tests/test_protocol_on_cpu.py | 1115 | except Exception: | CODE |
| LOW | tests/test_protocol_on_cpu.py | 1252 | except Exception as e: | CODE |
| LOW | tests/experimental/reward_loop/reward_fn.py | 50 | except Exception as e: | STRING |
| LOW | tests/experimental/reward_loop/reward_fn.py | 82 | except Exception: | CODE |
| LOW | tests/experimental/agent_loop/test_multi_modal.py | 129 | except Exception as e: | CODE |
| LOW | tests/experimental/agent_loop/test_basic_agent_loop.py | 154 | except Exception as e: | CODE |
| LOW | tests/experimental/agent_loop/test_basic_agent_loop.py | 186 | except Exception as e: | CODE |
| LOW | tests/experimental/agent_loop/function_tool_examples.py | 69 | except Exception as e: | CODE |
| LOW | tests/special_sanity/check_docstrings.py | 83 | except Exception as e: | CODE |
| MEDIUM | tests/special_sanity/check_docstrings.py | 84 | print(f"Error processing {filepath}: {e}") | CODE |
| LOW | tests/special_sanity/check_api_docs.py | 64 | except Exception as exc: | CODE |
| LOW | tests/special_sanity/check_pr_description.py | 54 | except Exception as e: | CODE |
| LOW | tests/special_sanity/check_pr_description.py | 63 | except Exception as e: | CODE |
| MEDIUM | tests/special_sanity/check_pr_description.py | 58 | def load_pr_body(event_path): | CODE |
| LOW | tests/special_sanity/validate_imported_docs.py | 88 | except Exception: # pragma: no cover – wide net for import quirks | CODE |
| MEDIUM | tests/special_sanity/check_docs_time_info.py | 54 | print(f"Error: Documentation directory '{DOCS_DIR}' does not exist.", file=sys.stderr) | CODE |
| LOW | tests/special_sanity/type_coverage_check.py | 50 | except Exception: | CODE |
| LOW | tests/utils/test_shared_memory.py | 218 | except Exception as e: | CODE |
| MEDIUM | tests/utils/test_shared_memory.py | 219 | print(f"Error in child process: {e}") | CODE |
| LOW⚡ | tests/utils/test_timeout_decorator_cpu.py | 77 | except Exception as e: | CODE |
| LOW | tests/utils/test_timeout_decorator_cpu.py | 170 | except Exception as e: | CODE |
| LOW | tests/utils/test_timeout_decorator_cpu.py | 182 | except Exception as e: | CODE |
| LOW | tests/utils/test_timeout_decorator_cpu.py | 211 | except Exception as e: | CODE |
| LOW | tests/utils/test_timeout_decorator_cpu.py | 223 | except Exception as e: | CODE |
| MEDIUM | tests/utils/test_timeout_decorator_cpu.py | 164 | def slow_task_in_thread(): | CODE |
| MEDIUM | tests/utils/test_timeout_decorator_cpu.py | 176 | def thread_target(): | CODE |
| MEDIUM | tests/utils/test_timeout_decorator_cpu.py | 205 | def slow_task_in_thread(): | CODE |
| MEDIUM | tests/utils/test_timeout_decorator_cpu.py | 217 | def thread_target(): | CODE |
| LOW | tests/utils/test_check_profiler_output.py | 161 | except Exception as e: | CODE |
| LOW⚡ | tests/utils/megatron/test_muon_optim_realcfg_on_cpu.py | 39 | except Exception as exc: # pragma: no cover - env dependent | CODE |
| LOW | …s/special_distributed/test_mcore_probe_differential.py | 148 | except Exception as e: # pragma: no cover | CODE |
| LOW⚡ | …kers/rollout/rollout_sglang/test_http_server_engine.py | 902 | except Exception: | CODE |
| LOW | tests/special_e2e/check_results.py | 30 | except Exception: | CODE |
| LOW | tests/trainer/ppo/test_rollout_corr.py | 458 | except Exception as e: | CODE |
| MEDIUM | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 162 | def run(self) -> None: | CODE |
| MEDIUM | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 212 | def run(self) -> None: | CODE |
| MEDIUM | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 243 | def run(self) -> None: | CODE |
| LOW | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 188 | except Exception as e: # surfaced to the test via join_and_check() | CODE |
| LOW | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 219 | except Exception as e: | CODE |
| LOW | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 250 | except Exception as e: | CODE |
| LOW | verl/__init__.py | 75 | except Exception as _e: | CODE |
| LOW | …rl/experimental/fully_async_policy/fully_async_main.py | 204 | except Exception as e: | CODE |
| LOW | …rl/experimental/fully_async_policy/fully_async_main.py | 212 | except Exception as e: | CODE |
| LOW | verl/experimental/fully_async_policy/detach_utils.py | 446 | except Exception as e: | CODE |
| LOW | verl/experimental/fully_async_policy/message_queue.py | 159 | except Exception: | CODE |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 257 | except Exception as e: | CODE |
| LOW | …perimental/fully_async_policy/fully_async_rollouter.py | 202 | except Exception as e: | CODE |
| LOW | …perimental/fully_async_policy/fully_async_rollouter.py | 1054 | except Exception as e: | CODE |
| LOW | …perimental/fully_async_policy/fully_async_rollouter.py | 1100 | except Exception as e: | CODE |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 366 | except Exception as e: | CODE |
| LOW | …l/fully_async_policy/unittest/simple_streaming_demo.py | 111 | except Exception as e: | CODE |
| LOW | …ly_async_policy/dynamic_schedule/fixed_ratio_policy.py | 88 | except Exception as e: | CODE |
| LOW | …/fully_async_policy/dynamic_schedule/default_policy.py | 234 | except Exception as e: | CODE |
| LOW | verl/experimental/reward_loop/reward_loop.py | 181 | except Exception as e: | CODE |
| LOW | verl/experimental/reward_loop/reward_manager/limited.py | 460 | except Exception as e: | CODE |
| LOW⚡ | verl/experimental/reward_loop/router/naive_router.py | 41 | except Exception: | CODE |
| LOW⚡ | verl/experimental/reward_loop/router/naive_router.py | 44 | except Exception: | CODE |
| LOW | verl/experimental/reward_loop/router/naive_router.py | 168 | except Exception as e: | CODE |
| LOW | verl/experimental/agent_loop/tool_parser.py | 123 | except Exception as e: | CODE |
| LOW | verl/experimental/agent_loop/tool_parser.py | 180 | except Exception as e: | CODE |
| 209 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | manage_envs.py | 152 | CODE | |
| LOW | …mental/agent_loop/test_audio_server_contract_on_cpu.py | 15 | CODE | |
| LOW | …ent_loop/test_agent_loop_extra_fields_schema_on_cpu.py | 15 | CODE | |
| LOW | tests/tools/test_mixed_tools_on_cpu.py | 22 | CODE | |
| LOW | tests/tools/test_function_tool_on_cpu.py | 16 | CODE | |
| LOW | tests/tools/_stub_search_tools.py | 18 | CODE | |
| LOW | tests/special_sanity/test_check_example_naming.py | 17 | CODE | |
| LOW | tests/special_sanity/check_api_docs.py | 28 | CODE | |
| LOW | tests/special_sanity/validate_imported_docs.py | 22 | CODE | |
| LOW | tests/special_sanity/check_uv_gpu_only.py | 47 | CODE | |
| LOW | tests/special_sanity/check_example_naming.py | 41 | CODE | |
| LOW | tests/special_sanity/validate_structure.py | 32 | CODE | |
| LOW | tests/utils/test_utils_skip.py | 17 | CODE | |
| LOW | tests/utils/test_rl_insight_logger_on_cpu.py | 15 | CODE | |
| LOW | …tils/megatron/test_muon_layerwise_bridge_ddp_on_cpu.py | 18 | CODE | |
| LOW | tests/utils/megatron/muon_optim_gpu_smoke.py | 50 | CODE | |
| LOW | tests/special_distributed/test_tensor_dict.py | 29 | CODE | |
| LOW | …ecial_distributed/test_megatron_dynamic_cp_features.py | 28 | CODE | |
| LOW | tests/workers/test_engine_workers_lora_sync.py | 26 | CODE | |
| LOW | …/rollout/test_llm_server_response_length_cap_on_cpu.py | 32 | CODE | |
| LOW | …/workers/rollout/test_vllm_pd_disaggregation_on_cpu.py | 27 | CODE | |
| LOW | …rkers/rollout/test_llm_server_routed_experts_on_cpu.py | 26 | CODE | |
| LOW | …rkers/rollout/test_vision_placeholder_tokens_on_cpu.py | 21 | CODE | |
| LOW | tests/workers/rollout/test_pd_disaggregation.py | 16 | CODE | |
| LOW | …/rollout/rollout_sglang/test_lora_merge_mode_on_cpu.py | 27 | CODE | |
| LOW | …orkers/rollout/rollout_sglang/test_lora_sleep_level.py | 22 | CODE | |
| LOW | tests/special_e2e/envs/__init__.py | 15 | CODE | |
| LOW | tests/special_e2e/envs/digit_completion/__init__.py | 17 | CODE | |
| LOW | tests/special_e2e/envs/digit_completion/__init__.py | 17 | CODE | |
| LOW | tests/checkpoint_engine/test_sglang_loader.py | 24 | CODE | |
| LOW | tests/checkpoint_engine/test_sharded_delta.py | 22 | CODE | |
| LOW | verl/__init__.py | 32 | CODE | |
| LOW | verl/experimental/teacher_loop/__init__.py | 15 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 15 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 15 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 15 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 15 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 16 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 17 | CODE | |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 18 | CODE | |
| LOW | …policy/dynamic_schedule/dynamic_resource_controller.py | 33 | CODE | |
| LOW | …policy/dynamic_schedule/dynamic_resource_controller.py | 33 | CODE | |
| LOW | …policy/dynamic_schedule/dynamic_resource_controller.py | 39 | CODE | |
| LOW | verl/experimental/reward_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/reward_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/reward_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/reward_loop/__init__.py | 16 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 15 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 15 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 16 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 17 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 18 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 19 | CODE | |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 20 | CODE | |
| LOW | verl/experimental/agent_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/agent_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/agent_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/agent_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/agent_loop/__init__.py | 15 | CODE | |
| LOW | verl/experimental/agent_loop/utils.py | 18 | CODE | |
| 197 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | manage_envs.py | 455 | CODE | |
| LOW | manage_envs.py | 662 | CODE | |
| LOW | tests/test_protocol_v2_on_cpu.py | 802 | CODE | |
| LOW | tests/test_protocol_on_cpu.py | 1092 | CODE | |
| LOW | …xperimental/reward_loop/run_determinism_e2e_with_rm.py | 65 | CODE | |
| LOW | tests/experimental/agent_loop/test_multi_modal.py | 83 | CODE | |
| LOW | …mental/agent_loop/test_audio_server_contract_on_cpu.py | 31 | CODE | |
| LOW | tests/special_sanity/check_pr_description.py | 40 | CODE | |
| LOW | tests/special_sanity/validate_imported_docs.py | 63 | CODE | |
| LOW | tests/special_sanity/check_uv_gpu_only.py | 91 | CODE | |
| LOW | tests/special_sanity/test_config_docs.py | 19 | CODE | |
| LOW | tests/special_sanity/type_coverage_check.py | 34 | CODE | |
| LOW | tests/special_sanity/type_coverage_check.py | 87 | CODE | |
| LOW | tests/utils/test_continuous_token_on_cpu.py | 397 | CODE | |
| LOW | tests/utils/test_continuous_token_on_cpu.py | 2110 | CODE | |
| LOW | tests/utils/test_continuous_token_on_cpu.py | 2180 | CODE | |
| LOW | tests/utils/test_linear_cross_entropy.py | 113 | CODE | |
| LOW | tests/utils/test_special_linear_cross_entropy_tp.py | 207 | CODE | |
| LOW | tests/utils/test_fsdp_lora_merge.py | 37 | CODE | |
| LOW | tests/utils/test_fsdp_lora_merge.py | 225 | CODE | |
| LOW | tests/utils/test_turn_separator_on_cpu.py | 117 | CODE | |
| LOW | tests/utils/test_turn_separator_on_cpu.py | 62 | CODE | |
| LOW | tests/utils/test_turn_separator_on_cpu.py | 138 | CODE | |
| LOW | tests/utils/test_special_megatron_kl_loss_tp.py | 55 | CODE | |
| LOW | …sts/utils/dataset/test_multiturn_sft_dataset_on_cpu.py | 245 | CODE | |
| LOW | …ts/utils/veomni/test_special_export_unfused_experts.py | 58 | CODE | |
| LOW | tests/utils/megatron/test_optimizer_offload_and_load.py | 76 | CODE | |
| LOW | …eward_score/reward_score/test_sandbox_fusion_on_cpu.py | 127 | CODE | |
| LOW | …s/special_distributed/test_mcore_probe_differential.py | 49 | CODE | |
| LOW | …s/special_distributed/test_mcore_probe_differential.py | 174 | CODE | |
| LOW | …s/special_distributed/test_mcore_probe_differential.py | 290 | CODE | |
| LOW | tests/special_distributed/test_tensor_dict.py | 32 | CODE | |
| LOW | tests/special_distributed/test_sglang_delta_loop.py | 41 | CODE | |
| LOW | tests/workers/test_engine_detach_tree_on_cpu.py | 121 | CODE | |
| LOW | …ts/workers/rollout/rollout_trtllm/test_trtllm_abort.py | 32 | CODE | |
| LOW | tests/workers/rollout/rollout_vllm/test_vllm_abort.py | 29 | CODE | |
| LOW | …kpoint_engine/test_fsdp_lora_only_checkpoint_on_cpu.py | 286 | CODE | |
| LOW | tests/checkpoint_engine/test_block_placement.py | 412 | CODE | |
| LOW | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 162 | CODE | |
| LOW | verl/protocol.py | 543 | CODE | |
| LOW | verl/protocol.py | 914 | CODE | |
| LOW | …rl/experimental/fully_async_policy/fully_async_main.py | 186 | CODE | |
| LOW | verl/experimental/fully_async_policy/detach_utils.py | 270 | CODE | |
| LOW | …perimental/fully_async_policy/fully_async_rollouter.py | 902 | CODE | |
| LOW | …l/fully_async_policy/unittest/simple_streaming_demo.py | 85 | CODE | |
| LOW | verl/experimental/reward_loop/router/naive_router.py | 137 | CODE | |
| LOW | …experimental/reward_loop/router/inner_sglang_router.py | 30 | CODE | |
| LOW | verl/experimental/agent_loop/tool_parser.py | 211 | CODE | |
| LOW | verl/experimental/agent_loop/tool_parser.py | 765 | CODE | |
| LOW | verl/experimental/agent_loop/tool_parser.py | 224 | CODE | |
| LOW | verl/experimental/agent_loop/tool_agent_loop.py | 131 | CODE | |
| LOW | verl/experimental/agent_loop/tool_agent_loop.py | 310 | CODE | |
| LOW | verl/experimental/separation/ray_trainer.py | 502 | CODE | |
| LOW | verl/plugin/platform/platform_rocm.py | 30 | CODE | |
| LOW | verl/plugin/platform/platform_cuda.py | 41 | CODE | |
| LOW | verl/model_merger/megatron_model_merger.py | 108 | CODE | |
| LOW | verl/model_merger/base_model_merger.py | 220 | CODE | |
| LOW | verl/model_merger/fsdp_model_merger.py | 167 | CODE | |
| LOW | verl/utils/torch_functional.py | 514 | CODE | |
| LOW | verl/utils/torch_functional.py | 1012 | CODE | |
| 204 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | docs/data/transfer_queue.md | 243 | Sample a batch of indices from the ready indices. Args: ready_indexes: List of global indices for which all | STRING |
| HIGH | verl/base_config.py | 56 | Implement the [] operator for the class. Allows accessing attributes like dictionary items. Args: k | STRING |
| HIGH | verl/protocol.py | 673 | Slice the DataProto and return a new DataProto object. This is an improved version of direct slicing wh | STRING |
| HIGH | verl/experimental/fully_async_policy/detach_utils.py | 87 | Assemble gen_batch_output from RolloutSample objects Assembles batches from RolloutSample objects, similar to t | STRING |
| HIGH | …perimental/fully_async_policy/fully_async_rollouter.py | 1350 | Block until the message queue contains at least ``required_count`` samples. Polls ``message_queue_client.get_qu | STRING |
| HIGH | verl/experimental/reward_loop/reward_manager/limited.py | 91 | Acquire tokens from the bucket, waiting if necessary. This method will block (using asyncio.sleep) until suffic | STRING |
| HIGH | verl/tools/base_tool.py | 63 | Execute the tool. Args: instance_id: The instance id of the tool. parameters: The json | STRING |
| HIGH | verl/utils/seqlen_balancing.py | 214 | Calculates partitions of indices from seqlen_list such that the sum of sequence lengths in each partition is ba | STRING |
| HIGH | verl/utils/torch_functional.py | 113 | Compute log-probabilities using Flash Attention's optimized cross-entropy. Uses the Flash Attention library's Trito | STRING |
| HIGH | verl/utils/py_functional.py | 57 | Decorator to add a timeout to a function. Args: seconds: The timeout duration in seconds. use_ | STRING |
| HIGH | verl/utils/device.py | 297 | Check if the given software and CANN versions support IPC. Compares the software version and CANN toolkit version a | STRING |
| HIGH | verl/utils/model.py | 639 | Extract LoRA rank from adapter configuration file. Args: adapter_path: Path to LoRA adapter directory | STRING |
| HIGH | verl/utils/import_utils.py | 92 | Resolve agent loop configuration file path. In multi-node Ray training, relative paths may not resolve correctly | STRING |
| HIGH | verl/utils/import_utils.py | 265 | Load a class from its fully qualified name. Args: fqn: Fully qualified class name (e.g., 'mypackage.module. | STRING |
| HIGH | verl/utils/tensordict_utils.py | 186 | Concatenate multiple nested tensors along the batch dimension. Takes a list of nested tensors with jagged layout an | STRING |
| HIGH | verl/utils/tensordict_utils.py | 222 | Handle concatenation of TensorDicts with empty batch size. For TensorDicts that contain only metadata (NonTensorDat | STRING |
| HIGH | verl/utils/tensordict_utils.py | 248 | Concatenate multiple TensorDicts along dimension zero. Combines a list of TensorDicts into a single TensorDict by c | STRING |
| HIGH | verl/utils/tensordict_utils.py | 303 | Split a TensorDict into equal-sized chunks with special nested tensor handling. Divides a TensorDict into the speci | STRING |
| HIGH | verl/utils/tensordict_utils.py | 459 | Select rows from a TensorDict using indices. Creates a new TensorDict containing only the rows specified by indices | STRING |
| HIGH | verl/utils/tensordict_utils.py | 514 | Merge two TensorDicts, adding keys from the second to the first. Performs an in-place union of two TensorDicts. Key | STRING |
| HIGH | verl/utils/tensordict_utils.py | 560 | Create an iterator that yields mini-batches from a TensorDict. Wraps a TensorDict in a DataLoader-style iterator th | STRING |
| HIGH | verl/utils/tensordict_utils.py | 699 | Extract a subset of keys from a TensorDict into a new TensorDict. Creates a new TensorDict containing only the spec | STRING |
| HIGH | verl/utils/tensordict_utils.py | 773 | Remove multiple keys from a TensorDict and return them as a new TensorDict. Removes the specified keys from the sou | STRING |
| HIGH | verl/utils/tensordict_utils.py | 814 | Pad a TensorDict's batch dimension to be divisible by a given divisor. If the TensorDict's length is not evenly div | STRING |
| HIGH | verl/utils/fsdp_utils.py | 453 | Get the full state dict from an FSDP model. Args: model (torch.nn.Module): The FSDP model to get state | STRING |
| HIGH | verl/utils/dataset/rl_dataset.py | 515 | split the dataset into num_splits sub-datasets Args: num_splits: specified number of splits | STRING |
| HIGH | verl/utils/tokenizer/continuous_token.py | 442 | Render messages with images through the processor. Unlike ``_render_tokens`` which uses only the tokenizer, thi | STRING |
| HIGH | verl/utils/kernel/fp8_kernel.py | 177 | High-performance FP8 blockwise quantization using Triton kernel. This is the recommended function to use for FP8 qu | STRING |
| HIGH | verl/utils/megatron/router_replay_utils.py | 606 | Reorder and merge per-VPP layer blocks into a contiguous layer dimension. Given a tensor shaped as [bs*vpp_siz | STRING |
| HIGH | verl/utils/reward_score/__init__.py | 29 | Compute the score for a given solution based on the data source. Args: data_source (str): The source datase | STRING |
| HIGH | verl/workers/config/optimizer.py | 299 | Build an optimizer based on the configuration. Dynamically imports and instantiates an optimizer class from the spe | STRING |
| HIGH | verl/workers/rollout/router.py | 39 | Acquire a server for the given request. Args: request_id: Request identifier for sticky session rou | STRING |
| HIGH | verl/workers/rollout/trtllm_rollout/trtllm_rollout.py | 178 | Make an async HTTP request with retry logic and consistent error handling. Args: endpoint (str): Th | STRING |
| HIGH | …l/workers/rollout/sglang_rollout/http_server_engine.py | 117 | Launch an SGLang HTTP server process and wait for it to be ready. This function starts a new process running an SGL | STRING |
| HIGH | …l/workers/rollout/sglang_rollout/http_server_engine.py | 303 | Make a HTTP request with retry logic and consistent error handling. Args: endpoint (str): The API e | STRING |
| HIGH | …l/workers/rollout/sglang_rollout/http_server_engine.py | 671 | Make an async HTTP request with retry logic and consistent error handling. Args: endpoint (str): Th | STRING |
| HIGH | verl/workers/engine/base.py | 433 | Function to create a new training engine instance based on the provided config. Args: key: | STRING |
| HIGH | verl/workers/engine/fsdp/utils.py | 57 | Determine the appropriate sharding strategy based on the number of dimensions of the device mesh. Args: | STRING |
| HIGH | verl/workers/engine/torchtitan/utils.py | 81 | Derive torchtitan model name and flavor from a HuggingFace config. The name is mapped from ``hf_config.model_type`` | STRING |
| HIGH | verl/trainer/ppo/reward.py | 51 | Load and return a custom reward function from external file. Dynamically imports a reward function from a specified | STRING |
| HIGH | verl/trainer/ppo/core_algos.py | 194 | Factory function to create appropriate KL controller based on configuration. Args: kl_ctrl: Configuration o | STRING |
| HIGH | verl/trainer/ppo/core_algos.py | 2272 | Compute importance weights for resampling based on scores. Args: scores (torch.Tensor): Tensor of s | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | docs/blog/v0.7.md | 0 | send the weights of the model. args: weights: a generator that yields the name of the weight tensor and the tensor itsel | STRING |
| HIGH | verl/checkpoint_engine/nccl_checkpoint_engine.py | 0 | send the weights of the model. args: weights: a generator that yields the name of the weight tensor and the tensor itsel | STRING |
| HIGH | verl/checkpoint_engine/nixl_checkpoint_engine.py | 0 | send the weights of the model. args: weights: a generator that yields the name of the weight tensor and the tensor itsel | STRING |
| HIGH | verl/checkpoint_engine/kimi_checkpoint_engine.py | 0 | send the weights of the model. args: weights: a generator that yields the name of the weight tensor and the tensor itsel | STRING |
| HIGH | verl/checkpoint_engine/hccl_checkpoint_engine.py | 0 | send the weights of the model. args: weights: a generator that yields the name of the weight tensor and the tensor itsel | STRING |
| HIGH | docs/blog/v0.7.md | 0 | receive the weights of the model. yields: a tuple of the name of the weight tensor and the tensor itself. | STRING |
| HIGH | verl/checkpoint_engine/nccl_checkpoint_engine.py | 0 | receive the weights of the model. yields: a tuple of the name of the weight tensor and the tensor itself. | STRING |
| HIGH | verl/checkpoint_engine/nixl_checkpoint_engine.py | 0 | receive the weights of the model. yields: a tuple of the name of the weight tensor and the tensor itself. | STRING |
| HIGH | verl/checkpoint_engine/kimi_checkpoint_engine.py | 0 | receive the weights of the model. yields: a tuple of the name of the weight tensor and the tensor itself. | STRING |
| HIGH | verl/checkpoint_engine/hccl_checkpoint_engine.py | 0 | receive the weights of the model. yields: a tuple of the name of the weight tensor and the tensor itself. | STRING |
| HIGH | …perimental/fully_async_policy/fully_async_rollouter.py | 0 | initialize distributed training workers using ray backend. creates: 1. ray resource pools from configuration 2. worker g | STRING |
| HIGH | …experimental/fully_async_policy/fully_async_trainer.py | 0 | initialize distributed training workers using ray backend. creates: 1. ray resource pools from configuration 2. worker g | STRING |
| HIGH | verl/experimental/separation/ray_trainer.py | 0 | initialize distributed training workers using ray backend. creates: 1. ray resource pools from configuration 2. worker g | STRING |
| HIGH | verl/trainer/ppo/ray_trainer.py | 0 | initialize distributed training workers using ray backend. creates: 1. ray resource pools from configuration 2. worker g | STRING |
| HIGH | …experimental/fully_async_policy/fully_async_trainer.py | 0 | the training loop of ppo. the driver process only need to call the compute functions of the worker group through rpc to | STRING |
| HIGH | verl/experimental/one_step_off_policy/ray_trainer.py | 0 | the training loop of ppo. the driver process only need to call the compute functions of the worker group through rpc to | STRING |
| HIGH | verl/trainer/ppo/ray_trainer.py | 0 | the training loop of ppo. the driver process only need to call the compute functions of the worker group through rpc to | STRING |
| HIGH | …experimental/fully_async_policy/fully_async_trainer.py | 0 | single-step training template method. handles all logic for one training step. flow: 1. pre-step processing -> 2. get ba | STRING |
| HIGH | verl/experimental/separation/ray_trainer.py | 0 | single-step training template method. handles all logic for one training step. flow: 1. pre-step processing -> 2. get ba | STRING |
| HIGH | verl/experimental/one_step_off_policy/ray_trainer.py | 0 | single-step training template method. handles all logic for one training step. flow: 1. pre-step processing -> 2. get ba | STRING |
| HIGH | verl/workers/reward_manager/prime.py | 0 | we will expand this function gradually based on the available datasets | STRING |
| HIGH | verl/workers/reward_manager/naive.py | 0 | we will expand this function gradually based on the available datasets | STRING |
| HIGH | verl/workers/reward_manager/dapo.py | 0 | we will expand this function gradually based on the available datasets | STRING |
| HIGH | …/workers/rollout/trtllm_rollout/trtllm_async_server.py | 0 | restore kv_cache gpu memory after a weight sync. counterpart to release_kv_cache(). | STRING |
| HIGH | …/workers/rollout/sglang_rollout/async_sglang_server.py | 0 | restore kv_cache gpu memory after a weight sync. counterpart to release_kv_cache(). | STRING |
| HIGH | verl/workers/rollout/vllm_rollout/vllm_async_server.py | 0 | restore kv_cache gpu memory after a weight sync. counterpart to release_kv_cache(). | STRING |
| HIGH | verl/workers/engine/base.py | 0 | whether the current rank is the first rank in model parallel group that contains model outputs | STRING |
| HIGH | verl/workers/engine/veomni/transformer_impl.py | 0 | whether the current rank is the first rank in model parallel group that contains model outputs | STRING |
| HIGH | verl/workers/engine/torchtitan/transformer_impl.py | 0 | whether the current rank is the first rank in model parallel group that contains model outputs | STRING |
| HIGH | verl/workers/engine/veomni/transformer_impl.py | 0 | move model parameters, optimizer states, or both to the specified device. note that this function executes irrespective | STRING |
| HIGH | verl/workers/engine/mindspeed/transformer_impl.py | 0 | move model parameters, optimizer states, or both to the specified device. note that this function executes irrespective | STRING |
| HIGH | verl/workers/engine/megatron/transformer_impl.py | 0 | move model parameters, optimizer states, or both to the specified device. note that this function executes irrespective | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tests/special_e2e/envs/__init__.py | 17 | __all__ = ["DigitCompletion"] | CODE |
| LOW | tests/special_e2e/envs/digit_completion/__init__.py | 22 | __all__ = ["DigitCompletion", "generate_ground_truth_response", "CharTokenizer"] | CODE |
| LOW | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 127 | def _set_prompt_status(partition_id: str, uid: str, status: str, global_steps: int) -> None: | CODE |
| LOW | verl/protocol.py | 43 | __all__ = ["DataProto", "union_tensor_dict"] | CODE |
| LOW | verl/__init__.py | 46 | __all__ = ["DataProto", "__version__"] | CODE |
| LOW | verl/experimental/teacher_loop/__init__.py | 17 | __all__ = ["MultiTeacherModelManager"] | CODE |
| LOW | …rl/experimental/fully_async_policy/fully_async_main.py | 159 | def _setup_hybrid_worker_group(self, config) -> None: | CODE |
| LOW | verl/experimental/fully_async_policy/message_queue.py | 23 | logger = logging.getLogger(__name__) | CODE |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 44 | logger = logging.getLogger(__name__) | CODE |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 302 | async def _setup_dynamic_resource_controller(self) -> None: | CODE |
| LOW | …ly_async_policy/dynamic_schedule/fixed_ratio_policy.py | 51 | def update_after_step(self, global_steps: int, ctx: DynamicScheduleContext) -> None: | CODE |
| LOW | …mental/fully_async_policy/dynamic_schedule/__init__.py | 20 | __all__ = [ | CODE |
| LOW | …policy/dynamic_schedule/dynamic_resource_controller.py | 41 | __all__ = [ | CODE |
| LOW | …c_policy/dynamic_schedule/static_fully_async_policy.py | 48 | def update_after_step(self, global_steps: int, ctx: DynamicScheduleContext) -> None: | CODE |
| LOW | …/fully_async_policy/dynamic_schedule/default_policy.py | 92 | def update_after_step(self, global_steps: int, ctx: DynamicScheduleContext) -> None: | CODE |
| LOW | …perimental/fully_async_policy/dynamic_schedule/base.py | 142 | def update_after_step(self, global_steps: int, ctx: DynamicScheduleContext) -> None: # noqa: B027 | CODE |
| LOW | verl/experimental/reward_loop/__init__.py | 18 | __all__ = ["RewardModelManager", "RewardLoopWorker", "RewardLoopManager", "migrate_legacy_reward_impl"] | CODE |
| LOW | …rl/experimental/reward_loop/reward_manager/registry.py | 19 | __all__ = ["register", "get_reward_manager_cls"] | CODE |
| LOW | …rl/experimental/reward_loop/reward_manager/__init__.py | 22 | __all__ = [ | CODE |
| LOW⚡ | verl/experimental/reward_loop/router/naive_router.py | 31 | logger = logging.getLogger(__name__) | CODE |
| LOW | …experimental/reward_loop/router/inner_sglang_router.py | 26 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/experimental/agent_loop/__init__.py | 27 | __all__ = [ | CODE |
| LOW | verl/experimental/separation/engine_workers.py | 33 | __all__ = ["DetachActorWorker"] | CODE |
| LOW | verl/plugin/platform/__init__.py | 19 | __all__ = ["PlatformBase", "PlatformRegistry", "get_platform", "set_platform"] | CODE |
| LOW | verl/plugin/platform/platform_manager.py | 23 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/plugin/platform/platform_manager.py | 150 | def set_platform(platform: PlatformBase) -> None: | CODE |
| LOW⚡ | verl/plugin/platform/platform_cuda.py | 78 | def set_device(self, device_index: int) -> None: | CODE |
| LOW⚡ | verl/plugin/platform/platform_cuda.py | 98 | def set_allocator_settings(self, settings: str) -> None: | CODE |
| LOW⚡ | verl/plugin/platform/platform_npu.py | 80 | def set_device(self, device_index: int) -> None: | CODE |
| LOW⚡ | verl/plugin/platform/platform_npu.py | 100 | def set_allocator_settings(self, settings: str) -> None: | CODE |
| LOW | verl/plugin/platform/platform_npu.py | 15 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | verl/plugin/platform/platform_base.py | 109 | def set_device(self, device_index: int) -> None: | CODE |
| LOW⚡ | verl/plugin/platform/platform_base.py | 137 | def set_allocator_settings(self, settings: str) -> None: | CODE |
| LOW | verl/utils/transferqueue_utils.py | 68 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/megatron_peft_utils.py | 82 | __all__ = [ | CODE |
| LOW | verl/utils/config.py | 20 | __all__ = ["omega_conf_to_dataclass", "validate_config"] | CODE |
| LOW⚡ | verl/utils/device.py | 29 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | verl/utils/device.py | 137 | def set_expandable_segments(enable: bool) -> None: | CODE |
| LOW | verl/utils/groupwise.py | 51 | __all__ = ["as_torch_index", "group_mean_std"] | CODE |
| LOW | verl/utils/fs.py | 29 | __all__ = ["copy", "exists", "makedirs"] | CODE |
| LOW | verl/utils/tracking.py | 32 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/attention_utils.py | 102 | __all__ = ["index_first_axis", "pad_input", "rearrange", "unpad_input"] | CODE |
| LOW | verl/utils/fsdp_utils.py | 40 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/fsdp_utils.py | 820 | def set_reshard_after_forward(module: FSDPModule, reshard_after_forward: bool, recurse: bool = True) -> None: | CODE |
| LOW | verl/utils/checkpoint/__init__.py | 17 | __all__ = ["CheckpointHandler", "OrchestrationMode"] | CODE |
| LOW | verl/utils/qat/linear.py | 27 | __all__ = ["QATLinear", "QATMode"] | CODE |
| LOW | verl/utils/qat/__init__.py | 46 | __all__ = [ | CODE |
| LOW | verl/utils/qat/core.py | 27 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/qat/quantizer.py | 40 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/qat/quantizer.py | 306 | __all__ = [ | CODE |
| LOW | verl/utils/qat/vllm_patch.py | 35 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/qat/vllm_patch.py | 824 | __all__ = [ | CODE |
| LOW | verl/utils/logger/__init__.py | 25 | __all__ = [ | CODE |
| LOW | verl/utils/dataset/__init__.py | 18 | __all__ = ["RLHFDataset", "RMDataset"] | CODE |
| LOW | verl/utils/dataset/rl_dataset.py | 38 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/tokenizer/continuous_token_wiring.py | 41 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/tokenizer/continuous_token.py | 34 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/tokenizer/__init__.py | 23 | __all__ = [ | CODE |
| LOW | verl/utils/tokenizer/chat_template.py | 9 | logger = logging.getLogger(__name__) | CODE |
| LOW | verl/utils/tokenizer/tokenizer.py | 19 | __all__ = [ | CODE |
| 90 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tests/utils/test_activation_offload.py | 110 | # Step 1: Initial update and save checkpoint | COMMENT |
| LOW | tests/utils/test_activation_offload.py | 123 | # Step 2: Second update and forward pass | COMMENT |
| LOW⚡ | tests/utils/test_activation_offload.py | 135 | # Step 3: wrap module with activation offloading and load checkpoint | COMMENT |
| LOW⚡ | tests/utils/test_activation_offload.py | 139 | # Step 4: Repeat the second update with same input | COMMENT |
| LOW | tests/utils/test_activation_offload.py | 151 | # Step 4: Verify outputs match | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 104 | # Step 1: Initial update and save checkpoint | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 118 | # Step 2: Second update and forward pass | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 130 | # Step 3: Load checkpoint and repeat second update | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 149 | # Step 4: Verify outputs match | COMMENT |
| LOW | docs/advance/rl_insight.md | 17 | ## Step 1: Install and Start RL-Insight | COMMENT |
| LOW | docs/advance/rl_insight.md | 68 | ## Step 2: Enable RL-Insight in verl | COMMENT |
| LOW | docs/advance/rl_insight.md | 97 | ## Step 3: Monitor Rollout and TransferQueue Metrics | COMMENT |
| LOW | docs/advance/rl_insight.md | 111 | ## Step 4: Add Hardware Metrics (Optional) | COMMENT |
| LOW | docs/advance/dynamic_schedule.md | 413 | ### Step 1: Subclass the base class | COMMENT |
| LOW | docs/advance/dynamic_schedule.md | 460 | ### Step 2: Register the import | COMMENT |
| LOW | docs/advance/dynamic_schedule.md | 471 | ### Step 3: Reference in config | COMMENT |
| LOW | docs/advance/dynamic_schedule.md | 481 | ### Step 4: Use `DynamicScheduleContext` fields | COMMENT |
| LOW | docs/advance/grafana_prometheus.md | 39 | ### Step 1: Environment Variables and Start Ray Cluster | COMMENT |
| LOW | docs/advance/grafana_prometheus.md | 64 | ### Step 2: Start Grafana (Visualization Dashboard) | COMMENT |
| LOW | docs/advance/grafana_prometheus.md | 80 | ### Step 3: Start Prometheus (Metrics Collection) | COMMENT |
| LOW | docs/advance/grafana_prometheus.md | 154 | ### Step 6: View Metrics in Prometheus | COMMENT |
| LOW | docs/advance/grafana_prometheus.md | 169 | ### Step 7: View Metrics in Grafana | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1239 | # Step 1: Clear sticky cache so retried requests use least-loaded routing. | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1246 | # Step 2: Abort in-flight requests on all active replicas. | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1252 | # Step 3: Wait until release_server() from all aborted requests has actually | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1267 | # Step 4: Resume generation so retried requests can be accepted. | COMMENT |
| LOW⚡ | …experimental/fully_async_policy/fully_async_trainer.py | 730 | # Step 1: NCCL broadcast from trainer to standalone rollout replicas. | COMMENT |
| LOW⚡ | …experimental/fully_async_policy/fully_async_trainer.py | 737 | # Step 2: When dynamic resource scheduling is enabled, the Trainer GPUs | COMMENT |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 290 | # Step 3: Sleep all hybrid replicas | COMMENT |
| LOW | …rimental/fully_async_policy/dynamic_schedule/README.md | 413 | ### Step 1: Subclass the base class | COMMENT |
| LOW | …rimental/fully_async_policy/dynamic_schedule/README.md | 460 | ### Step 2: Register the import | COMMENT |
| LOW | …rimental/fully_async_policy/dynamic_schedule/README.md | 471 | ### Step 3: Reference in config | COMMENT |
| LOW | …rimental/fully_async_policy/dynamic_schedule/README.md | 481 | ### Step 4: Use `DynamicScheduleContext` fields | COMMENT |
| LOW | verl/experimental/reward_loop/reward_manager/limited.py | 269 | # NOTE: This class owns a *global*, class-level set of rate limiters. Once the class has been | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 401 | # Step 1: Patch TransformerConfig to include the feature flag | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 452 | # Step 2: Patch TopKRouter only once to ensure idempotency. | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 458 | # Step 3: Define the new __init__ method | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 465 | # Step 4: Apply the patches | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 414 | # Step 3: patch input for multimodal sequence parallelism | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 420 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 453 | # Step 2: patch input for multimodal sequence parallelism | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 459 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 474 | # Step 2: patch attention to support ulysses parallelism | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 481 | # Step 3: patch input for multimodal sequence parallelism | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 366 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 395 | # Step 2: patch attention to support ulysses parallelism | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 501 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 537 | # Step 2: patch vision model to fix fsdp2 cpu_offload bug. | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 848 | # Step 1: Compute log ratio (log(π_train / π_rollout)) | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 852 | # Step 2: Compute IS weights (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 880 | # Step 4: Compute off-policy metrics (KL, PPL, χ², etc.) | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 888 | # Step 6: Add "rollout_corr/" prefix to all metrics for logging consistency | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 896 | # Step 7: Wrap IS weights in DataProto for consistency with API | COMMENT |
| LOW | verl/trainer/ppo/rollout_corr_helper.py | 864 | # Step 3: Compute rejection mask (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 926 | # Step 1: Compute w_per_timestep = 1 - 2π_t + Σπ²) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 930 | # Step 2: Apply rollout importance sampling correction (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 935 | # Step 3: Compute cumulative path-variance proxy: W_t = Σ_{j=1}^t w_j | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1042 | # Step 1: Compute w_per_timestep = 1 - 2π_t + Σπ²) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1046 | # Step 2: Apply rollout importance sampling correction (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1051 | # Step 3: Compute cumulative path-variance proxy: W_t = Σ_{j=1}^t w_j | COMMENT |
| 16 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | tests/test_protocol_v2_on_cpu.py | 434 | # Create a DataProto object with some batch and non-tensor data | COMMENT |
| MEDIUM | tests/test_protocol_v2_on_cpu.py | 629 | # Create a DataProto object with some batch and non-tensor data | COMMENT |
| MEDIUM | tests/test_protocol_on_cpu.py | 133 | # Create a circular reference and a non-circular one. | COMMENT |
| MEDIUM | tests/test_protocol_on_cpu.py | 371 | # Create a DataProto object with some batch and non-tensor data | COMMENT |
| MEDIUM | tests/test_protocol_on_cpu.py | 618 | # Create a DataProto object with some batch and non-tensor data | COMMENT |
| MEDIUM | tests/test_protocol_on_cpu.py | 1244 | # Create a DataProto with an empty TensorDict but with a batch size | COMMENT |
| MEDIUM | tests/experimental/agent_loop/test_multi_modal.py | 94 | # Create a simple colored image for testing | COMMENT |
| MEDIUM | tests/experimental/agent_loop/test_multi_modal.py | 339 | # Create a simple test image | COMMENT |
| MEDIUM | tests/experimental/agent_loop/test_basic_agent_loop.py | 398 | # Initialize the class to set up class-level attributes | COMMENT |
| MEDIUM | tests/utils/test_nvtx_profile.py | 135 | # Create a new ProfilerConfig instance | COMMENT |
| MEDIUM | tests/utils/test_model_on_cpu.py | 35 | # Create a fresh mock config object for each test case | COMMENT |
| MEDIUM⚡ | tests/utils/test_import_utils_on_cpu.py | 84 | # Create a temporary file with syntax errors | COMMENT |
| MEDIUM | tests/utils/test_special_linear_cross_entropy_tp.py | 273 | # Create a tensor to hold the gathered weights from all ranks | COMMENT |
| MEDIUM | tests/utils/test_special_linear_cross_entropy_tp.py | 277 | # Create a single contiguous tensor to hold all gathered weights | COMMENT |
| MEDIUM | …sts/utils/dataset/test_multiturn_sft_dataset_on_cpu.py | 49 | # Create a temporary parquet file with test data | COMMENT |
| MEDIUM⚡ | …eward_score/reward_score/test_sandbox_fusion_on_cpu.py | 338 | # Define a low global concurrency limit to test the semaphore's effect | STRING |
| MEDIUM⚡ | …eward_score/reward_score/test_sandbox_fusion_on_cpu.py | 340 | # Define the number of processes used in the test | STRING |
| MEDIUM⚡ | …eward_score/reward_score/test_sandbox_fusion_on_cpu.py | 342 | # Define the number of tasks processed by check_correctness in each process (i.e., internal | STRING |
| MEDIUM | …eward_score/reward_score/test_sandbox_fusion_on_cpu.py | 470 | # Create a multiprocessing.Semaphore instance, this is the global semaphore we are testing. | STRING |
| MEDIUM | …kers/rollout/rollout_sglang/test_http_server_engine.py | 17 | # This file is adapted from multiple sources: | COMMENT |
| MEDIUM | tests/single_controller/test_decorator_on_cpu.py | 38 | # Define a simple worker for testing | COMMENT |
| MEDIUM | tests/trainer/ppo/test_metric_utils_on_cpu.py | 297 | # Create a mock DataProto object | COMMENT |
| MEDIUM | tests/trainer/ppo/test_metric_utils_on_cpu.py | 358 | # Create a mock DataProto object | COMMENT |
| MEDIUM | tests/trainer/ppo/test_metric_utils_on_cpu.py | 435 | # Create a mock DataProto object | COMMENT |
| MEDIUM | verl/protocol.py | 700 | # Create a slice object | COMMENT |
| MEDIUM | …perimental/fully_async_policy/fully_async_rollouter.py | 1093 | # Create the main asynchronous task | COMMENT |
| MEDIUM⚡ | verl/experimental/one_step_off_policy/ray_trainer.py | 221 | # Create the initial batch from the data loader | COMMENT |
| MEDIUM | verl/utils/megatron_utils.py | 1096 | # Create a dummy total_loss_dict to collect MTP metrics | COMMENT |
| MEDIUM | verl/utils/tensordict_utils.py | 281 | # Create a list of tensordicts containing only non-nested tensors for concatenation | COMMENT |
| MEDIUM | verl/utils/tensordict_utils.py | 287 | # Create a new TensorDict with non-nested items without modifying the original | COMMENT |
| MEDIUM | verl/utils/tracking.py | 708 | # Create a new table with same columns and existing data | COMMENT |
| MEDIUM | verl/utils/vllm/vllm_fp8_utils.py | 330 | """This function is used to process the weights after loading for a Linear layer, it is used for vllm v0.18-v0.19. | STRING |
| MEDIUM | verl/utils/debug/performance.py | 16 | # This file is deprecated, for new features please develop in profiler/performance.py | COMMENT |
| MEDIUM | verl/workers/rollout/trtllm_rollout/trtllm_rollout.py | 153 | # Create a new session for each request to avoid resource competition | COMMENT |
| MEDIUM | …l/workers/rollout/sglang_rollout/http_server_engine.py | 17 | # This file is adapted from multiple sources: | COMMENT |
| MEDIUM | …l/workers/rollout/sglang_rollout/http_server_engine.py | 646 | # Create a new session for each request to avoid resource competition | COMMENT |
| MEDIUM | verl/trainer/main_ppo.py | 33 | # Define a function to run the PPO-like training process | COMMENT |
| MEDIUM | verl/trainer/main_ppo.py | 76 | # Create a remote instance of the TaskRunner class, and | COMMENT |
| MEDIUM | verl/trainer/main_ppo.py | 96 | # This file is used for performance analysis | COMMENT |
| MEDIUM | verl/trainer/ppo/reward.py | 140 | # Create a semaphore to control concurrent access to the sandbox | COMMENT |
| MEDIUM | .github/workflows/schedule_doc_translate_ascend.yaml | 15 | # This file is a part of the verl project. | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tests/utils/test_activation_offload.py | 110 | # Step 1: Initial update and save checkpoint | COMMENT |
| LOW | tests/utils/test_activation_offload.py | 123 | # Step 2: Second update and forward pass | COMMENT |
| LOW⚡ | tests/utils/test_activation_offload.py | 135 | # Step 3: wrap module with activation offloading and load checkpoint | COMMENT |
| LOW⚡ | tests/utils/test_activation_offload.py | 139 | # Step 4: Repeat the second update with same input | COMMENT |
| LOW | tests/utils/test_activation_offload.py | 151 | # Step 4: Verify outputs match | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 104 | # Step 1: Initial update and save checkpoint | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 118 | # Step 2: Second update and forward pass | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 130 | # Step 3: Load checkpoint and repeat second update | COMMENT |
| LOW | tests/special_distributed/test_fsdp_ckpt.py | 149 | # Step 4: Verify outputs match | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1239 | # Step 1: Clear sticky cache so retried requests use least-loaded routing. | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1246 | # Step 2: Abort in-flight requests on all active replicas. | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1252 | # Step 3: Wait until release_server() from all aborted requests has actually | COMMENT |
| LOW⚡ | …perimental/fully_async_policy/fully_async_rollouter.py | 1267 | # Step 4: Resume generation so retried requests can be accepted. | COMMENT |
| LOW⚡ | …experimental/fully_async_policy/fully_async_trainer.py | 730 | # Step 1: NCCL broadcast from trainer to standalone rollout replicas. | COMMENT |
| LOW⚡ | …experimental/fully_async_policy/fully_async_trainer.py | 737 | # Step 2: When dynamic resource scheduling is enabled, the Trainer GPUs | COMMENT |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 290 | # Step 3: Sleep all hybrid replicas | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 401 | # Step 1: Patch TransformerConfig to include the feature flag | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 452 | # Step 2: Patch TopKRouter only once to ensure idempotency. | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 458 | # Step 3: Define the new __init__ method | COMMENT |
| LOW⚡ | verl/utils/megatron/router_replay_patch.py | 465 | # Step 4: Apply the patches | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 414 | # Step 3: patch input for multimodal sequence parallelism | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 420 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 453 | # Step 2: patch input for multimodal sequence parallelism | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 459 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 474 | # Step 2: patch attention to support ulysses parallelism | COMMENT |
| LOW⚡ | verl/models/transformers/monkey_patch.py | 481 | # Step 3: patch input for multimodal sequence parallelism | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 366 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 395 | # Step 2: patch attention to support ulysses parallelism | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 501 | # Step 1: patch model to support image-text mixed data | COMMENT |
| LOW | verl/models/transformers/monkey_patch.py | 537 | # Step 2: patch vision model to fix fsdp2 cpu_offload bug. | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 848 | # Step 1: Compute log ratio (log(π_train / π_rollout)) | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 852 | # Step 2: Compute IS weights (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 880 | # Step 4: Compute off-policy metrics (KL, PPL, χ², etc.) | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 888 | # Step 6: Add "rollout_corr/" prefix to all metrics for logging consistency | COMMENT |
| LOW⚡ | verl/trainer/ppo/rollout_corr_helper.py | 896 | # Step 7: Wrap IS weights in DataProto for consistency with API | COMMENT |
| LOW | verl/trainer/ppo/rollout_corr_helper.py | 864 | # Step 3: Compute rejection mask (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 926 | # Step 1: Compute w_per_timestep = 1 - 2π_t + Σπ²) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 930 | # Step 2: Apply rollout importance sampling correction (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 935 | # Step 3: Compute cumulative path-variance proxy: W_t = Σ_{j=1}^t w_j | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1042 | # Step 1: Compute w_per_timestep = 1 - 2π_t + Σπ²) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1046 | # Step 2: Apply rollout importance sampling correction (if enabled) | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1051 | # Step 3: Compute cumulative path-variance proxy: W_t = Σ_{j=1}^t w_j | COMMENT |
| LOW⚡ | verl/trainer/ppo/core_algos.py | 1055 | # Step 4: Concatenate returns and w_cumulative for each trajectory | COMMENT |
| LOW⚡ | examples/data_preprocess/tinyllava_video_r1.py | 23 | # Step 1: Download | STRING |
| LOW⚡ | examples/data_preprocess/tinyllava_video_r1.py | 28 | # Step 2: Extract videos | STRING |
| LOW⚡ | examples/data_preprocess/tinyllava_video_r1.py | 31 | # Step 3: Preprocess | STRING |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 121 | # Step 1: Checkout the fork private repo's main branch. | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 135 | # Step 2: Sync the fork's main with the latest upstream verl source | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 157 | # Step 3: Setup Python and install dependencies. | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 168 | # Step 4: Clean stale translation memory for full translation. | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 178 | # Step 5: Detect whether to translate (full vs incremental). | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 198 | # Step 6: Translate via LLM API. | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 257 | # Step 7: Process translated files. | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 309 | # Step 8: Create clean branch and commit. | COMMENT |
| LOW⚡ | .github/workflows/schedule_doc_translate_ascend.yaml | 381 | # Step 9: Create Pull Request to the upstream verl source repo. | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tests/test_protocol_on_cpu.py | 1073 | # Check if it's a nested tensor | COMMENT |
| LOW | tests/experimental/agent_loop/test_multi_modal.py | 430 | # Check if this sample should have image pads (samples with index 1 and 2 in each repeat have images) | COMMENT |
| LOW | tests/special_sanity/test_config_docs.py | 35 | # Check if there's a comment above | COMMENT |
| LOW | tests/utils/test_special_linear_cross_entropy_tp.py | 494 | # Check if running with torchrun (distributed mode) | COMMENT |
| LOW | tests/models/test_tiled_mlp_accuracy.py | 95 | # Check if within tolerance (1e-2 for bf16) | COMMENT |
| LOW | verl/protocol.py | 386 | # Check if batch is empty to avoid torch.cat error in consolidate | COMMENT |
| LOW | …perimental/fully_async_policy/fully_async_rollouter.py | 877 | # Check if have reached the last step | COMMENT |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 816 | # Check if validation is needed | COMMENT |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 878 | # Check if the ESI (Elastic Server Instance)/training plan is close to expiration. | COMMENT |
| LOW | …experimental/fully_async_policy/fully_async_trainer.py | 883 | # Check if the conditions for saving a checkpoint are met. | COMMENT |
| LOW | verl/experimental/reward_loop/reward_manager/limited.py | 267 | # Check if already initialized before calling parent. | COMMENT |
| LOW | verl/experimental/separation/ray_trainer.py | 688 | # Check if the ESI (Elastic Server Instance)/training plan is close to expiration. | COMMENT |
| LOW | verl/experimental/separation/ray_trainer.py | 693 | # Check if the conditions for saving a checkpoint are met. | COMMENT |
| LOW | verl/utils/transferqueue_utils.py | 245 | # Check if collect_fn is a functools.partial and handle gracefully | COMMENT |
| LOW | verl/utils/fp8_utils.py | 56 | # Check if matches exclude patterns | COMMENT |
| LOW | verl/utils/fp8_utils.py | 76 | # Check if matches include patterns | COMMENT |
| LOW | verl/utils/fp8_utils.py | 106 | # Check if quantization is needed | COMMENT |
| LOW | verl/utils/fsdp_utils.py | 1038 | # Check if module is a PEFT model | COMMENT |
| LOW | verl/utils/qat/vllm_patch.py | 130 | # Check if we have metadata for this layer | COMMENT |
| LOW | verl/utils/qat/vllm_patch.py | 138 | # Check if this param needs rebuild | COMMENT |
| LOW | verl/utils/qat/vllm_patch.py | 182 | # Check if can rebuild from metadata | COMMENT |
| LOW | verl/utils/dataset/multiturn_sft_dataset.py | 334 | # Check if all tensors in the list have the same shape | COMMENT |
| LOW | verl/utils/dataset/rl_dataset.py | 580 | # Check if a custom dataset class is specified in the data configuration | COMMENT |
| LOW | verl/utils/kernel/fp8_kernel.py | 22 | # Check if Triton is available | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 272 | # Check if GPTModel has MTP and find the layers | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 278 | # Check if any layer in the model is MultiTokenPredictionLayer | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 311 | # Check if GPTModel has MTP and find the layers | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 317 | # Check if any layer in the model is MultiTokenPredictionLayer | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 356 | # Check if GPTModel has MTP and find the layers | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 362 | # Check if any layer in the model is MultiTokenPredictionLayer | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 394 | # Check if GPTModel has MTP and find the layers | COMMENT |
| LOW | verl/models/mcore/mtp_patch.py | 400 | # Check if any layer in the model is MultiTokenPredictionLayer | COMMENT |
| LOW | …third_party/torch/distributed/checkpoint/state_dict.py | 479 | # Verify if the model_state_dict and optim_state_dict are valid. This API | COMMENT |
| LOW | verl/workers/rollout/router.py | 182 | # Check if server is still in the active pool | COMMENT |
| LOW | …kers/rollout/trtllm_rollout/trtllm_worker_extension.py | 153 | # Check if model supports partial loading and use appropriate strategy | COMMENT |
| LOW | verl/workers/rollout/vllm_rollout/utils.py | 401 | # Check if FP8 quantization is enabled and apply appropriate weight loading | COMMENT |
| LOW | verl/workers/engine/veomni/utils.py | 87 | # Check if this is a MultiOptimizer (for ep and non-ep parameters when ep+fsdp2 is enabled) | COMMENT |
| LOW | verl/workers/engine/veomni/utils.py | 107 | # Check if this is a MultiOptimizer (for ep and non-ep parameters when ep+fsdp2 is enabled) | COMMENT |
| LOW | verl/trainer/main_ppo.py | 56 | # Check if Ray is not initialized | COMMENT |
| LOW | verl/trainer/ppo/metric_utils.py | 708 | # Check if we have the necessary data (sum_pi_squared is required for W-score) | COMMENT |
| LOW | verl/trainer/ppo/ray_trainer.py | 266 | # Check if sum_pi_squared is available | COMMENT |
| LOW | verl/trainer/ppo/ray_trainer.py | 1692 | # Check if the ESI (Elastic Server Instance)/training plan is close to expiration. | COMMENT |
| LOW | verl/trainer/ppo/ray_trainer.py | 1697 | # Check if the conditions for saving a checkpoint are met. | COMMENT |
| LOW | examples/tutorial/skypilot/verl-ppo.yaml | 22 | # Check if the script exists and use absolute path | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| CRITICAL⚡ | tests/trainer/ppo/test_core_algos_on_cpu.py | 42 | verl.trainer.ppo.core_algos.ADV_ESTIMATOR_REGISTRY.clear() | CODE |
| CRITICAL⚡ | tests/trainer/ppo/test_core_algos_on_cpu.py | 50 | verl.trainer.ppo.core_algos.ADV_ESTIMATOR_REGISTRY.clear() | CODE |
| CRITICAL | verl/trainer/ppo/v1/trainer_base.py | 730 | sync_refill_failed_groups = bool(self.config.trainer.v1.sampler.get("sync_refill_failed_groups", False)) | CODE |
| CRITICAL | verl/trainer/ppo/v1/trainer_base.py | 1848 | self.config.actor_rollout_ref.actor.policy_loss.get("rollout_correction", {}).get("score_centering", False) | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | …perimental/agent_loop/test_qwen3_tool_parser_on_cpu.py | 55 | CODE | |
| LOW | tests/single_controller/test_worker_group_basics.py | 103 | CODE | |
| LOW | tests/special_e2e/envs/digit_completion/task.py | 136 | CODE | |
| LOW | tests/trainer/ppo/test_score_centering_on_cpu.py | 461 | CODE | |
| LOW | tests/trainer/ppo/v1/test_replay_buffer_on_cpu.py | 51 | CODE | |
| LOW | …ainer/ppo/v1/test_separate_async_step_switch_on_cpu.py | 85 | CODE | |
| LOW | verl/protocol.py | 1123 | CODE | |
| LOW | verl/protocol.py | 1192 | CODE | |
| LOW | verl/experimental/agent_loop/agent_loop.py | 676 | CODE | |
| LOW | verl/experimental/separation/ray_trainer.py | 59 | CODE | |
| LOW | verl/utils/torch_functional.py | 70 | CODE | |
| LOW | verl/utils/torch_functional.py | 645 | CODE | |
| LOW | verl/utils/torch_functional.py | 673 | CODE | |
| LOW | verl/utils/torch_functional.py | 1008 | CODE | |
| LOW | verl/utils/checkpoint/megatron_checkpoint_manager.py | 307 | CODE | |
| LOW | verl/utils/qat/linear.py | 345 | CODE | |
| LOW | verl/utils/dataset/rm_dataset.py | 41 | CODE | |
| LOW | verl/utils/megatron/tensor_parallel.py | 233 | CODE | |
| LOW | verl/utils/megatron/router_replay_utils.py | 652 | CODE | |
| LOW | verl/utils/reward_score/rlla.py | 248 | CODE | |
| LOW | verl/models/mcore/patch.py | 374 | CODE | |
| LOW | verl/models/mcore/model_forward_fused.py | 300 | CODE | |
| LOW | verl/models/mcore/model_forward.py | 116 | CODE | |
| LOW | verl/models/mcore/mtp_patch.py | 97 | CODE | |
| LOW | verl/models/transformers/dense_common.py | 29 | CODE | |
| LOW | verl/models/transformers/dense_common.py | 71 | CODE | |
| LOW | verl/models/transformers/dense_common.py | 139 | CODE | |
| LOW | verl/models/transformers/qwen3_5.py | 335 | CODE | |
| LOW | verl/models/transformers/npu_patch.py | 103 | CODE | |
| LOW | verl/third_party/torch/distributed/_state_dict_utils.py | 394 | CODE | |
| LOW | verl/third_party/torch/distributed/_state_dict_utils.py | 526 | CODE | |
| LOW | verl/workers/utils/padding.py | 141 | CODE | |
| LOW | verl/workers/rollout/hf_rollout.py | 51 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 221 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 429 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 592 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 883 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 848 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 885 | CODE | |
| LOW | …l/workers/rollout/sglang_rollout/http_server_engine.py | 964 | CODE | |
| LOW | verl/workers/engine/utils.py | 219 | CODE | |
| LOW | verl/single_controller/ray/base.py | 434 | CODE | |
| LOW | verl/single_controller/base/worker.py | 348 | CODE | |
| LOW | verl/trainer/ppo/metric_utils.py | 857 | CODE | |
| LOW | examples/data_preprocess/full_hh_rlhf.py | 118 | CODE | |
| LOW | examples/data_preprocess/gsm8k.py | 85 | CODE | |
| LOW | examples/data_preprocess/gsm8k_multiturn_w_tool.py | 106 | CODE | |
| LOW | examples/data_preprocess/openr1mm.py | 75 | CODE | |
| LOW | examples/data_preprocess/gsm8k_multiturn_sft.py | 79 | CODE | |
| LOW | examples/data_preprocess/math_dataset.py | 78 | CODE | |
| LOW | examples/data_preprocess/gsm8k_tool_agent_loop.py | 107 | CODE | |
| LOW | examples/data_preprocess/geo3k.py | 83 | CODE | |
| LOW | examples/data_preprocess/hellaswag.py | 81 | CODE | |
| LOW | examples/data_preprocess/geo3k_multiturn_w_tool.py | 102 | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | tests/test_protocol_on_cpu.py | 98 | # 4D object array to ensure deep recursion is robust | COMMENT |
| MEDIUM | …fully_async_policy/config/fully_async_ppo_trainer.yaml | 95 | # And it can be used in conjunction with other rollout_correction algorithms. | COMMENT |
| MEDIUM | …nc_policy/config/fully_async_ppo_megatron_trainer.yaml | 94 | # And it can be used in conjunction with other rollout_correction algorithms. | COMMENT |
| MEDIUM | …f_policy/config/one_step_off_ppo_megatron_trainer.yaml | 28 | # And it can be used in conjunction with other rollout_correction algorithms. | COMMENT |
| MEDIUM | …e_step_off_policy/config/one_step_off_ppo_trainer.yaml | 28 | # And it can be used in conjunction with other rollout_correction algorithms. | COMMENT |
| MEDIUM | verl/utils/vllm/vllm_fp8_utils.py | 427 | # robust than version string comparison, especially for dev/pre-release versions. | COMMENT |
| MEDIUM | verl/utils/reward_score/math_dapo.py | 14 | # Adapted from https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/hendrycks_math/utils.py | COMMENT |
| MEDIUM | verl/utils/reward_score/math_reward.py | 14 | # Adapted from https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/hendrycks_math/utils.py | COMMENT |
| MEDIUM | verl/utils/reward_score/math_reward.py | 31 | # string normalization from https://github.com/EleutherAI/lm-evaluation-harness/blob/master/lm_eval/tasks/hendrycks_math | COMMENT |
| LOW | …third_party/torch/distributed/checkpoint/state_dict.py | 1180 | >>> # if we simply call ddp_model.state_dict() and fsdp_model.state_dict(), | STRING |
| MEDIUM | verl/single_controller/ray/base.py | 216 | # that can utilize different WorkerGroup for differnt models | COMMENT |
| MEDIUM | verl/trainer/ppo/rollout_corr_helper.py | 669 | """Compute comprehensive metrics for truncated importance sampling weights. | STRING |
| MEDIUM | verl/trainer/ppo/utils.py | 156 | # Use a sampler to facilitate checkpoint resumption. | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | README.md | 292 | - [DART-GUI](https://github.com/Computer-use-agents/dart-gui): a decoupled agentic RL framework for Computer Use Agents, | CODE |
| MEDIUM | tests/utils/test_turn_separator_on_cpu.py | 39 | # Thinking model (e.g. Qwen3): assistant turns are wrapped in ``<think></think>`` scaffolding. The | COMMENT |
| MEDIUM | tests/utils/test_turn_separator_on_cpu.py | 41 | # the scaffold, which is why the helper probes user turns instead. | COMMENT |
| MEDIUM⚡ | tests/utils/test_turn_separator_on_cpu.py | 200 | # The template really does inject scaffolding, so a naive assistant-turn probe is wrong: this is | COMMENT |
| MEDIUM⚡ | tests/utils/test_turn_separator_on_cpu.py | 209 | # The real helper avoids the scaffold and recovers the true separator. | COMMENT |
| MEDIUM | verl/utils/tokenizer/continuous_token.py | 718 | # Gemma4's generation scaffold depends on the preceding message type and | COMMENT |
| MEDIUM | verl/utils/tokenizer/continuous_token.py | 1357 | # The processor template always appends the scaffold; strip it unless a | COMMENT |
| MEDIUM | …les/grpo_trainer/run_kimi_k2_6_glm5_1_megatron_lite.sh | 4 | # Megatron Lite is Megatron's agentic experimental path. Its upstream home is | COMMENT |
| MEDIUM | examples/grpo_trainer/run_qwen3_5_35b_megatron_lite.sh | 4 | # Megatron Lite is Megatron's agentic experimental path. Its upstream home is | COMMENT |
| MEDIUM | examples/sft/gsm8k/run_deepseek_v4_megatron_lite.sh | 4 | # Megatron Lite is Megatron's agentic experimental path. Its upstream home is | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | tests/special_sanity/check_uv_gpu_only.py | 25 | RAY=(ray_kwargs.ray_init.runtime_env.py_executable=null) | STRING |
| HIGH | verl/trainer/ppo/rollout_corr_helper.py | 922 | - kl: Direct KL divergence estimator KL(π_rollout || π_training) | STRING |
| HIGH | verl/trainer/ppo/checkpoint_callback.py | 69 | Returns a no-op :class:`CheckpointCallback` when the key is unset or null, so | STRING |
| HIGH | verl/trainer/ppo/v1/replay_buffer.py | 172 | "[V1ReplayBuffer] off-policy version control disabled (max_off_policy_threshold is null): " | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | verl/models/mcore/mtp_patch.py | 472 | # embedding computation - you can modify this part | COMMENT |
| MEDIUM | verl/trainer/config/rollout/rollout.yaml | 59 | # you can use smaller batch size to save memory used in cuda graph, eg: [1 ,2, 4, 8, 16, 32] | COMMENT |
| MEDIUM | examples/grpo_trainer/run_gpt_oss_20b_fsdp.sh | 63 | # or you can use lmsys/gpt-oss-20b-bf16 | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tests/utils/test_rollout_trace_on_cpu.py | 86 | async def my_method(self, a, b="default"): | CODE |
| LOW | tests/utils/test_rollout_trace_on_cpu.py | 109 | async def my_method(self, x): | CODE |
| LOW | tests/utils/_test_module.py | 30 | def test_function(): | CODE |
| LOW | verl/utils/activation_offload.py | 457 | def my_function(*inputs): | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | verl/experimental/separation/ray_trainer.py | 741 | # TODO: implement actual tflpo and theoretical tflpo | COMMENT |
| LOW | verl/trainer/ppo/ray_trainer.py | 1784 | # TODO: implement actual tflpo and theoretical tflpo | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | …ples/ascend_extras/grpo_trainer/run_glm5_2_megatron.sh | 3 | # Usage: | COMMENT |