slime is an LLM post-training framework for RL Scaling.
This report summarizes scanner evidence for THUDM/slime, a Python project with 8,629 GitHub stars. SynthScan examined 128,975 lines of code across 622 source files, recording 2079 pattern matches distributed across 20 syntactic categories. The legacy adjusted pattern signal is 24.1 (Elevated pattern signal); it is not an authorship estimate.
Use the findings to ask whether maintainers can explain intent, verify dependencies, simplify unnecessary scaffolding, and focus review effort. Risk Payload v1 adds ownership, supply-chain, over-engineering, and review-effort vectors when available. These are prompts for human verification, not probabilities, security findings, or measured maintenance costs.
Use each vector to guide manual verification of the underlying findings.
Can maintainers explain intent, assumptions, and trade-offs? 1375 finding(s).
Are imports declared and package/API names verified? 238 finding(s).
Are wrappers, guards, and repeated structures necessary? 1546 finding(s).
Which changes need focused human review? 2079 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 2079 distinct pattern matches across 20 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 | tools/convert_torch_dist_to_hf_parallel.py | 60 | def collect_from_storage_metadata(storage_item, base_path=""): | CODE |
| LOW | tools/trace_timeline_viewer.py | 55 | def _make_dummy_pickle_global(module_name: str, name: str) -> type[_MissingPickleObject]: | CODE |
| LOW | tools/trace_timeline_viewer.py | 330 | def nearest_closed_ancestor_end(span: dict[str, Any]) -> float | None: | CODE |
| LOW | tools/convert_hf_to_int4_direct.py | 78 | def round_to_quantized_type_dtype( | CODE |
| LOW | slime/backends/megatron_utils/checkpoint.py | 32 | def _init_from_local_shards_and_global_metadata( # type: ignore[override] | CODE |
| LOW⚡ | slime/backends/megatron_utils/checkpoint.py | 104 | def get_cpu_save_function_and_args(writer): | CODE |
| LOW | slime/backends/megatron_utils/memory.py | 7 | def configure_buffer_allocation(args): | CODE |
| LOW⚡ | slime/backends/megatron_utils/arguments.py | 24 | def _is_allgather_cp_dsa_model(hf_config): | CODE |
| LOW⚡ | slime/backends/megatron_utils/arguments.py | 32 | def _validate_allgather_cp_supported(args, hf_config=None): | CODE |
| LOW | slime/backends/megatron_utils/arguments.py | 151 | def _set_default_megatron_args(args): | CODE |
| LOW⚡ | slime/backends/megatron_utils/model.py | 43 | def _disable_tqdm_for_non_main_rank() -> bool: | CODE |
| LOW⚡ | slime/backends/megatron_utils/model.py | 51 | def _should_update_microbatch_pbar(model) -> bool: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 63 | def _wrap_forward_step_with_microbatch_pbar(forward_step_func, pbar): | CODE |
| LOW | slime/backends/megatron_utils/model.py | 77 | def _with_rollout_top_p_token_keys(args: Namespace, keys: Sequence[str]) -> list[str]: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 83 | def _iter_critic_output_layers(model: Sequence[DDP]): | CODE |
| LOW | slime/backends/megatron_utils/model.py | 98 | def get_load_checkpoint_path_by_args(args, load_arg="load"): | CODE |
| LOW | slime/backends/megatron_utils/model.py | 128 | def _critic_output_layer_needs_reinit(args: Namespace, model: Sequence[DDP], role: str) -> bool: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 175 | def _reinitialize_critic_output_layer(args: Namespace, model: Sequence[DDP]) -> None: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 185 | def get_optimizer_param_scheduler(args: Namespace, optimizer: MegatronOptimizer) -> OptimizerParamScheduler: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 245 | def _disable_distributed_optimizer_state_initialization(optimizer: MegatronOptimizer) -> None: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 273 | def setup_model_and_optimizer( | CODE |
| LOW | slime/backends/megatron_utils/model.py | 719 | def should_disable_forward_pre_hook(args: Namespace) -> bool: | CODE |
| LOW | slime/backends/megatron_utils/model.py | 999 | def initialize_model_and_optimizer( | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 89 | def get_rollout_top_p_logprob_kwargs(args: Namespace, batch: dict[str, Any]) -> dict[str, Any]: | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 200 | def _allgather_cp_redistribute( | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 529 | def get_log_probs_and_entropy( | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 679 | def apply_opd_kl_to_advantages( | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 720 | def compute_advantages_and_returns(args: Namespace, rollout_data: RolloutBatch) -> None: | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 947 | def _slice_allgather_response_rows( | CODE |
| LOW | slime/backends/megatron_utils/loss.py | 973 | def get_score_centering_terms(args, batch, logits): | CODE |
| LOW | slime/backends/megatron_utils/model_provider.py | 282 | def wrap_model_provider_with_freeze(original_provider, args): | CODE |
| LOW | slime/backends/megatron_utils/transformer_engine.py | 10 | def patch_precision_aware_optimizer_checkpointing(optimizer): | CODE |
| LOW | slime/backends/megatron_utils/transformer_engine.py | 74 | def install_transformer_engine_extensions(): | CODE |
| LOW | slime/backends/megatron_utils/hf_checkpoint_saver.py | 216 | def _finalize_distributed_shards(path: Path, local_state: dict[str, Any]) -> None: | CODE |
| LOW | slime/backends/megatron_utils/hf_checkpoint_saver.py | 313 | def _clear_existing_hf_weights(path: Path) -> None: | CODE |
| LOW | slime/backends/megatron_utils/hf_checkpoint_saver.py | 360 | def _raise_if_rank_zero_failed(context: str, error: str | None) -> None: | CODE |
| LOW | slime/backends/megatron_utils/sglang.py | 24 | def _reduce_tensor_with_cpu_support(tensor): | CODE |
| LOW | slime/backends/megatron_utils/sglang.py | 32 | def monkey_patch_torch_reductions(): | CODE |
| LOW | slime/backends/megatron_utils/cp_utils.py | 13 | def get_logits_and_tokens_offset_with_cp( | CODE |
| LOW | slime/backends/megatron_utils/cp_utils.py | 279 | def _new_routed_experts_buffer( | CODE |
| LOW | slime/backends/megatron_utils/cp_utils.py | 318 | def _fill_padded_sample_range( | CODE |
| LOW | slime/backends/megatron_utils/cp_utils.py | 369 | def _fill_routed_experts_allgather_range( | CODE |
| LOW | slime/backends/megatron_utils/cp_utils.py | 408 | def _fill_routed_experts_cp_range( | CODE |
| LOW | slime/backends/megatron_utils/cp_utils.py | 458 | def prepare_routed_experts_for_routing_replay( | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 49 | def _normalize_label_token_ids(label_token_ids: Any, expected_length: int) -> list[list[int]] | None: | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 99 | def set_train_parallel_config(self, config: dict): | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 626 | def _run_warmup_via_private_http(sample_manager, args): | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 672 | def run_megatron_dp_models_loop_worker(dp_rank, pp_size, sample_manager, actor_models): | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 708 | def _build_update_from_disk_fn(actor_model): | CODE |
| LOW | slime/backends/megatron_utils/server/arguments.py | 34 | def add_megatron_server_arguments(parser): | CODE |
| LOW | slime/backends/megatron_utils/server/arguments.py | 87 | def configure_megatron_server_args(args): | CODE |
| LOW | slime/backends/megatron_utils/server/arguments.py | 103 | def validate_megatron_server_args(args): | CODE |
| LOW | slime/backends/megatron_utils/server/logprob_utils.py | 21 | def sample_from_vocab_parallel_logits_without_full_gather( | CODE |
| LOW | slime/backends/megatron_utils/server/logprob_utils.py | 158 | def get_label_token_log_probs_from_vocab_parallel_logits( | CODE |
| LOW | slime/backends/megatron_utils/server/logprob_utils.py | 268 | def _slice_response_rows_for_current_cp_rank( | CODE |
| LOW | slime/backends/megatron_utils/server/logprob_utils.py | 303 | def _merge_allgather_cp_tensors( | CODE |
| LOW | slime/backends/megatron_utils/server/logprob_utils.py | 352 | def _get_log_probs_and_optional_samples( | CODE |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 153 | def _compact_valid_rows_inplace( | CODE |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 289 | def _padded_expert_hidden_bytes( | CODE |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 1090 | def _configure_batch_invariant(deep_gemm: Any, deep_gemm_wrapper: Any) -> bool: | CODE |
| 1002 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM⚡ | tools/analyze_profile.py | 189 | # ─── Device Properties ────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tools/analyze_profile.py | 197 | # ─── CUDA/NCCL Info ───────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tools/analyze_profile.py | 207 | # ─── Kernel Analysis ──────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tools/analyze_profile.py | 320 | # ─── Communication ────────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tools/analyze_profile.py | 329 | # ─── CPU overhead ─────────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tools/analyze_profile.py | 337 | # ─── Pretty print ────────────────────────────────────────────────────────── | COMMENT |
| MEDIUM⚡ | tools/analyze_profile.py | 341 | # ── Config ────────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 23 | # ─── Color helpers ────────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 65 | # ─── Data structures ─────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 235 | # ─── GPU Utilization ──────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 278 | # ─── Per-stream analysis ──────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 295 | # ─── CUDA Graph analysis ─────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 352 | # ── Timeline ──────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 363 | # ── Kernel Breakdown ──────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 382 | # ── Top Kernels ───────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 390 | # ── CUDA Graph ────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 431 | # ── CUDA Graph Location in Timeline ───────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 456 | # ── GPU Idle Gaps ─────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 466 | # ── Per-Stream ────────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 484 | # ── Communication ─────────────────────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 492 | # ── Performance Bottleneck Summary ────────────────────────────────── | COMMENT |
| MEDIUM | tools/analyze_profile.py | 575 | # ── Optimization Recommendations ──────────────────────────────────── | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 35 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 37 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 54 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 56 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 124 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 126 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 149 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 151 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 340 | # =========================================================================== | COMMENT |
| MEDIUM | slime/agent/trajectory.py | 342 | # =========================================================================== | COMMENT |
| MEDIUM | tests/test_rm_gpqa.py | 31 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_rm_gpqa.py | 33 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_gpqa.py | 88 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_gpqa.py | 90 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_sample.py | 30 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_sample.py | 32 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_sample.py | 175 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_sample.py | 177 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/_cp_dist_helpers.py | 139 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/_cp_dist_helpers.py | 157 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 34 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 37 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 63 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 65 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 82 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 84 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 121 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 123 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 149 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 151 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 186 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math_dapo.py | 188 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_metric_report.py | 55 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM | tests/test_metric_report.py | 65 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math.py | 37 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math.py | 39 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math.py | 76 | # --------------------------------------------------------------------------- | COMMENT |
| MEDIUM⚡ | tests/test_rm_math.py | 78 | # --------------------------------------------------------------------------- | COMMENT |
| 80 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH⚡ | tools/trace_timeline_viewer.py | 950 | cursorTime: null, | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 951 | hoveredItem: null, | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 952 | selectedItem: null, | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1156 | const lane = rowExpanded(row) ? (item.lane || 0) : 0; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1165 | let current = null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1171 | clusters.push(current); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1174 | current.items.push(item); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1344 | const row = rowIdx >= 0 ? state.rows[rowIdx] : null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1345 | if (row && rowExpanded(row)) { | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1347 | y: laneTop(layout, item.lane || 0, scrollTop) + 2, | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1446 | if (raw.start_attrs || raw.end_attrs) { | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1455 | if (!attrs) return null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1459 | if (v != null && v > 0) segs.push({ phase, duration: v }); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1459 | if (v != null && v > 0) segs.push({ phase, duration: v }); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1501 | const pdVirtualRole = attrs.timeline_pd_virtual_role || null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1501 | const pdVirtualRole = attrs.timeline_pd_virtual_role || null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1503 | if (!phases || (expanded && !pdVirtualRole)) return null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1547 | const colorKey = attrs.timeline_pd_parent_name || item.name; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1552 | const pdVirtualRole = attrs.timeline_pd_virtual_role || null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1552 | const pdVirtualRole = attrs.timeline_pd_virtual_role || null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1781 | if (showText && labelWidthAvailable > 40) { | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1786 | if (a.prompt_tokens != null) pdParts.push(`P:${a.prompt_tokens}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1786 | if (a.prompt_tokens != null) pdParts.push(`P:${a.prompt_tokens}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1787 | if (a.completion_tokens != null) pdParts.push(`D:${a.completion_tokens}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1787 | if (a.completion_tokens != null) pdParts.push(`D:${a.completion_tokens}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1788 | if (a.cached_tokens != null && a.cached_tokens > 0) pdParts.push(`cache:${a.cached_tokens}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1788 | if (a.cached_tokens != null && a.cached_tokens > 0) pdParts.push(`cache:${a.cached_tokens}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1789 | if (a.pd_prefill_forward_duration != null) pdParts.push(`pf:${(a.pd_prefill_forward_duration*1000).toFix | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1789 | if (a.pd_prefill_forward_duration != null) pdParts.push(`pf:${(a.pd_prefill_forward_duration*1000).toFix | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1790 | if (a.pd_decode_forward_duration != null) pdParts.push(`df:${(a.pd_decode_forward_duration*1000).toFixed | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1790 | if (a.pd_decode_forward_duration != null) pdParts.push(`df:${(a.pd_decode_forward_duration*1000).toFixed | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1791 | if (a.pd_transfer_speed_gb_s != null) pdParts.push(`${a.pd_transfer_speed_gb_s.toFixed(1)}GB/s`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1791 | if (a.pd_transfer_speed_gb_s != null) pdParts.push(`${a.pd_transfer_speed_gb_s.toFixed(1)}GB/s`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1857 | lines.push(`${key}: ${value}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1862 | lines.push(`events: ${item.attrs?.total ?? 0}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1864 | lines.push(`${name}: ${count}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1868 | if (item.parent_span_name) lines.push(`parent: ${item.parent_span_name}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1870 | lines.push(`start: ${niceDuration(item.start_ts - state.globalStart)}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1871 | lines.push(`end: ${item.end_ts == null ? 'open' : niceDuration(item.end_ts - state.globalStart)}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1872 | if (item.end_ts == null) { | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1873 | lines.push(`display_end: ${niceDuration(item.display_end_ts - state.globalStart)}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1875 | lines.push(`duration: ${item.end_ts == null ? 'open' : niceDuration(item.end_ts - item.start_ts)}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1877 | lines.push(`time: ${niceDuration((item.ts ?? 0) - state.globalStart)}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1879 | if (item.state === 'orphan_end') lines.push('warning: unmatched end event'); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1886 | lines.push(`${key}: ${typeof value === 'object' ? JSON.stringify(value) : value}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1900 | lines.push('── PD disaggregation ──'); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1902 | lines.push(' [P] prefill instance'); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1903 | pPhases.forEach(p => lines.push(fmtPhase(p))); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1906 | lines.push(' [D] decode instance'); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1907 | dPhases.forEach(p => lines.push(fmtPhase(p))); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1909 | lines.push(` total phases: ${(totalDur * 1000).toFixed(1)}ms`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1917 | lines.push(` ${label}: ${value.toFixed(2)} GB/s`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1919 | lines.push(` ${label}: ${value.toFixed(1)} MB`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1921 | lines.push(` ${label}: ${typeof value === 'object' ? JSON.stringify(value) : value}`); | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 1929 | const activeItem = state.hoveredItem || state.selectedItem; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 2090 | if (absoluteY >= layout.top && absoluteY <= layout.top + layout.height) return idx; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 2096 | if (!layout) return null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 2102 | let closestPoint = null; | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 2111 | if (distance <= radius && distance < closestPointDistance) { | CODE |
| HIGH⚡ | tools/trace_timeline_viewer.py | 2124 | if (localY < itemTop || localY > itemBottom) continue; | CODE |
| 43 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | tools/convert_torch_dist_to_hf_parallel.py | 189 | except (ValueError, KeyError, OSError): | CODE |
| HIGH | tools/convert_hf_to_torch_dist.py | 36 | except Exception: | CODE |
| HIGH | tools/trace_timeline_viewer.py | 2341 | except KeyboardInterrupt: | CODE |
| HIGH | slime/backends/megatron_utils/checkpoint.py | 90 | except ImportError: | CODE |
| HIGH | slime/backends/megatron_utils/__init__.py | 38 | except RuntimeError: | CODE |
| HIGH | slime/backends/sglang_utils/sglang_engine.py | 419 | except requests.exceptions.RequestException: | CODE |
| HIGH | slime/backends/sglang_utils/sglang_engine.py | 89 | except requests.RequestException: | CODE |
| HIGH | slime/backends/sglang_utils/sglang_engine.py | 137 | except ValueError: | CODE |
| HIGH | slime/utils/arguments.py | 1785 | except (ValueError, TypeError): | CODE |
| HIGH | slime/utils/http_utils.py | 66 | except Exception: | CODE |
| HIGH | slime/utils/http_utils.py | 81 | except Exception: | CODE |
| HIGH | slime/agent/aiohttp_threaded.py | 40 | except Exception: | CODE |
| HIGH | slime/agent/adapters/common.py | 639 | except Exception: | CODE |
| HIGH | slime/rollout/fully_async_rollout.py | 238 | except Exception: # noqa: BLE001 | CODE |
| HIGH | slime/rollout/rm_hub/math_utils.py | 360 | except Exception: | CODE |
| HIGH | slime/rollout/rm_hub/math_utils.py | 306 | except Exception: | CODE |
| HIGH | slime/data/transport.py | 386 | except asyncio.CancelledError: | CODE |
| HIGH | tests/test_qwen3_4B_external_pd.py | 83 | except OSError: | CODE |
| HIGH | tests/test_qwen3_4B_external_pd.py | 208 | except Exception: | CODE |
| HIGH | tests/fanout_test_helpers.py | 37 | except OSError: | CODE |
| HIGH | tests/test_qwen2.5_0.5B_opd_sglang.py | 71 | except Exception: | CODE |
| HIGH | tests/test_qwen2.5_0.5B_fanout_short.py | 68 | except FileNotFoundError: | CODE |
| HIGH | tests/test_straw_fully_async_recovery.py | 226 | except psutil.NoSuchProcess: | CODE |
| HIGH | tests/test_straw_fully_async_recovery.py | 214 | except psutil.NoSuchProcess: | CODE |
| HIGH | tests/test_straw_fully_async_recovery.py | 236 | except psutil.NoSuchProcess: | CODE |
| HIGH | tests/test_qwen2.5_0.5B_training_recovery.py | 368 | except ray.exceptions.RayActorError: | CODE |
| HIGH | tests/test_qwen2.5_0.5B_training_recovery.py | 301 | except ValueError: | CODE |
| HIGH | tests/test_server_control.py | 253 | except (asyncio.IncompleteReadError, ConnectionError): | CODE |
| HIGH | tests/ci/gpu_lock_exec.py | 51 | except subprocess.TimeoutExpired: | CODE |
| HIGH | tests/ci/gpu_lock_exec.py | 56 | except ProcessLookupError: | CODE |
| HIGH | tests/ci/gpu_lock_exec.py | 129 | except BlockingIOError: | CODE |
| HIGH | tests/ci/gpu_lock_exec.py | 63 | except ProcessLookupError: | CODE |
| HIGH | docs/tests/test_browser.py | 24 | except (BrokenPipeError, ConnectionResetError): | CODE |
| HIGH | examples/geo3k_vlm_multi_turn/rollout.py | 386 | except Exception: | CODE |
| HIGH | examples/strands_sglang/subprocess_interpreter.py | 370 | except Exception: | CODE |
| HIGH | examples/coding_agent_rl/generate.py | 333 | except Exception: # pragma: no cover - diag must never crash | CODE |
| HIGH | examples/coding_agent_rl/swe.py | 436 | except OSError: | CODE |
| HIGH | examples/retool/tool_sandbox.py | 190 | except Exception: | CODE |
| HIGH | examples/retool/generate_with_retool.py | 127 | except (json.JSONDecodeError, KeyError, AttributeError): | CODE |
| HIGH | examples/retool/generate_with_retool.py | 296 | except ImportError: | CODE |
| HIGH | slime_plugins/models/qwen3_5.py | 16 | except ImportError: | CODE |
| HIGH | slime_plugins/models/qwen3_next.py | 19 | except ImportError: | CODE |
| HIGH | slime_plugins/models/qwen3_5_vl_utils.py | 150 | except StopIteration: | CODE |
| HIGH | slime_plugins/models/qwen3_5_vl_utils.py | 155 | except StopIteration: | CODE |
| HIGH | slime_plugins/models/glm5/ops/indexer.py | 15 | except Exception: | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | train.py | 120 | except Exception: | CODE |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 142 | except Exception as e: | CODE |
| MEDIUM | tools/convert_torch_dist_to_hf_parallel.py | 143 | print(f"Error reading {rel_path}: {e}") | CODE |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 291 | except Exception as e: | CODE |
| MEDIUM | tools/convert_torch_dist_to_hf_parallel.py | 292 | print(f"Error processing {name}: {e}") | CODE |
| MEDIUM | tools/convert_torch_dist_to_hf_parallel.py | 132 | def read_file(rel_path, full_path, ops): | CODE |
| LOW | tools/profile_rollout.py | 12 | except Exception as e: | CODE |
| MEDIUM | tools/profile_rollout.py | 13 | print(f"Error fetching workers from router: {e}") | CODE |
| LOW | tools/profile_rollout.py | 31 | except Exception as e: | CODE |
| LOW | tools/profile_rollout.py | 41 | except Exception as e: | CODE |
| MEDIUM | tools/profile_rollout.py | 7 | def get_workers(router_url): | CODE |
| MEDIUM | tools/profile_rollout.py | 35 | def stop_profile(worker_url): | CODE |
| LOW | tools/convert_hf_to_torch_dist.py | 36 | except Exception: | CODE |
| LOW | tools/convert_hf_to_int4.py | 37 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/model_provider.py | 230 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/hf_checkpoint_saver.py | 59 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/hf_checkpoint_saver.py | 78 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 351 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 487 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 560 | except Exception as e: | CODE |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 590 | except Exception as e: | CODE |
| LOW | slime/backends/sglang_utils/server_control.py | 200 | except Exception as e: | CODE |
| LOW | slime/backends/sglang_utils/server_control.py | 212 | except Exception as e: | CODE |
| LOW | slime/backends/sglang_utils/sglang_engine.py | 295 | except Exception as e: | CODE |
| LOW | slime/backends/sglang_utils/sglang_engine.py | 321 | except Exception as e: | CODE |
| LOW | slime/backends/sglang_utils/engine_group.py | 39 | except Exception as error: | CODE |
| LOW | slime/backends/sglang_utils/engine_group.py | 57 | except Exception as error: | CODE |
| LOW | slime/backends/sglang_utils/external.py | 81 | except Exception as exc: | CODE |
| LOW | slime/utils/reloadable_process_group.py | 518 | except Exception as e: | CODE |
| MEDIUM | slime/utils/reloadable_process_group.py | 511 | def _wrap_low_level_call(check_memory=True): | CODE |
| LOW | slime/utils/processing_utils.py | 127 | except Exception: | CODE |
| LOW | slime/utils/processing_utils.py | 147 | except Exception: | CODE |
| LOW | slime/utils/health_monitor.py | 150 | except Exception as error: | CODE |
| LOW | slime/utils/cleanup.py | 28 | except Exception as error: | CODE |
| MEDIUM | slime/utils/cleanup.py | 25 | def run(self, description, function, *args, **kwargs): | CODE |
| LOW | slime/utils/http_utils.py | 66 | except Exception: | CODE |
| LOW | slime/utils/http_utils.py | 81 | except Exception: | CODE |
| LOW | slime/utils/http_utils.py | 125 | except Exception as e: | CODE |
| LOW | slime/utils/http_utils.py | 209 | except Exception as e: | CODE |
| LOW | slime/utils/http_utils.py | 352 | except Exception as e: | CODE |
| MEDIUM | slime/utils/http_utils.py | 117 | def run_router(args): | CODE |
| LOW | slime/utils/accelerator/__init__.py | 87 | except Exception as exc: | CODE |
| LOW | slime/utils/accelerator/__init__.py | 140 | except Exception as exc: | CODE |
| LOW | slime/utils/accelerator/__init__.py | 193 | except Exception as exc: | CODE |
| MEDIUM | slime/utils/accelerator/__init__.py | 133 | def _import_torch_npu() -> bool: | CODE |
| MEDIUM | slime/utils/accelerator/__init__.py | 186 | def _import_torch_supa() -> bool: | CODE |
| LOW | slime/agent/parsing.py | 77 | except Exception: | CODE |
| LOW | slime/agent/sandbox.py | 250 | except Exception as e: | CODE |
| LOW | slime/agent/sandbox.py | 278 | except Exception as e: | CODE |
| LOW | slime/agent/sandbox.py | 306 | except Exception as e: | CODE |
| LOW | slime/agent/sandbox.py | 386 | except Exception: | CODE |
| MEDIUM | slime/agent/sandbox.py | 302 | def __aexit__(self, exc_type, exc, tb) -> None: | CODE |
| LOW | slime/agent/aiohttp_threaded.py | 40 | except Exception: | CODE |
| LOW | slime/agent/adapters/common.py | 306 | except Exception: | CODE |
| LOW | slime/agent/adapters/common.py | 379 | except Exception: | CODE |
| LOW | slime/agent/adapters/common.py | 639 | except Exception: | CODE |
| LOW | slime/agent/adapters/common.py | 660 | except Exception as exc: | CODE |
| LOW⚡ | slime/observability/trace_utils.py | 399 | except Exception as exc: | CODE |
| LOW⚡ | slime/observability/trace_utils.py | 409 | except Exception as trace_exc: | CODE |
| LOW⚡ | slime/observability/trace_utils.py | 415 | except Exception as exc: | CODE |
| 106 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | train.py | 103 | CODE | |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 57 | CODE | |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 84 | CODE | |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 165 | CODE | |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 60 | CODE | |
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 170 | CODE | |
| LOW | tools/analyze_profile.py | 184 | CODE | |
| LOW | tools/analyze_profile.py | 338 | CODE | |
| LOW | tools/trace_timeline_viewer.py | 168 | CODE | |
| LOW | tools/convert_k2_thinking_int4_to_bf16.py | 99 | CODE | |
| LOW | tools/fp8_cast_bf16.py | 44 | CODE | |
| LOW | tools/convert_torch_dist_to_hf.py | 106 | CODE | |
| LOW | tools/convert_to_hf.py | 28 | CODE | |
| LOW | slime/backends/megatron_utils/stateless_adam.py | 68 | CODE | |
| LOW | slime/backends/megatron_utils/model.py | 356 | CODE | |
| LOW | slime/backends/megatron_utils/model.py | 724 | CODE | |
| LOW | slime/backends/megatron_utils/model.py | 98 | CODE | |
| LOW | slime/backends/megatron_utils/loss.py | 103 | CODE | |
| LOW | slime/backends/megatron_utils/loss.py | 448 | CODE | |
| LOW | slime/backends/megatron_utils/loss.py | 720 | CODE | |
| LOW | slime/backends/megatron_utils/loss.py | 973 | CODE | |
| LOW | slime/backends/megatron_utils/model_provider.py | 115 | CODE | |
| LOW | slime/backends/megatron_utils/model_provider.py | 321 | CODE | |
| LOW | slime/backends/megatron_utils/model_provider.py | 139 | CODE | |
| LOW | slime/backends/megatron_utils/ci_utils.py | 11 | CODE | |
| LOW | slime/backends/megatron_utils/actor.py | 298 | CODE | |
| LOW | slime/backends/megatron_utils/actor.py | 449 | CODE | |
| LOW | slime/backends/megatron_utils/cp_utils.py | 408 | CODE | |
| LOW | slime/backends/megatron_utils/data.py | 14 | CODE | |
| LOW | slime/backends/megatron_utils/server/megatron_server.py | 672 | CODE | |
| LOW | slime/backends/megatron_utils/server/logprob_utils.py | 303 | CODE | |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 470 | CODE | |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 642 | CODE | |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 3100 | CODE | |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 910 | CODE | |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 2253 | CODE | |
| LOW | …/backends/megatron_utils/alignment/deepgemm_forward.py | 1057 | CODE | |
| LOW | …atron_utils/update_weight/update_weight_from_tensor.py | 378 | CODE | |
| LOW | …atron_utils/update_weight/update_weight_from_tensor.py | 195 | CODE | |
| LOW | slime/backends/megatron_utils/update_weight/common.py | 59 | CODE | |
| LOW | slime/backends/megatron_utils/update_weight/common.py | 129 | CODE | |
| LOW | …atron_utils/update_weight/hf_weight_iterator_direct.py | 174 | CODE | |
| LOW | …ime/backends/megatron_utils/megatron_to_hf/qwen3_vl.py | 5 | CODE | |
| LOW | …e/backends/megatron_utils/megatron_to_hf/minimax_m2.py | 6 | CODE | |
| LOW | …ime/backends/megatron_utils/megatron_to_hf/__init__.py | 37 | CODE | |
| LOW | …ime/backends/megatron_utils/megatron_to_hf/qwen3moe.py | 6 | CODE | |
| LOW | slime/backends/megatron_utils/megatron_to_hf/llama.py | 5 | CODE | |
| LOW | slime/backends/megatron_utils/megatron_to_hf/qwen3_5.py | 34 | CODE | |
| LOW | slime/backends/megatron_utils/megatron_to_hf/glm4.py | 5 | CODE | |
| LOW | …e/backends/megatron_utils/megatron_to_hf/deepseekv3.py | 6 | CODE | |
| LOW | slime/backends/megatron_utils/megatron_to_hf/glm4moe.py | 6 | CODE | |
| LOW | …e/backends/megatron_utils/megatron_to_hf/qwen3_next.py | 44 | CODE | |
| LOW | slime/backends/megatron_utils/megatron_to_hf/qwen2.py | 5 | CODE | |
| LOW | …atron_utils/megatron_to_hf/processors/quantizer_fp8.py | 10 | CODE | |
| LOW | slime/backends/sglang_utils/server_control.py | 17 | CODE | |
| LOW | slime/backends/sglang_utils/server_control.py | 106 | CODE | |
| LOW | slime/backends/sglang_utils/deployment.py | 79 | CODE | |
| LOW | slime/backends/sglang_utils/sglang_engine.py | 77 | CODE | |
| LOW | slime/backends/sglang_utils/engine_group.py | 477 | CODE | |
| LOW | slime/backends/sglang_utils/engine_group.py | 342 | CODE | |
| 124 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tools/trace_timeline_viewer.py | 9 | CODE | |
| LOW | slime/backends/megatron_utils/sglang.py | 18 | CODE | |
| LOW | slime/backends/megatron_utils/sglang.py | 45 | CODE | |
| LOW | slime/backends/megatron_utils/sglang.py | 47 | CODE | |
| LOW | slime/backends/megatron_utils/server/__init__.py | 3 | CODE | |
| LOW | slime/backends/megatron_utils/server/__init__.py | 3 | CODE | |
| LOW | slime/backends/megatron_utils/server/__init__.py | 3 | CODE | |
| LOW | slime/backends/megatron_utils/alignment/env.py | 10 | CODE | |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 14 | CODE | |
| LOW | …ckends/megatron_utils/alignment/layerwise_alignment.py | 3 | CODE | |
| LOW | …/backends/megatron_utils/alignment/deepgemm_forward.py | 15 | CODE | |
| LOW | …egatron_utils/alignment/deterministic_route_kernels.py | 8 | CODE | |
| LOW | …egatron_utils/update_weight/update_weight_from_disk.py | 1 | CODE | |
| LOW | slime/backends/megatron_utils/update_weight/__init__.py | 1 | CODE | |
| LOW | …n_utils/update_weight/update_weight_from_disk_delta.py | 1 | CODE | |
| LOW | …_utils/update_weight/update_weight_from_distributed.py | 1 | CODE | |
| LOW | …s/megatron_utils/megatron_to_hf/processors/__init__.py | 1 | CODE | |
| LOW | slime/backends/megatron_utils/hf_to_megatron/glm.py | 1 | CODE | |
| LOW | slime/backends/megatron_utils/hf_to_megatron/qwen3_5.py | 1 | CODE | |
| LOW | slime/backends/megatron_utils/hf_to_megatron/common.py | 1 | CODE | |
| LOW | slime/backends/megatron_utils/hf_to_megatron/qwen.py | 1 | CODE | |
| LOW | …ime/backends/megatron_utils/hf_to_megatron/deepseek.py | 1 | CODE | |
| LOW | …e/backends/megatron_utils/hf_to_megatron/qwen3_next.py | 1 | CODE | |
| LOW | slime/backends/sglang_utils/external.py | 3 | CODE | |
| LOW | slime/utils/routed_experts.py | 1 | CODE | |
| LOW | slime/utils/dp_schedule.py | 40 | CODE | |
| LOW | slime/utils/eval_config.py | 1 | CODE | |
| LOW | slime/utils/disk_delta.py | 1 | CODE | |
| LOW | slime/utils/accelerator/supa.py | 12 | CODE | |
| LOW | slime/utils/accelerator/__init__.py | 8 | CODE | |
| LOW | slime/utils/accelerator/npu.py | 7 | CODE | |
| LOW | slime/utils/accelerator/musa.py | 7 | CODE | |
| LOW | slime/utils/accelerator/cuda.py | 3 | CODE | |
| LOW | slime/utils/accelerator/base.py | 8 | CODE | |
| LOW | slime/utils/accelerator/torch_accelerator.py | 3 | CODE | |
| LOW | slime/agent/parsing.py | 3 | CODE | |
| LOW | slime/agent/trajectory.py | 10 | CODE | |
| LOW | slime/agent/sandbox.py | 9 | CODE | |
| LOW | slime/agent/aiohttp_threaded.py | 3 | CODE | |
| LOW | slime/agent/adapters/responses.py | 9 | CODE | |
| LOW | slime/agent/adapters/__init__.py | 3 | CODE | |
| LOW | slime/agent/adapters/__init__.py | 4 | CODE | |
| LOW | slime/agent/adapters/__init__.py | 5 | CODE | |
| LOW | slime/agent/adapters/__init__.py | 6 | CODE | |
| LOW | slime/agent/adapters/openai.py | 16 | CODE | |
| LOW | slime/agent/adapters/common.py | 13 | CODE | |
| LOW | slime/agent/adapters/anthropic.py | 17 | CODE | |
| LOW | slime/agent/harness/codex.py | 8 | CODE | |
| LOW | slime/agent/harness/__init__.py | 3 | CODE | |
| LOW | slime/agent/harness/__init__.py | 5 | CODE | |
| LOW | slime/agent/harness/__init__.py | 6 | CODE | |
| LOW | slime/agent/harness/__init__.py | 7 | CODE | |
| LOW | slime/agent/harness/__init__.py | 7 | CODE | |
| LOW | slime/agent/harness/common.py | 16 | CODE | |
| LOW | slime/agent/harness/claude_code.py | 3 | CODE | |
| LOW | slime/observability/trace_utils.py | 1 | CODE | |
| LOW | slime/rollout/sample_hooks.py | 1 | CODE | |
| LOW | slime/rollout/base_types.py | 1 | CODE | |
| LOW | slime/rollout/fully_async_rollout.py | 29 | CODE | |
| LOW | slime/rollout/rm_hub/ifbench.py | 1 | CODE | |
| 67 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW⚡ | slime/backends/megatron_utils/checkpoint.py | 93 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | slime/backends/megatron_utils/checkpoint.py | 95 | __all__ = ["save_checkpoint"] | CODE |
| LOW | slime/backends/megatron_utils/initialize.py | 15 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | slime/backends/megatron_utils/arguments.py | 13 | __all__ = ["validate_args", "megatron_parse_args", "set_default_megatron_args"] | CODE |
| LOW⚡ | slime/backends/megatron_utils/arguments.py | 15 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | slime/backends/megatron_utils/model.py | 40 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/megatron_utils/model_provider.py | 103 | def _set_critic_output_layer(model: torch.nn.Module, config: TransformerConfig) -> None: | CODE |
| LOW | slime/backends/megatron_utils/ci_utils.py | 8 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/megatron_utils/hf_checkpoint_saver.py | 13 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/megatron_utils/actor.py | 59 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/megatron_utils/actor.py | 628 | def update_weights(self) -> None: | CODE |
| LOW | slime/backends/megatron_utils/sglang.py | 49 | __all__ = [ | CODE |
| LOW | slime/backends/megatron_utils/server/__init__.py | 9 | __all__ = [ | CODE |
| LOW | …kends/megatron_utils/alignment/deepgemm_moe_forward.py | 38 | logger = logging.getLogger(__name__) | CODE |
| LOW | …ckends/megatron_utils/alignment/layerwise_alignment.py | 14 | logger = logging.getLogger(__name__) | CODE |
| LOW | …/backends/megatron_utils/alignment/deepgemm_forward.py | 26 | logger = logging.getLogger(__name__) | CODE |
| LOW | …egatron_utils/update_weight/update_weight_from_disk.py | 66 | def update_weights(self) -> None: | CODE |
| LOW | …ackends/megatron_utils/update_weight/expert_routing.py | 14 | __all__ = ["configure_expert_routing"] | CODE |
| LOW | …ackends/megatron_utils/update_weight/expert_routing.py | 17 | logger = logging.getLogger(__name__) | CODE |
| LOW | …atron_utils/update_weight/update_weight_from_tensor.py | 287 | def update_weights(self) -> None: | CODE |
| LOW | …n_utils/update_weight/update_weight_from_disk_delta.py | 30 | logger = logging.getLogger(__name__) | CODE |
| LOW | …n_utils/update_weight/update_weight_from_disk_delta.py | 90 | def update_weights(self) -> None: | CODE |
| LOW | …_utils/update_weight/update_weight_from_distributed.py | 103 | def update_weights(self) -> None: | CODE |
| LOW | …atron_to_hf/processors/quantizer_compressed_tensors.py | 15 | logger = logging.getLogger(__name__) | CODE |
| LOW | …atron_to_hf/processors/quantizer_compressed_tensors.py | 18 | __all__ = ["quantize_params_compressed_tensors"] | CODE |
| LOW | …s/megatron_utils/megatron_to_hf/processors/__init__.py | 3 | __all__ = ["quantize_params", "remove_padding"] | CODE |
| LOW | slime/backends/sglang_utils/server_control.py | 10 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/sglang_utils/deployment.py | 15 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/sglang_utils/sglang_engine.py | 19 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/sglang_utils/arguments.py | 8 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/sglang_utils/engine_group.py | 15 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/sglang_utils/external.py | 11 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/backends/sglang_utils/disaggregation.py | 11 | logger = logging.getLogger(__name__) | CODE |
| LOW⚡ | slime/backends/sglang_utils/sglang_config.py | 8 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/reloadable_process_group.py | 17 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/async_utils.py | 5 | __all__ = ["AsyncPacer", "get_async_loop", "run"] | CODE |
| LOW | slime/utils/routed_experts.py | 13 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/dp_schedule.py | 48 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/arguments.py | 17 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/processing_utils.py | 10 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/health_monitor.py | 7 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/cleanup.py | 6 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/http_utils.py | 12 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/data.py | 19 | __all__ = ["Dataset", "get_source"] | CODE |
| LOW | slime/utils/data.py | 21 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/memory_utils.py | 12 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/accelerator/__init__.py | 30 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/accelerator/__init__.py | 356 | def set_accelerator(accelerator: Accelerator) -> None: | CODE |
| LOW | slime/utils/accelerator/__init__.py | 393 | def set_device(index: int | str | torch.device) -> None: | CODE |
| LOW | slime/utils/accelerator/__init__.py | 491 | def set_rng_state(state: torch.Tensor, device_arg: int | str | torch.device | None = None) -> None: | CODE |
| LOW | slime/utils/accelerator/base.py | 37 | def set_device(self, index: int | str | torch.device) -> None: | CODE |
| LOW | slime/utils/accelerator/base.py | 127 | def set_rng_state(self, state: torch.Tensor, device: int | str | torch.device | None = None) -> None: | CODE |
| LOW | slime/utils/accelerator/torch_accelerator.py | 13 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/utils/accelerator/torch_accelerator.py | 42 | def set_device(self, index: int | str | torch.device) -> None: | CODE |
| LOW | slime/utils/accelerator/torch_accelerator.py | 144 | def set_rng_state(self, state: torch.Tensor, device: int | str | torch.device | None = None) -> None: | CODE |
| LOW | slime/agent/parsing.py | 12 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/agent/trajectory.py | 23 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/agent/trajectory.py | 581 | __all__ = [ | CODE |
| LOW | slime/agent/sandbox.py | 20 | logger = logging.getLogger(__name__) | CODE |
| LOW | slime/agent/adapters/__init__.py | 8 | __all__ = ["AnthropicAdapter", "BaseAdapter", "OpenAIAdapter", "ResponsesAdapter"] | CODE |
| 52 more matches not shown… | ||||
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | slime/utils/ppo_utils.py | 585 | # Step 3: Compute returns on full response kl tensor. | COMMENT |
| LOW⚡ | tests/ci/README.md | 11 | ### Step 1: Env | COMMENT |
| LOW⚡ | tests/ci/README.md | 18 | ### Step 2: Prepare `/home/runner/externals` | COMMENT |
| LOW⚡ | tests/ci/README.md | 28 | ### Step 3: Run | COMMENT |
| LOW | .claude/skills/add-rollout-function/SKILL.md | 20 | ### Step 1: Choose the Right Starting Point | COMMENT |
| LOW | .claude/skills/add-rollout-function/SKILL.md | 30 | ### Step 2: Create the New Rollout Module | COMMENT |
| LOW | .claude/skills/add-rollout-function/SKILL.md | 43 | ### Step 3: Implement Train and Eval Branches Explicitly | COMMENT |
| LOW | .claude/skills/add-rollout-function/SKILL.md | 70 | ### Step 4: Keep Data Contract Compatible | COMMENT |
| LOW | .claude/skills/add-rollout-function/SKILL.md | 81 | ### Step 5: Wire Through Arguments | COMMENT |
| LOW | .claude/skills/add-reward-function/SKILL.md | 20 | ### Step 1: Choose Reward Mode | COMMENT |
| LOW | .claude/skills/add-reward-function/SKILL.md | 29 | ### Step 2: Create Reward Module | COMMENT |
| LOW | .claude/skills/add-reward-function/SKILL.md | 47 | ### Step 3: Keep Reward Type Consistent | COMMENT |
| LOW | .claude/skills/add-reward-function/SKILL.md | 53 | ### Step 4: Optional Reward Post-Processing | COMMENT |
| LOW | .claude/skills/add-reward-function/SKILL.md | 71 | ### Step 5: Wire and Validate | COMMENT |
| LOW | .claude/skills/add-eval-dataset-config/SKILL.md | 20 | ### Step 1: Choose Config Entry Method | COMMENT |
| LOW | .claude/skills/add-eval-dataset-config/SKILL.md | 29 | ### Step 2: Build YAML with Required Fields | COMMENT |
| LOW | .claude/skills/add-eval-dataset-config/SKILL.md | 54 | ### Step 3: Understand Override Priority | COMMENT |
| LOW | .claude/skills/add-eval-dataset-config/SKILL.md | 68 | ### Step 4: Wire Eval Function if Needed | COMMENT |
| LOW | .claude/skills/add-eval-dataset-config/SKILL.md | 73 | ### Step 5: Validate Parsing and Runtime | COMMENT |
| LOW | .claude/skills/add-tests-and-ci/SKILL.md | 21 | ### Step 1: Pick the Right Test Pattern | COMMENT |
| LOW | .claude/skills/add-tests-and-ci/SKILL.md | 27 | ### Step 2: Keep CI Compatibility | COMMENT |
| LOW | .claude/skills/add-tests-and-ci/SKILL.md | 43 | ### Step 3: Register Tests in GitHub CI | COMMENT |
| LOW | .claude/skills/add-tests-and-ci/SKILL.md | 60 | ### Step 4: Run Local Validation | COMMENT |
| LOW | .claude/skills/add-tests-and-ci/SKILL.md | 67 | ### Step 5: Keep Workflow Template as Source of Truth | COMMENT |
| LOW | .claude/skills/add-tests-and-ci/SKILL.md | 80 | ### Step 6: Provide Verifiable PR Notes | COMMENT |
| LOW | .claude/skills/add-dynamic-filter/SKILL.md | 21 | ### Step 1: Pick the Correct Hook | COMMENT |
| LOW | .claude/skills/add-dynamic-filter/SKILL.md | 28 | ### Step 2: Implement the Function Signature | COMMENT |
| LOW | .claude/skills/add-dynamic-filter/SKILL.md | 66 | ### Step 3: Preserve Group Structure | COMMENT |
| LOW | .claude/skills/add-dynamic-filter/SKILL.md | 72 | ### Step 4: Wire and Validate | COMMENT |
| LOW | .claude/skills/add-dynamic-filter/SKILL.md | 83 | ### Step 5: Measure Side Effects | COMMENT |
| LOW | docs/en/advanced/sglang-config.md | 267 | # Step 1: Launch SGLang engines externally | COMMENT |
| LOW | docs/en/advanced/sglang-config.md | 271 | # Step 2: Connect slime to external engines | COMMENT |
| LOW | examples/search-r1/README.md | 187 | ### Step 1: Install Conda | COMMENT |
| LOW | examples/search-r1/README.md | 204 | ### Step 2: Create Retriever Environment | COMMENT |
| LOW | examples/search-r1/README.md | 222 | ### Step 3: Download Index and Corpus | COMMENT |
| LOW | examples/search-r1/README.md | 240 | ### Step 4: Start Local Retrieval Server | COMMENT |
| LOW | examples/search-r1/README.md | 275 | ### Step 5: Start Training | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | slime/backends/megatron_utils/model_provider.py | 176 | # Define the decoder block spec | COMMENT |
| MEDIUM | slime/backends/megatron_utils/model_provider.py | 184 | # Define the decoder layer spec | COMMENT |
| MEDIUM | slime/utils/async_utils.py | 45 | # Create a background event loop thread | COMMENT |
| MEDIUM | slime/utils/http_utils.py | 286 | # Define the async actor | COMMENT |
| MEDIUM | slime/rollout/rm_hub/__init__.py | 73 | # This function is intended for remote or time-consuming reward model evaluation. | COMMENT |
| MEDIUM | docs/build_all.sh | 11 | # Create a lightweight root index with auto redirect based on localStorage (done client side) | COMMENT |
| MEDIUM | examples/tau-bench/trainable_agents.py | 213 | # Initialize tracking variables | COMMENT |
| MEDIUM | examples/strands_sglang/subprocess_interpreter.py | 186 | # Create a temporary file with the modified source | STRING |
| MEDIUM | examples/strands_sglang/subprocess_interpreter.py | 352 | # Create a temporary directory first to ensure we have write | STRING |
| MEDIUM | .github/workflows/pr-test.yml | 2 | # This file is auto-generated from the .j2 file via generate_github_workflows.py. Do not edit manually. | COMMENT |
| MEDIUM | .github/workflows/generate_github_workflows.py | 27 | + "\n# This file is auto-generated from the .j2 file via generate_github_workflows.py. Do not edit manua | CODE |
| MEDIUM | slime_plugins/models/qwen3_5.py | 203 | # Define the decoder block spec | COMMENT |
| MEDIUM | slime_plugins/models/qwen3_next.py | 215 | # Define the decoder block spec | COMMENT |
| MEDIUM | slime_plugins/models/glm5/glm5.py | 1015 | # Define the decoder block spec | COMMENT |
| MEDIUM | …ime_plugins/rollout_buffer/generator/base_generator.py | 53 | # Create a fresh payload for each attempt | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | …atron_utils/megatron_to_hf/processors/quantizer_fp8.py | 115 | # for other parameters, we just return the original converted_named_params | COMMENT |
| LOW | slime/backends/sglang_utils/engine_group.py | 233 | # this group's engines may land on (conservative: just use global max). | COMMENT |
| MEDIUM | slime/agent/harness/codex.py | 34 | # an unprivileged PRoot sandbox. Never use this harness directly on the host. | COMMENT |
| MEDIUM | slime/rollout/rm_hub/math_dapo_utils.py | 14 | # Adapted from https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/hendrycks_math/utils.py | COMMENT |
| MEDIUM⚡ | tests/test_agent/test_harness.py | 209 | # §5 harness.run wires the steps in order | COMMENT |
| MEDIUM⚡ | examples/coding_agent_rl/swe.py | 181 | # Workspace prep (agent sandbox, before harness.run) | COMMENT |
| MEDIUM⚡ | examples/coding_agent_rl/swe.py | 235 | # Diff capture (agent sandbox, after harness.run) | COMMENT |
| MEDIUM | examples/coding_agent_rl/swe.py | 386 | # Mirror of swebench.harness.run_evaluation.GIT_APPLY_CMDS: try each in order, | COMMENT |
| MEDIUM | examples/retool/generate_with_retool.py | 100 | # Use a more robust regex that handles nested braces | COMMENT |
| MEDIUM | examples/retool/generate_with_retool.py | 370 | # collapse), but reward_func's regex is whitespace-robust and | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | slime/backends/sglang_utils/sglang_engine.py | 235 | Run /health_generate on the underlying SGLang HTTP server. Args: timeout: Timeout for the health re | STRING |
| HIGH | examples/tau-bench/openai_tool_adapter.py | 52 | Parse sglang response to OpenAI compatible format Args: response: Raw response text from s | STRING |
| HIGH | examples/tau-bench/generate_with_tau.py | 101 | Generate a complete agent-environment interaction trajectory for tau-bench. This is the main entry point for s | STRING |
| HIGH | examples/strands_sglang/subprocess_interpreter.py | 286 | Generates a temporary file with the given code, executes it, and deletes the file afterward. Args: | STRING |
| HIGH | examples/strands_sglang/subprocess_interpreter.py | 440 | Executes a shell command in a subprocess and captures its output. Args: command (str): The shell co | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | README.md | 97 | [Dressage](https://github.com/Accio-Lab/Dressage) — Alibaba Accio’s agentic RL framework, with Paddock, Sandbox, and Pro | CODE |
| MEDIUM | README.md | 109 | [Relax](https://github.com/redai-infra/Relax) — RedAI Infra’s multimodal agentic RL framework, using Ray Serve, Transfer | CODE |
| MEDIUM | tests/test_qwen3_4B_streaming_partial_rollout.py | 36 | # loop is still the stock sglang one (semaphore, abort orchestration). | COMMENT |
| MEDIUM | docs/zh/get_started/customization.md | 12 | | [`--custom-generate-function-path`](#custom-generate-function-path) | 仅覆盖生成步骤（例如用于 RAG 或工具调用）。 | | CODE |
| MEDIUM | docs/zh/get_started/customization.md | 32 | ## 通过 customization 接口实现 agentic workflow | COMMENT |
| MEDIUM | docs/en/get_started/customization.md | 12 | | [`--custom-generate-function-path`](#custom-generate-function-path) | Override only the generation step (e.g., for RAG | CODE |
| MEDIUM | examples/multi_agent/run-qwen3-30B-A3B-multi-agent.sh | 57 | # multi-agent do not support eval for now | COMMENT |
| MEDIUM | examples/tau-bench/README.md | 50 | "agent_strategy": "tool-calling", # Select between ["tool-calling", "act", "react", "few-shot"], only tool-calling | CODE |
| MEDIUM | examples/tau-bench/generate_with_tau.py | 25 | "agent_strategy": "tool-calling", # Select between ["tool-calling", "act", "react", "few-shot"] | CODE |
| MEDIUM | examples/strands_sglang/README.md | 3 | This example connects `slime` with [`strands-sglang`](https://github.com/horizon-rl/strands-sglang) (SGLang extension fo | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tools/convert_torch_dist_to_hf_parallel.py | 475 | # Check if this is a linear_q_down_proj or linear_kv_down_proj key | COMMENT |
| LOW | slime/backends/megatron_utils/model.py | 560 | # Set grad to zero. | COMMENT |
| LOW | slime/backends/megatron_utils/model_provider.py | 126 | # Check if the custom provider supports vp_stage parameter | COMMENT |
| LOW | slime/backends/megatron_utils/model_provider.py | 227 | # Check if fp8_model_init supports preserve_high_precision_init_val | COMMENT |
| LOW | examples/tau-bench/trainable_agents.py | 326 | # Check if done | COMMENT |
| LOW | examples/strands_sglang/subprocess_interpreter.py | 195 | # Check if command is available | STRING |
| LOW | examples/strands_sglang/subprocess_interpreter.py | 208 | # Check if command is available | STRING |
| LOW | examples/search-r1/qa_em_format.py | 86 | # Check if this is a tag | COMMENT |
| LOW | examples/retool/tool_sandbox.py | 242 | # Return result | COMMENT |
| LOW | examples/retool/generate_with_retool.py | 250 | # Check if total length exceeds max context length | COMMENT |
| LOW | examples/retool/generate_with_retool.py | 352 | # Check if maximum tool call count reached | COMMENT |
| LOW | slime_plugins/rollout_buffer/buffer.py | 77 | # Check if module has TASK_TYPE constant | COMMENT |
| LOW | slime_plugins/rollout_buffer/buffer.py | 82 | # Check if module has run_rollout function | COMMENT |
| LOW | …ime_plugins/rollout_buffer/generator/base_generator.py | 216 | # Check if instance_id should be skipped | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | tools/convert_hf_to_int4_direct.py | 120 | CODE | |
| LOW | slime/backends/megatron_utils/stateless_adam.py | 16 | CODE | |
| LOW | …atron_to_hf/processors/quantizer_compressed_tensors.py | 225 | CODE | |
| LOW | slime/backends/sglang_utils/sglang_engine.py | 395 | CODE | |
| LOW | slime/utils/distributed_utils.py | 55 | CODE | |
| LOW | slime/utils/data.py | 215 | CODE | |
| LOW | tests/test_qwen2.5_0.5B_pipeline_rl.py | 28 | CODE | |
| LOW | tests/test_qwen2.5_0.5B_training_recovery.py | 141 | CODE | |
| LOW | examples/tau-bench/openai_tool_adapter.py | 109 | CODE | |
| LOW | …es/search-r1/local_dense_retriever/retrieval_server.py | 290 | CODE | |
| LOW | examples/retool/tool_sandbox.py | 356 | CODE | |
| LOW | slime_plugins/models/qwen3_5_vl.py | 205 | CODE | |
| LOW | slime_plugins/models/qwen3_5.py | 153 | CODE | |
| LOW | slime_plugins/models/hf_attention.py | 65 | CODE | |
| LOW | slime_plugins/models/qwen3_next.py | 175 | CODE | |
| LOW | slime_plugins/models/glm5/glm5.py | 413 | CODE | |
| LOW | …ime_plugins/models/glm5/ops/tilelang_sparse_mla_bwd.py | 84 | CODE | |
| LOW | slime_plugins/rollout_buffer/buffer.py | 116 | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | slime/utils/seqlen_balancing.py | 1 | # Copied from https://github.com/volcengine/verl/blob/468adf22c43b744348051fccd7a5d830c6c3c36a/verl/utils/seqlen_balanci | COMMENT |
| LOW | slime/utils/routing_replay.py | 61 | # SGLang's deterministic DeepSeek/GLM biased top-k uses | COMMENT |
| LOW | slime/utils/data.py | 321 | # flattened rollout batch (== its index in the rollout debug dump's | COMMENT |
| LOW | slime/rollout/rm_hub/math_dapo_utils.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | tests/test_loss_cp_invariance.py | 161 | COMMENT | |
| LOW | tests/_cp_dist_helpers.py | 41 | import sys | COMMENT |
| LOW | tests/_cp_dist_helpers.py | 141 | # tests and the backward-grad-norm test. Keeping the data in one place so | COMMENT |
| LOW | examples/delta_weight_sync/run-glm4.7-30B-A3B-delta.sh | 1 | #!/bin/bash | COMMENT |
| LOW | examples/train_infer_mismatch_helper/mis.yaml | 1 | # Enable importance sampling, details refer to the comments of compute_mis_weights in mis.py | COMMENT |
| LOW | examples/strands_sglang/subprocess_interpreter.py | 1 | # Copyright 2025-2026 Strands RL Contributors | COMMENT |
| LOW | examples/search-r1/qa_em_format.py | 1 | # Adapt from https://github.com/PeterGriffinJin/Search-R1/blob/ceee7b89655ed52f205b9beb98e1190c3eedcfb0/verl/utils/rewar | COMMENT |
| LOW | …es/search-r1/local_dense_retriever/retrieval_server.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | examples/search-r1/local_dense_retriever/download.py | 1 | # Copyright 2024 Bytedance Ltd. and/or its affiliates | COMMENT |
| LOW | examples/retool/generate_with_retool.py | 361 | # Tool output is appended verbatim and can push total_length past | COMMENT |
| LOW | scripts/run-qwen2.5-0.5B-gb10-smoke.sh | 1 | #!/bin/bash | COMMENT |
| LOW | scripts/low_precision/run-qwen3-30b-a3b-fp8.sh | 1 | #!/bin/bash | COMMENT |
| LOW | slime_plugins/models/glm5/glm5.py | 121 | with torch.no_grad(): | COMMENT |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| HIGH | tests/test_qwen3_4B_ppo_train_critic_only.py | 0 | megatron: - name: default role: critic overrides: lr: 1e-5 - name: default role: actor overrides: lr: 1e-6 | STRING |
| HIGH | tests/test_qwen3_4B_ppo.py | 0 | megatron: - name: default role: critic overrides: lr: 1e-5 - name: default role: actor overrides: lr: 1e-6 | STRING |
| HIGH | tests/test_qwen3_4B_ppo_disaggregate.py | 0 | megatron: - name: default role: critic overrides: lr: 1e-5 - name: default role: actor overrides: lr: 1e-6 | STRING |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | slime/backends/sglang_utils/engine_group.py | 79 | worker_type: str = "regular" # "regular", "prefill", "decode", or "placeholder" | CODE |
| LOW | slime/backends/sglang_utils/engine_group.py | 139 | Placeholder groups (worker_type="placeholder") skip engine creation entirely. | STRING |
| LOW | slime/backends/sglang_utils/engine_group.py | 143 | if self.args.debug_train_only or self.worker_type == "placeholder": | CODE |
| LOW | slime/backends/sglang_utils/engine_group.py | 337 | values = {g.nodes_per_engine for g in self.server_groups if g.worker_type != "placeholder"} | CODE |
| LOW | slime/backends/sglang_utils/engine_group.py | 460 | all_engines=[None] * num_engines if group_config.worker_type != "placeholder" else [], | CODE |
| LOW⚡ | slime/backends/sglang_utils/sglang_config.py | 16 | worker_type: One of "regular", "prefill", "decode", "placeholder", | STRING |
| LOW⚡ | slime/backends/sglang_utils/sglang_config.py | 18 | "placeholder" reserves GPU slots without creating engines. | STRING |
| LOW | slime/backends/sglang_utils/sglang_config.py | 37 | valid_types = {"regular", "prefill", "decode", "placeholder", "encoder"} | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| MEDIUM | slime/rollout/on_policy_distillation.py | 64 | # If you have task rewards, you can add them here. | COMMENT |
| MEDIUM | examples/on_policy_distillation/run-qwen3-8B-opd.sh | 9 | TEACHER_IP="127.0.0.1" # Use localhost here, you can change it to your IP | CODE |
| Severity | File | Line | Snippet | Context |
|---|---|---|---|---|
| LOW | slime/utils/ppo_utils.py | 585 | # Step 3: Compute returns on full response kl tensor. | COMMENT |