NVIDIA/Megatron-LM

Pattern Findings

6402 matches across 19 categories. Click a row to expand file-level details.

Hyper-Verbose Identifiers3468 hits · 3431 pts

Severity	File	Line	Snippet
LOW	pretrain_vlm.py	227	def train_valid_test_datasets_provider(train_val_test_num_samples):
LOW	pretrain_vlm.py	267	def _preprocess_data_for_llava(data):
LOW	pretrain_vlm.py	465	def llava_position_embedding_ranks(pp_ranks):
LOW	train_rl.py	339	def train_valid_test_datasets_provider(train_val_test_num_samples):
LOW	pretrain_gpt.py	267	def core_gpt_dataset_config_from_args(args: Any) -> GPTDatasetConfig:
LOW	pretrain_gpt.py	332	def train_valid_test_datasets_provider(train_val_test_num_samples, vp_stage=None):
LOW	model_provider.py	62	def count_parameters_in_layer(model, layer_name):
LOW	pretrain_hybrid.py	259	def core_gpt_dataset_config_from_args(args: Any) -> GPTDatasetConfig:
LOW	pretrain_hybrid.py	301	def train_valid_test_datasets_provider(train_val_test_num_samples, vp_stage=None):
LOW	gpt_builders.py	103	def _get_transformer_layer_spec(use_te, config):
LOW	tasks/finetune_utils.py	53	def _cross_entropy_forward_step(batch, model):
LOW	tasks/finetune_utils.py	95	def _build_infinite_size_dataloader(dataloader):
LOW	tasks/finetune_utils.py	106	def _build_train_valid_dataloaders(train_dataset, valid_dataset,
LOW	tasks/data_utils.py	35	def build_tokens_types_paddings_from_text(text_a, text_b,
LOW	tasks/data_utils.py	49	def build_tokens_types_paddings_from_ids(text_a_ids, text_b_ids, max_seq_length,
LOW	tasks/eval_utils.py	65	def calculate_correct_answers(name, model, dataloader,
LOW	tasks/eval_utils.py	111	def correct_answers_forward_step(batch, model):
LOW	tools/run_vlm_text_generation.py	161	def generate_and_write_samples(model):
LOW	tools/prepare_cache.py	54	def _normalize_prepare_cache_args(args: Any) -> None:
LOW	tools/prepare_cache.py	65	def _validate_prepare_cache_args(args: Any) -> None:
LOW	tools/prepare_cache.py	82	def _disable_cache_load_only_flags(args: Any) -> Dict[str, bool]:
LOW	tools/prepare_cache.py	102	def _print_effective_configuration(
LOW	tools/prepare_cache.py	120	def core_gpt_dataset_config_from_args(args: Any) -> GPTDatasetConfig:
LOW	tools/run_inference_performance_test.py	48	def add_inference_benchmarking_args(parser):
LOW	tools/build_sequences_per_dataset.py	60	def build_sequences_per_dataset(args):
LOW	tools/run_dynamic_text_generation_server.py	21	def add_text_generation_server_args(parser: argparse.ArgumentParser):
LOW	tools/run_dynamic_text_generation_server.py	37	async def run_text_generation_server(
LOW	tools/checkpoint/saver_hf_llava.py	176	def receive_vision_projection(self):
LOW	tools/checkpoint/saver_hf_llava.py	352	def save_state_dict_to_hf_checkpoint(self):
LOW	tools/checkpoint/schema_core.py	10	def get_core_transformer_block_key(model_key):
LOW	tools/checkpoint/schema_hf.py	173	def get_language_model_schema(
LOW	tools/checkpoint/loader_base.py	30	def _maybe_parse_additional_megatron_args(self, margs, checkpoint_args):
LOW	tools/checkpoint/loader_base.py	90	def _maybe_ensure_additional_required_arguments(self):
LOW	tools/checkpoint/loader_base.py	107	def ensure_required_arguments(self):
LOW	tools/checkpoint/loader_base.py	193	def get_models_for_pipeline_stage(count, dtype):
LOW	tools/checkpoint/loader_base.py	425	def build_checkpoint_metadata(self, true_vocab_size):
LOW	tools/checkpoint/gpt_hybrid_conversion.py	404	def initialize_ssm_layer_params(
LOW	tools/checkpoint/gpt_hybrid_conversion.py	152	def parse_hybrid_layer_pattern(pattern):
LOW	tools/checkpoint/gpt_hybrid_conversion.py	183	def build_layer_index_mapping(layer_types, direction):
LOW	tools/checkpoint/gpt_hybrid_conversion.py	316	def validate_pattern_gpt_compatible(layer_types, direction):
LOW	tools/checkpoint/gpt_hybrid_conversion.py	361	def validate_source_args_gpt_compatible(source_args, direction):
LOW	tools/checkpoint/dist_checkpoint_io.py	69	def resolve_checkpoint_subdir(load_dir):
LOW	tools/checkpoint/dist_checkpoint_io.py	113	def ensure_single_rank_process_group():
LOW	tools/checkpoint/dist_checkpoint_io.py	159	def load_dist_checkpoint_full(load_dir):
LOW	tools/checkpoint/dist_checkpoint_io.py	221	def save_dist_checkpoint_full(
LOW	tools/checkpoint/dist_checkpoint_io.py	260	def write_latest_iteration_marker(save_dir, iteration):
LOW	tools/checkpoint/loader_mixtral_hf.py	28	def load_args_from_checkpoint(args):
LOW	tools/checkpoint/loader_mixtral_hf.py	63	def verify_transformers_version():
LOW	tools/checkpoint/saver_llava.py	134	def _maybe_parse_additional_megatron_args(self, margs):
LOW	tools/checkpoint/saver_llava.py	320	def receive_vision_projection(self):
LOW	tools/checkpoint/checkpoint_inspector.py	288	def save_checkpoint_with_pickle_protocol(state_dict, output_dir, pickle_protocol=4):
LOW	tools/checkpoint/checkpoint_inspector.py	292	def transform_object_override(write_item, obj):
LOW	tools/checkpoint/checkpoint_inspector.py	831	def convert_torch_dist_to_fsdp_dtensor(
LOW	tools/checkpoint/saver_base.py	31	def _maybe_parse_additional_megatron_args(self, margs):
LOW	tools/checkpoint/saver_base.py	38	def insert_megatron_path_and_check_te(self):
LOW	tools/checkpoint/saver_base.py	277	def receive_checkpoint_metadata(self):
LOW	tools/checkpoint/saver_base.py	341	def save_local_models_to_checkpoint(self):
LOW	tools/checkpoint/loader_llava.py	54	def _maybe_parse_additional_megatron_args(self, margs, checkpoint_args):
LOW	tools/checkpoint/loader_llava.py	88	def _maybe_ensure_additional_required_arguments(self):
LOW	tools/checkpoint/loader_llava.py	104	def build_checkpoint_metadata(self, true_vocab_size):
3408 more matches not shown…

Decorative Section Separators696 hits · 2438 pts

Severity	File	Line	Snippet
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	400	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	402	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	132	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	134	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	225	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	227	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	486	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	488	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	530	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	532	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	627	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	629	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	698	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	700	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	722	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/gpt_hybrid_conversion.py	724	# ---------------------------------------------------------------------------
MEDIUM	tools/checkpoint/saver_llava.py	225	#-----------
MEDIUM	tools/checkpoint/saver_llava.py	186	#-----------
MEDIUM	tools/checkpoint/saver_base.py	377	#-----------
MEDIUM	tools/checkpoint/saver_base.py	420	# --------------
MEDIUM	tools/checkpoint/saver_base.py	432	# ------------------
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	360	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	362	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	386	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	388	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	410	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	412	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	433	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	435	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	474	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	476	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	496	# ================================================================
MEDIUM	…s/common_pile_dataset/create_common_pile_ci_dataset.py	498	# ================================================================
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	20	# ── Parse KEY=VALUE positional args ───────────────────────────────────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	50	# ── Read model_config.yaml ────────────────────────────────────────────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	83	# ── Build model args (substituting ${CHECKPOINT_LOAD_PATH}) ───────────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	99	# ── Make image-bundled extras (mamba-ssm) visible to the cog venv ─────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	128	# ── Launch the inference server in the background ─────────────────────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	185	# ── Wait for server readiness ─────────────────────────────────────────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	205	# ── Benchmark sweep ───────────────────────────────────────────────────────────
MEDIUM	…ts/performance_tests/shell_test_utils/run_perf_test.sh	224	# ── Baseline comparison or recording ──────────────────────────────────────────
MEDIUM	tests/unit_tests/test_optimizer_param_scheduler.py	302	# ── get_canonical_lr_for_logging tests ──────────────────────────────────────
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	63	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	65	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	841	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	843	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	1288	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	1290	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	1575	# ===========================================================================
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	1577	# ===========================================================================
MEDIUM	tests/unit_tests/test_argument_utils.py	654	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/test_argument_utils.py	656	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/tokenizers/test_tokenizer.py	640	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/tokenizers/test_tokenizer.py	642	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/tokenizers/test_tokenizer.py	472	# ------------------------------------------------------------------------
MEDIUM	tests/unit_tests/tokenizers/test_tokenizer.py	474	# ------------------------------------------------------------------------
MEDIUM	tests/unit_tests/tokenizers/test_tokenizer.py	512	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/tokenizers/test_tokenizer.py	514	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/fusions/test_bias_dropout_fusion.py	8	# ---------------------------------------------------------------------------
MEDIUM	tests/unit_tests/fusions/test_bias_dropout_fusion.py	10	# ---------------------------------------------------------------------------
636 more matches not shown…

Cross-File Repetition168 hits · 840 pts

Severity	File	Line	Snippet
HIGH	pretrain_gpt.py	0	build the train test and validation datasets. args: train_val_test_num_samples : a list containing the number of samples
HIGH	pretrain_hybrid.py	0	build the train test and validation datasets. args: train_val_test_num_samples : a list containing the number of samples
HIGH	examples/post_training/modelopt/finetune.py	0	build the train test and validation datasets. args: train_val_test_num_samples : a list containing the number of samples
HIGH	examples/t5/pretrain_t5.py	0	build the train test and validation datasets. args: train_val_test_num_samples : a list containing the number of samples
HIGH	megatron/elastification/pretrain_hybrid_flex.py	0	build the train test and validation datasets. args: train_val_test_num_samples : a list containing the number of samples
HIGH	tools/checkpoint/saver_hf_llava.py	0	required top-level function that creates the saver and calls its .save().
HIGH	tools/checkpoint/saver_core.py	0	required top-level function that creates the saver and calls its .save().
HIGH	tools/checkpoint/saver_llava.py	0	required top-level function that creates the saver and calls its .save().
HIGH	tools/checkpoint/loader_base.py	0	orchestrates loading a megatron checkpoint and sending model parameters over a given multiprocessing queue. args: args:
HIGH	tools/checkpoint/loader_core.py	0	orchestrates loading a megatron checkpoint and sending model parameters over a given multiprocessing queue. args: args:
HIGH	tools/checkpoint/loader_llava.py	0	orchestrates loading a megatron checkpoint and sending model parameters over a given multiprocessing queue. args: args:
HIGH	tools/checkpoint/loader_base.py	0	parse megatron arguments by forcibly overwriting sys.argv. populates self.margs and self.checkpoint_args.
HIGH	tools/checkpoint/saver_base.py	0	parse megatron arguments by forcibly overwriting sys.argv. populates self.margs and self.checkpoint_args.
HIGH	tools/checkpoint/loader_llava.py	0	parse megatron arguments by forcibly overwriting sys.argv. populates self.margs and self.checkpoint_args.
HIGH	tools/checkpoint/loader_base.py	0	construct a sys.argv list for megatron's argument parser. this centralizes the hack of overwriting sys.argv.
HIGH	tools/checkpoint/loader_core.py	0	construct a sys.argv list for megatron's argument parser. this centralizes the hack of overwriting sys.argv.
HIGH	tools/checkpoint/saver_base.py	0	construct a sys.argv list for megatron's argument parser. this centralizes the hack of overwriting sys.argv.
HIGH	tools/checkpoint/loader_llava.py	0	construct a sys.argv list for megatron's argument parser. this centralizes the hack of overwriting sys.argv.
HIGH	tests/unit_tests/test_fp8_param.py	0	dp_overlap: (overlap_param_gather, overlap_grad_reduce)
HIGH	tests/unit_tests/test_fp8_param.py	0	dp_overlap: (overlap_param_gather, overlap_grad_reduce)
HIGH	tests/unit_tests/test_fp8_param.py	0	dp_overlap: (overlap_param_gather, overlap_grad_reduce)
HIGH	tests/unit_tests/test_hyper_comm_grid.py	0	set up distributed environment for the entire test class.
HIGH	…it_tests/pipeline_parallel/test_bridge_communicator.py	0	set up distributed environment for the entire test class.
HIGH	…sts/pipeline_parallel/test_multimodule_communicator.py	0	set up distributed environment for the entire test class.
HIGH	tests/unit_tests/test_argument_utils.py	0	config with argparse_meta metadata for testing overrides.
HIGH	tests/unit_tests/test_argument_utils.py	0	config with argparse_meta metadata for testing overrides.
HIGH	tests/unit_tests/test_argument_utils.py	0	config with argparse_meta metadata for testing overrides.
HIGH	…sts/unit_tests/transformer/test_submodule_callables.py	0	runs the model in reference mode and captures outputs and gradients. args: model: the transformer model to run. input_te
HIGH	…sts/unit_tests/a2a_overlap/test_schedule_layer_1f1b.py	0	runs the model in reference mode and captures outputs and gradients. args: model: the transformer model to run. input_te
HIGH	…sts/unit_tests/a2a_overlap/test_schedule_layer_1f1b.py	0	runs the model in reference mode and captures outputs and gradients. args: model: the transformer model to run. input_te
HIGH	…sts/unit_tests/transformer/test_submodule_callables.py	0	runs the model with all-to-all overlap optimization and captures outputs and gradients. args: model: the transformer mod
HIGH	…sts/unit_tests/a2a_overlap/test_schedule_layer_1f1b.py	0	runs the model with all-to-all overlap optimization and captures outputs and gradients. args: model: the transformer mod
HIGH	…sts/unit_tests/a2a_overlap/test_schedule_layer_1f1b.py	0	runs the model with all-to-all overlap optimization and captures outputs and gradients. args: model: the transformer mod
HIGH	…it_tests/pipeline_parallel/test_bridge_communicator.py	0	destroy all tracked grids and bridge communicator pgs.
HIGH	…_tests/pipeline_parallel/test_multimodule_schedules.py	0	destroy all tracked grids and bridge communicator pgs.
HIGH	tests/unit_tests/models/test_mimo_1f1b_schedule.py	0	destroy all tracked grids and bridge communicator pgs.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	helper method to get the model chunk id given the iteration number.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	helper method to get the model chunk id given the iteration number.
HIGH	megatron/core/pipeline_parallel/schedules.py	0	helper method to get the model chunk id given the iteration number.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	helper method to get the microbatch_id within model chunk given the iteration number.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	helper method to get the microbatch_id within model chunk given the iteration number.
HIGH	megatron/core/pipeline_parallel/schedules.py	0	helper method to get the microbatch_id within model chunk given the iteration number.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	check if an iteration is the first for a model chunk.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	check if an iteration is the first for a model chunk.
HIGH	megatron/core/pipeline_parallel/schedules.py	0	check if an iteration is the first for a model chunk.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	check if an iteration is the last for a model chunk.
HIGH	tests/unit_tests/pipeline_parallel/test_helpers.py	0	check if an iteration is the last for a model chunk.
HIGH	megatron/core/pipeline_parallel/schedules.py	0	check if an iteration is the last for a model chunk.
HIGH	tests/unit_tests/models/test_mimo_1f1b_schedule.py	0	return a transformerconfig for the vision projection mlp.
HIGH	examples/mimo/configs/llava_vlm.py	0	return a transformerconfig for the vision projection mlp.
HIGH	examples/mimo/configs/llava_avlm.py	0	return a transformerconfig for the vision projection mlp.
HIGH	…unit_tests/dist_checkpointing/models/test_gpt_model.py	0	test model loading with different vocab size (caused by tp padding).
HIGH	…nit_tests/dist_checkpointing/models/test_bert_model.py	0	test model loading with different vocab size (caused by tp padding).
HIGH	tests/unit_tests/dist_checkpointing/models/common.py	0	test model loading with different vocab size (caused by tp padding).
HIGH	…ultimodal/evaluation/evaluate_video_phys_game_bench.py	0	merge input files to a format compatible with the evaluator.
HIGH	examples/multimodal/evaluation/evaluate_realworldqa.py	0	merge input files to a format compatible with the evaluator.
HIGH	examples/multimodal/evaluation/evaluate_ai2d.py	0	merge input files to a format compatible with the evaluator.
HIGH	…es/multimodal/evaluation/evaluate_video_motionbench.py	0	merge input files to a format compatible with the evaluator.
HIGH	examples/multimodal/evaluation/evaluate_spdocvqa.py	0	merge input files to a format compatible with the evaluator.
HIGH	examples/multimodal/evaluation/evaluate_textvqa.py	0	merge input files to a format compatible with the evaluator.
108 more matches not shown…

Unused Imports795 hits · 750 pts

Severity	File	Line
LOW	train_rl.py	4
LOW	mamba_builders.py	14
LOW	mamba_builders.py	15
LOW	gpt_builders.py	19
LOW	tools/merge_datasets.py	3
LOW	tools/preprocess_data_nmt.py	6
LOW	tools/run_inference_performance_test.py	11
LOW	tools/run_inference_performance_test.py	12
LOW	tools/run_inference_performance_test.py	30
LOW	tools/run_inference_performance_test.py	36
LOW	tools/run_inference_performance_test.py	40
LOW	tools/run_inference_performance_test.py	41
LOW	tools/linter.py	11
LOW	tools/preprocess_mmdata.py	12
LOW	tools/checkpoint/saver_core.py	3
LOW	tools/checkpoint/saver_core.py	5
LOW	tools/checkpoint/saver_core.py	7
LOW	tools/checkpoint/saver_core.py	8
LOW	tools/checkpoint/remap_gpt_dsa_to_mamba.py	38
LOW	tools/checkpoint/remap_gpt_dsa_to_mamba.py	39
LOW	tools/checkpoint/remap_gpt_dsa_to_mamba.py	108
LOW	tools/checkpoint/gpt_hybrid_conversion.py	122
LOW	tools/checkpoint/dist_checkpoint_io.py	20
LOW	tools/checkpoint/dist_checkpoint_io.py	31
LOW	tools/checkpoint/utils.py	3
LOW	tools/checkpoint/hybrid_conversion.py	10
LOW	tools/checkpoint/loader_mixtral_hf.py	3
LOW	tools/checkpoint/loader_mixtral_hf.py	167
LOW	tools/checkpoint/saver_llava.py	2
LOW	tools/checkpoint/saver_llava.py	4
LOW	tools/checkpoint/saver_llava.py	173
LOW	tools/checkpoint/saver_llava.py	173
LOW	tools/checkpoint/checkpoint_inspector.py	37
LOW	tools/checkpoint/loader_core.py	3
LOW	tools/checkpoint/loader_core.py	4
LOW	tools/checkpoint/loader_core.py	5
LOW	tools/checkpoint/loader_core.py	6
LOW	tools/checkpoint/loader_core.py	7
LOW	tools/checkpoint/saver_base.py	2
LOW	tools/checkpoint/loader_llava.py	3
LOW	tools/checkpoint/loader_llava.py	7
LOW	tools/bert_embedding/embed.py	3
LOW	tools/bert_embedding/embed.py	8
LOW	tools/bert_embedding/embed.py	15
LOW	tools/bert_embedding/embed.py	16
LOW	tools/bert_embedding/embed.py	19
LOW	tools/bert_embedding/embed.py	21
LOW	tools/bert_embedding/embed.py	21
LOW	tools/bert_embedding/__init__.py	3
LOW	tools/bert_embedding/__init__.py	3
LOW	…formance_tests/shell_test_utils/compare_to_baseline.py	23
LOW	tests/unit_tests/test_fp8_param.py	3
LOW	tests/unit_tests/test_fp8_param.py	38
LOW	tests/unit_tests/test_inference.py	3
LOW	tests/unit_tests/test_basic.py	2
LOW	tests/unit_tests/conftest.py	4
LOW	tests/unit_tests/test_emerging_optimizers.py	11
LOW	tests/unit_tests/test_imports.py	14
LOW	tests/unit_tests/test_fp4_param.py	5
LOW	tests/unit_tests/test_fp4_param.py	11
735 more matches not shown…

Deep Nesting544 hits · 468 pts

Severity	File	Line
LOW	gpt_builders.py	24
LOW	tasks/finetune_utils.py	147
LOW	tasks/eval_utils.py	65
LOW	.gitlab/scripts/check_imports.py	66
LOW	.gitlab/scripts/check_imports.py	123
LOW	tools/preprocess_data.py	310
LOW	tools/run_vlm_text_generation.py	88
LOW	tools/run_text_generation_server.py	115
LOW	tools/checkpoint/saver_hf_llava.py	69
LOW	tools/checkpoint/loader_base.py	254
LOW	tools/checkpoint/gpt_hybrid_conversion.py	631
LOW	tools/checkpoint/gpt_hybrid_conversion.py	726
LOW	tools/checkpoint/dist_checkpoint_io.py	131
LOW	tools/checkpoint/hybrid_conversion.py	59
LOW	tools/checkpoint/hybrid_conversion.py	114
LOW	tools/checkpoint/hybrid_conversion.py	170
LOW	tools/checkpoint/hybrid_conversion.py	222
LOW	tools/checkpoint/saver_llava.py	183
LOW	tools/checkpoint/checkpoint_inspector.py	74
LOW	tools/checkpoint/checkpoint_inspector.py	335
LOW	tools/checkpoint/checkpoint_inspector.py	951
LOW	tools/checkpoint/saver_base.py	365
LOW	tools/checkpoint/loader_llava.py	161
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	63
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	109
LOW	…formance_tests/shell_test_utils/compare_to_baseline.py	36
LOW	tests/unit_tests/test_fp8_param.py	214
LOW	tests/unit_tests/test_optimizer.py	574
LOW	tests/unit_tests/test_utils.py	311
LOW	tests/unit_tests/test_imports.py	63
LOW	tests/unit_tests/test_layer_wise_optimizer.py	254
LOW	tests/unit_tests/test_layer_wise_optimizer.py	724
LOW	tests/unit_tests/ssm/ops/test_ssd_bmm.py	68
LOW	tests/unit_tests/tokenizers/test_tokenizer.py	290
LOW	…t_tests/tools/checkpoint/test_gpt_hybrid_conversion.py	461
LOW	tests/unit_tests/extension/test_kitchen_sdpa.py	60
LOW	…egatron_fsdp/test_mcore_fully_sharded_data_parallel.py	1067
LOW	…egatron_fsdp/test_mcore_fully_sharded_data_parallel.py	1139
LOW	…egatron_fsdp/test_mcore_fully_sharded_data_parallel.py	1142
LOW	…ts/distributed/megatron_fsdp/test_mfsdp_fully_shard.py	176
LOW	…ts/distributed/megatron_fsdp/test_mfsdp_fully_shard.py	888
LOW	…ts/distributed/megatron_fsdp/test_mfsdp_fully_shard.py	1058
LOW	tests/unit_tests/transformer/test_cuda_graphs.py	450
LOW	tests/unit_tests/transformer/test_cuda_graphs.py	642
LOW	tests/unit_tests/transformer/test_transformer_block.py	405
LOW	…mental_attention_variant/test_attention_variant_dsa.py	411
LOW	…mental_attention_variant/test_attention_variant_dsa.py	997
LOW	…mental_attention_variant/test_attention_variant_dsa.py	1031
LOW	…mental_attention_variant/test_attention_variant_dsa.py	1117
LOW	…mental_attention_variant/test_attention_variant_dsa.py	1245
LOW	…mental_attention_variant/test_attention_variant_dsa.py	1282
LOW	…mental_attention_variant/test_attention_variant_dsa.py	1496
LOW	tests/unit_tests/training/config/test_container_base.py	248
LOW	…it_tests/pipeline_parallel/test_bridge_communicator.py	311
LOW	…sts/pipeline_parallel/test_multimodule_communicator.py	170
LOW	…sts/pipeline_parallel/test_multimodule_communicator.py	609
LOW	…_tests/pipeline_parallel/test_multimodule_schedules.py	258
LOW	tests/unit_tests/pipeline_parallel/test_schedules.py	89
LOW	…ne_parallel/test_fine_grained_activation_offloading.py	335
LOW	…ts/unit_tests/models/test_dsa_gpt_mamba_equivalence.py	221
484 more matches not shown…

Hallucination Indicators30 hits · 415 pts

Severity	File	Line	Snippet
CRITICAL	tools/checkpoint/loader_mixtral_hf.py	69	model.embedding.word_embeddings.weight.data.copy_(
CRITICAL	tools/checkpoint/loader_mixtral_hf.py	74	model.decoder.final_layernorm.weight.data.copy_(hf_model.model.norm.weight)
CRITICAL	tools/checkpoint/loader_mixtral_hf.py	103	layer.mlp.router.weight.data.copy_(hf_layer.block_sparse_moe.gate.weight)
CRITICAL	tools/checkpoint/loader_mixtral_hf.py	127	layer.self_attention.linear_qkv.layer_norm_weight.data.copy_(hf_layer.input_layernorm.weight)
CRITICAL	tools/checkpoint/saver_llava.py	199	model.vision_model.conv1.weight.data.copy_(vit_embeddings_msg["conv1 weight"])
CRITICAL	tools/checkpoint/saver_llava.py	201	model.vision_model.conv1.bias.data.copy_(vit_embeddings_msg["conv1 bias"])
CRITICAL	tools/checkpoint/saver_llava.py	202	model.vision_model.position_embeddings.weight.data.copy_(vit_embeddings_msg["position embeddings"])
CRITICAL	tools/checkpoint/saver_llava.py	208	model.vision_model.embedder.weight.data.copy_(embedder_weight[tp_rank])
CRITICAL	tools/checkpoint/saver_llava.py	210	model.vision_model.embedder.bias.data.copy_(embedder_bias[tp_rank])
CRITICAL	tools/checkpoint/saver_llava.py	214	model.vision_model.ln_pre.weight.data.copy_(vit_embeddings_msg["ln pre weight"])
CRITICAL	tools/checkpoint/saver_llava.py	215	model.vision_model.ln_pre.bias.data.copy_(vit_embeddings_msg["ln pre bias"])
CRITICAL	tools/checkpoint/saver_llava.py	218	model.vision_model.ln_post.weight.data.copy_(vit_embeddings_msg["ln post weight"])
CRITICAL	tools/checkpoint/saver_llava.py	219	model.vision_model.ln_post.bias.data.copy_(vit_embeddings_msg["ln post bias"])
CRITICAL	tools/checkpoint/saver_llava.py	343	model.vision_projection.encoder.linear_fc1.weight.data.copy_(
CRITICAL	tools/checkpoint/saver_llava.py	345	model.vision_projection.encoder.linear_fc2.weight.data.copy_(
CRITICAL	tools/checkpoint/saver_llava.py	348	model.vision_projection.encoder.linear_fc1.layer_norm_weight.data.copy_(
CRITICAL	tools/checkpoint/saver_llava.py	351	model.vision_projection.encoder.linear_fc1.layer_norm_bias.data.copy_(
CRITICAL	tools/checkpoint/saver_llava.py	354	model.vision_projection.encoder.linear_fc1.bias.data.copy_(
CRITICAL	tools/checkpoint/saver_llava.py	356	model.vision_projection.encoder.linear_fc2.bias.data.copy_(vision_projection_l1_bias)
CRITICAL	tests/unit_tests/transformer/moe/test_routers.py	115	assert self.sequential_mlp.router.weight.grad.abs().sum() == 0
CRITICAL	tests/unit_tests/transformer/moe/test_routers.py	121	assert self.sequential_mlp.router.weight.grad.abs().sum() > 0
CRITICAL	tests/unit_tests/transformer/moe/test_routers.py	126	self.sequential_mlp.router.weight.grad.fill_(0)
CRITICAL	tests/unit_tests/transformer/moe/test_routers.py	129	assert self.sequential_mlp.router.weight.grad.abs().sum() > 0
CRITICAL	tests/unit_tests/models/test_gpt_model.py	87	assert self.gpt_model.embedding.word_embeddings.weight.std().cpu().item() == approx(
CRITICAL	tests/unit_tests/models/test_gpt_model.py	90	assert self.gpt_model.embedding.word_embeddings.weight.mean().cpu().item() == approx(
CRITICAL	megatron/core/parallel_state.py	2158	and torch.distributed.distributed_c10d._world.pg_map.get(_DATA_PARALLEL_GROUP_GLOO, None)
CRITICAL	megatron/core/parallel_state.py	2167	and torch.distributed.distributed_c10d._world.pg_map.get(
CRITICAL	megatron/core/parallel_state.py	2206	and torch.distributed.distributed_c10d._world.pg_map.get(
CRITICAL	megatron/core/parallel_state.py	2220	and torch.distributed.distributed_c10d._world.pg_map.get(
CRITICAL	megatron/core/transformer/attention.py	1731	self.linear_qkv.weight.main_param.data.copy_(

Self-Referential Comments131 hits · 381 pts

Severity	File	Line	Snippet
MEDIUM	gpt_builders.py	43	# Define the decoder block spec
MEDIUM	gpt_builders.py	55	# Define the decoder layer spec
MEDIUM	gpt_builders.py	68	# Define the decoder block spec
MEDIUM	.gitlab/stages/05.publish.yml	149	# Define the full refspec for the branch
MEDIUM	tools/run_inference_performance_test.py	112	# Create a list of valid token IDs
MEDIUM	tools/common_pile_dataset/setup_common_pile_dataset.sh	87	# Create a virtual environment to avoid system package conflicts
MEDIUM	tools/common_pile_dataset/setup_common_pile_dataset.sh	122	# Create the output directory
MEDIUM	tests/unit_tests/test_optimizer.py	790	# Create a new state_dict with all params set to 3.
MEDIUM	tests/unit_tests/test_optimizer.py	879	# Create a simple model for testing
MEDIUM	tests/unit_tests/test_optimizer.py	961	# Create a simple model
MEDIUM	tests/unit_tests/test_utils.py	454	# Create a straggler_detector with enabled set to false.
MEDIUM	tests/unit_tests/test_hyper_comm_grid.py	431	# Create a process group
MEDIUM	tests/unit_tests/test_hyper_comm_grid.py	434	# Create a tensor for communication test
MEDIUM	tests/unit_tests/test_hyper_comm_grid.py	535	# Create a unique tensor based on rank
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	70	# Create a simple linear model for testing
MEDIUM	tests/unit_tests/test_emerging_optimizers.py	640	# Create a model with QKV-like parameter
MEDIUM	tests/unit_tests/test_training.py	143	# Create a mock state_dict with gradients (use deterministic values for reproducibility).
MEDIUM	tests/unit_tests/ssm/ops/test_ssm_kernel.py	83	# Create the Mixer instance directly
MEDIUM	tests/unit_tests/fusions/test_torch_softmax.py	197	# Create a padding mask
MEDIUM	…/unit_tests/post_training/test_modelopt_module_spec.py	238	# Define the expected signature
MEDIUM	…/unit_tests/post_training/test_modelopt_module_spec.py	277	# Define the expected signature
MEDIUM	…tests/distributed/test_torch_fully_sharded_parallel.py	56	# Create a dummy model and configs.
MEDIUM	…tests/distributed/test_torch_fully_sharded_parallel.py	63	# Create the sharded model.
MEDIUM	…tests/distributed/test_torch_fully_sharded_parallel.py	84	# Create a dummy model and configs.
MEDIUM	…tests/distributed/test_torch_fully_sharded_parallel.py	90	# Create a custom process group (using the default world for testing)
MEDIUM	…tests/distributed/test_torch_fully_sharded_parallel.py	93	# Create the sharded model with explicit process group
MEDIUM	…distributed/megatron_fsdp/test_mfsdp_uneven_dtensor.py	675	# Create a manual uneven sharding along dim 1 with possible zero-length local on some ranks
MEDIUM	…/unit_tests/transformer/test_multi_token_prediction.py	774	# Create a dummy loss tensor
MEDIUM	…tests/transformer/test_transformer_block_custom_pgs.py	259	# Create a transformer block with default process groups
MEDIUM	…tests/transformer/test_transformer_block_custom_pgs.py	285	# Create a transformer block with custom process groups
MEDIUM	…tests/transformer/test_transformer_block_custom_pgs.py	737	# Create a single transformer block
MEDIUM	tests/unit_tests/transformer/test_cuda_graphs.py	718	# Create the CUDA graphs - this is where the is_last_layer logic is tested
MEDIUM	tests/unit_tests/transformer/test_cuda_graphs.py	965	# Create a mapping of sample_keys to indices
MEDIUM	tests/unit_tests/transformer/test_transformer_block.py	545	# Create a new build_layers method that uses interleaved attention
MEDIUM	…ts/unit_tests/transformer/moe/test_token_dispatcher.py	172	# Create the answer.
MEDIUM	tests/unit_tests/transformer/moe/test_shared_experts.py	96	# Create a dummy input tensor.
MEDIUM	tests/unit_tests/transformer/moe/test_aux_loss.py	211	# Create a new config with updated parameters
MEDIUM	tests/unit_tests/transformer/moe/test_aux_loss.py	214	# Create the router with the updated config
MEDIUM	…s/unit_tests/training/config/test_instantiate_utils.py	334	# Create a mock that raises an error when used with functools.partial
MEDIUM	tests/unit_tests/training/config/test_yaml_utils.py	215	# Create a mock torch dtype
MEDIUM	tests/unit_tests/training/config/test_yaml_utils.py	262	# Create a mock GenerationConfig
MEDIUM	tests/unit_tests/utils/test_experimental_log_once.py	23	# Define a fresh function with the decorator so it has its own closure state.
MEDIUM	tests/unit_tests/models/test_mimo_audio_submodules.py	212	# Create a time array
MEDIUM	tests/unit_tests/models/test_mimo_audio_submodules.py	215	# Create a simple sine wave at 440 Hz (A4)
MEDIUM	tests/unit_tests/models/test_gpt_model.py	237	# Define the expected signature
MEDIUM	…sts/unit_tests/models/test_mimo_embedding_alignment.py	22	# Create a minimal MimoModelConfig
MEDIUM	…sts/unit_tests/models/test_mimo_embedding_alignment.py	73	# Create a simple batch
MEDIUM	…sts/unit_tests/models/test_mimo_embedding_alignment.py	249	# Create a test case with 2 batches:
MEDIUM	…sts/unit_tests/models/test_mimo_embedding_alignment.py	285	# Create the unflattened embeddings that would come from a vision encoder
MEDIUM	tests/unit_tests/models/test_fastconformer_model.py	45	# Create a parameter with the target dtype so ``next(self.parameters()).dtype``
MEDIUM	tests/unit_tests/models/test_mimo_submodules.py	92	# Create the main module spec
MEDIUM	tests/unit_tests/models/test_mimo_submodules.py	351	# Create a data batch without images
MEDIUM	tests/unit_tests/dist_checkpointing/test_fp8.py	33	# Create a quantizer for FP8 conversion
MEDIUM	…s/unit_tests/dist_checkpointing/test_fully_parallel.py	601	# Create a mock that will do what it's supposed to do,
MEDIUM	…/unit_tests/inference/contexts/test_dynamic_context.py	317	# Initialize all variables
MEDIUM	…/unit_tests/inference/contexts/test_dynamic_context.py	896	# Create an active_requests_mask where requests 0, 2, and 4 are finished (0),
MEDIUM	…/unit_tests/inference/contexts/test_dynamic_context.py	986	# Create an active_requests_mask where all requests are finished
MEDIUM	…ts/unit_tests/inference/engines/test_dynamic_engine.py	1894	# Create a deterministic mock forward pass that returns logits
MEDIUM	…ts/unit_tests/inference/engines/test_dynamic_engine.py	1590	# Create a request with length 513
MEDIUM	tests/unit_tests/data/test_builder.py	93	# Define the class here to avoid pytest warnings
71 more matches not shown…

Excessive Try-Catch Wrapping192 hits · 228 pts

Severity	File	Line	Snippet
LOW	tasks/finetune_utils.py	61	except Exception:
LOW	tasks/eval_utils.py	114	except Exception:
MEDIUM	tasks/eval_utils.py	111	def correct_answers_forward_step(batch, model):
LOW	.gitlab/scripts/check_imports.py	117	except Exception:
LOW	tools/check_copyright.py	22	except Exception as e:
MEDIUM	tools/check_copyright.py	23	print(f"Error reading {file_path}: {e}")
MEDIUM	tools/text_generation_cli.py	21	print(f"Error {response.status_code}: {response.json()['message']}")
LOW	tools/checkpoint/saver_hf_llava.py	400	except Exception as e:
LOW	tools/checkpoint/loader_base.py	78	except Exception as e:
MEDIUM	tools/checkpoint/loader_base.py	79	print(f"Error validating Megatron arguments: {e}")
LOW	tools/checkpoint/saver_core.py	70	except Exception as e:
LOW	tools/checkpoint/gpt_hybrid_conversion.py	384	except Exception:
LOW	tools/checkpoint/dist_checkpoint_io.py	215	except Exception:
LOW	tools/checkpoint/hybrid_conversion.py	322	except Exception:
LOW	tools/checkpoint/loader_mixtral_hf.py	340	except Exception:
MEDIUM	tools/checkpoint/loader_mixtral_hf.py	337	def load_checkpoint(queue, args):
LOW	tools/checkpoint/saver_llava.py	404	except Exception as e:
LOW	tools/checkpoint/loader_core.py	90	except Exception as e:
LOW	tools/checkpoint/loader_llava.py	360	except Exception as e:
LOW	tools/bert_embedding/embed.py	217	except Exception:
LOW	tools/bert_embedding/embed.py	223	except Exception:
LOW	tests/unit_tests/conftest.py	48	except Exception:
LOW	tests/unit_tests/conftest.py	96	except Exception as e:
LOW	tests/unit_tests/test_hyper_comm_grid.py	325	except Exception as e:
LOW	tests/unit_tests/test_api_backwards_compat_setup.py	106	except Exception as e:
LOW	tests/unit_tests/test_api_backwards_compat_setup.py	134	except Exception as e:
LOW	tests/unit_tests/test_imports.py	58	except Exception:
LOW	tests/unit_tests/test_imports.py	77	except Exception:
LOW	tests/unit_tests/test_utilities.py	107	except Exception:
LOW	tests/unit_tests/tokenizers/test_tokenizer.py	649	except Exception:
LOW	tests/unit_tests/tokenizers/test_tokenizer.py	18	except Exception:
LOW	tests/unit_tests/tokenizers/test_tokenizer.py	151	except Exception:
LOW	tests/unit_tests/tokenizers/test_tokenizer.py	177	except Exception:
LOW	tests/unit_tests/tokenizers/test_tokenizer.py	319	except Exception:
LOW	…egatron_fsdp/test_mcore_fully_sharded_data_parallel.py	1156	except Exception:
LOW	…ts/distributed/megatron_fsdp/test_mfsdp_fully_shard.py	73	except Exception as e:
LOW	tests/unit_tests/transformer/moe/test_aux_loss.py	38	except Exception: # pragma: no cover - defensive
LOW	tests/unit_tests/transformer/moe/test_routers.py	25	except Exception: # pragma: no cover - defensive
LOW	tests/unit_tests/models/test_mimo_audio_submodules.py	123	except Exception as e:
LOW	tests/unit_tests/models/test_mimo_audio_submodules.py	197	except Exception as e:
LOW	tests/unit_tests/models/test_hybrid_moe_model.py	405	except Exception:
LOW	tests/unit_tests/models/test_mimo_submodules.py	55	except Exception as e:
LOW	tests/unit_tests/models/test_mimo_submodules.py	105	except Exception as e:
LOW	tests/unit_tests/models/test_mimo_submodules.py	229	except Exception as e:
LOW	tests/unit_tests/models/test_mimo_submodules.py	284	except Exception as e:
MEDIUM	tests/unit_tests/models/test_mimo_model.py	147	def setup_method(self, method):
MEDIUM	tests/unit_tests/models/test_mimo_model.py	163	def teardown_method(self, method):
MEDIUM	tests/unit_tests/models/test_mimo_model.py	528	def setup_method(self, method):
MEDIUM	tests/unit_tests/models/test_mimo_model.py	542	def teardown_method(self, method):
LOW	tests/unit_tests/models/test_mimo_model.py	150	except Exception:
LOW	tests/unit_tests/models/test_mimo_model.py	166	except Exception:
LOW	tests/unit_tests/models/test_mimo_model.py	531	except Exception:
LOW	tests/unit_tests/models/test_mimo_model.py	545	except Exception:
LOW	tests/unit_tests/dist_checkpointing/test_fp8.py	24	except Exception as e:
LOW	tests/unit_tests/dist_checkpointing/test_async_save.py	30	except Exception as e:
LOW	tests/unit_tests/resharding/test_model_swap.py	34	except Exception:
LOW	tests/unit_tests/resharding/test_model_swap.py	44	except Exception:
LOW	tests/unit_tests/resharding/test_model_swap.py	365	except Exception:
LOW	tests/unit_tests/resharding/test_model_swap.py	492	except Exception:
LOW	tests/unit_tests/resharding/test_model_swap.py	644	except Exception:
132 more matches not shown…

Redundant / Tautological Comments127 hits · 194 pts

Severity	File	Line	Snippet
LOW	train_rl.py	62	# Check if fp8_model_init supports preserve_high_precision_init_val
LOW	tools/check_copyright.py	20	# Check if the expected header is at the start of the file
LOW	tools/common_pile_dataset/setup_common_pile_dataset.sh	50	# Check if create_common_pile_ci_dataset.py was scp'd alongside this script
LOW	docker/common/install_source_wheels.sh	25	# Check if required arguments are provided
LOW	tests/unit_tests/test_optimizer.py	34	# Check if FP8 block scaling is available.
LOW	tests/unit_tests/test_utils.py	457	# Check if configuration was success.
LOW	tests/unit_tests/test_utils.py	460	# Check if the instance is in disabled state.
LOW	tests/unit_tests/test_utils.py	464	# Check if all ranks have straggler detector enabled.
LOW	tests/unit_tests/conftest.py	83	# Check if data directory exists and has content
LOW	tests/unit_tests/fusions/test_torch_softmax.py	145	# Check if output is a valid probability distribution
LOW	tests/unit_tests/transformer/test_cuda_graphs.py	516	# Check if cuda graph is correctly setting is first/last layer
LOW	tests/unit_tests/transformer/test_attention.py	416	# Check if output and bias have the correct shape
LOW	tests/unit_tests/transformer/test_attention.py	598	# Check if the output is close
LOW	…/unit_tests/transformer/test_multi_latent_attention.py	1551	# Check if the output is the same
LOW	…it_tests/transformer/moe/test_moe_layer_discrepancy.py	69	# Check if parameters are the same
LOW	…it_tests/transformer/moe/test_moe_layer_discrepancy.py	75	# Check if input is the same across all ranks
LOW	…it_tests/transformer/moe/test_moe_layer_discrepancy.py	150	# Check if output is the same across all ranks
LOW	…it_tests/transformer/moe/test_moe_layer_discrepancy.py	216	# Check if output is the same across all ranks
LOW	tests/unit_tests/transformer/moe/test_moe_layer.py	164	# Check if the moe layer is interleaved correctly
LOW	tests/unit_tests/models/test_llava_model.py	753	# Check if output shape is as expected
LOW	…_tests/dist_checkpointing/test_layer_wise_optimizer.py	147	# Check if optimizer is ChainedOptimizer (expected for standard setup)
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	879	# Assign blocks to the requests (one block per request)
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	955	# Assign blocks to the requests:
LOW	…ts/unit_tests/inference/engines/test_dynamic_engine.py	3183	# Check if any request was evicted during this step
LOW	…ts/test_utils/python_scripts/download_golden_values.py	282	# Check if we should skip based on only_failing flag
LOW	…nal_tests/test_cases/common/ckpt_converter/__main__.py	895	# Print results.
LOW	…unctional_tests/shell_test_utils/run_batch_ci_tests.sh	173	# Check if file is empty (job still running or not started)
LOW	tests/functional_tests/shell_test_utils/run_ci_test.sh	11	# Set umask to 0002 to allow group read/write permissions
LOW	tests/functional_tests/shell_test_utils/run_ci_test.sh	257	## Loop over the list of model configs in the params file and run each one in sequence, collecting
LOW	docs/conf.py	43	# Check if we should skip autodoc generation
LOW	examples/post_training/modelopt/finetune.py	271	# Check if this is OpenAI chat data?
LOW	examples/mimo/utils/logging.py	40	# Print output projections
LOW	examples/gptoss/02_train.sh	55	# Check if checkpoint path exists
LOW	examples/gptoss/02_train.sh	62	# Check if tensorboard logs path exists
LOW	examples/multimodal/dataset_helpers.py	121	# Check if all samples fit in the knapsack capacity.
LOW	examples/inference/utils.py	235	# Check if we have any prompts (from command line or JSONL)
LOW	examples/inference/advanced/gpt_dynamic_inference.py	263	# Check if all requests are finished.
LOW	examples/rl/benchmark_refit.py	216	# Print results
LOW	examples/rl/benchmark_refit.py	312	# Print results
LOW	examples/rl/environments/countdown/countdown.py	34	# Check if all numbers in equation are available
LOW	scripts/check_api_backwards_compatibility.py	240	# Check if this breakage kind should be ignored globally (not a signature change)
LOW	scripts/check_api_backwards_compatibility.py	253	# Check if this is a breakage kind we ignore for __init__ methods
LOW	scripts/check_api_backwards_compatibility.py	263	# Check if it's a child of a filtered object
LOW	scripts/check_api_backwards_compatibility.py	350	# Print results
LOW	.github/actions/check-nvidia-sso-membership/action.yml	94	# Check if SSO file is available
LOW	.github/actions/check-nvidia-sso-membership/action.yml	103	# Check if username exists as a key in the JSON object
LOW	megatron/post_training/model_builder.py	225	# Set num_layers to 0 for base model in offline mode
LOW	megatron/core/parallel_state.py	836	# Set NCCL_COLLNET_ENABLE to 1 to enable SHARP for the dp group.
LOW	megatron/core/parallel_state.py	1275	# Set NCCL_COLLNET_ENABLE to 1 to enable SHARP for the dp_replica group.
LOW	megatron/core/parallel_state.py	1304	# Set NCCL_COLLNET_ENABLE to 0 to restrict SHARP application to the dp_replica group.
LOW	megatron/core/fp4_utils.py	14	# Check if Transformer Engine is installed
LOW	megatron/core/fp4_utils.py	25	# Check if Transformer Engine has class for fp4 tensors.
LOW	megatron/core/fp8_utils.py	23	# Check if Transformer Engine is installed
LOW	megatron/core/fp8_utils.py	40	# Check if Transformer Engine has class for fp8 tensors.
LOW	megatron/core/fp8_utils.py	55	# Check if Transformer Engine has MXFP8Tensor class
LOW	megatron/core/fp8_utils.py	635	# Check if fp8_model_init supports setting recipe
LOW	megatron/core/fp8_utils.py	640	# Check if fp8_model_init supports preserve_high_precision_init_val
LOW	megatron/core/utils.py	2841	# Check if any deprecated key is present in kwargs
LOW	megatron/core/ssm/ops/causal_conv1d_triton.py	243	# Check if input is 2D, temporarily treat as 3D for uniform processing
LOW	megatron/core/tokenizers/megatron_tokenizer.py	67	# Check if metadata file exists
67 more matches not shown…

Docstring Block Structure35 hits · 175 pts

Severity	File	Line	Snippet
HIGH	tools/checkpoint/remap_gpt_dsa_to_mamba.py	46	Return the HybridModel state-dict key corresponding to key from GPTModel. Args: key: A key from the GPTMo
HIGH	…egatron_fsdp/test_mcore_fully_sharded_data_parallel.py	774	Run a small deterministic (optional) training loop using a mocked MoE/GPT model and optimizer. This hel
HIGH	megatron/core/process_groups_config.py	668	Get process group collection for a specific module. Args: module_name: Name of the module.
HIGH	megatron/core/hyper_comm_grid.py	121	Create a process group based on a list of dimension names Note: The unique key used to store the process group
HIGH	megatron/core/utils.py	120	Validates the request to the experimental function. Args: func (Callable): Callee max_l
HIGH	megatron/core/utils.py	186	Validates the request to the experimental function. Args: func (Callable): Callee max_l
HIGH	megatron/core/utils.py	218	Pass-through to callee attribute if experimental flag is enabled. Args: super (supe
HIGH	megatron/core/timers.py	385	Returns the output string with logged timer values according to configured options. Args: names (Li
HIGH	…ore/tokenizers/text/libraries/huggingface_tokenizer.py	189	Adds a dictionary of special tokens (eos, pad, cls...). If special tokens are NOT in the vocabulary, th
HIGH	megatron/core/fusions/fused_bias_swiglu.py	210	Implementation of biased SwiGLU that handles different input shapes. This function reshapes the input if necessary,
HIGH	megatron/core/fusions/fused_bias_geglu.py	154	Implementation of biased GEGLU that handles different input shapes. This function reshapes the input if necessary,
HIGH	megatron/core/optimizer/optimizer.py	414	Filter and reorder state_dict parameter groups to match current optimizer groups. Keys used for matching align w
HIGH	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	4187	Release the specified parameter bucket, freeing its associated buffer storage. This function marks or
HIGH	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	4598	Creates a distributed tensor (DTensor) from a local tensor with support for Megatron-FSDP and Tensor Parallel s
HIGH	…e/distributed/fsdp/src/megatron_fsdp/uneven_dtensor.py	140	Validates the chunk metadata of an uneven DTensor to ensure correctness and boundary coverage. Notes: - `g
HIGH	…e/distributed/fsdp/src/megatron_fsdp/uneven_dtensor.py	257	Gather a DTensor with potentially uneven sharding across ranks into a full tensor. This function handles DTens
HIGH	…core/distributed/fsdp/src/megatron_fsdp/fully_shard.py	446	Fully shard the optimizer for Megatron-FSDP. This is an in-place operation on the optimizer instance, which mod
HIGH	…tron/core/datasets/blended_megatron_dataset_builder.py	497	Build the DistributedDataset Return None if and only if the underlying dataset class is not built on the curren
HIGH	megatron/core/datasets/object_storage_utils.py	136	Ascertain whether the object at the given S3 path exists in S3 Args: client (S3Client): The S3 client
HIGH	megatron/core/datasets/indexed_dataset.py	88	Get the size of the dtype/code in bytes Args: key (Union[int, Type[numpy.number]]): The dtype or co
HIGH	megatron/core/datasets/indexed_dataset.py	798	Return from the dataset Args: idx (Union[int, numpy.integer, slice]): The index or index slice into
HIGH	…ron/core/pipeline_parallel/multimodule_communicator.py	496	Compute the total number of pipeline stages across a multi-module chain. Interprets ``topology`` as a directed
HIGH	megatron/core/models/hybrid/hybrid_layer_allocation.py	161	Count layers by type across the full hybrid pattern (main + MTP). Parses the pattern to extract main and MTP compon
HIGH	megatron/core/models/hybrid/hybrid_layer_allocation.py	200	Parse a unified hybrid pattern string into main and MTP components. The pattern uses "/" as a separator between the
HIGH	megatron/core/models/hybrid/hybrid_layer_allocation.py	301	Validate and convert a single pipeline segment pattern to a layer type list. This is used after the main pattern ha
HIGH	megatron/core/models/hybrid/hybrid_layer_allocation.py	337	Select and validate the pipeline segment for the given PP rank and VP stage. When the main pattern contains '\|' pip
HIGH	megatron/core/dist_checkpointing/validation.py	294	Raises or logs an error in case missing or unexpected keys are non-empty. Args: missing_keys (Set[str]): mi
HIGH	megatron/core/dist_checkpointing/validation.py	372	Validate if the ShardedTensors and ShardedObjects from multiple processes define correct sharding. Local ShardedTen
HIGH	…tron/core/dist_checkpointing/strategies/async_utils.py	669	Finalizes all available calls. This method must be called on all ranks. Args: blocking (bo
HIGH	…ng/nvshmem_copy_service/memory/tensor_pointer_utils.py	19	Extract the data pointer from a tensor. Args: tensor: Can be torch.Tensor, CuPy array, or
HIGH	megatron/core/export/trtllm/trtllm_layers.py	85	Helper function to rename model layer names to TRTLLM Layer names We go through each layer (keys) in the model
HIGH	megatron/training/arguments.py	183	Validate model config arguments from heterogeneous config. This function takes model arguments and validates them b
HIGH	megatron/training/config/instantiate_utils.py	128	Instantiate an object or callable from a config object. This function takes a configuration object (dictionary, lis
HIGH	megatron/training/config/instantiate_utils.py	232	Recursively instantiates a node within a configuration structure. This function handles the instantiation of indivi
HIGH	megatron/rl/rl_utils.py	1181	Pad trajectories and extract the generation masks. Args: rollouts: Rollouts to extract trajectories from.

Over-Commented Block112 hits · 98 pts

Severity	File	Line	Snippet
LOW	.gitlab/scripts/check_imports.py	1	# Copyright (c) 2025, NVIDIA CORPORATION.
LOW	tools/bisect.sh	1	#!/usr/bin/env bash
LOW	tools/trigger_internal_ci.py	1	#!/usr/bin/env python3
LOW	tools/checkpoint/convert.py	1	# Copyright (c) 2023, NVIDIA CORPORATION. All rights reserved.
LOW	tools/checkpoint/convert.py	21	# full model weights, nothing split.
LOW	tools/checkpoint/convert.py	41	# consumed_train_samples
LOW	tools/checkpoint/convert.py	61	# "mlp l0 bias"
LOW	tools/checkpoint/gpt_hybrid_conversion.py	221	else:
LOW	tools/checkpoint/gpt_hybrid_conversion.py	241	#
LOW	tools/common_pile_dataset/setup_common_pile_dataset.sh	1	#!/bin/bash
LOW	…ts/performance_tests/shell_test_utils/run_perf_test.sh	1	#!/usr/bin/env bash
LOW	…ts/performance_tests/shell_test_utils/run_perf_test.sh	101	# from /usr/lib/python3.12/dist-packages — but mamba-ssm + causal-conv1d (and
LOW	tests/unit_tests/test_imports.py	1	# Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
LOW	…s/checkpoint/test_gpt_hybrid_conversion_parallelism.py	41	# tiny synthetic DCP checkpoint and round-trips it through the converter on
LOW	…unit_tests/transformer/test_fsdp_dtensor_checkpoint.py	1	# Copyright (c) 2026, NVIDIA CORPORATION. All rights reserved.
LOW	…s/unit_tests/elastification/test_hybrid_flex_router.py	81	tensor_model_parallel_size=1, pipeline_model_parallel_size=1
LOW	…s/unit_tests/pipeline_parallel/test_pipeline_layout.py	361
LOW	tests/unit_tests/inference/test_hybrid_moe.py	61	# Combinatorial sweep: unordered combinations with repetition of ALL_STATES
LOW	tests/test_utils/python_scripts/recipe_parser.py	21	# cadence remains the trigger axis.
LOW	tests/test_utils/recipes/h100/moe.yaml	101	- environment: [lts]
LOW	tests/test_utils/recipes/h100/moe.yaml	221	# Super important mr, mr-github tests that run for both DEV and LTS per mr, mr-github #
LOW	tests/test_utils/recipes/h100/gpt.yaml	461	# scope: [nightly] # Requires PyT 2.4: #481
LOW	tests/test_utils/recipes/gb200/moe.yaml	101	platforms: [dgx_gb200]
LOW	tests/test_utils/recipes/gb200/moe.yaml	201	# - test_case: [gpt3_mcore_te_tp2_pp1_frozen_resume_torch_dist_te_8experts2parallel_dist_optimizer]
LOW	…unctional_tests/test_cases/common/moe_perf/__main__.py	401
LOW	…unctional_tests/test_cases/common/moe_perf/__main__.py	421	# test_moe_layer_performance(case, debug_mode=True)
LOW	…unctional_tests/shell_test_utils/run_batch_ci_tests.sh	1	#!/bin/bash
LOW	docs/autodoc2_docstrings_parser.py	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	docs/conf.py	1	# Copyright (c) 2025-2026, NVIDIA CORPORATION. All rights reserved.
LOW	examples/megatron_fsdp/train_llama3_8b_fsdp_h100_fp8.sh	121	--ckpt-format fsdp_dtensor
LOW	examples/megatron_fsdp/sbatch_checkpoint_convert.sh	21	SLURM_LOGS="${OUTPUT_PATH}/slurm_logs"
LOW	examples/mimo/train.py	121	# iterator exhausted on all ranks
LOW	examples/inference/run_inference_server.sh	1	#!/bin/bash
LOW	examples/inference/run_offline_inference.sh	1	#!/bin/bash
LOW	.github/workflows/sync-team-usergroups.yml	1	# Copyright (c) 2026, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/cicd-approve-test-queue.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/release-docs.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION.
LOW	.github/workflows/community-bot.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/nightly-sync-main-to-dev.yml	1	# Copyright (c) 2026, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/release.yaml	1	# Copyright (c) 2019-2026, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/sync-skills.yml	1	# Copyright (c) 2026, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/close-inactive-issue-pr.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/oncall-assign.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/oncall-rotation.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/cherry-pick-release-commit.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/build-docs.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION.
LOW	.github/workflows/release-nightly-docs.yml	1	# Copyright (c) 2026, NVIDIA CORPORATION.
LOW	.github/workflows/cicd-main.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/install-test.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/release-freeze.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/workflows/copyright-check.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION.
LOW	.github/scripts/oncall_manager.py	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	.github/scripts/sync_team_usergroups.py	1	# Copyright (c) 2026, NVIDIA CORPORATION. All rights reserved.
LOW	.github/actions/action.yml	1	# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
LOW	skills/nightly-sync/SKILL.md	261	# pyproject.toml is allowed to differ ONLY for git source reconciliation
LOW	megatron/core/parallel_state.py	41	_DATA_PARALLEL_GROUP_GLOO = None
LOW	megatron/core/parallel_state.py	1021	global _EMBEDDING_GROUP
LOW	megatron/core/parallel_state.py	1041	# UCC backend requires CUDA_DEVICE_MAX_CONNECTIONS variable to be larger than 1,
LOW	megatron/core/config_logger.py	1	# Copyright (c) 2025, NVIDIA CORPORATION.
LOW	megatron/core/rerun_state_machine.py	821	# caller isn't required to have wrapped its iterator in
52 more matches not shown…

Verbosity Indicators47 hits · 81 pts

Severity	File	Line	Snippet
LOW	tools/checkpoint/saver_hf_llava.py	17	# Step 1: Reshape back to (num_head, 3*head_dim, -1)
LOW	tools/checkpoint/saver_hf_llava.py	20	# Step 2: Slice along the head_dim dimension to get q, k, v
LOW	tools/checkpoint/saver_hf_llava.py	25	# Step 3: Reshape each back to (num_head * head_dim, -1)
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	361	# Step 1: Get vocabulary files
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	387	# Step 2: Get raw text data
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	411	# Step 3: Preprocess for GPT (GPT2BPETokenizer)
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	434	# Step 4: Preprocess for BERT (BertWordPieceLowerCase + split-sentences)
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	475	# Step 5: Preprocess for T5 (BertWordPieceCase)
LOW	…s/common_pile_dataset/create_common_pile_ci_dataset.py	497	# Step 6: Clean up and verify
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	1796	# Step 1: Forward pass for all 3 requests
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	1805	# Step 2: Forward pass where req 10 finishes, req 11 continues. Req 999 is NOT scheduled.
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	1827	# Step 3: Add the next chunk. It should sit exactly at the boundary (index 1) and inherit the state.
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	1894	# Step 1: All 3 requests are active, process forward pass
LOW	…/unit_tests/inference/contexts/test_dynamic_context.py	1903	# Step 2: Both decode requests finish, chunked prefill NOT scheduled this step.
LOW	…ts/unit_tests/inference/engines/test_dynamic_engine.py	2490	# Step 1: Prefill. Processes the 4 prompt tokens.
LOW	…l_tests/python_test_utils/compute_golden_statistics.py	11	# Step 1: Run batch tests (from megatron-rl directory):
LOW	…l_tests/python_test_utils/compute_golden_statistics.py	15	# Step 2: Wait for jobs to complete, then compute statistics:
LOW	.github/scripts/oncall_manager.py	218	# Step 1: Add new oncall first (include current members to avoid removing anyone yet)
LOW	.github/scripts/oncall_manager.py	228	# Step 2: Now set the usergroup to contain only the new oncall
LOW	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	1374	# Step 0: Register new FSDP unit modules.
LOW	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	1407	# Step 1: Group the parameters according to their execution order and attributes.
LOW	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	1445	# Step 2: Bucket the parameters based on the guide bucket size.
LOW	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	1494	# Step 3: Split parameter groups to meet communication segmentation requirements.
LOW	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	1539	# Step 4: Generate the groups of collective buckets, where each group aggregates
LOW	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	4050	# If use double buffer, we need to check if the next bucket
LOW	…e/transformer/custom_layers/batch_invariant_kernels.py	265	# Step 1: Find maximum value in the row for numerical stability
LOW	…e/transformer/custom_layers/batch_invariant_kernels.py	277	# Step 2: Compute sum of exp(x - max_val)
LOW	…e/transformer/custom_layers/batch_invariant_kernels.py	293	# Step 3: Compute final log_softmax values: x - max_val - log_sum_exp
LOW	megatron/core/pipeline_parallel/bridge_communicator.py	415	# Step 1: broadcast its shape so receivers can allocate
LOW	megatron/core/pipeline_parallel/bridge_communicator.py	421	# Step 2: broadcast the actual tensor
LOW	…n/core/dist_checkpointing/strategies/fully_parallel.py	238	# Step 3: load part of the checkpoint.
LOW	…n/core/dist_checkpointing/strategies/fully_parallel.py	264	# Step 4: exchange data between ranks
LOW	…gatron/core/resharding/nvshmem_copy_service/service.py	232	# Step 1: Segment tasks (break large tasks into chunks)
LOW	…gatron/core/resharding/nvshmem_copy_service/service.py	242	# Step 2: Pack tasks into workload groups
LOW	…gatron/core/resharding/nvshmem_copy_service/service.py	249	# Step 3: Schedule workloads to iterations
LOW	…gatron/core/resharding/nvshmem_copy_service/service.py	258	# Step 4: Prepare iteration schedules
LOW	…gatron/core/resharding/nvshmem_copy_service/service.py	264	# Step 5: Build GPU execution plans
LOW	…gatron/core/resharding/nvshmem_copy_service/service.py	273	# Step 6: Create double-buffered events
LOW	…harding/nvshmem_copy_service/core/pipeline_executor.py	153	# Step 1: Pack NEXT iteration (async)
LOW	…harding/nvshmem_copy_service/core/pipeline_executor.py	165	# Step 2: Unpack PRIOR iteration (async)
LOW	…harding/nvshmem_copy_service/core/pipeline_executor.py	180	# Step 3: Send CURRENT iteration
LOW	…harding/nvshmem_copy_service/core/pipeline_executor.py	223	# Step 5: Wait for async pack to complete (double-buffer safety)
LOW	…shmem_copy_service/planning/communication_scheduler.py	39	# Step 1: Collect all batches across all PE pairs
LOW	…shmem_copy_service/planning/communication_scheduler.py	44	# Step 2: Assign batches to iterations using greedy conflict-free algorithm
LOW	…shmem_copy_service/planning/communication_scheduler.py	49	# Step 3: Exchange detailed workload summaries (Task IDs/Sizes)
LOW	…shmem_copy_service/planning/communication_scheduler.py	55	# Step 4: Build schedule map for this PE
LOW	…e/communication/torch_symm_triton/fused_collectives.py	151	# Step 1: - reduce-scatter + residual add for this token + collect sq sum

AI Slop Vocabulary34 hits · 76 pts

Severity	File	Line	Snippet
LOW	tasks/finetune_utils.py	120	# shuffling so we can just use a simple infinite loop.
MEDIUM	tests/unit_tests/test_inference.py	107	# we are replicating what lm-eval-harness::TemplateLM::_encode_pair does
MEDIUM	…s/checkpoint/test_gpt_hybrid_conversion_parallelism.py	43	# harness launched pytest. When that default PG is multi-rank (e.g. Megatron's
LOW	…ts/distributed/megatron_fsdp/test_mfsdp_fully_shard.py	227	# DP: Only relevant when using HSDP, where we need the flattened DP group for data parallelism. (Otherwise, just pas
LOW	…ts/distributed/megatron_fsdp/test_mfsdp_fully_shard.py	229	# DP-Shard-CP: Only required if using CP. Otherwise, just pass dp_shard to FSDP.
MEDIUM	tests/unit_tests/transformer/test_utils.py	331	"""Test a comprehensive scenario with multiple configurations."""
MEDIUM	tests/unit_tests/training/models/test_base.py	171	"""from_dict() reconstructs configs from serialized dicts, handles nested dataclasses, and is robust to unknown keys
LOW	…s/unit_tests/dist_checkpointing/models/test_mlp_glu.py	129	# Load happens in-place, so we can just use the same tensors
MEDIUM	…/unit_tests/inference/contexts/test_dynamic_context.py	2292	# 6. Verify seamless append (no legacy offset math needed)
MEDIUM	…cipes/h100/gpt-dynamic-inference-with-coordinator.yaml	92	# skills/run-performance-tests/SKILL.md for the harness it runs under.
MEDIUM	…inference_server_smoke_tp1_pp1_dp8_583m/serve_smoke.py	43	# the JET harness expects at ``logs///attempt_0//std.log``) while still
LOW	…sts/functional_tests/shell_test_utils/_run_training.sh	85	# If value is "true", just use the key
MEDIUM	.github/workflows/claude_review.yml	94	# Strict review: comprehensive Megatron-LM focused analysis
LOW	megatron/core/optimizer_param_scheduler.py	235	# If the learning rate is constant, just return the initial value.
LOW	megatron/core/timers.py	476	# polutes the runs list, so we just add each as a scalar
MEDIUM	megatron/core/ssm/triton_cache_manager.py	69	# use temp dir to be robust against program interruptions
LOW	…ore/tokenizers/text/parsers/qwen3_coder_tool_parser.py	16	# These map to vLLM types but we just use dictionaries for now
LOW	megatron/core/tensor_parallel/random.py	108	# if not using cuda graphs, just use the builtin pytorch function
LOW	megatron/core/tensor_parallel/random.py	189	# already graphable, just return it.
LOW	megatron/core/tensor_parallel/random.py	195	# already non-graphable, just return it.
MEDIUM	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	1549	# Set aggregate buckets by FSDP units, i.e. buckets pertaining to the same
MEDIUM	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	2149	# to leverage NCCL UBR for high-precision gradient reduction with
MEDIUM	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	3111	# for a seamless user experience and coverage for ZeRO-1 and ZeRO-2?
MEDIUM	…ibuted/fsdp/src/megatron_fsdp/param_and_grad_buffer.py	4319	# TODO(@cspades): Clean up this logic in conjunction with
LOW	…atron/core/distributed/fsdp/src/megatron_fsdp/utils.py	198	# if not using cuda graphs, just use the builtin pytorch function
LOW	megatron/core/transformer/transformer_config.py	2237	# so just set both if either is specified.
LOW	megatron/core/transformer/transformer_layer.py	862	# elements in bias_chunks are the same for all chunks, so we can just use the first one
MEDIUM	megatron/core/transformer/mlp.py	207	# Weight resharding across TP sizes will have aforementioned problems.
MEDIUM	megatron/core/datasets/indexed_dataset.py	6	# Essentially re-written in entirety
MEDIUM	…e/models/common/embeddings/language_model_embedding.py	145	# the original tensor from being garbage collected. Clone to facilitate GC.
LOW	megatron/core/models/bert/bert_model.py	193	# For local layer spec we just use b1ss
LOW	megatron/core/extensions/transformer_engine.py	687	# TODO should we ditch normalization config and just use spec to choose LayerNorm vs RMSNorm?
LOW	…tron/core/dist_checkpointing/strategies/async_utils.py	479	# to simply call `sync_all_async_calls` to check if other ranks complete the writing
MEDIUM	…/core/inference/data_parallel_inference_coordinator.py	386	# Todo [Siddharth]: Make this more robust to handle invalid messages.

Cross-Language Confusion9 hits · 39 pts

Severity	File	Line	Snippet
HIGH	…core/transformer/heterogeneous/heterogeneous_config.py	177	"n_heads_in_group": null,
HIGH	…e/pipeline_parallel/fine_grained_activation_offload.py	504	self.push(chunk)
HIGH	…e/pipeline_parallel/fine_grained_activation_offload.py	655	self.push(cur_chunk)
HIGH	megatron/core/inference/inference_request.py	94	Each block hash is computed as SHA-256(parent_digest \|\| block_bytes), where
HIGH	megatron/core/inference/unified_memory.py	113	if (device != prev_device && device >= 0) cudaSetDevice(device);
HIGH	megatron/core/inference/unified_memory.py	149	if (device != prev_device && prev_device >= 0) cudaSetDevice(prev_device);
HIGH	megatron/core/inference/contexts/dynamic_context.py	3718	# : [ XX \| XX \| 16 XX \| 12 72 24 88 XX ] (XX = undefined)
HIGH	megatron/core/inference/moe/vllm_fused_moe.py	518	`input` are undefined).
HIGH	megatron/training/arguments.py	3220	"n_heads_in_group": null,

Dead Code5 hits · 10 pts

Severity	File	Line
MEDIUM	tools/checkpoint/loader_llava.py	52
MEDIUM	…nce_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py	183
MEDIUM	…nce_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py	184
MEDIUM	…nce_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py	185
MEDIUM	…nce_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py	186

Synthetic Comment Markers1 hit · 5 pts

Severity	File	Line	Snippet
HIGH	.github/workflows/claude-copy-to-main.yml	113	--body "🤖 This PR was auto-generated by Claude via the \`/claude copy\` command.\n\nCherry-picked f

Example Usage Blocks3 hits · 4 pts

Severity	File	Line	Snippet
LOW	tools/bisect.sh	5	# Usage:
LOW	tools/common_pile_dataset/setup_common_pile_dataset.sh	9	# Usage:
LOW	…unctional_tests/shell_test_utils/run_batch_ci_tests.sh	5	# Usage:

Slop Phrases2 hits · 2 pts

Severity	File	Line	Snippet
MEDIUM	megatron/core/process_groups_config.py	660	"""Check if this rank has a language model.
MEDIUM	megatron/core/process_groups_config.py	663	True if this rank has a language model, False otherwise.

Overly Generic Function Names3 hits · 2 pts

Severity	File	Line	Snippet
LOW	…s/unit_tests/training/config/test_instantiate_utils.py	52	def test_function(arg1=None, arg2=None, **kwargs):
LOW	tests/unit_tests/training/config/test_yaml_utils.py	45	def test_function():
LOW	megatron/core/utils.py	2544	def my_function():

Score History

Severity Breakdown

Pattern Findings