Functions — github.com/Agent-RL/ReCall

hub / github.com/Agent-RL/ReCall / functions

Functions1,539 in github.com/Agent-RL/ReCall

⨍ Functions 1,539◇ Types & classes 237↳ Endpoints 28

↓ 236 callersMethodget

Return a tensor with the input `shape` as a view into the 1-D data starting at `start_index`.

src/verl/utils/memory_buffer.py:43

↓ 157 callersMethodto

move the batch to device Args: device (torch.device, str): torch device Returns: DataProto: the current Data

src/verl/protocol.py:354

↓ 84 callersFunctionprint_rank_0

If distributed is initialized, print only on rank 0.

src/verl/utils/megatron_utils.py:263

↓ 70 callersMethodget

(self)

src/verl/protocol.py:750

↓ 64 callersMethodupdate_output

Update the output dict and keep a key in self.output can be used as an attribute.

src/flashrag/dataset/dataset.py:24

↓ 61 callersFunctionlog_gpu_memory_usage

(head: str, logger: logging.Logger = None, level=logging.DEBUG, rank: int = 0)

src/verl/utils/debug/performance.py:20

↓ 46 callersMethodchunk

Split the batch among dim=0 into chunks. The meta_info is passed to each DataProto after split. Args: chunks (int): the number of

src/verl/protocol.py:600

↓ 45 callersMethodupdate

(self, current_kl, n_steps)

src/verl/trainer/ppo/core_algos.py:52

↓ 42 callersMethodpop

Pop a subset of the DataProto via `batch_keys` and `meta_info_keys` Args: batch_keys (list, optional): a list of strings indicati

src/verl/protocol.py:484

↓ 34 callersMethoddecode

Converts a sequence of ids in a string, using the tokenizer and vocabulary with options to remove special tokens and clean up tokeniz

src/verl/workers/rollout/tokenizer.py:86

↓ 33 callersMethodencode

(self, query_list: List[str], batch_size=64, is_query=True)

src/flashrag/retriever/encoder.py:68

↓ 29 callersMethodgenerate

(self, input_ids, attention_mask, **kwargs)

src/flashrag/generator/fid.py:48

↓ 28 callersMethodget

Return a tensor with the input `shape` as a view into the 1-D data starting at `start_index`.

src/verl/utils/megatron/memory.py:33

↓ 24 callersFunctioncopy_to_local

Copy src from hdfs to local if src is on hdfs or directly return src. If cache_dir is None, we will use the default cache dir of the system. Note

src/verl/utils/fs.py:58

↓ 22 callersMethodconcat

Concat a list of DataProto. The batch is concatenated among dim=0. The meta_info is assumed to be identical and will use the first one.

src/verl/protocol.py:633

↓ 22 callersMethodeval

(self)

src/verl/workers/sharding_manager/megatron_vllm.py:171

↓ 21 callersMethodsearch

(self, *args, **kwargs)

src/flashrag/retriever/retriever.py:187

↓ 19 callersMethodget_string

(self, question=None, retrieval_result=None, formatted_reference=None, previous_gen=None, messages=None, **par

src/flashrag/prompt/base_prompt.py:96

↓ 18 callersMethod__init__

(self, config)

src/flashrag/evaluator/metrics.py:465

↓ 18 callersFunctionconvert_config

(hf_config: PretrainedConfig, megatron_config)

src/verl/utils/megatron_utils.py:157

↓ 18 callersFunctiongather_seq_scatter_heads

A func to sync embedding input with alltoall in sequence parallel gather sequence dimension and scatter head dim: e.g. seq_dim: 1, head_d

src/verl/utils/ulysses.py:61

↓ 16 callersMethodget_token_length

( self, text: str, add_special_tokens: bool = True, use_oai_tokenizer: bool =

src/flashrag/refiner/llmlingua_compressor.py:1049

↓ 15 callersFunctionget_retriever

r"""Automatically select retriever class based on config's retrieval method Args: config (dict): configuration with 'retrieval_method' ke

src/flashrag/utils/utils.py:73

↓ 15 callersFunctionhf_tokenizer

Create a huggingface pretrained tokenizer which correctness handles eos and pad tokens. Args: name (str): The name of the tokenizer.

src/verl/utils/tokenizer.py:35

↓ 14 callersMethodencode

(self, contents: list)

src/flashrag/judger/judger.py:67

↓ 14 callersMethodget_dataset_answer

(self, data)

src/flashrag/evaluator/metrics.py:33

↓ 14 callersFunctionget_generator

Automatically select generator class based on config.

src/flashrag/utils/utils.py:40

↓ 14 callersMethodselect

Select a subset of the DataProto via batch_keys and meta_info_keys Args: batch_keys (list, optional): a list of strings indicatin

src/verl/protocol.py:368

↓ 13 callersMethodevaluate

The evaluation process after finishing overall generation

src/flashrag/pipeline/pipeline.py:28

↓ 13 callersFunctionunwrap_model

(model, module_instances=ALL_MODULE_WRAPPER_CLASSNAMES)

src/verl/utils/megatron_utils.py:139

↓ 12 callersMethodadd

(self, idx: int, val: int)

src/verl/utils/seqlen_balancing.py:33

↓ 12 callersMethodbatch_search

(self, *args, **kwargs)

src/flashrag/retriever/retriever.py:190

↓ 12 callersFunctionfunc

(*args, **kwargs)

src/verl/single_controller/ray/base.py:39

↓ 12 callersMethodto_dtype

(precision)

src/verl/utils/torch_dtypes.py:63

↓ 11 callersFunction_timer

(name: str, timing_raw: Dict[str, float])

src/verl/trainer/ppo/ray_trainer.py:235

↓ 11 callersMethodfrom_dict

Create a DataProto from a dict of tensors. This assumes that 1. All the tensor in tensors have the same dim0 2. Only dim0 is the batch

src/verl/protocol.py:319

↓ 11 callersMethodget_megatron_rank_info

(self)

src/verl/single_controller/base/megatron/worker.py:32

↓ 11 callersFunctionget_ulysses_sequence_parallel_world_size

Get ulysses sequence parallel world size.

src/verl/utils/ulysses.py:45

↓ 11 callersMethodnamed_parameters

(self)

src/verl/utils/memory_buffer.py:208

↓ 11 callersMethodsave

Save the dataset into the original format.

src/flashrag/dataset/dataset.py:186

↓ 10 callersFunctionget_ulysses_sequence_parallel_group

Get ulysses sequence parallel process group.

src/verl/utils/ulysses.py:37

↓ 10 callersMethodlog

(self, data, step, backend=None)

src/verl/utils/tracking.py:104

↓ 10 callersFunctionnormalize_answer

(s)

src/flashrag/evaluator/utils.py:5

↓ 10 callersFunctionoffload_fsdp_model_to_cpu

(model: FSDP, empty_cache: bool = True)

src/verl/utils/fsdp_utils.py:115

↓ 10 callersFunctionpad_sequence_to_length

pad a 2D tensors (e.g. responses, logprobs) in the last dim to max_seq_length. input shape: [bs, seq_length] output shape: [bs, max_seq_l

src/verl/utils/torch_functional.py:245

↓ 10 callersMethodrepeat

Repeat the batch data a specified number of times. Args: repeat_times (int): Number of times to repeat the data.

src/verl/protocol.py:665

↓ 10 callersMethodtrain

(self)

src/verl/workers/sharding_manager/megatron_vllm.py:175

↓ 9 callersMethodget_rng_state

()

src/verl/utils/checkpoint/checkpoint_manager.py:107

↓ 9 callersFunctionload_fsdp_model_to_gpu

(model: FSDP)

src/verl/utils/fsdp_utils.py:136

↓ 9 callersMethodpreprocess_data

(self, data: DataProto)

src/verl/workers/sharding_manager/base.py:29

↓ 8 callersFunction_megatron_calc_global_rank

Calculate global rank with support for CP/EP parallelism

src/verl/models/mcore/saver.py:28

↓ 8 callersMethoddump

(self, data: io.BytesIO, name)

src/verl/utils/debug/trajectory_tracker.py:59

↓ 8 callersFunctionload_corpus

(corpus_path: str)

src/flashrag/retriever/utils.py:121

↓ 8 callersMethodpostprocess_data

(self, data: DataProto)

src/verl/workers/sharding_manager/base.py:32

↓ 8 callersMethodrank

(self)

src/verl/single_controller/base/worker.py:208

↓ 7 callersFunction_megatron_calc_global_rank

given TP,DP,PP rank to get the global rank.

src/verl/models/qwen2/megatron/checkpoint_utils/qwen2_saver.py:27

↓ 7 callersFunction_megatron_calc_global_rank

given TP,DP,PP rank to get the global rank.

src/verl/models/llama/megatron/checkpoint_utils/llama_saver.py:27

↓ 7 callersFunctionappend_to_dict

(data: Dict, new_data: Dict)

src/verl/utils/py_functional.py:41

↓ 7 callersFunctionclean_traceback

(error_traceback)

src/verl/utils/reward_score/prime_code/testing_util.py:101

↓ 7 callersFunctionoffload_fsdp_optimizer

(optimizer)

src/verl/utils/fsdp_utils.py:152

↓ 7 callersMethodunion

Union with another DataProto. Union batch and meta_info separately. Throw an error if - there are conflict keys in batch and they are

src/verl/protocol.py:542

↓ 6 callersMethod__init__

( self, config, threshold=0.2, look_ahead_steps=64, max_generation_len

src/flashrag/pipeline/active_pipeline.py:803

↓ 6 callersFunction_broadcast_tensor

broadcast tensor from rank0 across mp_group

src/verl/models/qwen2/megatron/checkpoint_utils/qwen2_loader_depracated.py:100

↓ 6 callersFunction_broadcast_tensor

broadcast tensor from rank0 across mp_group

src/verl/models/mcore/loader.py:100

↓ 6 callersFunction_broadcast_tensor

broadcast tensor from rank0 across mp_group

src/verl/models/llama/megatron/checkpoint_utils/llama_loader_depracated.py:104

↓ 6 callersFunction_fetch_tensor

fetch tensor

src/verl/models/qwen2/megatron/checkpoint_utils/qwen2_loader.py:100

↓ 6 callersFunction_fetch_tensor

fetch tensor

src/verl/models/llama/megatron/checkpoint_utils/llama_loader.py:104

↓ 6 callersMethodcalculate_rouge

(self, pred, golden_answers)

src/flashrag/evaluator/metrics.py:293

↓ 6 callersFunctioncombined_int_check

(val)

src/verl/utils/reward_score/prime_code/testing_util.py:97

↓ 6 callersFunctiongather_heads_scatter_seq

A func to sync attention result with alltoall in sequence parallel gather head dimension and scatter seq dim: e.g. seq_dim: 1, head_dim:

src/verl/utils/ulysses.py:85

↓ 6 callersFunctionis_non_local

(path)

src/verl/utils/fs.py:32

↓ 6 callersFunctionjudge_zh

(input_str: str)

src/flashrag/retriever/utils.py:27

↓ 6 callersFunctionlogprobs_from_logits

See: https://github.com/pytorch/pytorch/issues/563#issuecomment-330103591

src/verl/utils/torch_functional.py:48

↓ 6 callersFunctionmasked_mean

Compute mean of tensor with a masked values.

src/verl/utils/torch_functional.py:118

↓ 6 callersMethodpostprocess_prediction

(self, pred)

src/flashrag/pipeline/active_pipeline.py:462

↓ 6 callersFunctionrepeat_kv

This is the equivalent of torch.repeat_interleave(x, dim=2, repeats=n_rep). The hidden states go from (batch, seqlen, num_key_value_heads, he

src/verl/models/transformers/monkey_patch.py:32

↓ 6 callersFunctiontruncatefn

(s, length=300)

src/verl/utils/reward_score/prime_code/testing_util.py:42

↓ 6 callersFunctionulysses_pad_and_slice_inputs

Pad and slice input_ids to be divisible by sp_size Pad position_ids to be divisible by sp_size. Note both input_ids_rmpad and position_i

src/verl/utils/ulysses.py:252

↓ 5 callersMethod__init__

(self, config)

src/flashrag/generator/generator.py:22

↓ 5 callersMethod__init__

(self, config)

src/flashrag/retriever/retriever.py:118

↓ 5 callersMethod__init__

(self, config: Qwen2Config, megatron_config: ModelParallelConfig)

src/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:83

↓ 5 callersMethod__init__

(self, config: LlamaConfig, megatron_config: ModelParallelConfig)

src/verl/models/llama/megatron/modeling_llama_megatron.py:83

↓ 5 callersFunction_broadcast_tensor

broadcast tensor across mp_group

src/verl/models/qwen2/megatron/checkpoint_utils/qwen2_saver.py:123

↓ 5 callersFunction_broadcast_tensor

broadcast tensor across mp_group

src/verl/models/llama/megatron/checkpoint_utils/llama_saver.py:123

↓ 5 callersFunction_hdfs_cmd

(cmd: str)

src/verl/utils/hdfs_io.py:139

↓ 5 callersFunction_run_cmd

(cmd: str, timeout=None)

src/verl/utils/hdfs_io.py:135

↓ 5 callersMethod_search_or_batch_search

(self, query: Union[str, list], target_modal, num, return_score, method, retriever_list)

src/flashrag/retriever/retriever.py:653

↓ 5 callersFunctionagg_loss

Aggregate the loss matrix into a scalar. Args: loss_mat: `(torch.Tensor)` shape: (bs, response_length) loss_mask:

src/verl/trainer/ppo/core_algos.py:309

↓ 5 callersFunctionapply_rotary_pos_emb

(q, k, cos, sin, position_ids)

src/verl/models/qwen2/megatron/layers/parallel_attention.py:123

↓ 5 callersMethodencode

Converts a string to a sequence of ids (integer), using the tokenizer and vocabulary. Args: text (`str`, `List[str]` or

src/verl/workers/rollout/tokenizer.py:70

↓ 5 callersFunctiongather_outpus_and_unpad

(x: Tensor, gather_dim: int, unpad_dim: int = None,

src/verl/utils/ulysses.py:233

↓ 5 callersMethodgenerate

Get responses from the generater. Args: input_list: it contains input texts, each item represents a sample. Returns:

src/flashrag/generator/generator.py:51

↓ 5 callersFunctionget_dataset

Load dataset from config.

src/flashrag/utils/utils.py:8

↓ 5 callersFunctionget_init_weight_context_manager

(use_meta_tensor=True, mesh: DeviceMesh = None)

src/verl/utils/fsdp_utils.py:39

↓ 5 callersFunctionget_model

Build the model.

src/verl/utils/megatron_utils.py:42

↓ 5 callersMethodget_ppl

( self, text: str, granularity: str = "sentence", input_ids=None, atte

src/flashrag/refiner/llmlingua_compressor.py:303

↓ 5 callersFunctionget_response_mask

end of sentence token can be int or list: 1 or [1, 2] e.g. response_id = torch.tensor([[20, 10, 34, 1, 0, 0, 0],

src/verl/utils/torch_functional.py:150

↓ 5 callersMethodinit_model

(self)

src/verl/workers/fsdp_workers.py:805

↓ 5 callersFunctionload_docs

(corpus, doc_idxs: List[int])

src/flashrag/retriever/utils.py:148

↓ 5 callersFunctionrearrange_micro_batches

Split the batch into a list of micro_batches, where the max_token_len is smaller than max_token_len and the number of valid tokens in each micro b

src/verl/utils/seqlen_balancing.py:224

next →1–100 of 1,539, ranked by callers