MCPcopy Create free account

hub / github.com/ChenxinAn-fdu/POLARIS / functions

Functions1,563 in github.com/ChenxinAn-fdu/POLARIS

⨍ Functions 1,563◇ Types & classes 217↳ Endpoints 49
↓ 3 callersFunction_str_is_int
(x: str)
verl/verl/utils/reward_score/prime_math/__init__.py:89
↓ 3 callersFunction_str_is_int
(x: str)
deepscaler/rewards/math_utils/utils.py:222
↓ 3 callersFunction_unwrap_ray_remote
(cls)
verl/verl/single_controller/ray/base.py:480
↓ 3 callersFunctionbuild_memory_buffer
Build the memory buffer given weight_buffer_meta Args: weight_buffer_meta: contains mapping from name to a dictionary containing shape an
verl/verl/utils/memory_buffer.py:70
↓ 3 callersFunctioncalc_padded_numel
for cuda memory alignment, make sure alignment by 128-bits
verl/verl/utils/memory_buffer.py:53
↓ 3 callersMethodcollective_rpc
( self, method: Union[str, Callable], timeout: Optional[float] = None, args: T
verl/verl/workers/rollout/vllm_rollout/vllm_async_server.py:83
↓ 3 callersFunctioncompute_advantage
(data: DataProto, beta=1.0)
verl/recipe/sppo/sppo_ray_trainer.py:60
↓ 3 callersFunctioncompute_detach_dpo_loss_rm
(token_level_scores, acc, Q_bc, acc_bc, response_mask, beta, bon_mode="none")
verl/recipe/prime/prime_core_algos.py:81
↓ 3 callersFunctioncreate_colocated_worker_cls
This function should return a class instance that delegates the calls to every cls in cls_dict
verl/verl/single_controller/ray/base.py:500
↓ 3 callersMethoddump
(self, data: io.BytesIO, name)
verl/verl/utils/debug/trajectory_tracker.py:59
↓ 3 callersMethodexecute_all_async
(self, method_name: str, *args, **kwargs)
verl/verl/single_controller/ray/base.py:398
↓ 3 callersMethodfit
The training loop of PPO. The driver process only need to call the compute functions of the worker group through RPC to const
verl/verl/trainer/ppo/ray_trainer.py:881
↓ 3 callersMethodfrom_detached
( cls, name_prefix, worker_names=None, ray_cls_with_init=None, )
verl/verl/single_controller/ray/base.py:318
↓ 3 callersFunctionfsdp2_load_full_state_dict
Loads the full state dict (could be only on rank 0) into the sharded model. This is done by broadcasting the parameters from rank 0 to all ot
verl/verl/utils/fsdp_utils.py:385
↓ 3 callersFunctionget_default_kwargs_for_model_parallel_config
()
verl/verl/utils/megatron/tensor_parallel.py:33
↓ 3 callersFunctionget_mcore_forward_fn
Get the forward function for given model architecture.
verl/verl/models/mcore/registry.py:150
↓ 3 callersFunctionget_megatron_optimizer
( model, config: OptimizerConfig, no_weight_decay_cond=None, scale_lr_cond=None, lr_mult=1
verl/verl/utils/megatron/optimizer.py:20
↓ 3 callersFunctionget_micro_data_parallel_world_size
()
verl/verl/workers/sharding_manager/megatron_vllm.py:576
↓ 3 callersFunctionget_model_checkpoint_path
(checkpoint_path)
verl/verl/utils/megatron_utils.py:387
↓ 3 callersFunctionhf_to_mcore_config
(hf_config: PretrainedConfig, dtype: torch.dtype)
verl/verl/models/mcore/registry.py:112
↓ 3 callersFunctioninit_async_rollout_manager
(config: DictConfig, scheduler_kwargs: Dict[str, Any] = None)
verl/tests/rollout/async_rollout_utils.py:27
↓ 3 callersFunctioninit_megatron_optim_config
(optim_config: Dict)
verl/verl/utils/megatron_utils.py:190
↓ 3 callersMethodis_padding_enabled
Check if padding is enabled for the DataProto. Returns: bool: True if padding is enabled, False otherwise.
verl/verl/protocol.py:622
↓ 3 callersFunctionkl_penalty
Compute KL divergence given logprob and ref_logprob. Copied from https://github.com/huggingface/trl/blob/main/trl/trainer/ppo_trainer.py#L1104
verl/verl/trainer/ppo/core_algos.py:469
↓ 3 callersFunctionlast_boxed_only_string
(string)
verl/verl/utils/reward_score/math.py:63
↓ 3 callersMethodload_checkpoint
(self, local_path, hdfs_path=None, del_local_after_load=True)
verl/verl/workers/fsdp_workers.py:1008
↓ 3 callersFunctionload_fsdp_optimizer
(optimizer, device_id)
verl/verl/utils/fsdp_utils.py:191
↓ 3 callersFunctionload_reward_manager
(config, tokenizer, num_examine, **reward_kwargs)
verl/verl/trainer/ppo/reward.py:57
↓ 3 callersFunctionload_tensor_to_gpu
(tensor)
verl/verl/utils/megatron_utils.py:330
↓ 3 callersMethodlog
(self, func, *args, **kwargs)
verl/verl/utils/debug/performance.py:82
↓ 3 callersFunctionmake_batch_generator
(batches, vpp_size)
verl/verl/utils/megatron/pipeline_parallel.py:49
↓ 3 callersFunctionmake_map_fn
(split)
verl/examples/data_preprocess/hellaswag.py:54
↓ 3 callersFunctionnormalize_final_answer
Normalize a final answer to a quantitative reasoning question. Args: final_answer: The answer string to normalize Returns: N
verl/verl/utils/reward_score/math_dapo.py:125
↓ 3 callersFunctionoffload_tensor_to_cpu
(tensor)
verl/verl/utils/megatron_utils.py:296
↓ 3 callersFunctionremove_boxed
(s)
verl/verl/utils/reward_score/math.py:49
↓ 3 callersMethodselect_idxs
Select specific indices from the DataProto. Args: idxs (torch.Tensor or numpy.ndarray or list): Indices to select
verl/verl/protocol.py:429
↓ 3 callersMethodset_input_tensor
Set input tensor to be used instead of forward()'s input. When doing pipeline parallelism the input from the previous stage comes fro
verl/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:453
↓ 3 callersFunctionset_random_seed
(seed)
verl/verl/workers/megatron_workers.py:51
↓ 3 callersFunctionsplit_dict_tensor_into_batches
(tensors: TensorDict, batch_size)
verl/verl/utils/torch_functional.py:229
↓ 3 callersFunctiontimeout_limit
Decorator to add a timeout to a function. Args: seconds: The timeout duration in seconds. use_signals: (Deprecated) This is
verl/verl/utils/py_functional.py:51
↓ 3 callersMethodto_dict
(self)
verl/verl/single_controller/base/worker.py:87
↓ 3 callersFunctionunion_numpy_dict
(tensor_dict1: dict[str, np.ndarray], tensor_dict2: dict[str, np.ndarray])
verl/verl/protocol.py:115
↓ 3 callersFunctionunion_tensor_dict
Union two tensordicts.
verl/verl/protocol.py:103
↓ 3 callersMethodupdate_rm
(self, data: DataProto)
verl/recipe/prime/prime_dp_rm.py:211
↓ 3 callersFunctionvocab_parallel_log_probs_from_logits
TODO(zhangchi.usc1992): We may change the implementation later
verl/verl/utils/megatron/tensor_parallel.py:146
↓ 2 callersMethod__init__
(self, config)
verl/verl/workers/megatron_workers.py:484
↓ 2 callersMethod__init__
(self, config)
verl/verl/workers/fsdp_workers.py:715
↓ 2 callersMethod__init__
( self, input_size, num_heads, num_key_value_heads, head_dim,
verl/verl/models/llama/megatron/layers/parallel_linear.py:21
↓ 2 callersFunction_broadcast_tp_shard_tensor_qkv
broadcast tensor in tp shards across mp_group
verl/verl/models/qwen2/megatron/checkpoint_utils/qwen2_loader_depracated.py:265
↓ 2 callersFunction_broadcast_tp_shard_tensor_qkv
broadcast tensor in tp shards across mp_group
verl/verl/models/qwen2/megatron/checkpoint_utils/qwen2_saver.py:247
↓ 2 callersFunction_broadcast_tp_shard_tensor_qkv
broadcast tensor in tp shards across mp_group
verl/verl/models/mcore/saver.py:259
↓ 2 callersFunction_broadcast_tp_shard_tensor_qkv
broadcast tensor in tp shards across mp_group
verl/verl/models/mcore/loader.py:270
↓ 2 callersMethod_build_model_optimizer
(self, model_path, optim_config, override_model_config)
verl/verl/workers/megatron_workers.py:138
↓ 2 callersMethod_build_model_optimizer
( self, model_path, fsdp_config, optim_config, override_model_config,
verl/verl/workers/fsdp_workers.py:148
↓ 2 callersFunction_default_compute_score
(data_source, solution_str, ground_truth, extra_info=None)
verl/verl/utils/reward_score/__init__.py:17
↓ 2 callersMethod_download
(self, use_origin_parquet=False)
verl/verl/utils/dataset/rl_dataset.py:98
↓ 2 callersMethod_dump_generations
Dump rollout/validation samples as JSONL.
verl/verl/trainer/ppo/ray_trainer.py:530
↓ 2 callersFunction_fetch_tp_shard_tensor_qkv
fetch tensor in tp shards across mp_group
verl/verl/models/qwen2/megatron/checkpoint_utils/qwen2_loader.py:152
↓ 2 callersMethod_forward_micro_batch
(self, micro_batch)
verl/verl/workers/critic/dp_critic.py:54
↓ 2 callersMethod_forward_micro_batch
Returns: entropy: # (bs, response_len) log_probs: # (bs, response_len)
verl/verl/workers/actor/dp_actor.py:64
↓ 2 callersFunction_is_frac
(expr: str)
verl/verl/utils/reward_score/prime_math/__init__.py:85
↓ 2 callersFunction_is_frac
(expr: str)
deepscaler/rewards/math_utils/utils.py:218
↓ 2 callersMethod_load_checkpoint
(self)
verl/verl/trainer/ppo/ray_trainer.py:815
↓ 2 callersFunction_load_hf_model
Helper function containing the loading hf model logic
verl/verl/utils/model.py:284
↓ 2 callersFunction_materialize_futures
(*args, **kwargs)
verl/verl/single_controller/base/decorator.py:441
↓ 2 callersFunction_normalize
Normalize answer expressions.
verl/verl/utils/reward_score/prime_math/__init__.py:125
↓ 2 callersFunction_normalize
Normalize answer expressions.
deepscaler/rewards/math_utils/utils.py:258
↓ 2 callersMethod_offload_params_to_cpu
(self, pp_rank, to_empty=False)
verl/verl/workers/sharding_manager/megatron_vllm.py:142
↓ 2 callersMethod_optimizer_step
(self)
verl/recipe/prime/prime_dp_rm.py:152
↓ 2 callersFunction_pad_tensor
(x: Tensor, dim: int, padding_size: int)
verl/verl/utils/ulysses.py:104
↓ 2 callersMethod_pad_to_length
(self, input_ids, attention_mask)
verl/verl/utils/dataset/rm_dataset.py:99
↓ 2 callersFunction_parse
(s)
verl/verl/utils/reward_score/prime_math/grader.py:284
↓ 2 callersFunction_post_process_outputs
(tokenizer, output)
verl/verl/workers/rollout/sglang_rollout/sglang_rollout.py:70
↓ 2 callersMethod_postprocess
(self, batch: DataProto, batch_conversations: List[List[List[Dict[str, str]]]], n: int)
verl/examples/ppo_trainer/naive_chat_scheduler.py:90
↓ 2 callersMethod_read_files_and_tokenize
(self)
verl/verl/utils/dataset/rl_dataset.py:105
↓ 2 callersMethod_save_checkpoint
(self)
verl/verl/trainer/ppo/ray_trainer.py:783
↓ 2 callersMethod_set_cos_sin_cache
(self, seq_len, device, dtype)
verl/verl/models/qwen2/megatron/layers/parallel_attention.py:55
↓ 2 callersFunction_strip_properly_formatted_commas
(expr: str)
verl/verl/utils/reward_score/prime_math/__init__.py:114
↓ 2 callersFunction_strip_properly_formatted_commas
(expr: str)
deepscaler/rewards/math_utils/utils.py:247
↓ 2 callersFunction_unpad_tensor
(x: Tensor, dim: int, padding_size: int)
verl/verl/utils/ulysses.py:111
↓ 2 callersMethod_validate
(self)
verl/verl/trainer/ppo/ray_trainer.py:578
↓ 2 callersMethodadd
(self, x)
verl/tests/ray_cpu/test_fused_workers.py:28
↓ 2 callersMethodadd
(self, data: DataProto)
verl/tests/ray_gpu/test_colocated_workers_fused.py:34
↓ 2 callersMethodadd
(self, data: DataProto)
verl/tests/ray_gpu/test_colocated_workers.py:34
↓ 2 callersMethodadd
(self, a, b)
verl/tests/e2e/envs/digit_completion/task.py:76
↓ 2 callersFunctionall_to_all_tensor
( local_input: Tensor, scatter_dim: int, gather_dim: int, group: Optional[dist.ProcessGroup] =
verl/verl/utils/ulysses.py:133
↓ 2 callersFunctionallgather_dict_tensors
TODO: optimize this. - We can use async ops - We can use only one allgather Args: tensors: size: group:
verl/verl/utils/torch_functional.py:195
↓ 2 callersFunctionare_lists_similar
(a, b)
verl/tests/rollout/utils_sglang.py:40
↓ 2 callersFunctionbootstrap_metric
( data: list[Any], subset_size: int, reduce_fns: list[Callable[[np.ndarray], float]], n_bootst
verl/verl/trainer/ppo/metric_utils.py:237
↓ 2 callersFunctionbuild_memory_reference_from_module
(module: torch.nn.Module, memory_buffers: Dict[torch.dtype, MemoryBuffer], maintain_weight=True)
verl/verl/utils/memory_buffer.py:99
↓ 2 callersFunctionclean_torchelastic_env
()
verl/tests/rollout/utils_sglang.py:82
↓ 2 callersFunctioncompute_data_metrics
(batch, use_critic=True)
verl/recipe/prime/prime_ray_trainer.py:54
↓ 2 callersFunctioncompute_data_metrics
(batch, use_critic=True)
verl/verl/trainer/ppo/metric_utils.py:120
↓ 2 callersFunctioncompute_dpo_abs_accuracy
(token_level_scores, acc, response_mask, n_samples)
verl/recipe/prime/prime_core_algos.py:135
↓ 2 callersFunctioncompute_dpo_accuracy
(token_level_scores, acc, response_mask, n_samples)
verl/recipe/prime/prime_core_algos.py:112
↓ 2 callersFunctioncompute_policy_loss
Adapted from https://github.com/huggingface/trl/blob/main/trl/trainer/ppo_trainer.py#L1122 Args: old_log_prob: `(torch.Tensor)`
verl/verl/trainer/ppo/core_algos.py:354
↓ 2 callersFunctioncompute_response_mask
(data: DataProto)
verl/recipe/prime/prime_ray_trainer.py:113
↓ 2 callersFunctioncompute_response_mask
(data: DataProto)
verl/verl/trainer/ppo/ray_trainer.py:178
↓ 2 callersFunctioncompute_reward
Compute reward for a batch of data. Args: data: DataProto object containing the input data. reward_fn: Reward function to com
verl/verl/trainer/ppo/reward.py:88
↓ 2 callersFunctioncompute_throughout_metrics
(batch: DataProto, timing_raw: Dict[str, float], n_gpus: int)
verl/verl/trainer/ppo/metric_utils.py:224
← previousnext →201–300 of 1,563, ranked by callers