MCPcopy Create free account

hub / github.com/Alibaba-NLP/VRAG / functions

Functions1,037 in github.com/Alibaba-NLP/VRAG

⨍ Functions 1,037◇ Types & classes 150↳ Endpoints 26
↓ 1 callersMethod_forward_head
(self, hidden_states)
VRAG-RL/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:611
↓ 1 callersMethod_forward_head
(self, hidden_states)
VRAG-RL/verl/models/llama/megatron/modeling_llama_megatron.py:306
↓ 1 callersMethod_forward_head
(self, hidden_states)
VRAG-RL/verl/models/llama/megatron/modeling_llama_megatron.py:561
↓ 1 callersMethod_forward_micro_batch
(self, micro_batch)
VRAG-RL/verl/workers/fsdp_workers.py:1003
↓ 1 callersMethod_generate_minibatch
(self, prompts: DataProto)
VRAG-RL/verl/workers/rollout/hf_rollout.py:51
↓ 1 callersMethod_get_free_port
(self)
VRAG-RL/verl/single_controller/base/worker.py:56
↓ 1 callersMethod_get_node_ip
(self)
VRAG-RL/verl/single_controller/base/worker.py:39
↓ 1 callersFunction_get_parallel_model_architecture_from_config
(config: PretrainedConfig, value=False)
VRAG-RL/verl/utils/model.py:269
↓ 1 callersMethod_get_video_frames_base64
Extract video frames as base64 strings (in-memory, no disk I/O).
demo/vimrag_agent.py:205
↓ 1 callersMethod_handle_search_node
Handle add_search_node tool call.
demo/vimrag_agent.py:282
↓ 1 callersMethod_handle_summarize_memorize
Handle summarize_and_memorize tool call (in-memory processing).
demo/vimrag_agent.py:295
↓ 1 callersMethod_init_head
(self, config: Qwen2Config)
VRAG-RL/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:293
↓ 1 callersMethod_init_head
(self, config)
VRAG-RL/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:550
↓ 1 callersMethod_init_head
(self, config)
VRAG-RL/verl/models/llama/megatron/modeling_llama_megatron.py:294
↓ 1 callersMethod_init_head
(self, config)
VRAG-RL/verl/models/llama/megatron/modeling_llama_megatron.py:549
↓ 1 callersMethod_init_rope
(self)
VRAG-RL/verl/models/qwen2/megatron/layers/parallel_attention.py:208
↓ 1 callersMethod_init_rope
(self)
VRAG-RL/verl/models/llama/megatron/layers/parallel_attention.py:204
↓ 1 callersMethod_init_with_detached_workers
(self, worker_names)
VRAG-RL/verl/single_controller/ray/base.py:229
↓ 1 callersMethod_init_with_resource_pool
(self, resource_pool, ray_cls_with_init, bin_pack, detached)
VRAG-RL/verl/single_controller/ray/base.py:234
↓ 1 callersFunction_inject_implicit_mixed_number
Automatically make a mixed number evalable e.g. 7 3/4 => 7+3/4
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:126
↓ 1 callersFunction_inner_call_method
(_method)
VRAG-RL/verl/utils/reward_score/prime_code/testing_util.py:633
↓ 1 callersFunction_is_float
(num: str)
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:92
↓ 1 callersFunction_is_int
(x: float)
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:100
↓ 1 callersMethod_is_worker_alive
(self, worker)
VRAG-RL/verl/single_controller/base/worker_group.py:112
↓ 1 callersFunction_last_boxed_only_string
(string)
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:333
↓ 1 callersMethod_load_checkpoint
(self)
VRAG-RL/verl/trainer/ppo/ray_trainer.py:605
↓ 1 callersMethod_load_params_to_cuda
(self, pp_rank, to_empty=False)
VRAG-RL/verl/workers/sharding_manager/megatron_vllm.py:107
↓ 1 callersFunction_make_causal_mask
Make causal mask used for bi-directional self-attention.
VRAG-RL/verl/utils/torch_functional.py:489
↓ 1 callersFunction_make_causal_mask
Make causal mask used for bi-directional self-attention.
VRAG-RL/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:46
↓ 1 callersFunction_make_causal_mask
Make causal mask used for bi-directional self-attention.
VRAG-RL/verl/models/llama/megatron/modeling_llama_megatron.py:47
↓ 1 callersFunction_map_each_response
(l)
VRAG-RL/verl/workers/rollout/sglang_rollout/sglang_rollout.py:60
↓ 1 callersFunction_materialize_futures
(*args, **kwargs)
VRAG-RL/verl/single_controller/base/decorator.py:379
↓ 1 callersFunction_megatron_calc_layer_map
Calculate the mapping of global layer_idx to local layer_idx Returns: layer_map (Dict: int -> tuple(int, int, int)): mapping f
VRAG-RL/verl/models/qwen2/megatron/checkpoint_utils/qwen2_loader.py:21
↓ 1 callersFunction_megatron_calc_layer_map
Calculate the mapping of global layer_idx to local layer_idx Returns: layer_map (Dict: int -> tuple(int, int, int)): mapping f
VRAG-RL/verl/models/qwen2/megatron/checkpoint_utils/qwen2_saver.py:39
↓ 1 callersFunction_megatron_calc_layer_map
Calculate the mapping of global layer_idx to local layer_idx Returns: layer_map (Dict: int -> tuple(int, int, int)): mapping f
VRAG-RL/verl/models/llama/megatron/checkpoint_utils/llama_saver.py:42
↓ 1 callersFunction_megatron_calc_layer_map
Calculate the mapping of global layer_idx to local layer_idx Returns: layer_map (Dict: int -> tuple(int, int, int)): mapping f
VRAG-RL/verl/models/llama/megatron/checkpoint_utils/llama_loader.py:23
↓ 1 callersFunction_mkdir
hdfs mkdir
VRAG-RL/verl/utils/hdfs_io.py:75
↓ 1 callersMethod_model_generate
Call vLLM OpenAI-compatible API with streaming.
demo/vimrag_agent.py:66
↓ 1 callersMethod_normalize_config_bsz
(self)
VRAG-RL/verl/trainer/fsdp_sft_trainer.py:110
↓ 1 callersMethod_optimizer_step
(self)
VRAG-RL/verl/workers/critic/dp_critic.py:115
↓ 1 callersMethod_optimizer_step
(self)
VRAG-RL/verl/workers/actor/dp_actor.py:161
↓ 1 callersFunction_parse_latex
Attempts to parse latex to an expression sympy can read.
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:74
↓ 1 callersMethod_parse_response
Parse model response content and extract tool call.
demo/vimrag_agent.py:271
↓ 1 callersMethod_pooling_last
(hidden_state: torch.Tensor, attention_mask: torch.Tensor)
search_engine/models/Qwen3_VL_Embedding/qwen3_vl_embedding.py:365
↓ 1 callersFunction_post_process_outputs
(tokenizer, output)
VRAG-RL/verl/workers/rollout/sglang_rollout/sglang_rollout.py:58
↓ 1 callersMethod_post_process_params
For each param, if it is a tp-splited param, we all-gather from micro_dp group.
VRAG-RL/verl/workers/sharding_manager/megatron_vllm.py:334
↓ 1 callersFunction_pre_process_inputs
(pad_token_id, prompt_token_ids: torch.Tensor)
VRAG-RL/verl/workers/rollout/sglang_rollout/sglang_rollout.py:49
↓ 1 callersFunction_pre_process_inputs
(pad_token_id, prompt_token_ids: torch.Tensor)
VRAG-RL/verl/workers/rollout/vllm_rollout/fire_vllm_rollout.py:50
↓ 1 callersFunction_pre_process_inputs
(pad_token_id, prompt_token_ids: torch.Tensor)
VRAG-RL/verl/workers/rollout/vllm_rollout/vllm_rollout.py:49
↓ 1 callersFunction_pre_process_inputs
(pad_token_id, prompt_token_ids: torch.Tensor)
VRAG-RL/verl/workers/rollout/vllm_rollout/vllm_rollout_spmd.py:50
↓ 1 callersMethod_prepare_decoder_attention_mask
(self, attention_mask, input_shape, inputs_embeds)
VRAG-RL/verl/models/qwen2/megatron/modeling_qwen2_megatron.py:99
↓ 1 callersMethod_prepare_decoder_attention_mask
(self, attention_mask, input_shape, inputs_embeds)
VRAG-RL/verl/models/llama/megatron/modeling_llama_megatron.py:100
↓ 1 callersMethod_preprocess_inputs
(self, conversations: List[List[Dict]])
search_engine/models/Qwen3_VL_Embedding/qwen3_vl_embedding.py:329
↓ 1 callersMethod_process_next_obs
Process next observations from environment.
VRAG-RL/vrag_agent/generation.py:113
↓ 1 callersMethod_read_files_and_tokenize
(self)
VRAG-RL/verl/utils/dataset/rm_dataset.py:85
↓ 1 callersMethod_read_files_and_tokenize
(self)
VRAG-RL/verl/utils/dataset/sft_dataset.py:73
↓ 1 callersFunction_remove_right_units
(string)
VRAG-RL/verl/utils/reward_score/prime_math/math_normalize.py:104
↓ 1 callersMethod_save_checkpoint
(self)
VRAG-RL/verl/trainer/ppo/ray_trainer.py:572
↓ 1 callersFunction_str_to_int
(x: str)
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:120
↓ 1 callersFunction_strip_string
(string)
VRAG-RL/verl/utils/reward_score/prime_math/math_normalize.py:129
↓ 1 callersMethod_switch_chat_template
(self, data: DataProto)
VRAG-RL/verl/workers/fsdp_workers.py:1072
↓ 1 callersFunction_sympy_parse
Parses an expression with sympy.
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:65
↓ 1 callersFunction_transform_params_to_json_serializable
(x, convert_list_to_dict: bool)
VRAG-RL/verl/utils/tracking.py:140
↓ 1 callersMethod_update_messages_with_memory
Update messages with multimodal memory content.
demo/vimrag_agent.py:253
↓ 1 callersMethod_update_rolling_state
Update rolling state with new responses and observations.
VRAG-RL/vrag_agent/generation.py:216
↓ 1 callersMethod_validate_config
(self)
VRAG-RL/verl/trainer/ppo/ray_trainer.py:313
↓ 1 callersMethod_validate_config
Validate config options not implemented for Megatron backend
VRAG-RL/verl/workers/critic/megatron_critic.py:79
↓ 1 callersMethod_validate_config
Validate config options not implemented for Megatron backend
VRAG-RL/verl/workers/actor/megatron_actor.py:138
↓ 1 callersFunctionacc_reward
(predict_str: str, ground_truth: str)
VRAG-RL/verl/utils/reward_score/geo3k.py:25
↓ 1 callersFunctionadd_root_node
添加根节点
demo/vimrag_app.py:600
↓ 1 callersFunctionadd_search_node
添加搜索节点到图中
demo/vimrag_app.py:614
↓ 1 callersFunctionall_gather_tensor
(local_tensor: Tensor, group: Optional[dist.ProcessGroup] = None, async_op: bool = False)
VRAG-RL/verl/utils/ulysses.py:154
↓ 1 callersFunctionapply_kl_penalty
(data: DataProto, kl_ctrl: core_algos.AdaptiveKLController, kl_penalty='kl')
VRAG-RL/verl/trainer/ppo/ray_trainer.py:137
↓ 1 callersFunctionapply_rotary_pos_emb
(q, k, cos, sin, position_ids)
VRAG-RL/verl/models/llama/megatron/layers/parallel_attention.py:123
↓ 1 callersFunctionapply_rotary_pos_emb_rmpad_flash
(q, k, cos, sin, cu_seqlens, max_seqlen)
VRAG-RL/verl/models/qwen2/megatron/layers/parallel_attention.py:304
↓ 1 callersFunctionapply_rotary_pos_emb_rmpad_flash
(q, k, cos, sin, cu_seqlens, max_seqlen)
VRAG-RL/verl/models/llama/megatron/layers/parallel_attention.py:320
↓ 1 callersFunctionare_equal_under_sympy
(ground_truth_normalized: str, given_normalized: str)
VRAG-RL/verl/utils/reward_score/prime_math/__init__.py:240
↓ 1 callersFunctionarg_parse
()
VRAG-RL/scripts/data_construct_pipeline.py:360
↓ 1 callersFunctionbroadcast_params
(module)
VRAG-RL/verl/models/qwen2/megatron/checkpoint_utils/qwen2_loader.py:69
↓ 1 callersFunctionbroadcast_params
(module)
VRAG-RL/verl/models/llama/megatron/checkpoint_utils/llama_loader.py:73
↓ 1 callersFunctionbuild_graph_from_sample
从 answer 事件的 sample["graph"] 构建完整的图
demo/vimrag_app.py:874
↓ 1 callersMethodbuild_index
(self, input_dir, index_output_path, corpus_output_path, bs=2, save_interval=512)
search_engine/search_engine.py:90
↓ 1 callersFunctionbuild_memory_reference
Build the memory references. The memory buffers are built using the build_memory_buffer API. This API will allocate a weight buffer pointer to the
VRAG-RL/verl/utils/memory_buffer.py:116
↓ 1 callersFunctioncalculate_anls
(gold_labels, prediction, threshold=0.7)
VRAG-RL/verl/utils/reward_score/vrag.py:31
↓ 1 callersFunctioncalculate_intuitive_memory_energy
计算多模态记忆的能量值（直观版本）。 1. Phi_i 简化为出度（该动作引导了多少后续动作） 2. Gamma 简化为子节点能量的直接反馈
demo/vimrag_utils.py:156
↓ 1 callersFunctioncall_method
(method, inputs)
VRAG-RL/verl/utils/reward_score/prime_code/testing_util.py:617
↓ 1 callersFunctionceildiv
(a, b)
VRAG-RL/verl/utils/seqlen_balancing.py:220
↓ 1 callersMethodcheck_consistency
Check the consistency of the DataProto. Mainly for batch and non_tensor_batch We expose this function as a public one so that user can call th
VRAG-RL/verl/protocol.py:283
↓ 1 callersFunctioncollect_dp_compute
(worker_group, output)
VRAG-RL/verl/single_controller/base/decorator.py:265
↓ 1 callersFunctioncollect_megatron_compute
Only collect the data from the tp=0 and pp=last and every dp ranks
VRAG-RL/verl/single_controller/base/decorator.py:103
↓ 1 callersFunctioncollect_megatron_pp_as_dp
treat pp as dp. Only collect data on tp=0
VRAG-RL/verl/single_controller/base/decorator.py:209
↓ 1 callersFunctioncompute_advantage
(data: DataProto, adv_estimator, gamma=1.0, lam=1.0, num_repeat=1)
VRAG-RL/verl/trainer/ppo/ray_trainer.py:169
↓ 1 callersFunctioncompute_data_metrics
(batch: DataProto, use_critic: bool = True)
VRAG-RL/verl/trainer/ppo/metric_utils.py:46
↓ 1 callersFunctioncompute_format_reward_only
(predict_str: str, ground_truth: str, extra_info)
VRAG-RL/verl/utils/reward_score/vrag.py:67
↓ 1 callersMethodcompute_log_prob
(self, data: DataProto)
VRAG-RL/verl/workers/megatron_workers.py:425
↓ 1 callersMethodcompute_log_prob
(self, data: DataProto)
VRAG-RL/verl/workers/fsdp_workers.py:525
↓ 1 callersMethodcompute_log_prob
Compute logits given a batch of data. Args: data (DataProto): a batch of data represented by DataProto. It must contain key ```in
VRAG-RL/verl/workers/actor/base.py:39
↓ 1 callersMethodcompute_ref_log_prob
(self, data: DataProto)
VRAG-RL/verl/workers/fsdp_workers.py:559
↓ 1 callersMethodcompute_reward
Computing reward given input_ids. The transformers should output a tensor with shape [batch_size, sequence_length], and the value at [EOS]
VRAG-RL/verl/workers/reward_model/base.py:29
↓ 1 callersMethodcompute_rm_score
(self, data: DataProto)
VRAG-RL/verl/workers/fsdp_workers.py:1130
← previousnext →301–400 of 1,037, ranked by callers