Function to create a new training engine instance based on the provided config. Args: key: A configuration object contain

Unified entry point for `pad_input` across CUDA and NPU backends. Dynamically dispatches to the appropriate device-specific implementation:

Convert this DataProto to TensorDict. Note that this requires tensordict version at least 0.10 Returns:

Check correctness of code generation with a global timeout. The global timeout is to catch some extreme/rare cases not handled by the timeouts

Unified interface for computing IS weights and rejection masks. This function combines IS weight calculation (truncated) and rejection sampling (

Get the distributed communication backend based on device type. Returns the appropriate collective communication backend for the detected acc

Create a huggingface processor to process multimodal data. Args: name_or_path (str): The name of the processor. Returns: tra

发起请求

[Deprecated] convert config Args: hf_config (PretrainedConfig): _description_ megatron_config (_type_): _description_ Return

Get ulysses sequence parallel world size.

Unified entry point for `index_first_axis` across CUDA and NPU backends. Dynamically dispatches to the appropriate device-specific implement

移除并销毁环境

Convert a config dictionary to CLI arguments for vLLM server. Handles different value types appropriately: - None: skipped - bool Tr

Initialize class state shared across all instances.

Load an object from a module path. Args: module_path (str): See :func:`load_module`. object_name (str): The name of t

记录日志消息到文件和控制台，并过滤base64图片数据

Spawn to a dictionary of worker groups, each with a subset of method with prefix. Args: prefix_set: Set of prefixes to create wor

Make a HTTP request with retry logic and consistent error handling. Args: endpoint (str): The API endpoint to call (without leadi

More aggressive GPU memory cleanup function, tries to release PyTorch reserved but unallocated memory. Args: force_sync: Whether

Append values from new_data to lists in data. For each key in new_data, this function appends the corresponding value to a list stored under

计算a◎b对应的复数平方

Execute a method on all workers synchronously. Args: method_name: Name of the method to execute *args: Positional arg

Preprocess packed sequences CP splits sequence into CP*2 chunks, and each GPU gets 2 chunks (GPU0 gets first and last chunks, GPU1 gets s

Reduces a dictionary of metric lists by computing the mean, max, or min of each list. The reduce operation is determined by the key name:

Create a dataset. Arguments: data_paths: List of paths to data files. data_config: The data config. tokenizer (Tokenizer)

Helper to create DataProto for testing.

Generates a response for the current turn of interaction. Returns a tuple containing: - should_terminate_sequence (bool): Tru

Compute the mean of `values` over elements selected by `mask`. Args: values (Tensor): Input tensor. mask (Tensor): Boolean o

Determine if checkpoint should be saved based on capacity esi expiration. Args: max_steps_duration: Max estimated time (seconds) req

Create a DataProto from a dict of tensors and non_tensors

Get a value from a TensorDict with automatic unwrapping. Retrieves a value from the TensorDict and automatically converts it to a Python-nati

Get ulysses sequence parallel process group.

Load and initialize a reward manager based on the configuration. Args: config: PPO trainer configuration object containing reward_mo

_summary_ Thread-safe directory creation function that ensures the directory is created even if multiple processes attempt to create it simult

Compute per-token log-probabilities for the given labels. Uses a Flash-Attention–based cross-entropy (if available) for efficient backward,

Given the config, do we need ref policy.

Context manager for temporarily setting an environment variable. This context manager ensures that environment variables are properly set and res

Copied from ManiSkill! Maps any given sequence to a torch tensor on the CPU/GPU. If physx gpu is not enabled then we use CPU, otherwise G

Create a mock checkpoint directory.

Create a random mask given input_ids. Support left padding and right padding. Process: - Sample valid token length - Sample left_padding l

Create a sampler for the dataset. Arguments: data_config: The data config. dataset (Dataset): The dataset. Returns:

Gather a tensor across a process group and optionally unpad its padded elements. Args: x (Tensor): Input tensor to gather. g

加载数据集，支持 JSON、JSONL 和 Parquet 文件格式，并始终返回 list。 参数: dataset_path (str): 数据集文件路径。 dataset (list, optional): 如果提供了 dataset，

Split a batch into micro-batches by total token count, with optional DP sync and padding. Args: batch (TensorDict): must include "at

Execute the main PPO training workflow. This method sets up the distributed training environment, initializes workers, datasets, and

Sleep all rollout replica instances.

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

获取指定worker映射的instance数量

检查Worker是否健康（基于心跳时间）

查找可用端口（改进版） Args: host: 主机地址 start_port: 起始端口 max_port: 最大端口号，默认65535 randomize: 是否在端口范围内随机选择，默认True

Flush the cache of the server. This method repeatedly attempts to flush the server cache until successful. The flush operation will n

获取任务信息

Whether the current rank is the first rank in model parallel group that contains model outputs

A context manager to add attributes to a trace for the configured backend. Args: sample_index: Sample index for the trace. step: