Generates a memory snapshot and saves it as a pickle file in a specified directory. The files are organized by timestamp in subdirect

The keys of the environment variables that are used to configure the Worker.

Context manager entry for switching the engine and model into evaluation mode. Usage: with engine.eval_mode():

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

Execute a method on rank zero worker asynchronously. Args: method_name: Name of the method to execute *args: Position

Extract multi-modal inputs from TensorDict NonTensorStack.

Extract the equation from the solution string.

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

Args: input_ids: input ids. shape (batch_size, seq_length) attention_mask: attention_mask. shape (batch_size, seq_le

Args: input_ids: input ids. shape (batch_size, seq_length) attention_mask: attention_mask. shape (batch_size, seq_le

Perform a forward pass and optionally a backward pass on a batch of data. Args: data: The input data for the forward pas

We assume: - The model takes input: (input_ids, attention_mask, position_ids). No rmpad for the input - The communication sha

r""" Copy paste LLaMa's forward https://github.com/linkedin/Liger-Kernel/blob/main/src/liger_kernel/transformers/model/llama.py This func

Load configuration from a YAML file.

Generate tokens from prompt ids. Args: request_id (str): request id for sticky session. prompt_ids (List[int]): List

get extracter answers from model Args: prompt: str, the prompt answer_extractor: instance to extractor answer

Get all registered dispatch and collect dp_ranks. Returns: dict[str, int]: A dictionary mapping mesh names to the

Get the enum for the reduction method of cross entropy.

Get the first few rows of a dataset split.

Get the forward function for given model architecture.

Get the weight converter for given model architecture.

Build the model.

Generate and return an NPU profiler object. Args: contents (list[str]): A list of options to control the collection content,

Get PEFT class from model config. Args: model_config: Model configuration object. bridge: Megatron-Bridge AutoBridge instance.

Get rollout worker actor class for colocated and standalone mode.

Gets the position ids for GLM4V in padding-free format. The batch dim has been removed and the input_ids should be a 1D tensor representing a

Retrieve the available splits (train/test/validation) of a dataset.

Create a TensorDict from tensors and non-tensor data. Automatically handles nested structures in lists by converting them to NonTensorStack.

Get the index offset of any pipeline stage, given the level of pipelining. Make pipeline_rank and vp_stage as two arguments to make it more

Get ulysses sequence parallel rank.

Util to extract Vision Tower from a VL model instance

For vLLM, the smallest `max_lora_rank` is 8, and allowed values are (8, 16, 32, 64, 128, 256, 320, 512) This function automatically adjusts t

The answer will be considered correct if: (a) it normalizes to the same string as the ground truth answer OR (b) sympy can simplify t

Perform inference on a batch of data. Args: data: The input data for inference, typically containing tensors and metadat

Initialize a Mcore model. Args: tfconfig: The transformer config. hf_config: The HuggingFace config. pre_process: Op

Instantiate or load the model, optimizer, and learning rate scheduler. Should prepare all components necessary for training or evalu

Initialize tools from config file. Supports both NATIVE and MCP tool types. For MCP tools, a temporary event loop is created only when needed

Interpret pairwise GEval score to determine winner. Args: score: GEval score (0-1 scale, converted from 0-10) Returns:

Check if padding is enabled for the DataProto. Returns: bool: True if padding is enabled, False otherwise.

Return True if the current router_replay_action is REPLAY_FORWARD for the local router instances. This inspects the first local RouterReplay

Convert JSONL file to Parquet format. Simply converts format without modifying keys. verl will use prompt_key/response_key from config t

Extract the last LaTeX boxed expression from a string. Args: string: Input string containing LaTeX code Returns: The last bo

Load adapter parameters from checkpoint. Loads from Megatron's distributed checkpoint structure: reads from checkpoint_path/mp_rank_{tp:02d}_

Load test dataset from configured path.

Load demo examples from JSONL file. Args: demo_examples_path: Path to the JSONL file containing demo examples (optional,

Advance the learning rate scheduler by one step. Returns: current_lr (float or list[float]): Updated learning rate(s).

Make minibatch iterator for updating the actor Args: data (DataProto): a DataProto containing keys ``input_ids``

End a mark range in the profiler. Args: range_id (str): The id of the mark range to end.

Start a mark range in the profiler. Args: message (str, optional): The message to be displayed in the profiler. Defaults to N

Exact match of math if and only if: 1. numerical equal: both can convert to float and are equal 2. symbolic equal: both can convert to sy

Convert a MCP Tool to an OpenAI ChatCompletionTool.

Generate an MD5 hash of a path string. This function is used to create unique identifiers for paths, typically for creating cache directories

Given a config, do we need critic.

Given a role worker mapping, do we need ref policy.

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

Transform the model name in each model_chunk in each pp stage into the name in inference engine

Offload.

Generate data using the provided data generation strategy. Note: This method is intended to change the dataset after each training ba

On group commit forward.

Pad a DataProto to size divisible by size_divisor Args: size_divisor (int): size divisor Returns: data: (DataProto): the pad