Returns True if mobile data is working, False otherwise. The mobile data not working when any of the following are true: - Airplane m

Process payment for update reservation. - Wallet payments must be whole dollars and sufficient in balance. - Card/Other payme

Simulates running a speed test for mobile data based on current network conditions. Returns a tuple: (speed_mbps, description).

Search for direct flights Args: date: The date of the flight in the format 'YYYY-MM-DD', such as '2024-01-01'. origin

Search for direct trains on a given date with optional filters.

Remove residual stop-token strings from generated text.

Toggles the master Mobile Data switch. Returns the new state.

Lock → broadcast → clear → unlock → pbar++. Lock prevents NCCL deadlock.

Check if the task is valid. Only the tasks that require at least one action are valid.

Compute the approximate KL divergence between two distributions. Schulman blog: http://joschu.net/blog/kl-approx.html Args: log_

Destroy all reloadable process groups.

Disable forward pre-hooks for provided DDP-wrapped model chunks. Args: model_chunks (Sequence[DDP]): Sequence of DDP modules to disable h

Gather per-rank metrics, reduce by mean on the DP source rank, and log. Expects `log_dict` to contain plain scalars. The DP source rank prin

Generate the next message from an assistant message. Args: message: The agent message. state: The state of the user s

Get the function format of the action.

get order of seq lengths to make partitions balanced, this is used in balacing sum of seqlength across dp ranks and microbatches Parameter

Get a hash of the user database Returns None if the database is not available

Get the tools of the domain.

Similar to itertools.groupby, but do not require iterable to be sorted

Make a tool call based on the requestor. Args: tool_name: The name of the tool to call. requestor: The reques

Decorator to register a new Agent implementation

Reload all reloadable process groups.

Remove vocab padding: param[:vocab_size] for embedding/output layers, else unchanged.

Run a coroutine in the background event loop.

Runs any function available on agent environment or user environment.

Set the state of the environment given initialization data and a list of messages.

Update access time and count.

Update the mock device state.

Return an updated BaseModel instance based on the update_data.

Use a tool available to the user of the domain.

Initialize the LLMAgent.

Aggregate per-sample scores; semantically consistent with calc_metrics in ruler_hqa.py.

Redistribute response tensors from allgather-CP layout to zigzag ring-attn layout. After allgather context parallelism, each rank holds a contigu

Apply membership discount only on fare component (not fees/insurance). - gold: 10% - silver: 5% - regular: 0%

Dynamically build the call_expert tool based on the actual model_mapping and pricing.

tool definition for func_call orchestrator (call_expert).

QA correctness: normalized exact match. Handles numeric answers, short answers, and longer text.

Checks if a payment request has been made.

Clean a node and all its descendants. This is the core cleanup function. Args: node: Node to clean (including all

Arguments shared by both phases.

concessions spec: [{item_id, name, size, quantity, price_each}] Add tax_each and total per line.

Compute token log-probabilities for a list of packed batches. Parameters: model_tag: Which parameters to use, e.g. "actor" or "re

Compute a base price per passenger per segment as an integer (dollars). This is a simple model using service_type and travel_class.

Disconnects any active VPN connection.

Extract the final answer from solver's multi-turn response. Tries in order: 1. \\boxed{...} from the response 2. Last numeric value from

Extract reward_info from a tau2 simulation result. Mirrors Tau2Adapter._extract_from_sim to handle edge cases (e.g. tau_run_error, no

Find the dish that contains the given plate_ref.

Load full state dict into FSDP2 model with efficient broadcast from rank 0. This function loads weights from rank 0 and broadcasts to all oth

Build prompt_text using tokenizer.apply_chat_template(tools=...), then call SGLang /generate and return a GenerationOutput (containin

Returns the bills in the customer's bill_ids list that are in the AWAITING_PAYMENT status.

Get a catalogue group from the database.

YY-MM-DD-HH-MM in EST

Get a dish from the database.

Get context manager for model initialization. Returns a callable that creates a context manager. Uses meta device (no memory allocati