Shape: - y_pred: batch x time x vocab - y_true: batch x time - mask: batch x time

Returns web request result wrapped in BeautifulSoup object Arguments: url (`str`): If no url or html is provided, use the cu

Build tool use environments

Mirror *build_sokoban_envs* so higher‑level code can swap seamlessly.

Select an action randomly.

Determine which page (i.e. item_page, search_results) the given URL is pointing at

returns the object type of a task param

extract objects after "you see" and before "your task is to:"

Prepare the input for the encoder. Pass it through pad, embedding, and rnn.

Process action stage and execute tools.

Query memory for relevant information. Searches both in-memory data and file cache. Args: query: Search

Stores a new batch of records into memory.

Returns web request result wrapped in BeautifulSoup object Arguments: url (`str`): If no url or html is provided, use the cu

Returns a string action from poss_acts.

# BBoxes are [x1, y1, x2, y2]

Encode a list of lists of strs

Get the aggregated hidden state of the encoder. B x D

Returns list of available actions at the current step

Get cumulative reward score for purchased product and goal

Create enviroments

Parse action string to action name and its arguments.

Map action to the corresponding page

Returns result of applying `self.operation` to a contiguous subsequence of the array. self.operation(arr[start], operati

Reset memory with given batch size.

restore object locations and states

Set a custom output directory for the tool. Parameters: output_dir (str): The new output directory path.

Execute text actions and return the next state, rewards, done flags, and additional information. Parameters: - text_

Store memory to file and RAM - delegate to stage processor.

interact mask based action call

Build a Segment Tree data structure. https://en.wikipedia.org/wiki/Segment_tree Can be used as regular array, but with two

Build formatted history for the given environment index.

Normalize answer string for comparison