This function builds the text observation for the agent.

This function builds the text observation for the agent.

Build text observations for agent

Wrapper for `receive` handler for performing click action on current page

compress mask array

Strip HTML of tags and add separators to convert observation into simple mode

Create attention mask from input ids.

Create position ids from attention mask.

Duplicate the output based on the action sizes.

Execute the tool's main functionality. This method should be overridden by subclasses. Raises: NotImplementedError: If t

Heavily engineered admissible commands extraction. Lots of commonsense and heuristics used to prune list.

Collect and organize trajectory data, handling batch size adjustments to meet parallel training requirements. Parameters:

Get available actions

keys: `name`, `small_description`, `review` category_type: `category`, `query`

Ground-truth instance segemetations (with consistent object IDs) from THOR

Get corresponding instruction text for environment current step

Returns the metadata for the tool. Returns: dict: A dictionary containing the tool's metadata.

For masking out the subsequent info.

Collect all image observations from workers.

check if the overall task goal was satisfied.

Render and return the HTML for a product item page

The same as logkv(), but if called many times, values averaged.

look at a specific angle

mean pooling when there're paddings input: tensor: batch x time x h mask: batch x time output: tensor: bat

Extracts the first color found if exists

Parse staged format according to exact spec: Planning: <plan> with optional <memory query> inside Action: <action> followed b

Process a batch of observation samples, converting environment observations into model-processable format. Parameters:

Generate random index by sampling uniformly from sum of all weights, then selecting the `min` between the position to keep the list sorted (via bi

Create a new session and reset environment variables