> This page is for version 0.6.0. > For other versions, use one of these documentation indexes: > - Main (default): https://docs.nvidia.com/nemo/gym/main/llms.txt > - 0.7.0: https://docs.nvidia.com/nemo/gym/v0.7.0/llms.txt > - 0.6.0: https://docs.nvidia.com/nemo/gym/v0.6.0/llms.txt > - 0.5.1: https://docs.nvidia.com/nemo/gym/v0.5.1/llms.txt > - 0.5.0: https://docs.nvidia.com/nemo/gym/v0.5.0/llms.txt > - 0.4.0: https://docs.nvidia.com/nemo/gym/v0.4.0/llms.txt > - 0.3.0: https://docs.nvidia.com/nemo/gym/v0.3.0/llms.txt > - 0.2.1: https://docs.nvidia.com/nemo/gym/v0.2.1/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/nemo/gym/llms.txt. # nemo_gym.web.models Wire models for stateful, multimodal web environments. ## Module Contents ### Classes | Name | Description | | --------------------------------------------------------------------- | ---------------------------------------------------------------------- | | [`WebAction`](#nemo_gym-web-models-WebAction) | Model-independent computer-use action accepted by visual browsers. | | [`WebActionLimits`](#nemo_gym-web-models-WebActionLimits) | Runtime-owned waiting limits advertised when a browser session starts. | | [`WebActionProfile`](#nemo_gym-web-models-WebActionProfile) | - | | [`WebArtifactRef`](#nemo_gym-web-models-WebArtifactRef) | Stable reference to an artifact retained outside a model request. | | [`WebBenchmark`](#nemo_gym-web-models-WebBenchmark) | - | | [`WebImage`](#nemo_gym-web-models-WebImage) | An image can be transported inline, retained as an artifact, or both. | | [`WebObservation`](#nemo_gym-web-models-WebObservation) | JSON-safe observation sent from the browser runtime to the agent. | | [`WebObservationProfile`](#nemo_gym-web-models-WebObservationProfile) | - | | [`WebRuntimeProfile`](#nemo_gym-web-models-WebRuntimeProfile) | Browser execution protocol exposed by Gym. | | [`WebStepResult`](#nemo_gym-web-models-WebStepResult) | - | | [`WebTab`](#nemo_gym-web-models-WebTab) | - | | [`WebTask`](#nemo_gym-web-models-WebTask) | Normalized task envelope; original benchmark fields remain lossless. | | [`WebVerifierResult`](#nemo_gym-web-models-WebVerifierResult) | - | ### Data [`BROWSER_TARGET_CLOSED_STATUS`](#nemo_gym-web-models-BROWSER_TARGET_CLOSED_STATUS) [`CAPTCHA_BUDGET_EXHAUSTED_STATUS`](#nemo_gym-web-models-CAPTCHA_BUDGET_EXHAUSTED_STATUS) ### API ```python class nemo_gym.web.models.WebAction() ``` **Bases:** `BaseModel` Model-independent computer-use action accepted by visual browsers. **`answer`** `Optional[str] = None` --- **`arguments`** `dict[str, Any] = Field(default_factory=dict)` --- **`metadata`** `dict[str, Any] = Field(default_factory=dict)` --- **`name`** `str` --- **`raw_model_output`** `str = ''` --- **`script`** `str` --- **`terminal`** `bool = False` --- ```python class nemo_gym.web.models.WebActionLimits() ``` **Bases:** `BaseModel` Runtime-owned waiting limits advertised when a browser session starts. The default retains the historical tool contract. A benchmark profile may permit longer waits without rewriting the model's requested duration. **`max_step_wait_seconds`** `float | None` --- **`max_wait_seconds`** `float` --- ```python nemo_gym.web.models.WebActionLimits.validate_wait_budget() -> 'WebActionLimits' ``` ```python class nemo_gym.web.models.WebActionProfile ``` **Bases:** `enum.Enum` **`COMPUTER_USE`** `= 'computer_use'` --- ```python class nemo_gym.web.models.WebArtifactRef() ``` **Bases:** `BaseModel` Stable reference to an artifact retained outside a model request. **`mime_type`** `str` --- **`sha256`** `str = Field(pattern='^[0-9a-f]{64}$')` --- **`size_bytes`** `int = Field(ge=0)` --- **`uri`** `str` --- ```python class nemo_gym.web.models.WebBenchmark ``` **Bases:** `enum.Enum` **`VISUALWEBARENA`** `= 'visualwebarena'` --- **`WEBARENA`** `= 'webarena'` --- **`WEBVOYAGER`** `= 'webvoyager'` --- ```python class nemo_gym.web.models.WebImage() ``` **Bases:** `BaseModel` An image can be transported inline, retained as an artifact, or both. **`artifact`** `Optional[WebArtifactRef] = None` --- **`data_url`** `Optional[str] = None` --- ```python nemo_gym.web.models.WebImage.require_transport() -> 'WebImage' ``` ```python class nemo_gym.web.models.WebObservation() ``` **Bases:** `BaseModel` JSON-safe observation sent from the browser runtime to the agent. **`active_tab_index`** `int = Field(default=0, ge=0)` --- **`axtree_text`** `str = ''` --- **`elapsed_time`** `float = Field(default=0.0, ge=0.0)` --- **`element_map`** `dict[str, dict[str, Any]] = Field(default_factory=dict)` --- **`focused_element_id`** `str = ''` --- **`goal`** `list[dict[str, Any]] = Field(default_factory=list)` --- **`last_action`** `str = ''` --- **`last_action_error`** `str = ''` --- **`metadata`** `dict[str, Any] = Field(default_factory=dict)` --- **`screenshot`** `Optional[WebImage] = None` --- **`tabs`** `list[WebTab] = Field(default_factory=list)` --- **`url`** `str = ''` --- ```python class nemo_gym.web.models.WebObservationProfile ``` **Bases:** `enum.Enum` **`A11Y`** `= 'a11y'` --- **`SCREENSHOT`** `= 'screenshot'` --- **`SOM`** `= 'som'` --- ```python class nemo_gym.web.models.WebRuntimeProfile ``` **Bases:** `enum.Enum` Browser execution protocol exposed by Gym. Model-specific prompting is deliberately not represented here. Nano Omni, Qwen, and future policies all drive the same visual browser runtime after their policy adapters normalize actions into `WebAction`. **`VISUAL_BROWSER`** `= 'visual_browser'` --- ```python class nemo_gym.web.models.WebStepResult() ``` **Bases:** `BaseModel` **`benchmark_reward`** `Optional[float] = None` --- **`execution_ok`** `bool` --- **`info`** `dict[str, Any] = Field(default_factory=dict)` --- **`observation`** `WebObservation` --- **`terminated`** `bool = False` --- **`truncated`** `bool = False` --- ```python class nemo_gym.web.models.WebTab() ``` **Bases:** `BaseModel` **`active`** `bool = False` --- **`index`** `int = Field(ge=0)` --- **`title`** `str = ''` --- **`url`** `str = ''` --- ```python class nemo_gym.web.models.WebTask() ``` **Bases:** `BaseModel` Normalized task envelope; original benchmark fields remain lossless. **`action_profile`** `WebActionProfile = WebActionProfile.COMPUTER_USE` --- **`auth_profile`** `Optional[str] = None` --- **`benchmark`** `WebBenchmark` --- **`input_images`** `list[str] = Field(default_factory=list)` --- **`intent`** `str = ''` --- **`model_config`** `= ConfigDict(extra='allow')` --- **`observation_profile`** `Optional[WebObservationProfile] = WebObservationProfile.SCREENSHOT` --- **`original_metadata`** `dict[str, Any] = Field(default_factory=dict)` --- **`runtime_profile`** `WebRuntimeProfile = WebRuntimeProfile.VISUAL_BROWSER` --- **`seed`** `int = 0` --- **`sites`** `list[str] = Field(default_factory=list)` --- **`start_urls`** `list[str] = Field(default_factory=list)` --- **`task_id`** `str` --- **`task_kwargs`** `dict[str, Any] = Field(default_factory=dict)` --- **`verifier_profile`** `Optional[str] = None` --- ```python nemo_gym.web.models.WebTask.stringify_task_id( value: typing.Any ) -> str ``` classmethod ```python class nemo_gym.web.models.WebVerifierResult() ``` **Bases:** `BaseModel` **`evidence`** `list[WebArtifactRef] = Field(default_factory=list)` --- **`failure_kind`** `Optional[str] = None` --- **`metadata`** `dict[str, Any] = Field(default_factory=dict)` --- **`raw_score`** `float = 0.0` --- **`reward`** `float = 0.0` --- **`subscores`** `dict[str, float] = Field(default_factory=dict)` --- **`task_success`** `bool = False` --- **`valid_sample`** `bool = True` --- **`verifier_version`** `str = ''` --- ```python nemo_gym.web.models.BROWSER_TARGET_CLOSED_STATUS = 'browser_target_closed' ``` ```python nemo_gym.web.models.CAPTCHA_BUDGET_EXHAUSTED_STATUS = 'captcha_budget_exhausted' ```