Staged rollout that extends TrajectoryCollector for VERL compatibility. Supports: Planning stage <plan> xxx</plan> [opt]

From Multi-Head Attention module https://github.com/jadore801120/attention-is-all-you-need-pytorch

覆盖 __init__ 以绕过复杂依赖，仅用于测试无状态方法。

A wrapper of textEnv for models. Returns valid actions at each step of the game.

Mock configuration for testing.

Mock tokenizer for testing.

Environment manager for tool use tasks

Thin wrapper around OpenAI's Chat Completions API. The instance is callable and returns plain text. Images are not sent as binary by design t

Linear interpolation between initial_p and final_p over schedule_timesteps. After this many timesteps pass final_p is returned. :para

Memory manager: responsible for storing & fetching per‑environment history records.

an extension of ai2thor.controller.Controller for ALFRED tasks

Experiment configuration with sensible defaults.

A basic LSTM cell.

Mock actor worker for generating responses.

Mock environment for testing.

Configuration for test runs.

Simple tokenizer for testing.

Track a series of values and provide access to smoothed values over a window or the global series average.

inputs: p: batch x time x inp_p mask_p: batch x time q: batch x time x inp_q mask_q:

Memory manager with summarization capability. Inherits from SimpleMemory and adds optional history summarization.

Action stage - executes tools or returns environment actions.

Interface for Textworld Env

Interface for Embodied (THOR) environment

简易 ToolManager，仅返回固定的提示指令。

Extended memory that adds file persistence and query capabilities. Inherits from SimpleMemory for compatibility, adds memory.md support.

Google Programmable Search Engine (Custom Search JSON API) tool. Reads `GOOGLE_API_KEY` and `GOOGLE_CX` from environment and performs a search vi

Simple GCN layer, similar to https://arxiv.org/abs/1609.02907

Wraps another sampler to yield a mini-batch of indices. It enforces that the batch only contain elements from the same group. It also tri

Textworld instance of handcoded expert. Uses predicates from Textworld Engine for making decisions.

THOR instance of handcoded expert. Uses predicates from THOR metadata for making decisions.

Agent that interfaces with LLM APIs using chat-based conversation.

input: p: batch x inp_p p_mask: batch q: batch x time x inp_q q_mask:

Memory storage stage - saves important information.

Mock environment configuration.

Mock rollout configuration.

Main processor that orchestrates all stages.

OpenManus agent that can work with all environments

Planning stage - reads from memory, outputs plan.

Usage: with logger.ProfileKV("interesting_scope"): code