AgentGym-RL
Code and implementations for the paper "AgentGym-RL: Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning" by Zhiheng Xi et al.
File Explorer
Download Latest Version (.zip)- logo.png
- dpo_extension.rst
- fsdp_extension.rst
- megatron_extension.rst
- placement.rst
- config.rst
- gsm8k_example.rst
- ppo_code_architecture.rst
- ppo.rst
- faq.rst
- perf_tuning.rst
- prepare_data.rst
- reward_function.rst
- install.rst
- quickstart.rst
- fsdp_workers.rst
- megatron_workers.rst
- ray_trainer.rst
- conf.py
- hybrid_flow.rst
- index.rst
- Makefile
- README.md
- README_vllm0.7.md
- requirements-docs.txt
- format.sh
- model_merger.py
- multiple_model_merger.py
- single_model_merger.py
- evaluation.yaml
- generation.yaml
- ppo_trainer.yaml
- sft_trainer.yaml
- __init__.py
- core_algos.py
- ray_trainer.py
- __init__.py
- fsdp_sft_trainer.py
- main_eval.py
- main_generation.py
- main_ppo.py
- runtime_env.yaml
- __init__.py
- llama_loader.py
- llama_saver.py
- __init__.py
- parallel_attention.py
- parallel_decoder.py
- parallel_linear.py
- parallel_mlp.py
- parallel_rmsnorm.py
- __init__.py
- modeling_llama_megatron.py
- __init__.py
- __init__.py
- llama.py
- monkey_patch.py
- qwen2.py
- __init__.py
- README.md
- registry.py
- weight_loader_registry.py
- __init__.py
- worker.py
- worker_group.py
- __init__.py
- ray.py
- __init__.py
- decorator.py
- worker.py
- worker_group.py
- __init__.py
- base.py
- megatron.py
- __init__.py
- __init__.py
- dtensor_weight_loaders.py
- __init__.py
- arg_utils.py
- config.py
- llm.py
- llm_engine_sp.py
- model_loader.py
- model_runner.py
- parallel_state.py
- tokenizer.py
- weight_loaders.py
- worker.py
- __init__.py
- arg_utils.py
- config.py
- dtensor_weight_loaders.py
- hf_weight_loader.py
- llm.py
- llm_engine_sp.py
- megatron_weight_loaders.py
- model_loader.py
- model_runner.py
- parallel_state.py
- spmd_gpu_executor.py
- tokenizer.py
- worker.py
- __init__.py
- arg_utils.py
- config.py
- dtensor_weight_loaders.py
- hf_weight_loader.py
- llm.py
- llm_engine_sp.py
- megatron_weight_loaders.py
- model_loader.py
- model_runner.py
- parallel_state.py
- spmd_gpu_executor.py
- tokenizer.py
- worker.py
- __init__.py
- arg_utils.py
- config.py
- dtensor_weight_loaders.py
- hf_weight_loader.py
- llm.py
- llm_engine_sp.py
- megatron_weight_loaders.py
- model_loader.py
- model_runner.py
- parallel_state.py
- spmd_gpu_executor.py
- tokenizer.py
- worker.py
- __init__.py
- __init__.py
- __init__.py
- README.md
- rl_dataset.py
- sft_dataset.py
- client.py
- __init__.py
- checkpoint_manager.py
- fsdp_checkpoint_manager.py
- __init__.py
- performance.py
- trajectory_tracker.py
- __init__.py
- aggregate_logger.py
- __init__.py
- memory.py
- optimizer.py
- optimizer_config.py
- pipeline_parallel.py
- sequence_parallel.py
- tensor_parallel.py
- __init__.py
- ray_backend.py
- __init__.py
- testing_util.py
- utils.py
- __init__.py
- grader.py
- math_normalize.py
- __init__.py
- gsm8k.py
- math.py
- __init__.py
- config.py
- distributed.py
- flops_counter.py
- fs.py
- fsdp_utils.py
- hdfs_io.py
- import_utils.py
- logging_utils.py
- megatron_utils.py
- memory_buffer.py
- model.py
- py_functional.py
- ray_utils.py
- seqlen_balancing.py
- tokenizer.py
- torch_dtypes.py
- torch_functional.py
- tracking.py
- ulysses.py
- version
- __init__.py
- base.py
- dp_actor.py
- __init__.py
- base.py
- dp_critic.py
- __init__.py
- naive.py
- prime.py
- __init__.py
- reward_model.py
- __init__.py
- base.py
- __init__.py
- vllm_rollout.py
- __init__.py
- naive_rollout.py
- __init__.py
- base.py
- hf_rollout.py
- schemas.py
- tokenizer.py
- __init__.py
- base.py
- fsdp_ulysses.py
- fsdp_vllm.py
- megatron_vllm.py
- __init__.py
- agent_fsdp_workers.py
- megatron_workers.py
- __init__.py
- protocol.py
- .gitignore
- .readthedocs.yaml
- .style.yapf
- Notice.txt
- pyproject.toml
- requirements.txt
- setup.py
- VERL_LICENSE
- VERL_README.md
- AgentGym-RL-main.png
- bytedance.jpg
- env.jpg
- fudannlp_logo.png
- main_greedy_performance.jpg
- main_performance.jpg
- pseudo.jpg
- ScalingInter-RL-Method.png
- searchqa_performance.jpg
- shanghai_innovation_institute_logo.png
- webarena_performance.png
- babyai_eval.sh
- sciworld_eval.sh
- searchqa_eval.sh
- textcraft_eval.sh
- webarena_eval.sh
- babyai_train.sh
- sciworld_train.sh
- searchqa_train.sh
- textcraft_train.sh
- webarena_train.sh
- babyai_train.sh
- sciworld_train.sh
- searchqa_train.sh
- textcraft_train.sh
- webarena_train.sh
- .gitmodules
- AgentGym
- LICENSE
- README.md
๐ Installation Guide
1. Get the code
git clone https://github.com/WooooDyy/AgentGym-RL
Downloads the entire project code from GitHub to your computer.
cd AgentGym-RL
Moves into the project folder you just downloaded.
2. Official Install Script
Easy RecommendedPrerequisites
- Python 3 Python is required to use pip.
pip3 install torch==2.4.0 --index-url https://download.pytorch.org/whl/cu124
Installs the package published on PyPI directly โ no need to clone the source.
pip3 install $FLASH_ATTENTION_NAME
Installs the package published on PyPI directly โ no need to clone the source.
pip3 install transformers==4.51.3
Installs the package published on PyPI directly โ no need to clone the source.
After installing, open a new terminal and run the program's version command (e.g. --version) to confirm it worked.
Pulled directly from this repo's README.
3. Python
EasyPrerequisites
pip3 install torch==2.4.0 --index-url https://download.pytorch.org/whl/cu124
Installs the package published on PyPI directly โ no need to clone the source.
pip3 install $FLASH_ATTENTION_NAME
Installs the package published on PyPI directly โ no need to clone the source.
pip3 install -e .
Installs the Python libraries listed in requirements.txt (or similar).
pip3 install transformers==4.51.3
Installs the package published on PyPI directly โ no need to clone the source.
If it runs without errors and prints output in the terminal, it worked.
Pulled directly from this repo's README.
4. Make
MediumPrerequisites
- Git Needed to download the project code from GitHub.
- Make Usually pre-installed on Linux/macOS. On Windows, install separately (e.g. via MSYS2 or WSL).
cd AgentGym-RL/docs
This project's files live in a subfolder, so move into it first.
make
Compiles the code based on the generated build configuration to produce an executable.
If it finishes without errors, it worked. Try running the generated executable directly.
// repository documentation
Was this content helpful?
(0 ratings)