Unable to run the quick start example

[JustinTong0323]

Reproduce: follow the instruction in quickstart, at Step 4: Agent Training with GRPO, would have this error:

[2025-07-08 19:37:12,747] [INFO] [real_accelerator.py:254:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2025-07-08 19:37:14,247] [INFO] [logging.py:107:log_dist] [Rank -1] [TorchCheckpointEngine] Initialized with serialization = False
Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/root/xinyuan/sgl-workspace/rllm/examples/math_tool/train_math_with_tool.py", line 7, in <module>
    from rllm.trainer.agent_trainer import AgentTrainer
  File "/root/xinyuan/sgl-workspace/rllm/rllm/trainer/__init__.py", line 6, in <module>
    from .agent_trainer import AgentTrainer
  File "/root/xinyuan/sgl-workspace/rllm/rllm/trainer/agent_trainer.py", line 6, in <module>
    from rllm.trainer.verl.train_agent_ppo import train_agent
  File "/root/xinyuan/sgl-workspace/rllm/rllm/trainer/verl/train_agent_ppo.py", line 9, in <module>
    from rllm.trainer.env_agent_mappings import AGENT_CLASS_MAPPING, ENV_CLASS_MAPPING
  File "/root/xinyuan/sgl-workspace/rllm/rllm/trainer/env_agent_mappings.py", line 17, in <module>
    "swe": safe_import("rllm.environments.swe.swe", "SWEEnv"),
  File "/root/xinyuan/sgl-workspace/rllm/rllm/trainer/env_agent_mappings.py", line 3, in safe_import
    module = __import__(module_path, fromlist=[class_name])
  File "/root/xinyuan/sgl-workspace/rllm/rllm/environments/swe/swe.py", line 19, in <module>
    R2EGYM_PATH = os.path.dirname(r2egym.__file__)
AttributeError: 'NoneType' object has no attribute '__file__'. Did you mean: '__le__'?

[JustinTong0323]

After installing r2egym, got warning like:

ANTLR runtime and generated code versions disagree: 4.7.2!=4.9.3
ANTLR runtime and generated code versions disagree: 4.7.2!=4.9.3
---REPEAT MANY TIMES---
ANTLR runtime and generated code versions disagree: 4.7.2!=4.9.3
ANTLR runtime and generated code versions disagree: 4.7.2!=4.9.3

[shivakrishnareddy-ma]

I am facing this same issue, but when I installed r2egym, I get this error:

venv/lib/python3.10/site-packages/modal_proto/options_pb2.py", line 29, in

    google_dot_protobuf_dot_descriptor__pb2.FieldOptions.RegisterExtension(audit_target_attr)

AttributeError: type object 'FieldOptions' has no attribute 'RegisterExtension'

I tried downgrading protobuf, but looks like e2b-code-interpreter and modal have conflicting dependencies.

[shahofblah]

How are you installing r2egym? As a git submodule?
https://github.com/R2E-Gym/R2E-Gym does not seem like it's been published

It doesn't exist in pyproject.toml either

[michaelzhiluo]

You can install r2egym in https://github.com/agentica-project/R2E-Gym

[shivakrishnareddy-ma]

@shahofblah Let us know if you're successful in running the quick start example. Thanks!

[thejishnunair]

Used this, downgraded protobuf, and stubbed the e2b packages in a few places. Training seems to be working now.

[shahofblah]

@shivakrishnareddy-ma I just commented out the swe-agent loads since they were not being used for the math example.

Faced several other config errors though.(some params have to be divisible by other params, etc.)
It would be nice if these could be surfaced/sanity-checked early on and not ±15 minutes after I start running the script