docs: Added tool use examples.

README.md

@@ -124,6 +124,9 @@ See `docs/components.md` for more details on the API.
     - Open-Ended Environment: `examples/livecodebench/open_ended_environment.py`
     - Skill-List Environment: `examples/livecodebench/skill_list_environment.py`
     - Skill-Tree Environment: `examples/livecodebench/skill_tree_environment.py`
+- MnMs (Tool Use)
+    - Open-Ended Environment: `examples/tool_use/open_ended.py`
+    - Skill-List Environment: `examples/tool_use/skill_list.py`
 
 ## Running Examples
 Run the examples from the repository root. Set `CUDA_VISIBLE_DEVICES` to the GPUs you want to use and make sure to set `num_gpus` in `ray.init()` to the number of GPUs you have available.
@@ -140,6 +143,8 @@ This repository uses code from the following repositories:
 - [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory)
 - [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness)
 - [LiveCodeBench](https://github.com/LiveCodeBench/LiveCodeBench)
+- [MnMs](https://github.com/RAIVNLab/mnms)
+- [NaturalBench](https://github.com/Baiqi-Li/NaturalBench)
 
 We thank the authors of the above repositories for making their code freely available.
 

examples/tool_use/open_ended.py

@@ -0,0 +1,67 @@
+import warnings
+from pathlib import Path
+
+from dataenvgym.gym.accumulate_train_loop import (
+    IoProvider,
+    run_generic_accumulation_train_loop,
+)
+from dataenvgym.gym.data_generation_agents.tool_use.baselines.open_ended import (
+    DataGenerationAgent,
+)
+from dataenvgym.gym.tasks.tool_use.mnms.task import MnmsSplit, MnmsTask
+from dataenvgym.gym.trainable_predictors.tool_use.vllm_predictor import (
+    LLAMA3_8B_INSTRUCT_INFERENCE_CONFIG,
+    LLAMA3_8B_INSTRUCT_TRAINER_CONFIG,
+    CodeGenerationTrainablePredictor,
+    ParallelVllmCodeGenerationPredictor,
+    SftCodeGenerationTrainer,
+)
+from sklearn.exceptions import UndefinedMetricWarning
+
+warnings.filterwarnings("ignore", category=UndefinedMetricWarning)
+
+
+def main():
+    num_gpus = 8
+    experiment_dir = Path("workspace/tool_use_open_ended_example")
+    experiment_dir.mkdir(parents=True, exist_ok=True)
+
+    predictor_config = LLAMA3_8B_INSTRUCT_INFERENCE_CONFIG.set_with_gpu_count(num_gpus)
+    predictor = ParallelVllmCodeGenerationPredictor(config=predictor_config)
+    trainer_config = LLAMA3_8B_INSTRUCT_TRAINER_CONFIG.set_with_gpu_count(
+        num_gpus
+    ).set_working_directory(experiment_dir)
+    trainer_config.overrides = ["cutoff_len=1600"]
+    trainer = SftCodeGenerationTrainer(
+        config=trainer_config,
+    )
+
+    trainable_predictor = CodeGenerationTrainablePredictor(
+        predictor=predictor,
+        trainer=trainer,
+    )
+
+    val_task = MnmsTask(split=MnmsSplit.VAL)
+    test_task = MnmsTask(split=MnmsSplit.TEST)
+
+    data_strategy = DataGenerationAgent(
+        datum_to_generate_per_error=2,
+        logging_folder=experiment_dir / "data_strategy_outputs",
+        data_generation_per_invocation_limit=60,
+    )
+
+    io_provider = IoProvider(experiment_dir=experiment_dir)
+
+    run_generic_accumulation_train_loop(
+        validation_tasks=[val_task],
+        test_tasks=[test_task],
+        trainable_predictor=trainable_predictor,
+        training_data_production_strategy=data_strategy,
+        io_provider=io_provider,
+        num_cycles=10,
+        accumulation_iterations_per_cycle=1,
+    )
+
+
+if __name__ == "__main__":
+    main()

examples/tool_use/skill_list.py

@@ -0,0 +1,68 @@
+import warnings
+from pathlib import Path
+
+from dataenvgym.gym.accumulate_train_loop import (
+    IoProvider,
+    run_generic_accumulation_train_loop,
+)
+from dataenvgym.gym.data_generation_agents.tool_use.baselines.skill_list import (
+    DataGenerationAgent,
+)
+from dataenvgym.gym.tasks.tool_use.mnms.task import MnmsSplit, MnmsTask
+from dataenvgym.gym.trainable_predictors.tool_use.vllm_predictor import (
+    LLAMA3_8B_INSTRUCT_INFERENCE_CONFIG,
+    LLAMA3_8B_INSTRUCT_TRAINER_CONFIG,
+    CodeGenerationTrainablePredictor,
+    ParallelVllmCodeGenerationPredictor,
+    SftCodeGenerationTrainer,
+)
+from sklearn.exceptions import UndefinedMetricWarning
+
+warnings.filterwarnings("ignore", category=UndefinedMetricWarning)
+
+
+def main():
+    num_gpus = 8
+    experiment_dir = Path("workspace/tool_use_skill_list_example")
+    experiment_dir.mkdir(parents=True, exist_ok=True)
+
+    predictor_config = LLAMA3_8B_INSTRUCT_INFERENCE_CONFIG.set_with_gpu_count(num_gpus)
+    predictor = ParallelVllmCodeGenerationPredictor(config=predictor_config)
+    trainer_config = LLAMA3_8B_INSTRUCT_TRAINER_CONFIG.set_with_gpu_count(
+        num_gpus
+    ).set_working_directory(experiment_dir)
+    trainer_config.overrides = ["cutoff_len=1600"]
+    trainer = SftCodeGenerationTrainer(
+        config=trainer_config,
+    )
+
+    trainable_predictor = CodeGenerationTrainablePredictor(
+        predictor=predictor,
+        trainer=trainer,
+    )
+
+    val_task = MnmsTask(split=MnmsSplit.VAL)
+    test_task = MnmsTask(split=MnmsSplit.TEST)
+
+    data_strategy = DataGenerationAgent(
+        datum_to_generate_per_skill=2,
+        max_skills_to_use_during_generation=30,
+        logging_folder=experiment_dir / "data_strategy_outputs",
+        sampling_method="randomly_sample_imperfect_skills",
+    )
+
+    io_provider = IoProvider(experiment_dir=experiment_dir)
+
+    run_generic_accumulation_train_loop(
+        validation_tasks=[val_task],
+        test_tasks=[test_task],
+        trainable_predictor=trainable_predictor,
+        training_data_production_strategy=data_strategy,
+        io_provider=io_provider,
+        num_cycles=10,
+        accumulation_iterations_per_cycle=1,
+    )
+
+
+if __name__ == "__main__":
+    main()