Option 'eps' is not accepted for the example run command in README.md + Incompabilities in dependencies

[emirceyani]

Dear authors,

Thanks for sharing the codebase. I've tried to follow the installation instructions on a Linux x64 server; however, there are some incompatibilities related to the Numpy version and sentencepiece. That is, just using pip install -r requirements.txt or conda command to install the environment did not work for me.
Under a Python 3.10, had to install all dependencies by hand and had to downgrade numpy to 1.26.4 (not mentioned).

After some hassles, I could manage to resolve import issues and run the code as instructed. Here's the output i get for the example command (no change in provided YAML files):

python train.py --config=configs/pllm_phi.yaml

/home/ceyani/anaconda3/envs/gollum/lib/python3.10/site-packages/rxnfp/transformer_fingerprints.py:8: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  import pkg_resources
usage: train.py [--config CONFIG] [--seed SEED] [--benchmark BENCHMARK]
                [--n_iters N_ITERS] [--group GROUP]
                [--data CONFIG | CLASS_PATH_OR_NAME | .INIT_ARG_NAME VALUE]
                [--surrogate_model CONFIG | CLASS_PATH_OR_NAME | .INIT_ARG_NAME VALUE]
                [--acquisition CONFIG | CLASS_PATH_OR_NAME | .INIT_ARG_NAME VALUE]
                [--bo CONFIG | CLASS_PATH_OR_NAME | .INIT_ARG_NAME VALUE]
tip: For details of accepted options run: train.py --help
error: Parser key "surrogate_model":
  Does not validate against any of the Union subtypes
  Subtypes: [<class 'NoneType'>, <class 'gollum.surrogate_models.gp.SurrogateModel'>]
  Errors:
    - Expected a <class 'NoneType'>
    - Problem with given class_path 'gollum.surrogate_models.gp.GP':
        Parser key "covar_module":
          Does not validate against any of the Union subtypes
          Subtypes: [<class 'NoneType'>, <class 'gpytorch.module.Module'>]
          Errors:
            - Expected a <class 'NoneType'>
            - Problem with given class_path 'gpytorch.kernels.ScaleKernel':
                Parser key "base_kernel":
                  Problem with given class_path 'gpytorch.kernels.MaternKernel':
                    Option 'eps' is not accepted
          Given value type: <class 'dict'>
          Given value: {'class_path': 'gpytorch.kernels.ScaleKernel', 'init_args': {'base_kernel': {'class_path': 'gpytorch.kernels.MaternKernel', 'init_args': {'nu': 2.5, 'ard_num_dims': None, 'batch_shape': None, 'active_dims': None, 'lengthscale_prior': None, 'lengthscale_constraint': None, 'eps': 1e-06}}, 'outputscale_prior': None, 'outputscale_constraint': None, 'ard_num_dims': None, 'batch_shape': None, 'active_dims': None, 'lengthscale_prior': None, 'lengthscale_constraint': None, 'eps': 1e-06}}
  Given value type: <class 'dict'>
  Given value: {'class_path': 'gollum.surrogate_models.gp.GP', 'init_args': {'likelihood': {'class_path': 'gpytorch.likelihoods.GaussianLikelihood', 'init_args': {'noise_prior': None, 'noise_constraint': None, 'batch_shape': {'class_path': 'torch.Size'}}}, 'covar_module': {'class_path': 'gpytorch.kernels.ScaleKernel', 'init_args': {'base_kernel': {'class_path': 'gpytorch.kernels.MaternKernel', 'init_args': {'nu': 2.5, 'ard_num_dims': None, 'batch_shape': None, 'active_dims': None, 'lengthscale_prior': None, 'lengthscale_constraint': None, 'eps': 1e-06}}, 'outputscale_prior': None, 'outputscale_constraint': None, 'ard_num_dims': None, 'batch_shape': None, 'active_dims': None, 'lengthscale_prior': None, 'lengthscale_constraint': None, 'eps': 1e-06}}, 'mean_module': {'class_path': 'gpytorch.means.ConstantMean'}, 'standardize': True, 'normalize': False, 'initial_noise_val': 1.0, 'noise_constraint': 0.0001, 'initial_outputscale_val': 1.0, 'initial_lengthscale_val': 1.0}}

I am also leaving my conda environment for reproducing of error env.yml.

I appreciate your suppor, alongside with renewed environment and much clearer installation instructions as there are multiple conflicts within the packages due to the use of FastFood kernel and sentencepiece package.

Thanks again for open-sourcing the code base. This project opens interesting questions in LLM+BO domain!