AttributeError when using GritLM with HippoRAG

[PascualMeritaTorres]

AttributeError when using GritLM with HippoRAG

Error Description

When running HippoRAG with the GritLM model using the following command:

python main.py --dataset sample --llm_base_url https://api.openai.com/v1 --llm_name gpt-4o-mini --embedding_name GritLM/GritLM-7B

We encounter this error:

AttributeError: 'BaseConfig' object has no attribute 'embedding_return_as_cpu'

Root Cause

The error occurs in the configuration dictionary of GritLLMEmbeddingModel:

config_dict = {
    "embedding_model_name": self.embedding_model_name,
    "return_cpu": self.global_config.embedding_return_as_cpu,         # Error here
    "return_numpy": self.global_config.embedding_return_as_numpy,     # Error here
    "norm": self.global_config.embedding_return_as_normalized,
    "model_init_params": {
        "model_name_or_path": self.embedding_model_name,
        "torch_dtype": "auto",
        "device_map": "auto",  # added this line to use multiple GPUs
        # **kwargs
    },
    "encode_params": {
        "batch_size": self.global_config.embedding_batch_size,
        # "instruction": "",
    },
    "generate_params": {
        # "max_new_tokens": 
        # "do_sample": 
    }
}

The attributes embedding_return_as_cpu and embedding_return_as_numpy are being referenced but were never defined in the BaseConfig class.

Possible Solutions

1. Add these missing attributes to BaseConfig by doing:

embedding_return_as_cpu: bool = field(
        default=True,
        metadata={"help": "Whether to return embeddings on CPU."}
 )
  embedding_return_as_numpy: bool = field(
      default=True,
      metadata={"help": "Whether to return embeddings as numpy arrays."}
  )

3. Remove these settings in the config dictionary

[bernaljg]

Thanks for pointing this out!

We have removed the unnecessary attributes from the GritLM class config to avoid this issue. The commit is 2f05c27.