MNLP_M3_document_encoder_old

This repository hosts a document encoder model intended for the MNLP M3 project.

Important Update Regarding Model Compatibility

Reason for the recent repository update:

Initially, this repository contained a model that was pushed as a standard Hugging Face transformers model. When attempting to load this model using the sentence-transformers library, I encountered an error indicating that it was not recognized as a sentence-transformers compatible model.

This issue arose because the sentence-transformers library expects specific configuration files (such as modules.json and sentence_bert_config.json) within the model's directory. These files are crucial for defining the model's specific architecture, including how its token embeddings are converted into a single sentence or document embedding (e.g., via mean pooling, CLS token pooling, etc.). Without these configurations, sentence-transformers cannot properly interpret the model's structure.

The intended model for this repository has always been the pre-trained sentence-transformers/all-MiniLM-L12-v2 model. The previous upload method inadvertently created a transformers model structure that lacked the necessary sentence-transformers-specific configurations for direct loading.

What has been changed:

To rectify this, I have re-uploaded the original sentence-transformers/all-MiniLM-L12-v2 model directly to a new MNLP_M3_document_encoder repository. This was done using the sentence-transformers library's save_to_hub function, which ensures that all the correct and required sentence-transformers configuration files are present.

Impact of the change:

Please note that the underlying model weights and architecture have not changed from the originally intended sentence-transformers/all-MiniLM-L12-v2 model. The update purely addresses the repository's internal file structure to ensure proper loading and seamless compatibility with the sentence-transformers library.

You can verify the previous model content by checking the commit history of this repository.

I sincerely apologize for this initial oversight and any confusion it may have caused.