bge-reranker-v2-m3 training details

[sreenija07]

Could you please provide details about the process used to train the reranker model available at https://huggingface.co/BAAI/bge-reranker-v2-m3? Specifically, I'm interested in the pipeline employed for training, the dataset utilized, as well as information on the learning rate, base model, and training group size.

[545999961]

The training pipeline is the same as the https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/reranker The dataset utilized is the same as BAAI/bge-reranker-v2-gemma and BAAI/bge-reranker-v2-minicpm-layerwise (see https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/llm_reranker) The learning rate is 4e-5, the base model is BAAI/bge-m3 (https://huggingface.co/BAAI/bge-m3) And other settings are simliar to BAAI/bge-reranker-v2-gemma and BAAI/bge-reranker-v2-minicpm-layerwise (see https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/llm_reranker)

[passionate11]

Could you please provide details about the process used to train the reranker model available at https://huggingface.co/BAAI/bge-reranker-v2-minicpm-layerwise? I'm mostly interested in how the model is trained