Is your feature request related to a problem? Please describe.
Hello!
How about improving the system to allow the use of other embedding models dynamically, in addition to the predefined models?
|
self.embedding = embedding_models[embedding_model]() |
|
embedding_models = { |
|
# llama index |
|
"openai": LazyInit( |
|
OpenAIEmbedding |
|
), # default model is OpenAIEmbeddingModelType.TEXT_EMBED_ADA_002 |
|
"openai_embed_3_large": LazyInit( |
|
OpenAIEmbedding, model_name=OpenAIEmbeddingModelType.TEXT_EMBED_3_LARGE |
|
), |
|
"openai_embed_3_small": LazyInit( |
|
OpenAIEmbedding, model_name=OpenAIEmbeddingModelType.TEXT_EMBED_3_SMALL |
|
), |
|
"mock": LazyInit(MockEmbeddingRandom, embed_dim=768), |
|
# langchain |
|
"openai_langchain": LazyInit(OpenAIEmbeddings), |
|
} |
|
|
|
try: |
|
# you can use your own model in this way. |
|
from llama_index.embeddings.huggingface import HuggingFaceEmbedding |
|
|
|
embedding_models["huggingface_baai_bge_small"] = LazyInit( |
|
HuggingFaceEmbedding, model_name="BAAI/bge-small-en-v1.5" |
|
) |
|
embedding_models["huggingface_cointegrated_rubert_tiny2"] = LazyInit( |
|
HuggingFaceEmbedding, model_name="cointegrated/rubert-tiny2" |
|
) |
|
embedding_models["huggingface_all_mpnet_base_v2"] = LazyInit( |
|
HuggingFaceEmbedding, |
|
model_name="sentence-transformers/all-mpnet-base-v2", |
|
max_length=512, |
|
) |
|
embedding_models["huggingface_bge_m3"] = LazyInit( |
|
HuggingFaceEmbedding, model_name="BAAI/bge-m3" |
|
) |
|
except ImportError: |
|
logger.info( |
|
"You are using API version of AutoRAG." |
|
"To use local version, run pip install 'AutoRAG[gpu]'" |
|
) |
