SystemSh0cker/private-gpt
1
0
Fork 0
mirror of https://github.com/zylon-ai/private-gpt.git synced 2025-12-22 20:12:55 +01:00
Code Issues Projects Releases Packages Activity

added max_length for embedding models. Some models crash if not provided, see https://huggingface.co/T-Systems-onsite/cross-en-de-roberta-sentence-transformer/discussions/7#65ce1ca0102df4e9e00ff823

Browse source
This commit is contained in:
Robert Hirsch 2024-06-06 20:56:09 +02:00
parent 94712824d6
commit 757a8c79fd
No known key found for this signature in database
GPG key ID: A9D9D1205DBED12C
3 changed files with 6 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
private_gpt/components/embedding/embedding_component.py
View file

@ -31,6 +31,7 @@ class EmbeddingComponent:
self.embedding_model = HuggingFaceEmbedding( self.embedding_model = HuggingFaceEmbedding(
model_name=settings.huggingface.embedding_hf_model_name, model_name=settings.huggingface.embedding_hf_model_name,
cache_folder=str(models_cache_path), cache_folder=str(models_cache_path),
max_length=settings.huggingface.embedding_hf_max_length,
) )
case "sagemaker": case "sagemaker":
try: try:

4
private_gpt/settings/settings.py
View file

@ -151,6 +151,10 @@ class HuggingFaceSettings(BaseModel):
embedding_hf_model_name: str = Field( embedding_hf_model_name: str = Field(
description="Name of the HuggingFace model to use for embeddings" description="Name of the HuggingFace model to use for embeddings"
) )
embedding_hf_max_length: int = Field(
None,
description="Some embedding models have a maximum length for input, provide here for not crashing"
)
access_token: str = Field( access_token: str = Field(
None, None,
description="Huggingface access token, required to download some models", description="Huggingface access token, required to download some models",

1
settings.yaml
View file

@ -69,6 +69,7 @@ embedding:
huggingface: huggingface:
embedding_hf_model_name: BAAI/bge-small-en-v1.5 embedding_hf_model_name: BAAI/bge-small-en-v1.5
embedding_hf_max_length: 512 # some models have a maximum length for input
access_token: ${HUGGINGFACE_TOKEN:} access_token: ${HUGGINGFACE_TOKEN:}
vectorstore: vectorstore: