mirror of
https://github.com/khoj-ai/khoj.git
synced 2024-11-27 17:35:07 +01:00
989526ae54
- The all-MiniLM-L6-v2 is more accurate
- The exact previous model isn't benchmarked but based on the
performance of the closest model to it. Seems like the new model
maybe similar in speed and size
- On very preliminary evaluation of the model, the new model seems
faster, with pretty decent results
47 lines
No EOL
1.5 KiB
YAML
47 lines
No EOL
1.5 KiB
YAML
content-type:
|
|
# The /data/folder/ prefix to the folders is here because this is
|
|
# the directory to which the local files are copied in the docker-compose.
|
|
# If changing, the docker-compose volumes should also be changed to match.
|
|
org:
|
|
input-files: null
|
|
input-filter: "/data/notes/*.org"
|
|
compressed-jsonl: "/data/embeddings/notes.jsonl.gz"
|
|
embeddings-file: "/data/embeddings/note_embeddings.pt"
|
|
|
|
ledger:
|
|
input-files: null
|
|
input-filter: /data/ledger/*.beancount
|
|
compressed-jsonl: /data/embeddings/transactions.jsonl.gz
|
|
embeddings-file: /data/embeddings/transaction_embeddings.pt
|
|
|
|
image:
|
|
input-directory: "/data/images/"
|
|
embeddings-file: "/data/embeddings/image_embeddings.pt"
|
|
batch-size: 50
|
|
use-xmp-metadata: true
|
|
|
|
music:
|
|
input-files: ["/data/music/music.org"]
|
|
input-filter: null
|
|
compressed-jsonl: "/data/embeddings/songs.jsonl.gz"
|
|
embeddings-file: "/data/embeddings/song_embeddings.pt"
|
|
|
|
search-type:
|
|
symmetric:
|
|
encoder: "sentence-transformers/all-MiniLM-L6-v2"
|
|
cross-encoder: "cross-encoder/ms-marco-MiniLM-L-6-v2"
|
|
model_directory: "/data/models/symmetric"
|
|
|
|
asymmetric:
|
|
encoder: "sentence-transformers/multi-qa-MiniLM-L6-cos-v1"
|
|
cross-encoder: "cross-encoder/ms-marco-MiniLM-L-6-v2"
|
|
model_directory: "/data/models/asymmetric"
|
|
|
|
image:
|
|
encoder: "clip-ViT-B-32"
|
|
model_directory: "/data/models/image_encoder"
|
|
|
|
processor:
|
|
conversation:
|
|
openai-api-key: null
|
|
conversation-logfile: "/data/embeddings/conversation_logs.json" |