highperfocused/danswer
1
0
Fork 0
mirror of https://github.com/danswer-ai/danswer.git synced 2025-09-26 11:58:28 +02:00
Code Issues Packages Projects Releases Wiki Activity

Fix multilingual .env embedding dimension (#1976)

Browse Source
This commit is contained in:
Bart Schuller
2024-09-08 19:25:07 +02:00
committed by GitHub
parent 1e4b27185d
commit 6012a7cbd9
1 changed files with 6 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
deployment/docker_compose/env.multilingual.template
View File

@@ -1,33 +1,8 @@
# This env template shows how to configure Danswer for multilingual use
# In this case, it is configured for French and English
# To use it, copy it to .env in the docker_compose directory.
# Feel free to combine it with the other templates to suit your needs
# This env template shows how to configure Danswer for custom multilingual use
# Note that for most use cases it will be enough to configure Danswer multilingual purely through the UI
# See "Search Settings" -> "Advanced" for UI options.
# To use it, copy it to .env in the docker_compose directory (or the equivalent environment settings file for your deployment)
# Rephrase the user query in specified languages using LLM, use comma separated values
MULTILINGUAL_QUERY_EXPANSION="English, French"
# Change the below to suit your specific needs, can be more explicit about the language of the response
LANGUAGE_HINT="IMPORTANT: Respond in the same language as my query!"
# The following is included with the user prompt. Here's one example but feel free to customize it to your needs:
LANGUAGE_HINT="IMPORTANT: ALWAYS RESPOND IN FRENCH! Even if the documents and the user query are in English, your response must be in French."
LANGUAGE_CHAT_NAMING_HINT="The name of the conversation must be in the same language as the user query."
# A recent MIT license multilingual model: https://huggingface.co/intfloat/multilingual-e5-small
DOCUMENT_ENCODER_MODEL="intfloat/multilingual-e5-small"
# The model above is trained with the following prefix for queries and passages to improve retrieval
# by letting the model know which of the two type is currently being embedded
ASYM_QUERY_PREFIX="query: "
ASYM_PASSAGE_PREFIX="passage: "
# Depends model by model, the one shown above is tuned with this as True
NORMALIZE_EMBEDDINGS="True"
# Use LLM to determine if chunks are relevant to the query
# May not work well for languages that do not have much training data in the LLM training set
# If using a common language like Spanish, French, Chinese, etc. this can be kept turned on
DISABLE_LLM_DOC_RELEVANCE="True"
# Enables fine-grained embeddings for better retrieval
# At the cost of indexing speed (~5x slower), query time is same speed
# Since reranking is turned off and multilingual retrieval is generally harder
# it is advised to turn this one on
ENABLE_MULTIPASS_INDEXING="True"