RAG (Retrieval Augmented Generation) is a method to combine a search
engine with a LLM (Large Language Model). When a new prompt is
submitted, a search engine injects knowledge from a search into the
content. This is done using a reverse proxy between the Chat Client and
the LLM. In this case, we used the following software:
LLM Backend - Ollama:
https://github.com/ollama/ollama
Install ollama and then load two required LLM models
with the following commands:
ollama pull phi3:3.8b
ollama pull llama3:8b
Chat Client - susi_chat:
https://github.com/susiai/susi_chat
just clone the repository and the open the file
susi_chat/chat_terminal/index.html
in your browser. This displays a chat terminal.
In this terminal, run the following command:
host http://localhost:8090
This sets the LLM backend to your YaCy peer.
Then start YaCy. It will provide the LLM endpoint to the client
while using ollama in the backend. It then injects search results
only from the local Solr index, not from the p2p network (so far).
to //download.yacy.net/ in network.*.unit files
@Orbiter for the latest avail. releases (v1.924 ...tar.gz) the *.tar.gz.sig file is missing,
so download fails with error "Download of releas .... failed"
a main problem when crawling is long waiting time cuased by crawl-delay
values from robots.txt entries. that attribute is not supported by
google and interpreted by yandex and bing in different ways. In large
crawls there is always one host which blocks the whole crawl with
extreme large values. YaCy now still obeys crawl-delay but limits them
to 10 seconds.
Additionally the blocking logic when loading new robots.txt was analyzed
and a deadlock was removed. Furthermore the construction of new queue
lists was redesigned and it was ensured that always a large list of
different hosts for host-balancing is provided for the loader.
We will use the default value for now on.
This is much better for resource economy and fits better into a
container/docker/kubernetes strategy.
Furthermore, a small memory footprint is essential for the usage on
small devices like RaspberryPi.
- replace new guava 30 with older 25 because that is the correct
dependency for solr 8.8.1. The newer one did actually not work!
- index will be crated in a DATA/INDEX/freeworld/SEGMENTS/solr_8_8_1
subfolder. The older solr_6_6 index is not touched but also not
migrated. The index starts with fresh (empty) content.
- Older indexes must be migrated by hand (export/import) so far until a
better solution is found.
- Large schema adoptions for lucene 8.8.1
variables
To use that feature, set an environment variable with prefix "yacy." and
suffix identical to the yacy configuration attribute name.
Additionaly we implemented a way to set a peer name using the setting
"network.unit.agent". This can therefore now be used to set a peer name
with the java call parameter
-Dyacy.network.unit.agent=anonymous
The purpose for this feature is the ability to set peer names in
mass-deployed kubernetes clusters to the same name to prevent that we
are flooding peer name statistics with auto-deployment-generated names.
Acces rate limitations to this search mode by unauthenticated users are
set low by default to prevent unwanted server overload but can be
customized through the SearchAccessRate_p.html configuration page
Fixes#291
Michael Peter Christen