really cool! maybe this is an old technique but what if we use traditional RAG but add another layer of meta-data entity generation, generating meta-data per semantic chunk which we would than prioritize with filter **before** doing vector search (kind of hybrid-RAG I guess).
In the first example with San Francisco who defined the entity and predicate types (such as 'Location' for entity and 'population' for predicate)? It seems like the definition of these was not automated, but maybe I'm missing something. In your second example it looked like the types were defined by the system, which makes sense because defining it manually would be impossible.
@Mervin: Thanks for a great video. Am I right to assume that this toolkit can theoretically be deployed locally? (as long as your embedding and LLM models are also available)? What would be the vGPU limitation here?
( polish ) Jak uważasz- czy lepiej stworzyć zarówno grafy wiedzy jak i wektory? Czy po prostu grafy wiedzy są o wiele lepsze od wektorów? Istnieje coś oprócz tych dwóch metod?
hey i am getting an error while running "r2r ingest-files test.txt" the error says " raise R2RException( r2r.base.abstractions.exception.R2RException: An error 'Max retries reached. Last error: ' occurred during ingest_files_app "
is there any guide document how to use r2r for this ? it gives error both: r2r --config-name=local_neo4j_kg serve --docker --docker-ext-neo4j Usage: r2r serve [OPTIONS] Try 'r2r serve --help' for help. Error: No such option: --docker-ext-neo4j (Possible options: --docker, --exclude-neo4j) OR r2r --config-name=local_neo4j_kg serve --docker --exclude-neo4j true Traceback (most recent call last): File "/home/redel/Projects/RAG/venvR2R/bin/r2r", line 8, in sys.exit(main()) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/r2r/cli/cli.py", line 44, in main cli() File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1157, in __call__ return self.main(*args, **kwargs) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1078, in main rv = self.invoke(ctx) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1688, in invoke return _process_result(sub_ctx.command.invoke(sub_ctx)) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1434, in invoke return ctx.invoke(self.callback, **ctx.params) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 783, in invoke return __callback(*args, **kwargs) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/decorators.py", line 45, in new_func return f(get_current_context().obj, *args, **kwargs) File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/r2r/cli/commands/server_operations.py", line 205, in serve run_docker_serve( File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/r2r/cli/utils/docker_utils.py", line 104, in run_docker_serve R2RBuilder.CONFIG_OPTIONS[obj["config_name"] or "default"] any fix
I tried it but it needs the entities name specified and I can't find the list of whole entities and predicates named supported. So cannot use to extract triplets for my use case with animal and plants documents. It sure not supported entity name like "plant", "or animal" and so on hence when I do it, it return the empty output. The model does not know what is it. But it's different with Microsoft graphrag because it detects automatically the triplet entity name and predicate name.
File "C:\Users\prajw\AppData\Local\Programs\Python\Python312\Lib\site-packages 2r\main\api\client.py", line 56, in handle_request_error raise R2RException( r2r.base.abstractions.exception.R2RException: An error 'Max retries reached. Last error: ' occurred during ingest_files_app i am getting this error after r2r ingest-files data.txt could you please help
really cool! maybe this is an old technique but what if we use traditional RAG but add another layer of meta-data entity generation, generating meta-data per semantic chunk which we would than prioritize with filter **before** doing vector search (kind of hybrid-RAG I guess).
Someone just shared this in our Discord and we wanted to stop by and say thanks @mervin for taking the time to record this awesome video!
Love this guys content. Truly useful stuff!
In the first example with San Francisco who defined the entity and predicate types (such as 'Location' for entity and 'population' for predicate)? It seems like the definition of these was not automated, but maybe I'm missing something.
In your second example it looked like the types were defined by the system, which makes sense because defining it manually would be impossible.
Awesome. Can’t wait to try this out.
As always wonderful video and very informative. I learn so much from your videos
Great video but the real question is where are the models that generate graph models?
Great video Mervin. Now waiting for your next video!.
May I suggest you go use the knowledge to build something in the meantime.
@Mervin: Thanks for a great video. Am I right to assume that this toolkit can theoretically be deployed locally? (as long as your embedding and LLM models are also available)? What would be the vGPU limitation here?
( polish ) Jak uważasz- czy lepiej stworzyć zarówno grafy wiedzy jak i wektory? Czy po prostu grafy wiedzy są o wiele lepsze od wektorów? Istnieje coś oprócz tych dwóch metod?
Hey, thanks for your work.
Is there a way to extract all the existing entities?
Is Neo4j only for visualization? I thought it was a db.
it s a ghrap db
hey i am getting an error while running "r2r ingest-files test.txt" the error says " raise R2RException(
r2r.base.abstractions.exception.R2RException: An error 'Max retries reached. Last error: ' occurred during ingest_files_app
"
love your vids
j
can you show the application of this for working on large code bases?
is there any guide document how to use r2r for this ? it gives error both:
r2r --config-name=local_neo4j_kg serve --docker --docker-ext-neo4j
Usage: r2r serve [OPTIONS]
Try 'r2r serve --help' for help.
Error: No such option: --docker-ext-neo4j (Possible options: --docker, --exclude-neo4j)
OR
r2r --config-name=local_neo4j_kg serve --docker --exclude-neo4j true
Traceback (most recent call last):
File "/home/redel/Projects/RAG/venvR2R/bin/r2r", line 8, in
sys.exit(main())
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/r2r/cli/cli.py", line 44, in main
cli()
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1157, in __call__
return self.main(*args, **kwargs)
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1078, in main
rv = self.invoke(ctx)
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1688, in invoke
return _process_result(sub_ctx.command.invoke(sub_ctx))
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 1434, in invoke
return ctx.invoke(self.callback, **ctx.params)
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/core.py", line 783, in invoke
return __callback(*args, **kwargs)
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/click/decorators.py", line 45, in new_func
return f(get_current_context().obj, *args, **kwargs)
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/r2r/cli/commands/server_operations.py", line 205, in serve
run_docker_serve(
File "/home/redel/Projects/RAG/venvR2R/lib/python3.10/site-packages/r2r/cli/utils/docker_utils.py", line 104, in run_docker_serve
R2RBuilder.CONFIG_OPTIONS[obj["config_name"] or "default"]
any fix
Would be nice to see it with networkX
do we need to pass entity type and relationship ?
anyone who faced this ??
r2r --config-name=local_neo4j_kg serve --docker --docker-ext-neo4j
Usage: r2r serve [OPTIONS]
Try 'r2r serve --help' for help.
Error: No such option: --docker-ext-neo4j (Possible options: --docker, --exclude-neo4j)
+1
@@love12xfuture +1
I tried it but it needs the entities name specified and I can't find the list of whole entities and predicates named supported. So cannot use to extract triplets for my use case with animal and plants documents. It sure not supported entity name like "plant", "or animal" and so on hence when I do it, it return the empty output. The model does not know what is it. But it's different with Microsoft graphrag because it detects automatically the triplet entity name and predicate name.
did you figure it out or went with a different approach? I am supposing SciPhi triplex seems to be open source, could it not be tweaked?
Couldn’t we just use gpt-4o mini? That should reduce the cost by 50-60%?
But if GPT-4o sucks at it GPT-4o mini would really suck lol
Does it work with any languages?
You might need to fine tune further for it to work for languages
After actually using it, the performance is significantly lower compared to Microsoft's, and it is practically useless for non-English documents.
File "C:\Users\prajw\AppData\Local\Programs\Python\Python312\Lib\site-packages
2r\main\api\client.py", line 56, in handle_request_error
raise R2RException(
r2r.base.abstractions.exception.R2RException: An error 'Max retries reached. Last error: ' occurred during ingest_files_app
i am getting this error after r2r ingest-files data.txt
could you please help
+1