Outstanding! An absolute masterclass. Thank you so much Mervin. Personally, I think that this is one of your best videos to-date. I watched enthralled!
Best 15 mins of my day. Excited to try the agentic chunker. Would love to hear your thoughts on what if the source contents contain a mix of texts and lot of tables.
I applied the semantic approach for topic modelling approach and it's produce a better topicallity a better titling propositions when we use hf LLM with. I will try the last method for the db creation Thanks Mille
Thanks for an amazing, educational and insightful video. One comment please: It seems you're skipping the "Chunk #1" along the process, which has the "This capability" in its beginning, so it's not categorized correctly to the first chunk. This seems to be caused by an error of the LLM, not un-contextualizing that sentence correctly as it's instructed by the prompt. Am I right? Any idea why it happened? Thanks again for a great video!
Nice tutorials Great stuff. How do we store the json data as chucking is actually not doing great job with key values instead thinking random brackets and so on? Do you have a solution ?
Thank you very much! This video is helpful for those interested in building RAG apps. I'm not sure if it can also be helpful with chunking legal or contract documents in PDF format.
So once youve used this tactic to create a vector database with all the stuff inside it and save it to a chroma db, how would you use an llm to query the database afterwards ? It seems the chunking is part of the retrival process aswell based on how you presented this, running this script everytime doesnt seem that great
Thank you so much for the video!! When using models like "all-mpnet-base-v2" from the Sentence Transformers library to generate embeddings and store them in Vector DB, which are based on architectures like BERT, the maximum token limit per input sequence is typically 512 tokens. As long as my input text does not exceed the maximum input sequence which is 512 tokens then we don't need to deal with different chunking strategies covered in this video. I meant chunking strategies are meant for long input text (1000s of words), is that correct understanding?
You're right, chunking is only relevant for longer input. Though, I'm not sure if it would be correct to say that it's ONLY relevant when the context length of your LLM is exceeded or if a RAG makes sense nonetheless.
@@ilianosYeah, we will be retrieving the top k closest vectors, ranking them if necessary then fed to LLM so that we will not exceed the context window of our LLM.
@@MervinPraisonI installed langchain-experimental and it is listed when I checked it with pip list. However, when I tried to import it, it could not be resolved. I tried to rerun the terminal, conda, reopen the folder, but nothing seems to work.
Good one Mervin. Shouldn't you use chunks instead of documents in this snippet? # RAG def rag(chunks, collection_name): vectorstore = Chroma.from_documents( documents=documents, #
awesome as always! i have a question, can we use any other llm for agentic chunking, can you suggest any open source free alternative to gpt3.5? I have the same setup with local ollama mistral, can i again use that for agentic chunkling?
@@jawadmansoor6064 If you mean 'expensive' as a computing power - you're right. About expensive in money - maybe it can be implemented using one of the great free available LLMs? What do you say, dear @MervinParison ?
This is so mind-blowing!! Is there any option for coder beginners? I'd be happy to pay a one-time subscription to get a software that helps me organize my data this way. Especially if the rag system is able to get the references from the retrieved info. That'd be awesome for a deeper understanding of multiple (not hundreds, but thousands) papers in academic reaearch. Thanks for the video!
I am a fan of preparing data beforehand and semantic chunking - this random 1k token chunking methods does not work for me. In medical field you need at least 95% accuracy:)
Coz the problem I’m facing is that the data is very scattered… like there is a webpage(1000url) that contains everything. But looking simply at one url and do the chunk won’t yield valuable information for query. So imagine I will have to access some parts of 5-6 url to create an useful document for retrieval. I think it’s because I don’t have a structured and organised documents so I resort to manual chunking. I know it’s very not efficient so I’m looking for ways to solve this problem.
You are correct. Thanks for the feedback First I recorded the whole tutorial and at the end realised I forgot to press the record button 🤦♂️ Then out of frustration I recorded the second time
![RAG Chunking Strategies [Top 11] | Semantic Chunking to LLM Chunking | Learn RAG from Scratch](http://i.ytimg.com/vi/-hsaiU6fmtQ/mqdefault.jpg)
This is a shorter version of Greg's video, thanks for sharing!
Outstanding! An absolute masterclass. Thank you so much Mervin. Personally, I think that this is one of your best videos to-date. I watched enthralled!
Thank you 😊
@@MervinPraison How can we use the agentic chunking with groq api, i tried to replace there in the code but it is throwing error
Thank you for the video!
I recently built my first RAG application and chunking + retrieval was the main challenge.
🙏
this is best high quality channel and videos no intor/no long stuff just to point thank you very much
Thank you
This is an absolutely brilliant video Mervin. Exactly what I needed and implemented and explained brilliantly.
Thank you
Really a good job Mr Mervin !!!!!! we really appreciate your effort for this valuable knwoledge
Thank you
That's really excellent! I watched Greg's video last week and have been thinking since about implementing it. Thank you so much!
Thank you
I am glad you put all these on open ,Thanks ,I subscribed instantly Underrated channel I must say
Thank you
That’s what I m looking for ! Thanks so much again , Mervin 🙏🏻
Thank you
Best 15 mins of my day. Excited to try the agentic chunker. Would love to hear your thoughts on what if the source contents contain a mix of texts and lot of tables.
Im loving your videos!
Great video. Short and practical. All the best.
Thanks a lot for your work, it is very interesting and you are very clear in your explanation
Thank you
This is very nice of you to make. Thank you.
Thanks
Fantastic ideas and breakdown! Thank you!
Well done, it expleained briefly. Thanks
Thank you
Superb!!!
Really amazing...Thanks a lot
I applied the semantic approach for topic modelling approach and it's produce a better topicallity a better titling propositions when we use hf LLM with.
I will try the last method for the db creation
Thanks Mille
Great!
You did what I recommended! I appreciate you 🎉
Thank you
hello @Mervin praison Sir
what we can do if we want to use it in free tier
Thanks a lot for the video! I have a question? how can use the same code but directly to a pdf file and not a string in the code?
thanks a lot
Quality Content, Thanks
Thanks for an amazing, educational and insightful video.
One comment please: It seems you're skipping the "Chunk #1" along the process, which has the "This capability" in its beginning, so it's not categorized correctly to the first chunk. This seems to be caused by an error of the LLM, not un-contextualizing that sentence correctly as it's instructed by the prompt.
Am I right?
Any idea why it happened?
Thanks again for a great video!
Excellent, bravo
Thank you
That is great! Mervin.
Thank you
Excellent Video! Thanks for sharing!
What would be the best chunking strategy for a large code base?
Thank you! Any IOT use cases with CrewAI?
I will look in to this
Thanks for the video, trying this out
👍
Great video
Thank you for the informative video. How do you handle pictures or plots in documents ? Is there a special way to handle them?
u got anything on that?
@@MrAhsan99 no, but ollama has a model for that, Llava. It does ok. I guess I would need to fine tune a model for better results
@@Hoxle-87 Thanks for sharing this
Really great stuff
❤ , excellent tutorial/infos . thanks bro
Thank you
Nice tutorials
Great stuff. How do we store the json data as chucking is actually not doing great job with key values instead thinking random brackets and so on? Do you have a solution ?
really great video and well explained
Amazing Video !!
Thank you very much! This video is helpful for those interested in building RAG apps. I'm not sure if it can also be helpful with chunking legal or contract documents in PDF format.
Yes , you need to extend this document chunking code to include PDF docs .
How to increase this code for PDF?
Great explanation... Can you make a video on how to chunk a large json file.
does this is for all cases ,or for this specific case
😮Thank you so much!
love with this video
very usefull sir
So the best technique is to use agentic chunking with grouping ?
Yes, but it also costs a bit more compared to others
@@MervinPraison so which next best for free
Great video!
Thank you
can we try agentic chunking with ollama?
you are excellent! thank u
Thank you
Great content ❤
Thank you
So once youve used this tactic to create a vector database with all the stuff inside it and save it to a chroma db, how would you use an llm to query the database afterwards ? It seems the chunking is part of the retrival process aswell based on how you presented this, running this script everytime doesnt seem that great
Amazing one
Thank you so much for the video!!
When using models like "all-mpnet-base-v2" from the Sentence Transformers library to generate embeddings and store them in Vector DB, which are based on architectures like BERT, the maximum token limit per input sequence is typically 512 tokens.
As long as my input text does not exceed the maximum input sequence which is 512 tokens then we don't need to deal with different chunking strategies covered in this video. I meant chunking strategies are meant for long input text (1000s of words), is that correct understanding?
You're right, chunking is only relevant for longer input. Though, I'm not sure if it would be correct to say that it's ONLY relevant when the context length of your LLM is exceeded or if a RAG makes sense nonetheless.
@@ilianosYeah, we will be retrieving the top k closest vectors, ranking them if necessary then fed to LLM so that we will not exceed the context window of our LLM.
Are you from South Africa?
value is not a valid list (type=type_error.list) I get this error
Is it possible to use nomic-embed-text with autogen ?
Yes you can
Can I use Gemini embedding instead of openai?
Yes. But you might need to modify the code accordingly.
@@MervinPraisonI installed langchain-experimental and it is listed when I checked it with pip list. However, when I tried to import it, it could not be resolved. I tried to rerun the terminal, conda, reopen the folder, but nothing seems to work.
Good one Mervin. Shouldn't you use chunks instead of documents in this snippet?
# RAG
def rag(chunks, collection_name):
vectorstore = Chroma.from_documents(
documents=documents, #
This is amazing
awesome as always!
i have a question, can we use any other llm for agentic chunking, can you suggest any open source free alternative to gpt3.5? I have the same setup with local ollama mistral, can i again use that for agentic chunkling?
Nexus Raven is the closest. You need funtion calling to do that task . But you might need to modify the agentic chunker.py code
Thakyou so much sir
hello Mervin Praison，
Your coding environment is my favorite style.
colud you tell me how to setup in ubuntu?
thank you.
It’s just VS code
Very nice, how about achieving the same using open source LLMs?
agentic chunking, good idea, but you can top that apply vactorization of these chunks i.e. apply previous (semantic chunking) now.
BTW, agentic chunking is extremely expensive, though on MTEB only LLMs stand atop.
Yes you can again implement semantic back again . Yes it’s expensive
@@jawadmansoor6064 If you mean 'expensive' as a computing power - you're right. About expensive in money - maybe it can be implemented using one of the great free available LLMs?
What do you say, dear @MervinParison ?
Thank you for the video. That's amazing. Could you please give us the name of the tool that makes autocomplete for your code ? Amazing content :D
I use GitHub copilot and continue interchangeably
this is good sh..t.. thank you for your public service !
This is so mind-blowing!! Is there any option for coder beginners? I'd be happy to pay a one-time subscription to get a software that helps me organize my data this way.
Especially if the rag system is able to get the references from the retrieved info. That'd be awesome for a deeper understanding of multiple (not hundreds, but thousands) papers in academic reaearch.
Thanks for the video!
I like it 👍
Thank you
1- How are you handling the tables and diagrams in the documents?
2- You are loading a text document, How to do this on a pdf file?
Good explanation. But it is impossible to read line 185 and by consequence, impossible to test the best snippet that you are soposed to teaching
I am a fan of preparing data beforehand and semantic chunking - this random 1k token chunking methods does not work for me. In medical field you need at least 95% accuracy:)
True. Preparing data beforehand would be more ideal
Yep data prepping is key especially if you're extracting data from pdfs and such!
Hello! How do we limit the output tokens in each document to say 512 tokens in semantic chunking. I am trying to implement it but it fails.
Hey what do you mean by preparing data beforehand here can you explain me
Hello
Thanks. What do you think about chunking manually? Human brain understands the context more accurately ..
Haha Good idea . But what if you have 1 million pages to chunk? And you have 1 day to do it
Coz the problem I’m facing is that the data is very scattered… like there is a webpage(1000url) that contains everything. But looking simply at one url and do the chunk won’t yield valuable information for query.
So imagine I will have to access some parts of 5-6 url to create an useful document for retrieval. I think it’s because I don’t have a structured and organised documents so I resort to manual chunking. I know it’s very not efficient so I’m looking for ways to solve this problem.
Little slower might be better?
You are correct. Thanks for the feedback
First I recorded the whole tutorial and at the end realised I forgot to press the record button 🤦♂️
Then out of frustration I recorded the second time
ok ok chunking ..
Too fast to understand 😢
Sorry for that, yes I did speed up little this time. Will try to be normal next time
Great video.
Thank you