This channel is a real gem. Excellent teaching style, and the tutorials are concise yet filled with a treasure trove of information and application. Thank you for making these, and I hope you will continue to explore AI with us. Will definitely look forward for your first masterclass.
Really helpful your videos ❤!! I would like to add in min 11:10, if you are getting this warning 'LangChainDeprecationWarning: The function `__call__` was deprecated in LangChain 0.1.7 and will be removed in 0.2.0. Use invoke instead. ' change the line --> response = chat(messages) by --> response = chat.invoke(messages) :)
This is awesome thank! I wonder how to add recent announced function calling functionality in our project, since it makes much more easier to work with structured json objects.
Excellent instruction. How would you approach getting SEO keywords from a blog article using LangChain? Perhaps make a video about how that would be done.
@@codingcrashcourses8533 I believe there a libraries that mimick a browser to first render the page then get the page content. I kinda have that figured out in my mind but I'm just not sure how one would architect the flow using LangChain. Do I need to use Agents, the new OpenAI function calling feature, or something else which Chaining could handle alone. I'm too new to LangChain to wrap my head around a basic programmatic workflow for this use case using LangChain.
Amazing video! I have one question. In the Index part of the video, you go ahead and create the prompt_template with two arguments, 'question' and 'context'. But then, you didn't specify the 'question' and 'context values', so I think your query is not taking into consideration your template. To verify it, I added in the prompt 'In all your answers, start them by saying "Dear Pedro" ' and it didn't follow the instructions. I'm looking for ways in which to combine PromptTemplate and Embeddings, to be able to give the model a prompt and at the same time using data from .txt files
The High Level chains like RetrievelQAChain handles that for you. Context is used under the hood, but you as "user" of that chain only have to use "question". Forget to provide "context" in your template and you will see an error. For some chains it´s a little bit tricky to pass the template to the RetrievelChain.
Yes, a version is planned. I´ll be honest: I will release a Udemy couse and release the new crashcourse at the same time to get some traffic from the updated Crashcourse. Will be finished probably in ~6-7 weeks.
Bomben Video! What about a locally running opensource LLM? Would be a Gamechanger for German companies especially in finance since the burdens and regulations are very strict
I tried multiple open source LLMs. I have not seen a single one that comes closes to what OpenAI offers. The problem is that GPT performance is expected when you introduce a service with an LLM. I work in finance in germany and our solution is currently not to use critical information from users and use a private endpoint on azure.
Ich bin prozessmanager (nicht IT) bei einem deutschen Versicherer und bin genau an diesem Thema dran. Azure kam ich irgendwie nicht zurecht. Würde mich zu gern mit dir darüber austauschen - oder du machst dazu auch ein Tutorial😃
@@monkeyy992 Azure müsste dein Unternehmen schon im größeren Stil einbinden. Bei diesem Thema passiert auf Infrastrukturseite bei uns sehr viel, wo ich allerdings nicht genau weiß, was alles geprüft wird. Firewalls und der Ganze Kram eben. Groß austauschen kann ich mich da nicht, mehr als das was ich gesagt habe, kann ich da nicht als Info weitergeben. Es gibt noch einen deutschen Anbieter für LLMs, aleph alpha oder so ähnlich, vielleicht wäre das ja eine Alternative. Ich halte nicht viel davon Modelle selbst zu hosten, das ist erstens ganz schön teuer und bei der Geschwindigkeit aktuell hat man ruck-zuck eine veraltete Technologie am Bein.
appreciated but allot of these modules are outdated which is very frustrating. Is the code from your course updated to the latest ? e.g.: openai.ChatCompletion is replaced with the Python API.
Excellent... however, if I can add, I think you are not using the latest langchain version ? Just a minor thing really, I think the imports are different.
can you tell me what exactly does not work? Since the langchain version almost changes every day, I probably did not use the latest one. I used 0.0.190 for the video.
@@codingcrashcourses8533 sure, so for example when i run the code in the notebook called models_prompts_parsers I get the following error: type object 'ChatPromptTemplate' has no attribute 'from_template' I checked the latest documentation and indeed it doesn't have that attribute/method
@@ldandco i had that Problem when i used an old Version. Can you Tell me your Version you are using? The latest one definately has got this classmethod
@@codingcrashcourses8533 I used a new kernel in the notebook, and it now worked. So, perhaps something that was cached, i am still not completely sure. But it works. thanks
Problem will always be the tokens usage. Even though its already reduced by a lot but it will still sending it to openAI API in full blown so it will use your token credits fast depending on how big your embedded data are.
I think its quite cheap, I never used more than 3$ for the API. But you can also use an open source modell from Huggingface. In another video I show how to use the falcon-7b model
@@codingcrashcourses8533 The oss version are garbage even cgpt 4 still garbage if we dont know how to properly used it 😁 I'm not talking about personal usages here but in a more professional way. cgpt took quite a long time to produce result too albeit its a paid product, the performance are awful right now.
Your video is good; however, I am quite confused about the output_parser, as you did not explain the uses of the module, the method, or the overall logic behind it. I need to reread the documentation again."
Does this require a paid subscription? I'm asking because I am new to using langchain, and I tried using my own text file as a custom dataset to asks questions on, but when I got it to run after fixing all dependency errors, it showed rate limit exceeded, but I am using an API key from a new account with the very first key. Would be great if anyone could point out what I am getting wrong?
@@codingcrashcourses8533 I see, so what you're saying is - create an open ai account, go to profile section, get a secret key(api key), and the key should work. Right?
I have OpenAI to generate SQL query which can take Mysql schema as a prompt. How can i leverage Langchain in this use case. Can you please help me on this.
I prefer a new apporach. Use functions and make requests via a dedicated REST Service. my experience with letting the AI create queries was pretty aweful. You can try an SQL Agent from Langchain to test that yourself
@@codingcrashcourses8533 So, you said you prefer a new apporach. Use functions and make requests via a dedicated REST Service. How to do this for my usecase or please suggest. because with OPEN AI facing token limit issue, sql query syntax error when openai generates sql query, and random SQL query generated with different outputs and also due to token limit, not able to generate chart for the report
Issues while using RetrievalQA ``` ValidationError: 2 validation errors for RetrievalQA retriever field required (type=value_error.missing) retriver extra fields not permitted (type=value_error.extra) ```
python.langchain.com/docs/modules/memory/how_to/conversational_customization I would have to look at the implementation details, but probably yes. You can use a custom prompt and I assume under the hood Langhchain converts that into a System Message
just run: git clone github.com/Coding-Crashkurse/Langchain-Full-Course.git inside a terminal (make sure git is installed!) Then you will create a new directory Langchain-Full-Course. Inside this folder you will find all notebooks. I would recommend using Visual Studio Code to work with them, but plain Jupyter will also do the job :-)
Python or JavaScript, since these two languages are supported. You should have intermedia knowledge with the language itself and also have some understanding about language models ifself.
Thank you 🙏 so much for your super comprehensive and massively detailed tutorial. I can’t believe this is free. Knowing you are an expert in this area, I have a quick question for you, and I’d appreciate it if you could give your advice on it. I’m a non-tech founder with good amount of conceptual knowledge about Python, LangChain, and other relevant approaches, frameworks and tools. I’ve been trying to develop a market research app similar to but more complex than this (th-cam.com/video/zP8JUbyCUk0/w-d-xo.html). If I just think about its mechanics, I.e., data flow, building it doesn’t seem awfully difficult. But the engineer I hired and others I was going to hire have trouble in coming up with practical development/design plan, let alone, building it. Do you think the development of something like mine is a pipe dream? Or am I searching engineers in wrong places? Thank you for reading this lengthy message. ❤
thanks for the video! I tried to execute the prompt template in the model_prompts_parsers but it return error. Can you help me on this? thank you --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) Cell In[10], line 3 1 from langchain.prompts import ChatPromptTemplate ----> 3 prompt_template = ChatPromptTemplate.from_template(template=template) 4 chain = LLMChain(llm=llm, prompt=prompt_template) 5 chain.predict(input="I ordered Pizza Salami and it was awesome!") AttributeError: type object 'ChatPromptTemplate' has no attribute 'from_template'
I am pretty sure you use an old version of langchain. Update it to a new version. from_template came at ~0.0.190 or so. pip install --upgrade langchain. Should solve that problem
@@0xgenji Maybe you use the wrong environment? import pkg_resources def print_version(package_name): try: version = pkg_resources.get_distribution(package_name).version print(f"The version of the {package_name} library is {version}.") except pkg_resources.DistributionNotFound: print(f"The {package_name} library is not installed.") print_version("langchain") try checking the result or what code
UPDATE: PLEASE USE THE UPDATED (2024) VERSION OF THIS COURSE: th-cam.com/video/-maRNgjEvLU/w-d-xo.html
This channel is a real gem. Excellent teaching style, and the tutorials are concise yet filled with a treasure trove of information and application. Thank you for making these, and I hope you will continue to explore AI with us. Will definitely look forward for your first masterclass.
Best Langchain video I have ever watched. 🎉
thank you so much for the comment. Glad you liked it :-)
Exactly what i was looking for! Awesome job. Keep going on 🎉
Thank you, great to hear that! 🙂
That was a lot of work. Thank you for making this very enlightening video.
For me or for you? ;-). Great you liked it, thank you for your comment
Excellent information, very helpful
2024 version will soon be released :)
Good Stuff- Explanation is great!
2024 Will follow :)
Thank you so much for sharing your knowledge and expertise.
answers all my doubts, great job sir!!
Really helpful your videos ❤!! I would like to add in min 11:10, if you are getting this warning 'LangChainDeprecationWarning: The function `__call__` was deprecated in LangChain 0.1.7 and will be removed in 0.2.0. Use invoke instead. ' change the line --> response = chat(messages) by --> response = chat.invoke(messages) :)
The Video is quite old already :)
Thanks for bringing so much value!
Thank you for your nice comment :)
Are you a Web Developer and more interested in JavaScript? Check out my LangChainJS Course: th-cam.com/video/mAYS4d0hrek/w-d-xo.html
Awesome explanations!
This is awesome thank! I wonder how to add recent announced function calling functionality in our project, since it makes much more easier to work with structured json objects.
I already worked with them - they are quite of a big deal. I uploaded a Video with the basics: th-cam.com/video/vJot6WCUlGE/w-d-xo.html .
Its a very good video to understand langchain.
That was excellent.Thank you!
This was so helpful. Thank you!
Great job!
Great video! Keep up the good work!
Thank you so much
Excellent instruction. How would you approach getting SEO keywords from a blog article using LangChain? Perhaps make a video about how that would be done.
Should that be Dynamic? Website scraping can be quite difficult because of Client side rendered js frameworks.
@@codingcrashcourses8533 I believe there a libraries that mimick a browser to first render the page then get the page content. I kinda have that figured out in my mind but I'm just not sure how one would architect the flow using LangChain. Do I need to use Agents, the new OpenAI function calling feature, or something else which Chaining could handle alone. I'm too new to LangChain to wrap my head around a basic programmatic workflow for this use case using LangChain.
@@StephenPasco yeah, the content will still be highly dynamic. Not easy
Sign up for my complete LangChain course: www.udemy.com/course/langchain-in-action-develop-llm-powered-applications/?couponCode=557BD137E003DD93E491
Amazing video! I have one question. In the Index part of the video, you go ahead and create the prompt_template with two arguments, 'question' and 'context'. But then, you didn't specify the 'question' and 'context values', so I think your query is not taking into consideration your template.
To verify it, I added in the prompt 'In all your answers, start them by saying "Dear Pedro" ' and it didn't follow the instructions.
I'm looking for ways in which to combine PromptTemplate and Embeddings, to be able to give the model a prompt and at the same time using data from .txt files
The High Level chains like RetrievelQAChain handles that for you. Context is used under the hood, but you as "user" of that chain only have to use "question". Forget to provide "context" in your template and you will see an error. For some chains it´s a little bit tricky to pass the template to the RetrievelChain.
Thank you sooooo much!!!
Nice! Can you run the notebooks on google colab as is, or are modification needed?
the code itself should work fine in colab too
marcus!! are u going to make a 2024 version of this with the LCEL and updated things?
A hug from spain!
Yes, a version is planned. I´ll be honest: I will release a Udemy couse and release the new crashcourse at the same time to get some traffic from the updated Crashcourse. Will be finished probably in ~6-7 weeks.
Thanks dude!
you´re welcome
great video thank you
Great job
thanks for the free key! 😉
Your welcome:p
Great video and explanations.............................:) bye
What modifications are needed to run against another LLM models, like those on hugging face?
python.langchain.com/docs/modules/model_io/models/llms/integrations/huggingface_hub
You need to load models differently, that´s pretty much it
@@codingcrashcourses8533
Is this approach fits for chatOpenAI also? At 39:11
Should I fill OpenAI balance to activate OpenAI Key? Or is it working free?
I would recommend to Set a rate limit. The 3.5 turbo model is very cheap
Bomben Video! What about a locally running opensource LLM? Would be a Gamechanger for German companies especially in finance since the burdens and regulations are very strict
I tried multiple open source LLMs. I have not seen a single one that comes closes to what OpenAI offers. The problem is that GPT performance is expected when you introduce a service with an LLM.
I work in finance in germany and our solution is currently not to use critical information from users and use a private endpoint on azure.
Ich bin prozessmanager (nicht IT) bei einem deutschen Versicherer und bin genau an diesem Thema dran. Azure kam ich irgendwie nicht zurecht. Würde mich zu gern mit dir darüber austauschen - oder du machst dazu auch ein Tutorial😃
@@monkeyy992 Azure müsste dein Unternehmen schon im größeren Stil einbinden. Bei diesem Thema passiert auf Infrastrukturseite bei uns sehr viel, wo ich allerdings nicht genau weiß, was alles geprüft wird. Firewalls und der Ganze Kram eben. Groß austauschen kann ich mich da nicht, mehr als das was ich gesagt habe, kann ich da nicht als Info weitergeben.
Es gibt noch einen deutschen Anbieter für LLMs, aleph alpha oder so ähnlich, vielleicht wäre das ja eine Alternative. Ich halte nicht viel davon Modelle selbst zu hosten, das ist erstens ganz schön teuer und bei der Geschwindigkeit aktuell hat man ruck-zuck eine veraltete Technologie am Bein.
@@codingcrashcourses8533 danke dir fürs Feedback!
appreciated but allot of these modules are outdated which is very frustrating. Is the code from your course updated to the latest ? e.g.: openai.ChatCompletion is replaced with the Python API.
I will release a 2024 Version in a few Weeks:)
@@codingcrashcourses8533 alright, keep us updated!
I love you bro
Nice tutorial what theme is used for VS code
Material Theme
Excellent... however, if I can add, I think you are not using the latest langchain version ?
Just a minor thing really, I think the imports are different.
can you tell me what exactly does not work? Since the langchain version almost changes every day, I probably did not use the latest one. I used 0.0.190 for the video.
@@codingcrashcourses8533 sure, so for example when i run the code in the notebook called models_prompts_parsers I get the following error:
type object 'ChatPromptTemplate' has no attribute 'from_template'
I checked the latest documentation and indeed it doesn't have that attribute/method
@@ldandco i had that Problem when i used an old Version. Can you Tell me your Version you are using? The latest one definately has got this classmethod
@@codingcrashcourses8533 I used a new kernel in the notebook, and it now worked. So, perhaps something that was cached, i am still not completely sure. But it works. thanks
@@ldandco great :)
Does langchain have caching? What if the same questions are asked by users of the bot?
python.langchain.com/docs/modules/model_io/models/llms/llm_caching
Take a look at that. Have ot used it yet :)
Problem will always be the tokens usage. Even though its already reduced by a lot but it will still sending it to openAI API in full blown so it will use your token credits fast depending on how big your embedded data are.
I think its quite cheap, I never used more than 3$ for the API. But you can also use an open source modell from Huggingface. In another video I show how to use the falcon-7b model
@@codingcrashcourses8533 The oss version are garbage even cgpt 4 still garbage if we dont know how to properly used it 😁
I'm not talking about personal usages here but in a more professional way. cgpt took quite a long time to produce result too albeit its a paid product, the performance are awful right now.
Thank you for this video but I don't get what is the difference between the .predict and .run methods , can you please explain?
It is normally doing the same, one might even call the other. I recommend looking in the source Code for getting a deeper understanding
31:00 Why we don't need to pass parameter in the "context" variable, how does the qa object knows our query goes into "question" variables?
thats handles by Langchain under the hood with these high level chains
Your video is good; however, I am quite confused about the output_parser, as you did not explain the uses of the module, the method, or the overall logic behind it. I need to reread the documentation again."
thanks for your feedback
Does this require a paid subscription? I'm asking because I am new to using langchain, and I tried using my own text file as a custom dataset to asks questions on, but when I got it to run after fixing all dependency errors, it showed rate limit exceeded, but I am using an API key from a new account with the very first key. Would be great if anyone could point out what I am getting wrong?
You need an OpenAI Account, which has nothing to do with the ChatGPTPlus Subsscription. Without the subscription to OpenAI, you can not use an API Key
@@codingcrashcourses8533 I see, so what you're saying is - create an open ai account, go to profile section, get a secret key(api key), and the key should work. Right?
@@mriganktiwari3782 yep :)
@@codingcrashcourses8533 Yeah that is how I expect it to work as well. But for some reason I am hitting a rate limit exceeded error... :(
@@mriganktiwari3782 same with me . Did you find any solution?
minute 2: it needs atualization
from openai import OpenAI
client = OpenAI(api_key=api_key)
def chat(input):
response = client.chat.completions.create(
model="gpt-3.5-turbo",
messages = [{"role": "user", "content": input}],
temperature = 0.7,
max_tokens = 256,
top_p = 1
)
print(dict(response).get('usage'))
return response.choices[0].message.content
My 2024 course will be released monday
can you teach onion architecture in python with OOP?
I have OpenAI to generate SQL query which can take Mysql schema as a prompt. How can i leverage Langchain in this use case. Can you please help me on this.
I prefer a new apporach. Use functions and make requests via a dedicated REST Service. my experience with letting the AI create queries was pretty aweful. You can try an SQL Agent from Langchain to test that yourself
If possible can u please architecture for this SQL Agent from langchain
@@arunakumari71795 my recommendation is to use something easier to control and also more reliable
@@codingcrashcourses8533
So, you said you prefer a new apporach. Use functions and make requests via a dedicated REST Service. How to do this for my usecase or please suggest. because with OPEN AI facing token limit issue, sql query syntax error when openai generates sql query, and random SQL query generated with different outputs and also due to token limit, not able to generate chart for the report
Issues while using RetrievalQA
```
ValidationError: 2 validation errors for RetrievalQA
retriever
field required (type=value_error.missing)
retriver
extra fields not permitted (type=value_error.extra)
```
You missspelled retriever
Can we change ConversationChain's system role?
python.langchain.com/docs/modules/memory/how_to/conversational_customization
I would have to look at the implementation details, but probably yes. You can use a custom prompt and I assume under the hood Langhchain converts that into a System Message
Why you call the LLM = chat OpenAI(temperature 0.0) without referring to any model (like gpt3.5)? It is not clear to me.
thats the default model for the OpenAI class
Hey a Langchain JS tutorial by any chance?
Since I am currently working on that, yes, 100% ;-). Gonna release it next weekend
As a beginner in GitHub I did not understand how to download your book
just run: git clone github.com/Coding-Crashkurse/Langchain-Full-Course.git inside a terminal (make sure git is installed!)
Then you will create a new directory Langchain-Full-Course. Inside this folder you will find all notebooks. I would recommend using Visual Studio Code to work with them, but plain Jupyter will also do the job :-)
@@codingcrashcourses8533 thank you very much. Will try. Hope to manage 🙂😉
@@acap2025 if it does not work, you can of course ask follow up questions
what skills and language i have to know first before getting into langchain
Python or JavaScript, since these two languages are supported. You should have intermedia knowledge with the language itself and also have some understanding about language models ifself.
@@codingcrashcourses8533 thanks
@@codingcrashcourses8533 is this langchain the future? Should one really get into this field? Cuz Im a web developer too,will it help me in this?
@@powerball200 LLMs are the future for sure. LangChain maybe, but that is just a wrapper to work with LLMs. Go for it if you are interested ;-)
You're just made comments about the langchain notebooks....
What?
The ChatGPT API keys are not free.
Where did I say that the openai api is for free?
@@codingcrashcourses8533 If you read what I wrote. It's a statement, not a question.
Thank you 🙏 so much for your super comprehensive and massively detailed tutorial. I can’t believe this is free.
Knowing you are an expert in this area, I have a quick question for you, and I’d appreciate it if you could give your advice on it.
I’m a non-tech founder with good amount of conceptual knowledge about Python, LangChain, and other relevant approaches, frameworks and tools.
I’ve been trying to develop a market research app similar to but more complex than this (th-cam.com/video/zP8JUbyCUk0/w-d-xo.html). If I just think about its mechanics, I.e., data flow, building it doesn’t seem awfully difficult. But the engineer I hired and others I was going to hire have trouble in coming up with practical development/design plan, let alone, building it.
Do you think the development of something like mine is a pipe dream? Or am I searching engineers in wrong places?
Thank you for reading this lengthy message. ❤
Hi, thank you but i dont do freelance work.
Good video, a bit too much content and little context. Huge subject and limited time.
yes. I will also create a project based course in the next few days :)
thanks for the video! I tried to execute the prompt template in the model_prompts_parsers but it return error. Can you help me on this? thank you
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
Cell In[10], line 3
1 from langchain.prompts import ChatPromptTemplate
----> 3 prompt_template = ChatPromptTemplate.from_template(template=template)
4 chain = LLMChain(llm=llm, prompt=prompt_template)
5 chain.predict(input="I ordered Pizza Salami and it was awesome!")
AttributeError: type object 'ChatPromptTemplate' has no attribute 'from_template'
I am pretty sure you use an old version of langchain. Update it to a new version. from_template came at ~0.0.190 or so. pip install --upgrade langchain. Should solve that problem
Thanks for the input. I Already upgraded langchain version ro 0.0.206 but oddly still give me the same error..
@@0xgenji Maybe you use the wrong environment?
import pkg_resources
def print_version(package_name):
try:
version = pkg_resources.get_distribution(package_name).version
print(f"The version of the {package_name} library is {version}.")
except pkg_resources.DistributionNotFound:
print(f"The {package_name} library is not installed.")
print_version("langchain")
try checking the result or what code