Machines with processors that can crunch 250+ GFlops are accessible to everyone who has access to a mail account, and a browser. I believe that making these machines work for you will be best thing to work on at this moment. I enable you to do that.
Hi Kamalaraj, when using the unstructured library for extracting images and text,tables from the scanned pdf files in windows. I got the error, poppler is not found and in path?. i'm installed poppler in the program files and add the path in the system variables also.. stll im getting the same error... can i install poppler in local or vs code ? please help me in this issue sir...
@@lakshmipanguluri6939 I reviewed the challenge you are facing, and it's most likely you are using windows. stackoverflow.com/questions/53481088/poppler-in-path-for-pdf2image The above SO post provides different solutions for your issue. Try them out and update. The most easiest solution is to move to Ubuntu OS.... WSL creates some weird challenges...
@@insightbuilder Thanks for your response. Will try and update you… still if it is not working then I’ll move to ubuntu os… thank you so much for your valuable advice…🙂
If I wanted to use Groq to improve the embeddings and better LLM what would be the change? And wouldn't the vector database be local to be used as RAG?
Can you help me understand, how the examples being chosen out of trainset provided in Botstrapfewshot. Also, I understand that it add rationale in the questions but like how Bootstrapfewshot works. Can you go in the code and try to gather the information, the teacher and student model, it was not clear.
@@AmanIndia-m5l github.com/stanfordnlp/dspy/blob/main/dspy%2Fteleprompt%2Fbootstrap.py The above script implements the bootstrapwitheandomsearch and there is random_search in the same path which implements the BsFsWRs algo. To understand the working, you have to get comfortable with how metrics works, and how bootstrap withfewshots works..try that first.
Can you show the input it goes to LLM's, is it the same being shown in last videos. Well if then it is just re writing the signature in a certain way, not optimizing it or generating new one!
I have been working on DSPy from a month, and truly saying, I uncover that, it just we write the stuff in a certain format LLM better understand, nothing much, can you please comment your views!
@@AmanIndia-m5l DSPy framework's primary objective is to automatically tune the prompts for better reply from the LLM. LLMs give better reply when good examples outputs are provided in the prompts. So DSPy adds these examples to the signature when you compile the Modules. I hope the above makes sense. The compilation, and it's optimization are the keys that you have to dive deep and understand.
docker is possible, and there are tutorials provided by matrix team themselves. Much of the configuration is automated there. My intention is provide in depth understanding for any one who might need alternate route.
Hi Kamalraj, I need your help with a project I'm working on. The project involves storing dataset records that come from different sources like Hugging Face, imports, etc. I have stored these dataset records on GCP, and now I need to send these records to Argilla. I noticed that to push records to Argilla, we need to convert the data into Argilla's data structure. The problem is that I don't have a specific record format because Hugging Face uses different record formats for different models. Could you please help me with this? Thank you!
Hi Kamalraj, I need your help with a project I'm working on. The project involves storing dataset records that come from different sources like Hugging Face, imports, etc. I have stored these dataset records on GCP, and now I need to send these records to Argilla. I noticed that to push records to Argilla, we need to convert the data into Argilla's data structure. The problem is that I don't have a specific record format because Hugging Face uses different record formats for different models. Could you please help me with this? Thank you!
Could you teach me how the "trace" argument can be used to improve training? All it says on the official document is that trace is turned off during evaluation and optimization, but turned on during compiling and bootstrapping. Can trace be used to monitor the evaluation metric somehow?
@@sesburg During metrics calculation, you can add traces that you want, when you are going to make your own metrics. There are some community examples that touch this idea
@@sesburg DSPy is AI framework, so the ecosystem surrounding it will have such examples. Weaviate has here in their cookbook github.com/weaviate/recipes/blob/main/integrations/llm-frameworks/dspy/ Also search for "DSPy receipe" that might provide some links.
Why download the model? Can I directly run from the HF? something like - 'HF_MODEL_ID':'distilbert/distilbert-base-uncased-finetuned-sst-2-english', instead of storing in s3? Wouldnt that automatically download model into s3?
It's a great explanation for ollama thank you sir. But i am getting a ```{"error":"model 'llama2' not found, try pulling it first"}``` even i already pull and run the model in my server.
can you actually put mutlilple questions in a sing tavily search . So perhaps a) I want to find X b) Where is X located c) What does it work on d) What is the USP of X ?
Hi Kamalraj, I have a question. Whats the point of using Evaluator Function ? Its not going to optimize the prompt rite.. so, its purpose is just to see how model responds to a dataset. Or , is it meant to be used again once the prompts are optimized so that we can verify whether the optimization is good or not ?
Hi Kamalraj, One question. At 8.54 In the code you were using print("Response by OpenAI: ", resp.answer). ie, you are trying to print the answer. But in the output that got printed, the Question and answer was present. why is that ? problem happened with your question "Whos is the worlds fastest man and women" . But the answer was ok for your previous question " President of India"
Check out playlist on argilla in this channel, you will find a better more streamlined way of building datasets. I will update on DSPy datasets in sometime
hello, i'll like to setup stablediffusion comfy ui using modal but there's no video about modal for absiolute beginners let alone stbldiffusion. they have something about stablediffusion in their docs but its just outdated, the images their trial stablediffusion demo makes look very outdated. pls if you can, can you make a video about using stblediffusion in modal
Metrics in DSPy are basically "test cases" built with assert statements. One can use the readily available metrics like Accuracy, F1 Score function inside HF's evaluate library. However one has to wire the outputs of your prompts as inputs to the said functions.
Kindly share the error that it threw when you tried to "Import Torch" on your 32-bit OS. We need to look at your full error trace, and then only we can find the solution. Its better to use 64-bit OS only.