LLMWare by Ai Bloks innovates the latest in cutting edge solutions in AI for enterprises. Our open source AI Platform delivers the most integrated and easy to use out of the box Retrieval Augmented Generation. We also have over 60 models in Hugging Face that are small specialized language models for private cloud or on prem for enterprise use cases. Check us out!
Hey there, this is all interesting and I tried running your fast start RAG notebooks on google colab. This 5th notebook in the RAG series (Semantic query ) when run on colab throws an error. In fact, you can find the error in the original notebook itself which was uploaded. It seems this notebook never ran well as a notebook but was still uploaded on Github. I have opened an issue there. Hope somebody can have a look.
Hi Thank you for your feedback. We used to support it for months then we deprecated it about. a month ago because very few people were using it...😅 We will reconsider this in the future if we get more requests!
Is there a way to combine this with your "example 5" program that runs a semantic query with RAG? This would be such a better interface. Also, it is desirable to avoid the chatbot drop down menu that asks which specific pdf you want analyzed for the semantic query, and instead have it search all of the pdfs in your library OR provide an answer for each pdf just like the example 5 program.
This is absolutely brilliant! The utility of this is off the charts! I will have to review all your vids and find out how to run this, and what model suits it best.
Very Very NIce, Darren is it possible, to have LLMWare automagically boz bpt tale a URL address of a webpage and output it to a format: audio, text, pdf or csv then automatically load it into the biz bot?
hi great stuff by the way it seems though your not getting greatly noticed !!! perhaps make a few gradio spaces ? Personally i feel that you have nice method with building the slim models (in fact real agent experts) ... but also the wrapper performs a task: which could basically be overfit to the slim model : therby fixing the model in this specific mode: SO: i Think that creating a model from all these models or subsets of these experts as mixture of experts models : with the expert slim models as the experts : (trying to add up to a 7b(total parameters) moe: ie a slim moe? this would actually be very usefull model as you could use the creation of the model prompts internally for the merge directing the specific querys to the correct expert: the concept of slim models is great but the wrappers them selves are also greater .... as thier own sidepeices: hence geven the wraper we could essentially patch and openAI component or a Transformers(pretrained), Or a llamaCpp(gguf) model... hence the customized wrappers that perform the tasks would only need a model and tokenizer to plug in (utilize the embeddings from the model please).... here you can see the product produced .... now the service is designing these overfit experts , and wrapers : hence an object model which i think you have ... the object model is the Free Library... for open source community to build thier own wrappers .... and your comercial wrappers either as example (to show case your departments skillls hence business models would come direct to have thier custom models produced!) ... i can see why you have not been popular and even skipped by many !
very useful but where exactly is the code for all this ? specific all this !!! Some people are not good with coding and want to test these examples, but this specific has everything one needs....Will you be able to produce some ready examples just like this one ?
@@llmware please also make it generic meaning give the option to upload the files also the option to upload maybe the small lm? for each usage? just ideas !
@@gnosisdg8497 Hi we have uploaded previous videos that perform these individual tasks (this is representing some of our capabilities in one bot)... Please check out our repo or join our discord and we can answer more questions there as well!
I tried to run a chat in Colab several times and it downloaded the model and it did not complete. Should I run the example from the same folder without modifying its real path? I don’t know the problem.
this is beautiful content! exactly what I was looking for but summarizing this work in a blog with evaluation table and some diagram of the workflow of the benchmarking would have been great.
Hi Thank you so much for your kind feedback! While it is not exactly a perfect match for this particular video, we do have a blog post on this topic. Please take a look! 😃medium.com/@darrenoberst/how-accurate-is-rag-8f0706281fd9
🎯 Key points for quick navigation: 00:18 *📄 Introduction to document parsing, chunking, and data extraction.* 00:33 *🛠️ Advanced techniques for extracting images, tables, and automating workflows.* 01:17 *📚 Preparing datasets for self-supervised learning and fine-tuning.* 01:31 *💡 Focus on data wrangling and Microsoft Office documents.* 02:14 *🗂️ Accessing public Microsoft Word, PowerPoint, and Excel documents.* 03:22 *📂 Downloading and preparing Microsoft Office documents.* 04:03 *🛠️ Setting up the environment to parse and chunk documents.* 05:12 *🔍 Smart chunking strategies and their configurations.* 06:22 *📑 Parsing tables and images from documents.* 07:32 *🗃️ Exporting tables into CSV files.* 08:28 *🖼️ Running OCR on extracted images.* 09:54 *📄 Creating a consolidated JSONL file.* 10:35 *📊 Building a dataset for unsupervised testing.* 11:14 *⚡ Parsing 152 files in 6 seconds using a local Mac M1.* 12:37 *🔍 Running OCR and storing text in the library.* 13:17 *⏱️ Comparing the speed of digital parsing versus OCR.* 14:23 *📁 Exploring file artifacts created during parsing.* 16:29 *📄 Reviewing the created dataset.* 19:44 *🎥 Closing remarks and upcoming example videos.* Made with HARPA AI
Loved the video! The step-by-step guide on parsing docs and data was super helpful. I was really impressed by how you used OCR to pull text from images in Microsoft Office files - that was cool. The smart chunking strategy explanation was also 👌.
This is really a nice way of extracting data and converting the unstructured data into structured form. I believe the data after extraction can be used as a data source for the RAG pipeline and probably LLMs can give more accurate answers.
That's the best video and simplest program I have come across ingesting pdfs at large scale. I immediately looked into your playlist but couldn't identify the next video to digest the extracted data. Can you kindly guide me to the right video?
Hi @kushis4ever thank you so much for the kind comment! Yes we actually have a playlist for getting started with LLMWare: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-uW3fElxcri4.html&pp=gAQBiAQB
Our Fast Start to RAG playlist in our channel will help you get started with RAG using LLMWare - please also join our discord community to interact with us and to learn more tips and tricks! discord.gg/4pMEYHxR2K
i run into an issue when the model generates multiple answers, everything is by default, i just said hi to it and here is the answer i get. Hello! How can I assist you today? Whether it's answering questions, providing information, or helping with tasks, feel free to let me know what you need help with. <|assistant|> Hi there! I'm here to help you with any inquiries or assistance you may require. Just ask away! <|assistant|> Greetings! I'm ready and eager to provide any assistance you may need. What can I do for you today? <|assistant|> Hey! I'm your virtual helper. Just say the word and I'll be at your service for any assistance you may need. What can I do for you today? <|assistant|> Good day! As your virtual assistant, I'm ready and able to provide any support you may require. Please go ahead and ask your question or describe the task at hand. I'll do my best to assist you.<|end|><|assistant|> Hello! It's great to have you here. As your virtual assistant, I'm ready and available to help answer your questions and provide any assistance you may need. Just let me know how I can be of service!<|end|><|assistant|> Hey! Welcome! As your virtual assistant, I'm here to provide any support and assistance you may require. Just tell me how I can help and we'll get started right away!<|end|><|assistant|> Hey there! It's great to have you here. As your virtual assistant, I'm ready and eager to provide any assistance you may need. Just ask away and I'll do my best to help!<|end|><|assistant|> Good day! As your virtual assistant, I'm ready and able to provide any support and assistance you may require. Please go ahead and ask your question or describe the task at hand. I'll do my best to assist you in any way I can.<|end|><|assistant|> Welcome! It's great to have you here. As your virtual assistant, I'm ready and able to provide any support and assistance you may require. Just tell me how I can be of service and we'll get started right away!<|end|><|assistant|> Hey! It's great to have you here. As your virtual assistant, I'm ready and eager to provide any assistance and support you it is running on my nvidia 4060ti. ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no ggml_cuda_init: CUDA_USE_TENSOR_CORES: yes ggml_cuda_init: found 1 CUDA devices: Device 0: NVIDIA GeForce RTX 4060 Ti, compute capability 8.9, VMM: yes
Hi The theme we use is just basic "dark" theme and nothing else special... I hope this answers your question but please let me know if there are others!
I tried to pass bank statement which is in pdf format. but the tables within the pdf is not getting extracted. any change I need to make to improve parsing?
Would like to see a tutorial to use it on an Android smartphone: local FOSS LLM, speech and image recognition with sufficient performance IMO are crucial
IMHO, the game changer will be a local language model that is an interface to a graph database that is local for personal and private data, and merges personal and public data with digital agents running on millions of other devices into a decentralized graph structure world model for planning collective actions.
I am having issues with this line: source = prompter.add_source_document(contracts_path, contract, query=key) The documents are not loaded and facing below error: source = prompter.add_source_document(contracts_path, contract, query=key) WARNING:root:No source materials attached to the Prompt. Running prompt_with_source inference without source may lead to unexpected results. Note: Made sure the docs are in the directory Please help!
Hi can you pls check if the sample files downloaded in the path /llmware_data/sample_files/ If not there then you can force download refresh as option in the pull sample files... If you are still having problems, please find us in discord so we can help you more there! discord.gg/pUvKzYujdM
I would love to see performance measures for llmware models running on a Oneplus 11 with 16 GB of RAM. It has as good of hardware as any Android smartphone sold in 2023.
Yes, if you cloned the repo, then please copy the examples out of their folder tree into the main path as a peer to the /llmware source - and you should be good to go!
