Harshit Kedia
@SoftDevHarshitK
Followers
57
Following
49
Media
75
Statuses
486
I'm into Coding, Tech, Spirituality, Philosophy, Science, Astrophysics and much more 😎 | Remote Work 🤩 | Generative AI | Let's Connect 😊
Remote
Joined September 2022
A detailed and interesting introduction about myself 😁: https://t.co/29GmXfvzL4 Attributes mentioned: Career (technical), Health, Personality, Life Insights, Passion/Hobbies, Spiritual Side and much more 😄
0
0
0
JUST IN: THE FINAL PHOTO OF SAMOURAI #BITCOIN DEVELOPER KEONNE RODRIGUEZ BEFORE HE REPORTS TO PRISON RT IF YOU WANT TRUMP TO PARDON HIM IMMEDIATELY 👏
58
533
2K
🚨 BITCOINERS UNITE! 🚨 Let’s keep the momentum going! Tonight live at 8pm EST. Join the epic @_BitcoinCenter LIVESTREAM-A-THON🔥 We’re rallying support to pardon @Keonne & Bill, persecuted for building privacy tools. #PardonSamourai 📺👉🏻 https://t.co/TOohO4EP5i
2
8
55
You need to take time out of your day to watch this interview. It is only 57min long, or about 37min if you watch it in 1.5x, this interview will tell you everything you need to know about the Samourai case. Once again, the government is going after coders, builders, people who
0
0
0
I used is MistralAI LLM with Langchain framework and ChromaDB database. Guidance for running the program is given in the README file in the following repo link. Associated github repository:
github.com
Contribute to Harshitnitw/Source-Code-Analysis-Generative-AI development by creating an account on GitHub.
0
0
0
Then the users can, for instance, ask about a function written in their code and LLM will respond with an explanation of that function. The chat remembers conversation history.
1
0
0
The code internally downloads the repo and feeds the data to a vectorstore with context based text splitting, because code should be provided to the vectorstore not randomly but such that the embeddings characterize which chunk belongs to which function.
1
0
0
Worked on an End to End generative AI based source code analysis project. Users are required to provide a link to the public github repository for which they want source code analysis to be done. #100DaysOfCode #DataScience #GenerativeAI
1
0
1
I finished working on a Retrieval Augmented Generation (RAG) custom data (medical book) based medical chatbot with LLM using Langchain, Flask, Mistral AI and Pinecone vector database! Associated github repo: https://t.co/fmwyNQoc5E
#100DaysOfCode #DataScience #GenerativeAI
0
0
2
Link of associated code repository: https://t.co/314zoliaIA
#100DaysOfCode #DataScience #GenerativeAI
0
0
2
Associated repository for the stocks data is also there in the code from which Vectorstore can be made in a single go by running the python file. You need to add your MistralAi API key in the .env file before running the app.
1
0
0
Future predictions of these stocks can be made on the streamlit app. We have options of both, knowing about an individual stock as well as comparing stocks of 2 companies.
1
0
0
I was able to create a financial stocks analyst using LlamaIndex and Mistral AI LLM. Information of past stock data of major tech giants like Apple, Google, NVIDIA, etc. was shared as PDFs and based on the retrieved information about these companies.
1
0
1
Used LlamaIndex to successfully Inference Mistral-7B-Instruct-v0.1 downloaded from Huggingface, with custom data retrieved from Vectorstore saved on local disk 😄 Source code: https://t.co/DecS8EgL77
#100DaysOfCode #DataScience #GenerativeAI
github.com
Contribute to Harshitnitw/LlamaIndex-with-self-hosted-MistralAi development by creating an account on GitHub.
0
0
2
Switched to gemini flash and the code worked like charm! 😄 Here’s the associated github repo of my code: https://t.co/7i5VipoN6y
#100DaysOfCode #DataScience #GenerativeAI
github.com
Contribute to Harshitnitw/llamaindex-practise-with-google-gemini development by creating an account on GitHub.
0
0
3
Following the tutorial, I got stuck trying to work my code on google PaLM API. Spent my evening trying to workaround the 404 error for inferencing the API, but then rolled my eyes to find that the model has been decommissioned in 2024.
1
0
0
Got first hand experience of LlamaIndex! I’d say that although fast, it seems fairly less intuitive than Langchain because its classes’ names convey what they intend to do. Or it’s maybe because I am accustomed to langchain so maybe I’m biased towards its naming convention? 🤔
1
0
0
Finally, I was successfully able to merge the original model with LoRA weights and upload the fine tuned model to github. Source code link: https://t.co/FWv5MvGFNQ Fine tuned model available on huggingface:
lnkd.in
This link will take you to a page that’s not on LinkedIn
0
0
0
As this exercise was for practice purpose and performed in free Google colab with T4 GPU, I trained the model for one epoch only. Tensorboard visualizations were also done for model accuracy, training and validation losses, etc.
1
0
0
Thanks to Nous research and Maxime Labonne for making the model and the dataset publicly available: -> NousResearch/Llama-2-7b-chat-hf -> mlabonne/guanaco-llama2-1k
1
0
0