Pinned Repositories
100-days-of-code
lingua
prompt compression using LLMLingua.
meditron
medical RAG QA app using meditron 7B LLM, qdrant vector database, and PubMedBERT embedding model.
Modal-window
A modal is a dialog box/popup window that is displayed on top of the current page.
primary_llm
built my own large language model, from scratch (getting used for other reasons for now)
qa
QA generator application where you upload a pdf (chunk of text), creates embeddings and generates open-ended questions and answers
shopping_0
from the api call which connects to the scraper - the reminding progress and tweaked a bit due to bugs. If in future there are bugs, pull out branches from here
solana-pay
A point-of-sale web app that allows us to take payments from customers in person.first, we present a QR code that they can scan with a Solana wallet to make the payment. As soon as they do our app will update to show that they’ve paid.
starling
Starling-7B is a open LLM developed through Reinforcement Learning from AI Feedback (RLAIF), scoring 8.09 in the MT Bench evaluation with GPT-4 as the judge.
token-count
LLM cost calculator
sohomx's Repositories
sohomx/lingua
prompt compression using LLMLingua.
sohomx/solana-pay
A point-of-sale web app that allows us to take payments from customers in person.first, we present a QR code that they can scan with a Solana wallet to make the payment. As soon as they do our app will update to show that they’ve paid.
sohomx/starling
Starling-7B is a open LLM developed through Reinforcement Learning from AI Feedback (RLAIF), scoring 8.09 in the MT Bench evaluation with GPT-4 as the judge.
sohomx/meditron
medical RAG QA app using meditron 7B LLM, qdrant vector database, and PubMedBERT embedding model.
sohomx/shopping_0
from the api call which connects to the scraper - the reminding progress and tweaked a bit due to bugs. If in future there are bugs, pull out branches from here
sohomx/mymoneydapp
Developed our own crypto-currency on the Solana blockchain or our own spl-token in the Solana
sohomx/chatwithpdf
sohomx/emojiLLM
emoji LLM uses medical diagnosis questions and their corresponding emoji responses through ReFT
sohomx/flashrank
re-ranking with flashrank
sohomx/gemini
gemini model demonstrates remarkable versatility in it's visual understanding, classification, summarization, or generating content from images and videos,
sohomx/giskard
Giskard is an AI-powered personal assistant that helps users manage tasks, schedule appointments, and provides personalized recommendations, ensuring that the models are not just high-performing but also ethical and secure
sohomx/idefics
fine-tuning the image-aware IDEFICS 9B multimodal language model on a custom dataset for enhanced conversational abilities.
sohomx/LCM-LoRA
Image Generation: SDXL with LCM LoRA.
sohomx/lidagraph
sohomx/matmul
MatMul language modeling.
sohomx/meme-coin-dapp
A meme coin is a cryptocurrency that has almost no value and doesn't do anything interesting. It's just a coin, like Shiba Inu or dogecoin. Meme coins are basically just culture coins, and people trade them because they like memes.
sohomx/metadata
transforming RAG pipeline through metadata
sohomx/phi
how to implement and infer using the Microsoft/phi-2 base model, which is designed to harness high-quality, textbook-level datasets to achieve remarkable results.
sohomx/powerfinder
activation locality to optimize neural network computations charging LLMs for performace
sohomx/quartz
sohomx/rag-vector-dbs
few ways to implement advanced RAG
sohomx/runpod
Script sets up an environment using RunPod's API to create and manage a pod hosting Falcon-40B LLM, on two NVIDIA A100 80GB GPUs
sohomx/shopping
A web scraper built using OxyLabs. It fetches products with their prices and descriptions and sorts_by with the given information tally provided
sohomx/slim
SLIM models by LLMWare
sohomx/solvent
a simple program to transform tokens to and from the program (which can be released after _token_time_ mentioned as in the form of vault)
sohomx/training-w-gpt-2
using distilgpt2 to comprehend diseases and symptoms, exploring dataset loading, tokenization, hyperparameter tuning, and text generation, empowering contextual understanding in medical contexts.
sohomx/unsloth-
utilize the imdb dataset for fine-tuning a pre-trained 7b mistral language model (quantized at 4 bits), targeting improvements in text generation performance.
sohomx/vectara
using the Word Embedding Association Test (WEAT) and Demographic Diversity Analysis and using vectara's hallucination evaluation model to understand how it transforms the LLMs
sohomx/vision
sohomx/wandb
leveraging Weights & Biases to monitor RAG pipeline