CODE

→ View original post on X — @learnopencv

The Complexity and Unreliability of Prompt Engineering Explained

By

@learnopencv

–

01 March 2024 19h25

Prompt engineering gets very complicated and unreliable very soon. You are essentially programming with the convenience of natural language but without the precision of computer code. It also suffers from all the problems of solving a machine learning problem using logical

→ View original post on X — @langchain

Advanced RAG Series: Indexing Optimization Techniques

By

@langchain

–

01 March 2024 19h01

Advanced RAG series: Indexing Indexing is a huge part or RAG. Lot's of different stuff to optimize: 1. Chunk Optimization
2. Multi-representation indexing
3. Specialized Embeddings
4. Hierarchical Indexing Great article by @divyanshu_van diving into all these topics! Blog:

→ View original post on X — @langchain

Models Support Function Calling Documentation Updates

By

@langchain

–

01 March 2024 18h51

As more and more models support function calling, we're updating our docs to highlight them all! Function calling

bm25_pt: Fast BM25 Search Package for Python

By

–

01 March 2024 16h25

without bm25_pt you either have to use really slow string-based code (rank-bm25) or run a server (elasticsearch) to do bm25 in python now, it's a pip package, and two lines; i made everything fast using huggingface & pytorch

BM25 Search Implementation Performance Comparison with ElasticSearch

By

–

01 March 2024 16h25

also shout out to @clifapt for running BM25 via ES on LoCo first: https://
x.com/clifapt/status
/1755303087873429667
… my implementation seems to be slightly less performant than ElasticSearch; not sure why. maybe because i'm using a subword tokenizer, or suboptimal hyperparams?

Fast GPU-Enabled BM25 Implementation in PyTorch Achieves SOTA

By

–

01 March 2024 16h25

implemented a fast, GPU-enabled BM25 in pytorch! BM25 is a simple search algorithm from the 70s that works as well as neural networks for most search problems; for all the advances we've made in neural text retrieval, it's still around got near SOTA on stanford LoCO benchmark

BM25_PT: Matrix Multiplication for Efficient Search Scoring

By

–

01 March 2024 16h25

my library is here, bm25_pt: http://
github.com/jxmorris12/bm2
5_pt
… the key insight is that you can reduce the BM25 scoring to a single matrix multiplication. everything in the big fraction on the right side here can be stored in a big matrix of scores then, bm25(q) = bag(q) @ scores.T