Memory Efficient Python Packages

madoka

Memory-efficient Count-Min Sketch Counter (based on Madoka C++ library)

adv-optm

A family of highly efficient, lightweight yet powerful optimizers.

expanse-trie

256-ary radix trie and Judy arrays in Rust — drop-in libjudy C ABI, MIT/Apache-2.0

came-pytorch

[ACL 2023] The official implementation of "CAME: Confidence-guided Adaptive Memory Optimization"

gigaxml

Extract structured data from multi-gigabyte XML files with bounded memory, as a command line tool and a desktop application.

apollo-torch

APOLLO: SGD-like Memory, AdamW-level Performance; MLSys'25 Oustanding Paper Honorable Mention

adafactor8bit

Memory-efficient PyTorch optimizers with Adaptive Log-Space quantization, fused CUDA kernels, and Adafactor, CAME, APOLLO support.

savov-compact-tree

Compact tree representation in storage implemented in Python3 and C

neurostream

Run transformers larger than memory. Weights stream from NVMe at neuron-block granularity under a hard RAM/VRAM ceiling - 142 GB across 3 shards on an 8 GB GPU. GGUF, MoE selective expert fetch, Triton fused dequant-GEMV kernels.

fast-csv-loader

A fast and memory efficient way to load large CSV files (Timeseries data) in Pandas.

zenreg

ZenReg is a Python-based high-throughput, memory-efficient pipeline for N-dimensional image registration, designed for large microscopy and imaging datasets.

memory-efficient-attention-pytorch

Implementation of a memory efficient multi-head attention as proposed in the paper, "Self-attention Does Not Need O(n²) Memory"

flatwire

Streaming JSON serialization that keeps memory flat and time linear - one identical API across Python, Node, .NET, Rust, Go, and Java.

inf-cl

[CVPR 2025 Highlight] The official CLIP training codebase of Inf-CL: "Breaking the Memory Barrier: Near Infinite Batch Size Scaling for Contrastive Loss". A super memory-efficiency CLIP training scheme.

faceflash

Face search engine: 100% recall at 96x less memory than HNSW. Register photos, identify people from camera/video/images. AVX-512 + Rust. No GPU needed.

diet-pandas

Automated memory optimization for Pandas DataFrames. Reduces memory by 50-80% and loads CSV 5-10x faster. Drop-in replacement that returns standard DataFrames - works with sklearn, matplotlib, and all your favorite libraries. Safe, fast, and zero refactoring required.

optimized-transducer

Memory efficient transducer loss computation

fastflash-attention

Drop-in exact bf16 flash-attention for CUDA with a deterministic backward, tuned for Blackwell (sm_120 / RTX 5090).

ringattention

Large Context Attention

morris-counter

Memory-efficient probabilistic counter namely Morris Counter

ssma

Structured State Matrix Architecture (SSMA) is a high-performance framework designed for efficient sequence modeling, combining structured state space models with adaptive attention mechanisms.