exllama logo

exllama

turboderp/exllama

More memory-efficient rewrite of HF transformers for Llama with quantized weights

GraphCanon updated today · GitHub synced today

2.9k
Stars
223
Forks
65
Open issues
34
Watchers
2y
Last push
Python MITCreated May 4, 2023

Trust & integrity

Full report
Maintenance
Dormant (1014d since push)
As of today · Source: github_public_v1
Provenance
Not a fork · Personal account
As of today · Source: github_public_v1
Security (OSV)
29 low (29 low)
As of today · Source: osv@v1

Public GitHub metadata and optional OSV dependency scans. Signals, not a guarantee. Trust methodology.

Overview

ExLlama provides an optimized and more memory-efficient alternative to the HF transformers implementation specifically tailored for use with quantized Llama models, targeting hardware such as RTX series GPUs.

Capability facts

Deploy
Self-host

Source: dockerfile:Dockerfile · Jul 12, 2026

Docker
Dockerfile present

Source: dockerfile:Dockerfile · Jul 12, 2026

Languages
python

Source: github.language · Jul 12, 2026

Categories

Tags

README

Hardware requirements

I am developing on an RTX 4090 and an RTX 3090-Ti. 30-series and later NVIDIA GPUs should be well supported, but anything Pascal or older with poor FP16 support isn't going to perform well. AutoGPTQ or GPTQ-for-LLaMa are better options at the moment for older GPUs. ROCm is also theoretically supported (via HIP) though I currently have no AMD devices to test or optimize on.


Docker

For security benefits and easier deployment, it is also possible to run the web UI in an isolated docker container. Note: the docker image currently only supports NVIDIA GPUs.


Requirements

It is recommended to run docker in rootless mode.