# NVIDIA/FasterTransformer

Transformer related optimization, including BERT, GPT

Repository: https://github.com/NVIDIA/FasterTransformer
Canonical: https://ross.abutalabs.com/products/fastertransformer
Language: C++
License: Apache-2.0
License Family: permissive
Topics: pytorch, transformer, gpt, bert
Last push: 2024-03-27T11:25:30+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1979, "days_push": 889, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6447, forks 933 (observed 2026-08-28T04:09:43.472025+00:00)

## What it is
NVIDIA's highly optimized C++/CUDA library for fast inference of Transformer-based models such as BERT, GPT, and encoder-decoder models, with TensorFlow, PyTorch, and Triton APIs. It leverages Tensor Cores with FP16/INT8/FP8 precision and supports tensor and pipeline parallelism, but development has moved to TensorRT-LLM.

## Use cases
- run fast BERT inference on NVIDIA GPUs
- serve GPT models with tensor parallelism
- speed up transformer encoder and decoder inference
- use FP16 or INT8 transformer inference with Tensor Cores
- benchmark optimized transformer decoding performance
- integrate optimized transformer kernels into PyTorch or TensorFlow

## When to choose
- you need maximum transformer inference throughput on NVIDIA GPUs and can use an older stack
- you want multi-GPU tensor or pipeline parallel inference for GPT-style models
- you need FP16/INT8 optimized BERT or GPT kernels in TensorFlow, PyTorch, or Triton

## When to avoid
- you are starting a new project - use TensorRT-LLM instead, since FasterTransformer is no longer developed
- you need the latest model support or FP8/Hopper optimizations
- you do not have NVIDIA Volta or newer GPUs

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-inference, deep-learning, machine-learning, gpu-computing
- domain: large-language-models, deep-learning, gpu-computing
- platform: cpp, python
- tags: transformer, bert, gpt, cuda, tensor-cores, inference-optimization, tensor-parallelism, pipeline-parallelism, nvidia, superseded-by-tensorrt-llm, natural-language-processing, linux, gpu, docker

## Member repositories
- NVIDIA/FasterTransformer (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:43.472025+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:44:40.097132+00:00, confidence not recorded.
  - readme: https://github.com/NVIDIA/FasterTransformer (fetched 2026-08-28T04:09:43.472025+00:00, sha 8a7123d2da2c)
- Data as of 2026-08-30T08:39:29.467469+00:00.