PyPI page
Home page
Author:
NVIDIA
Summary:
DTensor-native pretraining and fine-tuning for LLMs/VLMs with day-0 Hugging Face support, GPU-acceleration, and memory efficiency.
Latest version:
0.7.0
Required dependencies:
datasets
|
flashoptim
|
megatron-fsdp
|
mistral-common
|
ninja
|
pyyaml
|
quack-kernels
|
tiktoken
|
torch
|
torchao
|
torchdata
|
transformers
Optional dependencies:
albumentations
|
alembic
|
apache-tvm-ffi
|
av
|
backoff
|
bitsandbytes
|
boto3
|
causal-conv1d
|
databricks-sql-connector
|
decord
|
deep-ep
|
deltalake
|
diffusers
|
ffpa-attn
|
flash-attn
|
flash-linear-attention
|
fmha-sm100
|
ftfy
|
imageio
|
imageio-ffmpeg
|
kernels
|
mamba-ssm
|
mistral-common
|
mixture-of-kittens
|
mlflow
|
mlflow-skinny
|
multi-storage-client
|
nemo-automodel
|
nemo-run
|
numba
|
numpy
|
nvidia-cudnn-frontend
|
nvidia-cutlass-dsl
|
onnxscript
|
open-clip-torch
|
opencv-python-headless
|
perceptron
|
pillow
|
pyyaml
|
qwen-omni-utils
|
qwen-vl-utils
|
sentencepiece
|
soundfile
|
sqlalchemy
|
tile-kernels
|
tilelang
|
timm
|
torchvision
|
transformer-engine
|
wandb
Downloads last day:
244
Downloads last week:
3,238
Downloads last month:
16,025
7-day smoothing