datasets>=4.0.0
megatron-fsdp>=0.2.3
mistral-common[audio,hf-hub,image,sentencepiece]
opencv-python-headless==4.10.0.84
pybind11
pyyaml
tiktoken
torch<=2.10.0,>=2.6.0
torchdata
transformers==5.5.0
wandb
torchao
mlflow
flashoptim>=0.1.3

[all]
nemo_automodel[cuda]
nemo_automodel[delta-databricks]
nemo_automodel[diffusion]
nemo_automodel[extra]
nemo_automodel[vlm]

[cli]
pyyaml

[cuda]
causal-conv1d
mamba-ssm
nv-grouped-gemm
onnxscript>=0.5.6
transformer-engine[pytorch]<=2.11.0

[cuda_source]
bitsandbytes

[delta-databricks]
deltalake>=1.0.0
databricks-sql-connector>=3.0.0

[diffusion]
diffusers>=0.37.0
ftfy
imageio
imageio-ffmpeg
opencv-python-headless
torchvision

[diffusion_kernels]
nemo_automodel[diffusion]
kernels

[extra]
perceptron
sentencepiece

[fa]
flash-attn<=2.8.3

[moe]
nemo_automodel[cuda]
deep_ep

[vlm]
albumentations
backoff
mistral_common[opencv]>=1.11.0
numpy
numba
open-clip-torch
pillow
qwen-omni-utils
timm<=1.0.22

[vlm:platform_machine == "x86_64" and platform_system != "Darwin"]
qwen-vl-utils[decord]
torchcodec
