kernel-tonic/requirements.txt at main · TeamTonic/kernel-tonic · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
# =========================
# ROCm/MI300X Requirements
# =========================

# PyTorch, TorchVision, and Torchaudio for ROCm/MI300X
# DO NOT install from PyPI. Instead, follow the official ROCm install instructions:
# https://rocm.docs.amd.com/projects/install-on-linux/en/develop/install/3rd-party/pytorch-install.html
# Example (for ROCm 6.x, Python 3.10):
# pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0

# =========================
# Core dependencies
# =========================

# Hugging Face Transformers (validated on ROCm/MI300X)
transformers

# Hugging Face Datasets (for data loading)
datasets

# Hugging Face Hub (for model uploads)
huggingface_hub

# Optimum-AMD for ROCm acceleration (recommended for AMD GPUs)
optimum[amd]

# =========================
# Model Acceleration Libraries
# =========================

# Flash Attention 2 (ROCm/CK version, must be built from source)
# See: https://rocm.docs.amd.com/en/latest/how-to/rocm-for-ai/inference-optimization/model-acceleration-libraries.html
# Install with:
#   git clone https://github.com/ROCm/flash-attention.git
#   cd flash-attention
#   GPU_ARCHS=gfx942 python setup.py install

# xFormers (ROCm/CK version, must be built from source)
#   git clone https://github.com/ROCm/xformers.git
#   cd xformers
#   git submodule update --init --recursive
#   PYTORCH_ROCM_ARCH=gfx942 python setup.py install

# =========================
# vLLM (LLM Inference)
# =========================
# vLLM ROCm support: must be built from source for MI300X
# See: https://docs.vllm.ai/en/latest/getting_started/amd-installation.html
#   git clone https://github.com/vllm-project/vllm.git
#   cd vllm
#   pip install -e .
# Or use AMD's prebuilt Docker image for MI300X

# =========================
# Quantization Libraries
# =========================

# AMD Quark (recommended for FP8/INT8 quantization on MI300X)
amd-quark

# AutoGPTQ (ROCm wheels available, see docs)
# pip install auto-gptq --no-build-isolation --extra-index-url https://huggingface.github.io/autogptq-index/whl/rocm573/

# bitsandbytes (ROCm fork, must be built from source)
#   git clone --recurse https://github.com/ROCm/bitsandbytes.git
#   cd bitsandbytes
#   git checkout rocm_enabled_multi_backend
#   pip install -r requirements-dev.txt
#   cmake -DBNB_ROCM_ARCH="gfx942" -DCOMPUTE_BACKEND=hip -S .
#   make
#   python setup.py install

# =========================
# Transformer Engine (ROCm fork)
# =========================
# See: https://github.com/ROCm/TransformerEngine
#   git clone --recursive https://github.com/ROCm/TransformerEngine.git
#   cd TransformerEngine
#   export NVTE_ROCM_ARCH=gfx942
#   pip install .

# =========================
# Training, Logging, Utilities
# =========================

# Weights & Biases (wandb) - supported
wandb
# TensorBoard - supported
tensorboard
# PyYAML - supported
pyyaml
# Hydra - supported
hydra-core
# OmegaConf - supported
omegaconf
# pytest-cov - supported
pytest-cov
# PEFT (Parameter-Efficient Fine-Tuning) - supported
peft

# =========================
# Project-specific modules
# =========================
# (Add your own modules here)

# =========================
# Notes
# =========================
# - Only include packages above that are available on PyPI or have a clear ROCm install method.
# - For vLLM, transformer-engine, flash-attn, xformers, bitsandbytes, and auto-gptq, follow the ROCm-specific build instructions.
# - Remove or comment out any package not validated for ROCm/MI300X.