Repository navigation
Expand file tree
/
Copy pathenvironment.yml
More file actions
97 lines (96 loc) · 3.33 KB
/
Copy pathenvironment.yml
File metadata and controls
97 lines (96 loc) · 3.33 KB
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
# conda env create -f environment.yml && conda activate genfirst
#
# The pip section is pinned to the versions these results were produced with.
# The conda section deliberately does not pin OS-level build strings, so the
# environment resolves on other machines; only the Python version is fixed.
#
# torch 2.3.1 here is built for CUDA 12.1. On a different CUDA runtime, install
# torch/torchvision from the matching index first, then let pip resolve the rest.
name: genfirst
channels:
- defaults
dependencies:
- python=3.10
- pip
- pip:
- accelerate==1.1.1
- antlr4-python3-runtime==4.9.3
- certifi==2025.1.31
- charset-normalizer==3.4.1
- click==8.1.8
- git+https://github.com/openai/CLIP.git
# Fork: adds `fid_statistics_file` so FID can be scored against the
# precomputed statistics in fid_stats/. Upstream torch-fidelity lacks it.
- git+https://github.com/LTH14/torch-fidelity.git@master#egg=torch-fidelity
- lpips==0.1.4
- dictdot==1.5.1
- diffusers==0.31.0
- docker-pycreds==0.4.0
- einops==0.8.0
- filelock==3.18.0
- fsspec==2025.3.0
- ftfy==6.3.1
- gitdb==4.0.12
- gitpython==3.1.44
- h5py==3.12.1
- huggingface-hub==0.29.3
- idna==3.10
- jinja2==3.1.6
- markupsafe==3.0.2
- mpmath==1.3.0
- networkx==3.2.1
- numpy==1.26.4
- nvidia-cublas-cu12==12.1.3.1
- nvidia-cuda-cupti-cu12==12.1.105
- nvidia-cuda-nvrtc-cu12==12.1.105
- nvidia-cuda-runtime-cu12==12.1.105
- nvidia-cudnn-cu12==8.9.2.26
- nvidia-cufft-cu12==11.0.2.54
- nvidia-curand-cu12==10.3.2.106
- nvidia-cusolver-cu12==11.4.5.107
- nvidia-cusparse-cu12==12.1.0.106
- nvidia-nccl-cu12==2.20.5
- nvidia-nvjitlink-cu12==12.8.93
- nvidia-nvtx-cu12==12.1.105
- omegaconf==2.3.0
- packaging==24.2
- pillow==11.1.0
- platformdirs==4.3.7
- protobuf==5.29.4
- psutil==7.0.0
- pyarrow==17.0.0
- pyspng==0.1.1
- pyyaml==6.0.2
- regex==2024.11.6
- requests==2.32.3
- safetensors==0.5.3
- sentry-sdk==2.24.0
- setproctitle==1.3.5
- six==1.17.0
- smmap==5.0.2
- sympy==1.13.3
- timm==1.0.11
- tokenizers==0.20.3
- torch==2.3.1
- torchmetrics==1.4.3
- torchvision==0.18.1
- tqdm==4.67.1
- transformers==4.46.3
- triton==2.3.1
- typing-extensions==4.12.2
- urllib3==2.3.0
- wandb==0.18.7
- wcwidth==0.2.13
# Optional, for the Qwen3-VL paths of the understanding stage (docs/UND.md):
# `--vae qwenvit` and `--und-type nll` need transformers >= 4.57, which is newer
# than the pin above. `nll` is the default understanding objective; `--und-type
# siglip` is the arm that runs on the pin. The understanding loss is off unless
# --und-coeff is set, so upgrade inside the env only if you enable it:
#
# pip install -U 'transformers>=4.57' 'tokenizers>=0.22'
#
# This applies to this file only. It is the archival record of the versions the
# published numbers were produced with, so its pins are deliberately frozen and
# include packages with known CVEs. requirements.txt carries the security floors
# and already pins a transformers that has Qwen3-VL; install from that instead
# unless you specifically need to reproduce bit-for-bit.