https://arxiv.org/abs/2606.28344. The end of web parsing. The beginning of scalable pixel-native search. link: https://pixelrag.ai/
[project]
name = "wiki-screenshot-training"
version = "0.1.0"
description = "Wiki screenshot embedding training with Qwen3-VL + LoRA"
requires-python = ">=3.12,<3.14"
dependencies = [
"torch==2.9.1",
"torchvision",
"transformers==4.57.1",
"nvidia-cudnn-cu12==9.20.0.48",
"peft>=0.15.0",
"accelerate>=1.0.0",
"Pillow",
"tqdm",
"numpy",
"faiss-cpu",
"wandb",
"safetensors",
"huggingface-hub",
"qwen-vl-utils",
"datasets",
"openai>=2.30.0",
]
[tool.uv]
package = false
override-dependencies = ["nvidia-cudnn-cu12==9.20.0.48"]
[[tool.uv.index]]
name = "pytorch-cu129"
url = "https://download.pytorch.org/whl/cu129"
explicit = true
[tool.uv.sources]
torch = [{ index = "pytorch-cu129" }]
torchvision = [{ index = "pytorch-cu129" }]