FastVideo Training Infra Quick StartFastVideo Training Infra Quick Start
A quick-start guide for the new FastVideo training framework (fastvideo/train), covering code checkout, how to run training, example commands, and links to the related PR and RFC.FastVideo 新训练框架（fastvideo/train）的快速上手指南，包括代码拉取、训练运行示例及相关 PR/RFC 链接。
Pull the Code:
Try running:
git fetch origin pull/1159/head:train-clean-refactor && git checkout train-clean-refactorOr:
git clone git@github.com:FoundationResearch/FastVideo.git && git checkout train-clean-refactorRun Training:
- Edit lines 39-40 in
examples/train/run.shto point to your own Conda environment path (or simply delete those two lines).
- (Optional) If you want to run a Finetune:
bash examples/training/finetune/wan_t2v_1.3B/crush_smol/download_dataset.shPS: You don't need to wait for the full download to finish — Ctrl-C halfway through is fine. You'll end up with a subset of the dataset, which is enough.
- Run:
WANDB_MODE=online WANDB_API_KEY='your wandb api key' \
bash examples/train/run.sh <run_yaml_path>Example:
WANDB_MODE=online FASTVIDEO_ATTENTION_BACKEND=VIDEO_SPARSE_ATTN \
WANDB_API_KEY='your key' \
bash examples/train/run.sh examples/train/finetune_wan2.1_t2v_1.3B_vsa_phase3.4_0.9sparsity.yamlAll new architecture training-related code lives in the fastvideo/train directory. Feedback is very welcome!
拉取代码：
尝试运行：
git fetch origin pull/1159/head:train-clean-refactor && git checkout train-clean-refactor或者：
git clone git@github.com:FoundationResearch/FastVideo.git && git checkout train-clean-refactor运行训练：
- 将
examples/train/run.sh文件中的第 39-40 行修改为你自己的 Conda 环境路径（或者你也可以直接删除这两行）。
- (可选) 如果你要跑 Finetune：
bash examples/training/finetune/wan_t2v_1.3B/crush_smol/download_dataset.shPS: 你不需要等他全下完，可以下到一半就 Ctrl-C 停掉，相当于有一个数据集的子集即可。
- 运行：
WANDB_MODE=online WANDB_API_KEY='你的 wandb api 密钥' \
bash examples/train/run.sh <run_yaml_path>示例：
WANDB_MODE=online FASTVIDEO_ATTENTION_BACKEND=VIDEO_SPARSE_ATTN \
WANDB_API_KEY='你的密钥' \
bash examples/train/run.sh examples/train/finetune_wan2.1_t2v_1.3B_vsa_phase3.4_0.9sparsity.yaml