first commit

Dockerfile

+FROM nvidia/cuda:12.8.0-cudnn-devel-ubuntu22.04 AS base
+
+WORKDIR /app
+
+ENV DEBIAN_FRONTEND=noninteractive
+ENV LANG=C.UTF-8
+ENV LC_ALL=C.UTF-8
+
+RUN sed -i 's|http://archive.ubuntu.com/ubuntu/|https://mirrors.tuna.tsinghua.edu.cn/ubuntu/|g' /etc/apt/sources.list && \
+    sed -i 's|http://security.ubuntu.com/ubuntu/|https://mirrors.tuna.tsinghua.edu.cn/ubuntu/|g' /etc/apt/sources.list
+
+RUN apt-get update && \
+    apt-get install -y vim tmux zip unzip wget git cmake build-essential software-properties-common curl libibverbs-dev ca-certificates iproute2 ffmpeg libsm6 libxext6 && \
+    add-apt-repository ppa:deadsnakes/ppa && \
+    apt-get update && \
+    apt-get install -y python3.11 python3.11-venv python3.11-dev python3-pip && \
+    apt-get clean && rm -rf /var/lib/apt/lists/*
+
+RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.11 1 && \
+    update-alternatives --install /usr/bin/python python /usr/bin/python3.11 1
+
+RUN pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
+
+RUN pip install packaging ninja
+
+RUN pip install vllm
+
+RUN pip install torch torchvision
+
+# FROM tmp-image AS base
+
+WORKDIR /workspace
+
+# download flash-attention source code
+# git clone https://github.com/Dao-AILab/flash-attention.git --recursive
+COPY flash-attention /workspace/flash-attention
+
+RUN cd flash-attention && pip install --no-cache-dir -v -e .
+
+RUN cd flash-attention/hopper && pip install --no-cache-dir -v -e .
+
+RUN pip install diffusers transformers tokenizers accelerate safetensors opencv-python numpy imageio imageio-ffmpeg einops loguru
+
+RUN pip install sgl-kernel
+
+
+# FROM registry.cn-sh-01.sensecore.cn/devsft-ccr-2/video-gen:25030702 AS base
+
+RUN pip install qtorch ftfy

README.md

-# lightx2v
+# lightx2v
\ No newline at end of file
+
+这是一个视频生成推理框架
+
+
+## 运行环境
+
+```
+# 内网镜像
+docker pull registry.cn-sh-01.sensecore.cn/devsft-ccr-2/video-gen:25031303
+
+docker run --gpus all -itd --ipc=host --name [name] -v /mnt:/mnt --entrypoint /bin/bash [image id]
+```
+
+## 运行方式
+
+```
+git clone https://gitlab.bj.sensetime.com/video-gen/lightx2v.git
+cd lightx2v
+
+# 修改运行脚本的参数
+bash run_hunyuan_t2v.sh
+```

examples/vae_trt/convert_trt.sh

+#!/bin/bash
+export PYTHONPATH="./":$PYTHONPATH
+# trtexec \
+#     --onnx="/mnt/nvme0/wq/project/sd/code/lightx2v/vae_decoder_hf_sim.onnx" \
+#     --saveEngine="./vae_decoder_hf_sim.engine" \
+#     --allowWeightStreaming \
+#     --stronglyTyped \
+#     --fp16 \
+#     --weightStreamingBudget=100 \
+#     --minShapes=inp:1x16x9x18x16 \
+#     --optShapes=inp:1x16x17x32x16 \
+#     --maxShapes=inp:1x16x17x32x32
+
+python examples/vae_trt/convert_vae_trt_engine.py --model_path "/mnt/nvme1/yongyang/models/hy/ckpts"
\ No newline at end of file

examples/vae_trt/convert_vae_trt_engine.py

+from pathlib import Path
+import os
+import argparse
+
+import torch
+from loguru import logger
+
+from lightx2v.text2v.models.video_encoders.hf.autoencoder_kl_causal_3d.autoencoder_kl_causal_3d import AutoencoderKLCausal3D
+from lightx2v.text2v.models.video_encoders.trt.autoencoder_kl_causal_3d.trt_vae_infer import HyVaeTrtModelInfer
+
+
+def parse_args():
+    args = argparse.ArgumentParser()
+    args.add_argument("--model_path", help="", type=str)
+    args.add_argument("--dtype", default=torch.float16)
+    args.add_argument("--device", default="cuda", type=str)
+    return args.parse_args()
+
+
+def convert_vae_trt_engine(args):
+    vae_path = os.path.join(args.model_path, 'hunyuan-video-t2v-720p/vae')
+    assert Path(vae_path).exists(), f"{vae_path} not exists."
+    config = AutoencoderKLCausal3D.load_config(vae_path)
+    model = AutoencoderKLCausal3D.from_config(config)
+    assert Path(os.path.join(vae_path, 'pytorch_model.pt')).exists(), f"{os.path.join(vae_path, 'pytorch_model.pt')} not exists."
+    ckpt = torch.load(os.path.join(vae_path, 'pytorch_model.pt'), map_location='cpu', weights_only=True)
+    model.load_state_dict(ckpt)
+    model = model.to(dtype=args.dtype, device=args.device)
+    onnx_path = HyVaeTrtModelInfer.export_to_onnx(model.decoder, vae_path)
+    del model
+    torch.cuda.empty_cache()
+    engine_path = onnx_path.replace(".onnx", ".engine")
+    HyVaeTrtModelInfer.convert_to_trt_engine(onnx_path, engine_path)
+    logger.info(f"ONNX: {onnx_path}")
+    logger.info(f"TRT Engine: {engine_path}")
+    return
+
+
+def main():
+    args = parse_args()
+    convert_vae_trt_engine(args)
+
+
+if __name__ == "__main__":
+    main()

lightx2v/attentions/__init__.py

+from lightx2v.attentions.common.torch_sdpa import torch_sdpa
+from lightx2v.attentions.common.flash_attn2 import flash_attn2
+from lightx2v.attentions.common.flash_attn3 import flash_attn3
+from lightx2v.attentions.common.sage_attn2 import sage_attn2
+
+def attention(
+    attention_type="flash_attn2",
+    *args, **kwargs
+):
+    if attention_type == "torch_sdpa":
+        return torch_sdpa(*args, **kwargs)
+    elif attention_type == "flash_attn2":
+        return flash_attn2(*args, **kwargs)
+    elif attention_type == "flash_attn3":
+        return flash_attn3(*args, **kwargs)
+    elif attention_type == 'sage_attn2':
+        return sage_attn2(*args, **kwargs)
+    else:
+        raise NotImplementedError(f"Unsupported attention mode: {attention_type}")

lightx2v/attentions/common/flash_attn2.py

+try:
+    from flash_attn.flash_attn_interface import flash_attn_varlen_func
+except ImportError:
+    flash_attn_varlen_func = None
+
+def flash_attn2(
+    q,
+    k,
+    v,
+    cu_seqlens_q=None,
+    cu_seqlens_kv=None,
+    max_seqlen_q=None,
+    max_seqlen_kv=None
+):
+    x = flash_attn_varlen_func(
+        q,
+        k,
+        v,
+        cu_seqlens_q,
+        cu_seqlens_kv,
+        max_seqlen_q,
+        max_seqlen_kv,
+    ).reshape(max_seqlen_q, -1)
+    return x

lightx2v/attentions/common/flash_attn3.py

+try:
+    from flash_attn_interface import flash_attn_varlen_func as flash_attn_varlen_func_v3
+except ImportError:
+    flash_attn_varlen_func_v3 = None
+
+def flash_attn3(
+    q,
+    k,
+    v,
+    cu_seqlens_q=None,
+    cu_seqlens_kv=None,
+    max_seqlen_q=None,
+    max_seqlen_kv=None
+):
+    x = flash_attn_varlen_func_v3(
+        q,
+        k,
+        v,
+        cu_seqlens_q,
+        cu_seqlens_kv,
+        max_seqlen_q,
+        max_seqlen_kv,
+    )[0].reshape(max_seqlen_q, -1)
+    return x

lightx2v/attentions/common/sage_attn2.py

+import torch
+try:
+    from sageattention import sageattn
+except ImportError:
+    sageattn = None
+
+
+def sage_attn2(
+    q,
+    k,
+    v,
+    cu_seqlens_q=None,
+    cu_seqlens_kv=None,
+    max_seqlen_q=None,
+    max_seqlen_kv=None,
+):
+    q, k, v = (
+        q.transpose(1, 0).contiguous(),
+        k.transpose(1, 0).contiguous(),
+        v.transpose(1, 0).contiguous(),
+    )
+    x1 = sageattn(
+        q[:, : cu_seqlens_q[1], :].unsqueeze(0),
+        k[:, : cu_seqlens_q[1], :].unsqueeze(0),
+        v[:, : cu_seqlens_kv[1], :].unsqueeze(0),
+    )
+    x2 = sageattn(
+        q[:, cu_seqlens_q[1] :, :].unsqueeze(0),
+        k[:, cu_seqlens_kv[1] :, :].unsqueeze(0),
+        v[:, cu_seqlens_kv[1] :, :].unsqueeze(0),
+    )
+    x = torch.cat((x1, x2), dim=-2).transpose(2, 1).contiguous()
+    x = x.view(max_seqlen_q, -1)
+    return x

lightx2v/attentions/common/torch_sdpa.py

+import torch
+import torch.nn.functional as F
+
+
+def torch_sdpa(
+    q,
+    k,
+    v,
+    drop_rate=0,
+    attn_mask=None,
+    causal=False,
+):
+    q = q.transpose(1, 2)
+    k = k.transpose(1, 2)
+    v = v.transpose(1, 2)
+    if attn_mask is not None and attn_mask.dtype != torch.bool:
+        attn_mask = attn_mask.to(q.dtype)
+    x = F.scaled_dot_product_attention(
+        q, k, v, attn_mask=attn_mask, dropout_p=drop_rate, is_causal=causal
+    )
+    x = x.transpose(1, 2)
+    b, s, a, d = x.shape
+    out = x.reshape(b, s, -1)
+    return out

lightx2v/attentions/distributed/comm/all2all.py

+import torch
+import torch.distributed as dist
+
+
+def all2all_seq2head(input):
+    '''
+    将输入张量从 [seq_len/N, heads, hidden_dims] 转换为 [seq_len, heads/N, hidden_dims] 的格式。
+
+    参数:
+        input (torch.Tensor): 输入张量，形状为 [seq_len/N, heads, hidden_dims]
+
+    返回:
+        torch.Tensor: 转换后的输出张量，形状为 [seq_len, heads/N, hidden_dims]
+    '''
+    # 确保输入是一个3D张量
+    assert (input.dim() == 3), f"input must be 3D tensor"
+
+    # 获取当前进程的世界大小
+    world_size = dist.get_world_size()
+
+    # 获取输入张量的形状
+    shard_seq_len, heads, hidden_dims = input.shape
+    seq_len = shard_seq_len * world_size  # 计算总序列长度
+    shard_heads = heads // world_size  # 计算每个进程处理的头数
+
+    # 重塑输入张量以便进行 all-to-all 操作
+    input_t = (
+        input.reshape(shard_seq_len, world_size, shard_heads, hidden_dims)  # 重塑为 [shard_seq_len, world_size, shard_heads, hidden_dims]
+        .transpose(0, 1)  # 转置以便进行 all-to-all 操作
+        .contiguous()  # 确保内存连续
+    )
+
+    # 创建一个与输入张量相同形状的输出张量
+    output = torch.empty_like(input_t)
+
+    # 执行 all-to-all 操作，将输入张量的内容分发到所有进程
+    dist.all_to_all_single(output, input_t)
+
+    # 重塑输出张量为 [seq_len, heads/N, hidden_dims] 形状
+    output = output.reshape(seq_len, shard_heads, hidden_dims).contiguous()
+
+    return output  # 返回转换后的输出张量
+
+
+def all2all_head2seq(input):
+    '''
+    将输入张量从 [seq_len, heads/N, hidden_dims] 转换为 [seq_len/N, heads, hidden_dims] 的格式。
+
+    参数:
+        input (torch.Tensor): 输入张量，形状为 [seq_len, heads/N, hidden_dims]
+
+    返回:
+        torch.Tensor: 转换后的输出张量，形状为 [seq_len/N, heads, hidden_dims]
+    '''
+    # 确保输入是一个3D张量
+    assert (input.dim() == 3), f"input must be 3D tensor"
+
+    # 获取当前进程的世界大小
+    world_size = dist.get_world_size()
+
+    # 获取输入张量的形状
+    seq_len, shard_heads, hidden_dims = input.shape
+    heads = shard_heads * world_size  # 计算总头数
+    shard_seq_len = seq_len // world_size  # 计算每个进程处理的序列长度
+
+    # 重塑输入张量以便进行 all-to-all 操作
+    input_t = (
+        input.reshape(world_size, shard_seq_len, shard_heads, hidden_dims)  # 重塑为 [world_size, shard_seq_len, shard_heads, hidden_dims]
+        .transpose(1, 2)  # 转置以便进行 all-to-all 操作
+        .contiguous()  # 确保内存连续
+        .reshape(world_size, shard_heads, shard_seq_len, hidden_dims)  # 再次重塑为 [world_size, shard_heads, shard_seq_len, hidden_dims]
+    )
+
+    # 创建一个与输入张量相同形状的输出张量
+    output = torch.empty_like(input_t)
+
+    # 执行 all-to-all 操作，将输入张量的内容分发到所有进程
+    dist.all_to_all_single(output, input_t)
+
+    # 重塑输出张量为 [heads, shard_seq_len, hidden_dims] 形状
+    output = output.reshape(heads, shard_seq_len, hidden_dims)
+
+    # 转置输出张量并重塑为 [shard_seq_len, heads, hidden_dims] 形状
+    output = output.transpose(0, 1).contiguous().reshape(shard_seq_len, heads, hidden_dims)
+
+    return output  # 返回转换后的输出张量
+

lightx2v/attentions/distributed/partial_heads_attn/attn.py

+import torch
+import torch.distributed as dist
+from lightx2v.attentions import attention
+
+
+def partial_heads_attn(attention_type, q, k, v, cu_seqlens_qkv, max_seqlen_qkv):
+    num_heads = q.shape[-2]
+    cur_rank = dist.get_rank()
+    world_size = dist.get_world_size()
+    num_chunk_heads = int(num_heads / dist.get_world_size())
+
+    if cur_rank == world_size-1:
+        q = q[:, num_chunk_heads*cur_rank:, :]
+        k = k[:, num_chunk_heads*cur_rank:, :]
+        v = v[:, num_chunk_heads*cur_rank:, :]
+    else:
+        q = q[:, num_chunk_heads*cur_rank:num_chunk_heads*(cur_rank+1), :]
+        k = k[:, num_chunk_heads*cur_rank:num_chunk_heads*(cur_rank+1), :]
+        v = v[:, num_chunk_heads*cur_rank:num_chunk_heads*(cur_rank+1), :]
+
+    output = attention(
+        attention_type=attention_type,
+        q=q,
+        k=k,
+        v=v,
+        cu_seqlens_q=cu_seqlens_qkv,
+        cu_seqlens_kv=cu_seqlens_qkv,
+        max_seqlen_q=max_seqlen_qkv,
+        max_seqlen_kv=max_seqlen_qkv,
+    )
+
+    gathered_outputs = [torch.empty_like(output) for _ in range(world_size)]
+    dist.all_gather(gathered_outputs, output)
+
+    combined_output = torch.cat(gathered_outputs, dim=1)
+
+    return combined_output
\ No newline at end of file

lightx2v/attentions/distributed/partial_heads_attn/tests/test.sh

+export PYTHONPATH=/workspace/lightx2v:$PYTHONPATH
+export CUDA_VISIBLE_DEVICES=0,1
+torchrun --nproc_per_node=2 test_acc.py
\ No newline at end of file