flash-attention-v100

KO | EN

gitlite — search

#ai-agent #python #dsh-plugin #mcp-server #agentic-ai #automation #genpark #llm-tools #deepseek-harness #ai-agents #dsh #claude-code

flash-attention-v100

★ 59 Open GitHub ↗

forked from vllm-project/flash-attention

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

### FlashAttention V100 (SM70) 移植版

这个项目把 FlashAttention-2 移植到了 NVIDIA V100 (SM70) 上。目前前向性能已经比 `TRITON_ATTN` 快了，在 `q8k` 场景下快了 580%，而且直接能在 `vllm` 里跑起来。

![V100 vLLM Screenshot](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/attention_benchmark_results.png)

![Qwen3.5-27B q8k Performance](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/Perf_qwen3.6-27b_c1_2026-05-08_17_38_20.png)

## 使用方法

```bash
git clone https://github.com/zhinianqin/flash-attention-v100.git
cd flash-attention-v100

uv venv --python 3.12
source .venv/bin/activate
uv pip install "cmake>=3.26.1" ninja "packaging>=24.2" "setuptools>=77.0.3,<81.0.0" wheel jinja2
uv pip install torch==2.10.0 --index-url https://download.pytorch.org/whl/cu128

# 安装 vllm
uv pip install vllm==0.19.1 --torch-backend=cu128

# 构建安装
./build.sh
```

安装完成后，需要修改 vllm 的 `flash_attn` backend，让它支持 SM70：

```bash
vi /path/to/.venv/lib/python3.12/site-packages/vllm/v1/attention/backends/flash_attn.py
```

找到 `supports_compute_capability` 方法，把 SM 版本检查从 8.0 改为 7.0：

```python
    @classmethod
    def supports_compute_capability(cls, capability: DeviceCapability) -> bool:
-        return capability >= DeviceCapability(8, 0)
+        return capability >= DeviceCapability(7, 0)
```

然后正常启动 vllm 即可，默认会使用 FLASH_ATTN。

### FlashAttention V100 (SM70) Port

This project ports FlashAttention-2 to NVIDIA V100 (SM70). The forward path is already faster than `TRITON_ATTN`, shows a 580% improvement in the `q8k` case, and can run directly in `vllm`.

## Usage

```bash
git clone https://github.com/zhinianqin/flash-attention-v100.git
cd flash-attention-v100

# Install vllm
uv pip install vllm==0.19.1 --torch-backend=cu128

# Build and install
./build.sh
```

After installation, patch vllm's `flash_attn` backend to support SM70:

```bash
vi /path/to/.venv/lib/python3.12/site-packages/vllm/v1/attention/backends/flash_attn.py
```

Find the `supports_compute_capability` method and change the SM version check from 8.0 to 7.0:

Then start vllm normally. It will use FLASH_ATTN by default.

### 1) 构建流程

#### 推荐方式（项目内统一方式）
```bash
./build.sh
```

`build.sh` 已包含以下关键设置：
- 激活 `.venv` 虚拟环境
- 使用 `CUDA_HOME=/usr/local/cuda-12.8`
- 开启 `ccache`，并设置上限 `100G`
- 使用 `uv pip install --no-build-isolation . -v` 进行本地安装

### 2) 测试流程

#### 通用回归入口
```bash
./test.sh
```

#### Dense 测试入口
```bash
./test_dense.sh
.venv/bin/python tests/test_dense.py
.venv/bin/python -m pytest -q tests/test_dense.py
```

#### 常用定点测试（建议用于 debug）
```bash
CASE_IDS=119 ./test_dense.sh
CASE_IDS=183,184 ./test_dense.sh
HEAD_DIMS=96 CASE_IDS=183 .venv/bin/python tests/test_dense.py
HEAD_DIMS=32,64 .venv/bin/python -m pytest -q tests/test_dense.py
```

#### 测试说明
- `./test.sh` 是仓库级的基础 pytest 入口，当前实际执行的是 `tests/test_flash_attn.py`。
- 当你想快速确认通用 FlashAttention 行为没有回退时，优先使用 `./test.sh`。
- 当你在排查 V100 / SM70 分支上的 dense 数值、split-kv 或不同 `head_dim` 组合问题时，使用 `test_dense.sh` / `tests/test_dense.py`。
- `tests/test_dense.py` 现在同时支持脚本入口和 `pytest` 参数化入口，二者共用同一套执行逻辑。
- `dense` 默认会对主测试矩阵覆盖 `head_dim = 32, 64, 96, 128, 192, 256`，并保留 smoke 回归与 split-kv 路径校验。
- `HEAD_DIMS` 可用于选择 `head_dim` 子集，例如 `HEAD_DIMS=96` 或 `HEAD_DIMS=32,64,128`。
- `CASE_IDS` 仍用于复现最小失败集，过滤顺序为先筛 case，再与 `HEAD_DIMS` 做组合。
- 调试阶段建议先跑最小失败集（`CASE_IDS` + 必要的 `HEAD_DIMS`），确认后再回归全量。
- 如需只检查收集结果，可执行 `.venv/bin/python -m pytest --collect-only -q tests/test_dense.py`。

#### Dense test notes (English)
- `./test.sh` is the lightweight repository-level regression entrypoint and currently runs `tests/test_flash_attn.py`.
- Use `./test.sh` when you want a quick sanity check for the general FlashAttention path.
- Use `test_dense.sh` / `tests/test_dense.py` when you specifically need V100 / SM70 dense validation, split-kv checks, or multi-`head_dim` coverage.
- `tests/test_dense.py` now supports both the script entrypoint and parametrized `pytest` execution, backed by the same implementation logic.
- By default, the dense matrix runs across `head_dim = 32, 64, 96, 128, 192, 256`, while still covering the smoke regression and split-kv path checks.
- Use `HEAD_DIMS` to select a subset of head dimensions, for example `HEAD_DIMS=96` or `HEAD_DIMS=32,64,128`.
- `CASE_IDS` still reproduces the minimal failing case set; filtering happens on cases first, then crosses with the selected `HEAD_DIMS`.
- For debug loops, start with a small `CASE_IDS` / `HEAD_DIMS` slice and then expand to the full regression.
- To inspect collection only, run `.venv/bin/python -m pytest --collect-only -q tests/test_dense.py`.

### 3) 项目依赖

### 硬件 / 系统
- NVIDIA Tesla V100（计算能力 `sm_70`）
- Linux 环境

### CUDA / 编译工具
- CUDA 12.8（路径约定：`/usr/local/cuda-12.8`）
- `ccache`
- `cmake >= 3.26.1`
- `ninja`

### Python / 包依赖
- Python 虚拟环境：`.venv`
- `uv`
- `torch == 2.10.0`
- `packaging >= 24.2`
- `setuptools >= 77.0.3, < 81.0.0`
- `wheel`, `jinja2`

### 4) 测试状态

以下为最近一次完整回归（2026-03-15）结果快照：

**所有测试用例均已通过！**

- `dense` 测试：全部通过
- `splitkv` 测试：`60/60 PASS`

项目已完成在 V100 (SM70) 上的 FA2 完整功能验证。

### 5) Dense Test Coverage Notes (English)

For the V100 / SM70 branch, dense regression is maintained in `tests/test_dense.py`.

- Script mode: `.venv/bin/python tests/test_dense.py` or `./test_dense.sh`
- Pytest mode: `.venv/bin/python -m pytest -q tests/test_dense.py`
- Default coverage: every selected dense case is exercised with `head_dim = 32, 64, 96, 128, 192, 256`
- Common selectors:
  - `HEAD_DIMS=96`
  - `HEAD_DIMS=32,64,128`
  - `CASE_IDS=183`
  - `DENSE_SUITE=numerical`
- Example:

```bash
HEAD_DIMS=96 CASE_IDS=183 .venv/bin/python -m pytest -q tests/test_dense.py
```

This repository provides the official implementation of FlashAttention and
FlashAttention-2 from the
following papers.

**FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness**  
Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher Ré  
Paper: https://arxiv.org/abs/2205.14135  
IEEE Spectrum [article](https://spectrum.ieee.org/mlperf-rankings-2022) about our submission to the MLPerf 2.0 benchmark using FlashAttention.
![FlashAttention](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/flashattn_banner.jpg)

**FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning**  
Tri Dao

Paper: https://tridao.me/publications/flash2/flash2.pdf

![FlashAttention-2](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/flashattention_logo.png)

## Usage

We've been very happy to see FlashAttention being widely adopted in such a short
time after its release. This [page](https://github.com/Dao-AILab/flash-attention/blob/main/usage.md)
contains a partial list of places where FlashAttention is being used.

FlashAttention and FlashAttention-2 are free to use and modify (see LICENSE).
Please cite and credit FlashAttention if you use it.

## FlashAttention-3 beta release
FlashAttention-3 is optimized for Hopper GPUs (e.g. H100).

Blogpost: https://tridao.me/blog/2024/flash3/

Paper: https://tridao.me/publications/flash3/flash3.pdf

![FlashAttention-3 speedup on H100 80GB SXM5 with FP16](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/flash3_fp16_fwd.png)

This is a beta release for testing / benchmarking before we integrate that with
the rest of the repo.

Currently released:
- FP16 / BF16 forward and backward, FP8 forward

Requirements: H100 / H800 GPU, CUDA >= 12.3.

We highly recommend CUDA 12.8 for best performance.

To install:
```sh
cd hopper
python setup.py install
```
To run the test:
```sh
export PYTHONPATH=$PWD
pytest -q -s test_flash_attn.py
```
Once the package is installed, you can import it as follows:
```python
import flash_attn_interface
flash_attn_interface.flash_attn_func()
```

## Installation and features
**Requirements:**
- CUDA toolkit or ROCm toolkit
- PyTorch 2.2 and above.
- `packaging` Python package (`pip install packaging`)
- `psutil` Python package (`pip install psutil`)
- `ninja` Python package (`pip install ninja`) *
- Linux. Might work for Windows starting v2.3.2 (we've seen a few positive [reports](https://github.com/Dao-AILab/flash-attention/issues/595)) but Windows compilation still requires more testing. If you have ideas on how to set up prebuilt CUDA wheels for Windows, please reach out via Github issue.

\* Make sure that `ninja` is installed and that it works correctly (e.g. `ninja
--version` then `echo $?` should return exit code 0). If not (sometimes `ninja
--version` then `echo $?` returns a nonzero exit code), uninstall then reinstall
`ninja` (`pip uninstall -y ninja && pip install ninja`). Without `ninja`,
compiling can take a very long time (2h) since it does not use multiple CPU
cores. With `ninja` compiling takes 3-5 minutes on a 64-core machine using CUDA toolkit.

**To install:**
```sh
pip install flash-attn --no-build-isolation
```
Alternatively you can compile from source:
```sh
python setup.py install
```

If your machine has less than 96GB of RAM and lots of CPU cores, `ninja` might
run too many parallel compilation jobs that could exhaust the amount of RAM. To
limit the number of parallel compilation jobs, you can set the environment
variable `MAX_JOBS`:
```sh
MAX_JOBS=4 pip install flash-attn --no-build-isolation
```

**Interface:** `src/flash_attention_interface.py`

### NVIDIA CUDA Support
**Requirements:**
- CUDA 12.0 and above.

We recommend the
[Pytorch](https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch)
container from Nvidia, which has all the required tools to install FlashAttention.

FlashAttention-2 with CUDA currently supports:
1. Ampere, Ada, or Hopper GPUs (e.g., A100, RTX 3090, RTX 4090, H100). Support for Turing
   GPUs (T4, RTX 2080) is coming soon, please use FlashAttention 1.x for Turing
   GPUs for now.
2. Datatype fp16 and bf16 (bf16 requires Ampere, Ada, or Hopper GPUs).
3. All head dimensions up to 256. ~~Head dim > 192 backward requires A100/A800 or H100/H800~~. Head dim 256 backward now works on consumer GPUs (if there's no dropout) as of flash-attn 2.5.5.

### AMD ROCm Support
ROCm version has two backends. There is [composable_kernel](https://github.com/ROCm/composable_kernel) (ck) which is the default backend and a [Triton](https://github.com/triton-lang/triton) backend. They provide an implementation of FlashAttention-2.

**Requirements:**
- ROCm 6.0 and above.

We recommend the
[Pytorch](https://hub.docker.com/r/rocm/pytorch)
container from ROCm, which has all the required tools to install FlashAttention.

#### Composable Kernel Backend
FlashAttention-2 ROCm CK backend currently supports:
1. MI200x, MI250x, MI300x, and MI355x GPUs.
2. Datatype fp16 and bf16
3. Both forward's and backward's head dimensions up to 256.

#### Triton Backend
The Triton implementation of [Flash Attention](https://tridao.me/publications/flash2/flash2.pdf) supports AMD's CDNA (MI200, MI300) and RDNA GPUs using fp16, bf16, and fp32 datatypes. It provides forward and backward passes with causal masking, variable sequence lengths, arbitrary Q/KV sequence lengths and head sizes, MQA/GQA, dropout, rotary embeddings, ALiBi, paged attention, and FP8 (via the Flash Attention v3 interface). Sliding window attention is currently a work in progress.

To install, first get PyTorch for ROCm from https://pytorch.org/get-started/locally/, then install Triton and Flash Attention:
```sh
pip install triton==3.5.1
cd flash-attention
FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" python setup.py install
```

To run the tests (note: full suite takes hours):
```sh
FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" pytest tests/test_flash_attn_triton_amd.py
```

For better performance, enable autotune with `FLASH_ATTENTION_TRITON_AMD_AUTOTUNE="TRUE"`.

Alternativly, if _not_ autotuning, `FLASH_ATTENTION_FWD_TRITON_AMD_CONFIG_JSON` may be used to set a single triton config overriding the hardcoded defaults for `attn_fwd`. E.g.
```sh
FLASH_ATTENTION_FWD_TRITON_AMD_CONFIG_JSON='{"BLOCK_M":128,"BLOCK_N":64,"waves_per_eu":1,"PRE_LOAD_V":false,"num_stages":1,"num_warps":8}'
```

For a quick start with Docker:
```dockerfile
FROM rocm/pytorch:latest

WORKDIR /workspace

# install triton
RUN pip install triton==3.5.1

# build flash attention with triton backend
RUN git clone https://github.com/Dao-AILab/flash-attention &&\ 
    cd flash-attention &&\
    FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" python setup.py install

# set working dir
WORKDIR /workspace/flash-attention

# set env variable to use triton backend
ENV FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
```

Build and run:
```sh
docker build -t flash-attn-triton .
docker run -it --network=host --user root --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ipc=host --shm-size 16G --device=/dev/kfd --device=/dev/dri flash-attn-triton
```

## How to use FlashAttention

The main functions implement scaled dot product attention (softmax(Q @ K^T *
softmax_scale) @ V):
```python
from flash_attn import flash_attn_qkvpacked_func, flash_attn_func
```

```python
flash_attn_qkvpacked_func(qkv, dropout_p=0.0, softmax_scale=None, causal=False,
                          window_size=(-1, -1), alibi_slopes=None, deterministic=False):
"""dropout_p should be set to 0.0 during evaluation
If Q, K, V are already stacked into 1 tensor, this function will be faster than
calling flash_attn_func on Q, K, V since the backward pass avoids explicit concatenation
of the gradients of Q, K, V.
If window_size != (-1, -1), implements sliding window local attention. Query at position i
will only attend to keys between [i - window_size[0], i + window_size[1]] inclusive.
Arguments:
    qkv: (batch_size, seqlen, 3, nheads, headdim)
    dropout_p: float. Dropout probability.
    softmax_scale: float. The scaling of QK^T before applying softmax.
        Default to 1 / sqrt(headdim).
    causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling).
    window_size: (left, right). If not (-1, -1), implements sliding window local attention.
    alibi_slopes: (nheads,) or (batch_size, nheads), fp32. A bias of (-alibi_slope * |i - j|) is added to
        the attention score of query i and key j.
    deterministic: bool. Whether to use the deterministic implementation of the backward pass,
        which is slightly slower and uses more memory. The forward pass is always deterministic.
Return:
    out: (batch_size, seqlen, nheads, headdim).
"""
```

```python
flash_attn_func(q, k, v, dropout_p=0.0, softmax_scale=None, causal=False,
                window_size=(-1, -1), alibi_slopes=None, deterministic=False):
"""dropout_p should be set to 0.0 during evaluation
Supports multi-query and grouped-query attention (MQA/GQA) by passing in KV with fewer heads
than Q. Note that the number of heads in Q must be divisible by the number of heads in KV.
For example, if Q has 6 heads and K, V have 2 heads, head 0, 1, 2 of Q will attention to head
0 of K, V, and head 3, 4, 5 of Q will attention to head 1 of K, V.
If window_size != (-1, -1), implements sliding window local attention. Query at position i
will only attend to keys between
[i + seqlen_k - seqlen_q - window_size[0], i + seqlen_k - seqlen_q + window_size[1]] inclusive.

Arguments:
    q: (batch_size, seqlen, nheads, headdim)
    k: (batch_size, seqlen, nheads_k, headdim)
    v: (batch_size, seqlen, nheads_k, headdim)
    dropout_p: float. Dropout probability.
    softmax_scale: float. The scaling of QK^T before applying softmax.
        Default to 1 / sqrt(headdim).
    causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling).
    window_size: (left, right). If not (-1, -1), implements sliding window local attention.
    alibi_slopes: (nheads,) or (batch_size, nheads), fp32. A bias of
        (-alibi_slope * |i + seqlen_k - seqlen_q - j|)
        is added to the attention score of query i and key j.
    deterministic: bool. Whether to use the deterministic implementation of the backward pass,
        which is slightly slower and uses more memory. The forward pass is always deterministic.
Return:
    out: (batch_size, seqlen, nheads, headdim).
"""
```

```python
def flash_attn_with_kvcache(
    q,
    k_cache,
    v_cache,
    k=None,
    v=None,
    rotary_cos=None,
    rotary_sin=None,
    cache_seqlens: Optional[Union[(int, torch.Tensor)]] = None,
    cache_batch_idx: Optional[torch.Tensor] = None,
    block_table: Optional[torch.Tensor] = None,
    softmax_scale=None,
    causal=False,
    window_size=(-1, -1),  # -1 means infinite context window
    rotary_interleaved=True,
    alibi_slopes=None,
):
    """
    If k and v are not None, k_cache and v_cache will be updated *inplace* with the new values from
    k and v. This is useful for incremental decoding: you can pass in the cached keys/values from
    the previous step, and update them with the new keys/values from the current step, and do
    attention with the updated cache, all in 1 kernel.

If you pass in k / v, you must make sure that the cache is large enough to hold the new values.
    For example, the KV cache could be pre-allocated with the max sequence length, and you can use
    cache_seqlens to keep track of the current sequence lengths of each sequence in the batch.

Also apply rotary embedding if rotary_cos and rotary_sin are passed in. The key @k will be
    rotated by rotary_cos and rotary_sin at indices cache_seqlens, cache_seqlens + 1, etc.
    If causal or local (i.e., window_size != (-1, -1)), the query @q will be rotated by rotary_cos
    and rotary_sin at indices cache_seqlens, cache_seqlens + 1, etc.
    If not causal and not local, the query @q will be rotated by rotary_cos and rotary_sin at
    indices cache_seqlens only (i.e. we consider all tokens in @q to be at position cache_seqlens).

See tests/test_flash_attn.py::test_flash_attn_kvcache for examples of how to use this function.

Supports multi-query and grouped-query attention (MQA/GQA) by passing in KV with fewer heads
    than Q. Note that the number of heads in Q must be divisible by the number of heads in KV.
    For example, if Q has 6 heads and K, V have 2 heads, head 0, 1, 2 of Q will attention to head
    0 of K, V, and head 3, 4, 5 of Q will attention to head 1 of K, V.

If causal=True, the causal mask is aligned to the bottom right corner of the attention matrix.
    For example, if seqlen_q = 2 and seqlen_k = 5, the causal mask (1 = keep, 0 = masked out) is:
        1 1 1 1 0
        1 1 1 1 1
    If seqlen_q = 5 and seqlen_k = 2, the causal mask is:
        0 0
        0 0
        0 0
        1 0
        1 1
    If the row of the mask is all zero, the output will be zero.

If window_size != (-1, -1), implements sliding window local attention. Query at position i
    will only attend to keys between
    [i + seqlen_k - seqlen_q - window_size[0], i + seqlen_k - seqlen_q + window_size[1]] inclusive.

Note: Does not support backward pass.

Arguments:
        q: (batch_size, seqlen, nheads, headdim)
        k_cache: (batch_size_cache, seqlen_cache, nheads_k, headdim) if there's no block_table,
            or (num_blocks, page_block_size, nheads_k, headdim) if there's a block_table (i.e. paged KV cache)
            page_block_size must be a multiple of 256.
        v_cache: (batch_size_cache, seqlen_cache, nheads_k, headdim) if there's no block_table,
            or (num_blocks, page_block_size, nheads_k, headdim) if there's a block_table (i.e. paged KV cache)
        k [optional]: (batch_size, seqlen_new, nheads_k, headdim). If not None, we concatenate
            k with k_cache, starting at the indices specified by cache_seqlens.
        v [optional]: (batch_size, seqlen_new, nheads_k, headdim). Similar to k.
        rotary_cos [optional]: (seqlen_ro, rotary_dim / 2). If not None, we apply rotary embedding
            to k and q. Only applicable if k and v are passed in. rotary_dim must be divisible by 16.
        rotary_sin [optional]: (seqlen_ro, rotary_dim / 2). Similar to rotary_cos.
        cache_seqlens: int, or (batch_size,), dtype torch.int32. The sequence lengths of the
            KV cache.
        block_table [optional]: (batch_size, max_num_blocks_per_seq), dtype torch.int32.
        cache_batch_idx: (batch_size,), dtype torch.int32. The indices used to index into the KV cache.
            If None, we assume that the batch indices are [0, 1, 2, ..., batch_size - 1].
            If the indices are not distinct, and k and v are provided, the values updated in the cache
                 might come from any of the duplicate indices.
        softmax_scale: float. The scaling of QK^T before applying softmax.
            Default to 1 / sqrt(headdim).
        causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling).
        window_size: (left, right). If not (-1, -1), implements sliding window local attention.
        rotary_interleaved: bool. Only applicable if rotary_cos and rotary_sin are passed in.
            If True, rotary embedding will combine dimensions 0 & 1, 2 & 3, etc. If False,
            rotary embedding will combine dimensions 0 & rotary_dim / 2, 1 & rotary_dim / 2 + 1
            (i.e. GPT-NeoX style).
        alibi_slopes: (nheads,) or (batch_size, nheads), fp32. A bias of
            (-alibi_slope * |i + seqlen_k - seqlen_q - j|)
            is added to the attention score of query i and key j.

Return:
        out: (batch_size, seqlen, nheads, headdim).
    """
```

To see how these functions are used in a multi-head attention layer (which
includes QKV projection, output projection), see the MHA [implementation](https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/modules/mha.py).

## Changelog

### 2.0: Complete rewrite, 2x faster
Upgrading from FlashAttention (1.x) to FlashAttention-2

These functions have been renamed:
- `flash_attn_unpadded_func` -> `flash_attn_varlen_func`
- `flash_attn_unpadded_qkvpacked_func` -> `flash_attn_varlen_qkvpacked_func`
- `flash_attn_unpadded_kvpacked_func` -> `flash_attn_varlen_kvpacked_func`

If the inputs have the same sequence lengths in the same batch, it is simpler
and faster to use these functions:
```python
flash_attn_qkvpacked_func(qkv, dropout_p=0.0, softmax_scale=None, causal=False)
```
```python
flash_attn_func(q, k, v, dropout_p=0.0, softmax_scale=None, causal=False)
```
### 2.1: Change behavior of causal flag

If seqlen_q != seqlen_k and causal=True, the causal mask is aligned to the
bottom right corner of the attention matrix, instead of the top-left corner.

For example, if seqlen_q = 2 and seqlen_k = 5, the causal mask (1 = keep, 0 =
masked out) is:  
v2.0:  
    1 0 0 0 0  
    1 1 0 0 0  
v2.1:  
    1 1 1 1 0  
    1 1 1 1 1

If seqlen_q = 5 and seqlen_k = 2, the causal mask is:  
v2.0:  
    1 0  
    1 1  
    1 1  
    1 1  
    1 1  
v2.1:  
    0 0  
    0 0  
    0 0  
    1 0  
    1 1  
If the row of the mask is all zero, the output will be zero.

### 2.2: Optimize for inference

Optimize for inference (iterative decoding) when query has very small sequence
length (e.g., query sequence length = 1). The bottleneck here is to load KV
cache as fast as possible, and we split the loading across different thread
blocks, with a separate kernel to combine results.

See the function `flash_attn_with_kvcache` with more features for inference
(perform rotary embedding, updating KV cache inplace).

Thanks to the xformers team, and in particular Daniel Haziza, for this
collaboration.

### 2.3: Local (i.e., sliding window) attention

Implement sliding window attention (i.e., local attention). Thanks to [Mistral
AI](https://mistral.ai/) and in particular Timothée Lacroix for this
contribution. Sliding window was used in the [Mistral 7B](https://mistral.ai/news/announcing-mistral-7b/) model.

### 2.4: ALiBi (attention with linear bias), deterministic backward pass.

Implement ALiBi (Press et al., 2021). Thanks to Sanghun Cho from Kakao Brain for this contribution.

Implement deterministic backward pass. Thanks to engineers from [Meituan](www.meituan.com) for this contribution.

### 2.5: Paged KV cache.

Support paged KV cache (i.e., [PagedAttention](https://arxiv.org/abs/2309.06180)).
Thanks to @beginlner for this contribution.

### 2.6: Softcapping.

Support attention with softcapping, as used in Gemma-2 and Grok models.
Thanks to @Narsil and @lucidrains for this contribution.

### 2.7: Compatibility with torch compile

Thanks to @ani300 for this contribution.

## Performance

We present expected speedup (combined forward + backward pass) and memory savings from using FlashAttention against PyTorch standard attention, depending on sequence length, on different GPUs (speedup depends on memory bandwidth - we see more speedup on slower GPU memory).

We currently have benchmarks for these GPUs:
* [A100](#a100)
* [H100](#h100)

### A100

We display FlashAttention speedup using these parameters:
* Head dimension 64 or 128, hidden dimension 2048 (i.e. either 32 or 16 heads).
* Sequence length 512, 1k, 2k, 4k, 8k, 16k.
* Batch size set to 16k / seqlen.

#### Speedup

![FlashAttention speedup on A100 80GB SXM5 with FP16/BF16](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/flash2_a100_fwd_bwd_benchmark.png)

#### Memory

![FlashAttention memory](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/flashattn_memory.jpg)

We show memory savings in this graph (note that memory footprint is the same no matter if you use dropout or masking).
Memory savings are proportional to sequence length -- since standard attention has memory quadratic in sequence length, whereas FlashAttention has memory linear in sequence length.
We see 10X memory savings at sequence length 2K, and 20X at 4K.
As a result, FlashAttention can scale to much longer sequence lengths.

### H100

![FlashAttention speedup on H100 SXM5 with FP16/BF16](https://raw.githubusercontent.com/zhinianqin/flash-attention-v100/main/assets/flash2_h100_fwd_bwd_benchmark.png)

## Full model code and training script

We have released the full GPT model
[implementation](https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/models/gpt.py).
We also provide optimized implementations of other layers (e.g., MLP, LayerNorm,
cross-entropy loss, rotary embedding). Overall this speeds up training by 3-5x
compared to the baseline implementation from Huggingface, reaching up to 225
TFLOPs/sec per A100, equivalent to 72% model FLOPs utilization (we don't need
any activation checkpointing).

We also include a training
[script](https://github.com/Dao-AILab/flash-attention/tree/main/training) to
train GPT2 on Openwebtext and GPT3 on The Pile.

## Triton implementation of FlashAttention

Phil Tillet (OpenAI) has an experimental implementation of FlashAttention in Triton:
https://github.com/openai/triton/blob/master/python/tutorials/06-fused-attention.py

As Triton is a higher-level language than CUDA, it might be easier to understand
and experiment with. The notations in the Triton implementation are also closer
to what's used in our paper.

We also have an experimental implementation in Triton that support attention
bias (e.g. ALiBi):
https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/flash_attn_triton.py

## Tests
We test that FlashAttention produces the same output and gradient as a reference
implementation, up to some numerical tolerance. In particular, we check that the
maximum numerical error of FlashAttention is at most twice the numerical error
of a baseline implementation in Pytorch (for different head dimensions, input
dtype, sequence length, causal / non-causal).

To run the tests:
```sh
pytest -q -s tests/test_flash_attn.py
```
## When you encounter issues

This new release of FlashAttention-2 has been tested on several GPT-style
models, mostly on A100 GPUs.

If you encounter bugs, please open a GitHub Issue!

## Tests
To run the tests:
```sh
pytest tests/test_flash_attn_ck.py
```

## Citation
If you use this codebase, or otherwise found our work valuable, please cite:
```
@inproceedings{dao2022flashattention,
  title={Flash{A}ttention: Fast and Memory-Efficient Exact Attention with {IO}-Awareness},
  author={Dao, Tri and Fu, Daniel Y. and Ermon, Stefano and Rudra, Atri and R{\'e}, Christopher},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2022}
}
@inproceedings{dao2023flashattention2,
  title={Flash{A}ttention-2: Faster Attention with Better Parallelism and Work Partitioning},
  author={Dao, Tri},
  booktitle={International Conference on Learning Representations (ICLR)},
  year={2024}
}
```

flash-attention-v100

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README