flashinfer-python 0.5.2__tar.gz → 0.5.3__tar.gz

flashinfer_python-0.5.3/PKG-INFO

@@ -0,0 +1,247 @@
+Metadata-Version: 2.4
+Name: flashinfer-python
+Version: 0.5.3
+Summary: FlashInfer: Kernel Library for LLM Serving
+Author: FlashInfer team
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/flashinfer-ai/flashinfer
+Requires-Python: <4.0,>=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: apache-tvm-ffi<0.2,>=0.1
+Requires-Dist: click
+Requires-Dist: einops
+Requires-Dist: ninja
+Requires-Dist: numpy
+Requires-Dist: nvidia-cudnn-frontend>=1.13.0
+Requires-Dist: nvidia-cutlass-dsl>=4.2.1
+Requires-Dist: nvidia-ml-py
+Requires-Dist: packaging>=24.2
+Requires-Dist: requests
+Requires-Dist: tabulate
+Requires-Dist: torch
+Requires-Dist: tqdm
+Dynamic: license-file
+
+<p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://github.com/flashinfer-ai/web-data/blob/main/logo/FlashInfer-black-background.png?raw=true">
+    <img alt="FlashInfer" src="https://github.com/flashinfer-ai/web-data/blob/main/logo/FlashInfer-white-background.png?raw=true" width=55%>
+  </picture>
+</p>
+<h1 align="center">
+Kernel Library for LLM Serving
+</h1>
+
+<p align="center">
+| <a href="https://flashinfer.ai"><b>Blog</b></a> | <a href="https://docs.flashinfer.ai"><b>Documentation</b></a> | <a href="https://join.slack.com/t/flashinfer/shared_invite/zt-379wct3hc-D5jR~1ZKQcU00WHsXhgvtA"><b>Slack</b></a> |  <a href="https://github.com/orgs/flashinfer-ai/discussions"><b>Discussion Forum</b></a> |
+</p>
+
+[![Build Status](https://ci.tlcpack.ai/job/flashinfer-ci/job/main/badge/icon)](https://ci.tlcpack.ai/job/flashinfer-ci/job/main/)
+[![Documentation](https://github.com/flashinfer-ai/flashinfer/actions/workflows/build-doc.yml/badge.svg)](https://github.com/flashinfer-ai/flashinfer/actions/workflows/build-doc.yml)
+
+FlashInfer is a library and kernel generator for Large Language Models that provides high-performance implementation of LLM GPU kernels such as FlashAttention, SparseAttention, PageAttention, Sampling, and more. FlashInfer focuses on LLM serving and inference, and delivers state-of-the-art performance across diverse scenarios.
+
+Check our [v0.2 release blog](https://flashinfer.ai/2024/12/16/flashinfer-v02-release.html) for new features!
+
+The core features of FlashInfer include:
+
+1. **Efficient Sparse/Dense Attention Kernels**: Efficient single/batch attention for sparse(paged)/dense KV-storage on CUDA Cores and Tensor Cores (both FA2 & FA3) templates. The vector-sparse attention can achieve 90% of the bandwidth of dense kernels with same problem size.
+2. **Load-Balanced Scheduling**: FlashInfer decouples `plan`/`run` stage of attention computation where we schedule the computation of variable-length inputs in `plan` stage to alleviate load-imbalance issue.
+3. **Memory Efficiency**: FlashInfer offers [Cascade Attention](https://docs.flashinfer.ai/api/cascade.html#flashinfer.cascade.MultiLevelCascadeAttentionWrapper) for hierarchical KV-Cache, and implements Head-Query fusion for accelerating Grouped-Query Attention, and efficient kernels for low-precision attention and fused-RoPE attention for compressed KV-Cache.
+4. **Customizable Attention**: Bring your own attention variants through JIT-compilation.
+5. **CUDAGraph and torch.compile Compatibility**: FlashInfer kernels can be captured by CUDAGraphs and torch.compile for low-latency inference.
+6. **Efficient LLM-specific Operators**: High-Performance [fused kernel for Top-P, Top-K/Min-P sampling](https://docs.flashinfer.ai/api/sampling.html) without the need to sorting.
+
+FlashInfer supports PyTorch, TVM and C++ (header-only) APIs, and can be easily integrated into existing projects.
+
+## News
+
+- [Mar 10, 2025] [Blog Post](https://flashinfer.ai/2025/03/10/sampling.html) Sorting-Free GPU Kernels for LLM Sampling, which explains the design of sampling kernels in FlashInfer.
+- [Mar 1, 2025] Checkout flashinfer's [intra-kernel profiler](https://github.com/flashinfer-ai/flashinfer/tree/main/profiler) for visualizing the timeline of each threadblock in GPU kernels.
+- [Dec 16, 2024] [Blog Post](https://flashinfer.ai/2024/12/16/flashinfer-v02-release.html) FlashInfer 0.2 - Efficient and Customizable Kernels for LLM Inference Serving
+- [Sept 2024] We've launched a [Slack](https://join.slack.com/t/flashinfer/shared_invite/zt-2r93kj2aq-wZnC2n_Z2~mf73N5qnVGGA) workspace for Flashinfer users and developers. Join us for timely support, discussions, updates and knowledge sharing!
+- [Jan 31, 2024] [Blog Post](https://flashinfer.ai/2024/01/08/cascade-inference.html) Cascade Inference: Memory-Efficient Shared Prefix Batch Decoding
+- [Jan 31, 2024] [Blog Post](https://flashinfer.ai/2024/01/03/introduce-flashinfer.html) Accelerating Self-Attentions for LLM Serving with FlashInfer
+
+## Getting Started
+
+Using our PyTorch API is the easiest way to get started:
+
+### Install from PyPI
+
+FlashInfer is available as a Python package for Linux. Install the core package with:
+
+```bash
+pip install flashinfer-python
+```
+
+**Package Options:**
+
+- **flashinfer-python**: Core package that compiles/downloads kernels on first use
+- **flashinfer-cubin**: Pre-compiled kernel binaries for all supported GPU architectures
+- **flashinfer-jit-cache**: Pre-built kernel cache for specific CUDA versions
+
+**For faster initialization and offline usage**, install the optional packages to have most kernels pre-compiled:
+
+```bash
+pip install flashinfer-python flashinfer-cubin
+# JIT cache package (replace cu129 with your CUDA version: cu128, cu129, or cu130)
+pip install flashinfer-jit-cache --index-url https://flashinfer.ai/whl/cu129
+```
+
+This eliminates compilation and downloading overhead at runtime.
+
+### Install from Source
+
+Build the core package from source:
+
+```bash
+git clone https://github.com/flashinfer-ai/flashinfer.git --recursive
+cd flashinfer
+python -m pip install -v .
+```
+
+**For development**, install in editable mode:
+
+```bash
+python -m pip install --no-build-isolation -e . -v
+```
+
+**Build optional packages:**
+
+`flashinfer-cubin`:
+
+```bash
+cd flashinfer-cubin
+python -m build --no-isolation --wheel
+python -m pip install dist/*.whl
+```
+
+`flashinfer-jit-cache` (customize `FLASHINFER_CUDA_ARCH_LIST` for your target GPUs):
+
+```bash
+export FLASHINFER_CUDA_ARCH_LIST="7.5 8.0 8.9 9.0a 10.0a 10.3a 11.0a 12.0f"
+cd flashinfer-jit-cache
+python -m build --no-isolation --wheel
+python -m pip install dist/*.whl
+```
+
+For more details, see the [Install from Source documentation](https://docs.flashinfer.ai/installation.html#install-from-source).
+
+### Install Nightly Build
+
+Nightly builds are available for testing the latest features:
+
+```bash
+# Core and cubin packages
+pip install -U --pre flashinfer-python --index-url https://flashinfer.ai/whl/nightly/ --no-deps # Install the nightly package from custom index, without installing dependencies
+pip install flashinfer-python  # Install flashinfer-python's dependencies from PyPI
+pip install -U --pre flashinfer-cubin --index-url https://flashinfer.ai/whl/nightly/
+# JIT cache package (replace cu129 with your CUDA version: cu128, cu129, or cu130)
+pip install -U --pre flashinfer-jit-cache --index-url https://flashinfer.ai/whl/nightly/cu129
+```
+
+### Verify Installation
+
+After installation, verify that FlashInfer is correctly installed and configured:
+
+```bash
+flashinfer show-config
+```
+
+This command displays:
+
+- FlashInfer version and installed packages (flashinfer-python, flashinfer-cubin, flashinfer-jit-cache)
+- PyTorch and CUDA version information
+- Environment variables and artifact paths
+- Downloaded cubin status and module compilation status
+
+### Trying it out
+
+Below is a minimal example of using FlashInfer's single-request decode/append/prefill attention kernels:
+
+```python
+import torch
+import flashinfer
+
+kv_len = 2048
+num_kv_heads = 32
+head_dim = 128
+
+k = torch.randn(kv_len, num_kv_heads, head_dim).half().to(0)
+v = torch.randn(kv_len, num_kv_heads, head_dim).half().to(0)
+
+# decode attention
+
+num_qo_heads = 32
+q = torch.randn(num_qo_heads, head_dim).half().to(0)
+
+o = flashinfer.single_decode_with_kv_cache(q, k, v) # decode attention without RoPE on-the-fly
+o_rope_on_the_fly = flashinfer.single_decode_with_kv_cache(q, k, v, pos_encoding_mode="ROPE_LLAMA") # decode with LLaMA style RoPE on-the-fly
+
+# append attention
+append_qo_len = 128
+q = torch.randn(append_qo_len, num_qo_heads, head_dim).half().to(0) # append attention, the last 128 tokens in the KV-Cache are the new tokens
+o = flashinfer.single_prefill_with_kv_cache(q, k, v, causal=True) # append attention without RoPE on-the-fly, apply causal mask
+o_rope_on_the_fly = flashinfer.single_prefill_with_kv_cache(q, k, v, causal=True, pos_encoding_mode="ROPE_LLAMA") # append attention with LLaMA style RoPE on-the-fly, apply causal mask
+
+# prefill attention
+qo_len = 2048
+q = torch.randn(qo_len, num_qo_heads, head_dim).half().to(0) # prefill attention
+o = flashinfer.single_prefill_with_kv_cache(q, k, v, causal=False) # prefill attention without RoPE on-the-fly, do not apply causal mask
+```
+
+Check out [documentation](https://docs.flashinfer.ai/) for usage of batch decode/append/prefill kernels and shared-prefix cascading kernels.
+
+## Custom Attention Variants
+
+Starting from FlashInfer v0.2, users can customize their own attention variants with additional parameters. For more details, refer to our [JIT examples](https://github.com/flashinfer-ai/flashinfer/blob/main/tests/utils/test_jit_example.py).
+
+## GPU Support
+
+FlashInfer currently provides support for NVIDIA SM architectures 75 and higher and beta support for 103, 110, 120, and 121.
+
+## Adoption
+
+We are thrilled to share that FlashInfer is being adopted by many cutting-edge projects, including but not limited to:
+
+- [MLC-LLM](https://github.com/mlc-ai/mlc-llm)
+- [Punica](https://github.com/punica-ai/punica)
+- [SGLang](https://github.com/sgl-project/sglang)
+- [ScaleLLM](https://github.com/vectorch-ai/ScaleLLM)
+- [vLLM](https://github.com/vllm-project/vllm)
+- [TGI](https://github.com/huggingface/text-generation-inference)
+- [lorax](https://github.com/predibase/lorax)
+- [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM)
+- [LightLLM](https://github.com/ModelTC/lightllm)
+
+## Acknowledgement
+
+FlashInfer is inspired by [FlashAttention 1&2](https://github.com/dao-AILab/flash-attention/), [vLLM](https://github.com/vllm-project/vllm), [stream-K](https://arxiv.org/abs/2301.03598), [cutlass](https://github.com/nvidia/cutlass) and [AITemplate](https://github.com/facebookincubator/AITemplate) projects.
+
+## Citation
+
+If you find FlashInfer helpful in your project or research, please consider citing our [paper](https://arxiv.org/abs/2501.01005):
+
+```bibtex
+@article{ye2025flashinfer,
+    title = {FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving},
+    author = {
+      Ye, Zihao and
+      Chen, Lequn and
+      Lai, Ruihang and
+      Lin, Wuwei and
+      Zhang, Yineng and
+      Wang, Stephanie and
+      Chen, Tianqi and
+      Kasikci, Baris and
+      Grover, Vinod and
+      Krishnamurthy, Arvind and
+      Ceze, Luis
+    },
+    journal = {arXiv preprint arXiv:2501.01005},
+    year = {2025},
+    url = {https://arxiv.org/abs/2501.01005}
+}
+```

flashinfer_python-0.5.3/README.md

@@ -0,0 +1,222 @@
+<p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://github.com/flashinfer-ai/web-data/blob/main/logo/FlashInfer-black-background.png?raw=true">
+    <img alt="FlashInfer" src="https://github.com/flashinfer-ai/web-data/blob/main/logo/FlashInfer-white-background.png?raw=true" width=55%>
+  </picture>
+</p>
+<h1 align="center">
+Kernel Library for LLM Serving
+</h1>
+
+<p align="center">
+| <a href="https://flashinfer.ai"><b>Blog</b></a> | <a href="https://docs.flashinfer.ai"><b>Documentation</b></a> | <a href="https://join.slack.com/t/flashinfer/shared_invite/zt-379wct3hc-D5jR~1ZKQcU00WHsXhgvtA"><b>Slack</b></a> |  <a href="https://github.com/orgs/flashinfer-ai/discussions"><b>Discussion Forum</b></a> |
+</p>
+
+[![Build Status](https://ci.tlcpack.ai/job/flashinfer-ci/job/main/badge/icon)](https://ci.tlcpack.ai/job/flashinfer-ci/job/main/)
+[![Documentation](https://github.com/flashinfer-ai/flashinfer/actions/workflows/build-doc.yml/badge.svg)](https://github.com/flashinfer-ai/flashinfer/actions/workflows/build-doc.yml)
+
+FlashInfer is a library and kernel generator for Large Language Models that provides high-performance implementation of LLM GPU kernels such as FlashAttention, SparseAttention, PageAttention, Sampling, and more. FlashInfer focuses on LLM serving and inference, and delivers state-of-the-art performance across diverse scenarios.
+
+Check our [v0.2 release blog](https://flashinfer.ai/2024/12/16/flashinfer-v02-release.html) for new features!
+
+The core features of FlashInfer include:
+
+1. **Efficient Sparse/Dense Attention Kernels**: Efficient single/batch attention for sparse(paged)/dense KV-storage on CUDA Cores and Tensor Cores (both FA2 & FA3) templates. The vector-sparse attention can achieve 90% of the bandwidth of dense kernels with same problem size.
+2. **Load-Balanced Scheduling**: FlashInfer decouples `plan`/`run` stage of attention computation where we schedule the computation of variable-length inputs in `plan` stage to alleviate load-imbalance issue.
+3. **Memory Efficiency**: FlashInfer offers [Cascade Attention](https://docs.flashinfer.ai/api/cascade.html#flashinfer.cascade.MultiLevelCascadeAttentionWrapper) for hierarchical KV-Cache, and implements Head-Query fusion for accelerating Grouped-Query Attention, and efficient kernels for low-precision attention and fused-RoPE attention for compressed KV-Cache.
+4. **Customizable Attention**: Bring your own attention variants through JIT-compilation.
+5. **CUDAGraph and torch.compile Compatibility**: FlashInfer kernels can be captured by CUDAGraphs and torch.compile for low-latency inference.
+6. **Efficient LLM-specific Operators**: High-Performance [fused kernel for Top-P, Top-K/Min-P sampling](https://docs.flashinfer.ai/api/sampling.html) without the need to sorting.
+
+FlashInfer supports PyTorch, TVM and C++ (header-only) APIs, and can be easily integrated into existing projects.
+
+## News
+
+- [Mar 10, 2025] [Blog Post](https://flashinfer.ai/2025/03/10/sampling.html) Sorting-Free GPU Kernels for LLM Sampling, which explains the design of sampling kernels in FlashInfer.
+- [Mar 1, 2025] Checkout flashinfer's [intra-kernel profiler](https://github.com/flashinfer-ai/flashinfer/tree/main/profiler) for visualizing the timeline of each threadblock in GPU kernels.
+- [Dec 16, 2024] [Blog Post](https://flashinfer.ai/2024/12/16/flashinfer-v02-release.html) FlashInfer 0.2 - Efficient and Customizable Kernels for LLM Inference Serving
+- [Sept 2024] We've launched a [Slack](https://join.slack.com/t/flashinfer/shared_invite/zt-2r93kj2aq-wZnC2n_Z2~mf73N5qnVGGA) workspace for Flashinfer users and developers. Join us for timely support, discussions, updates and knowledge sharing!
+- [Jan 31, 2024] [Blog Post](https://flashinfer.ai/2024/01/08/cascade-inference.html) Cascade Inference: Memory-Efficient Shared Prefix Batch Decoding
+- [Jan 31, 2024] [Blog Post](https://flashinfer.ai/2024/01/03/introduce-flashinfer.html) Accelerating Self-Attentions for LLM Serving with FlashInfer
+
+## Getting Started
+
+Using our PyTorch API is the easiest way to get started:
+
+### Install from PyPI
+
+FlashInfer is available as a Python package for Linux. Install the core package with:
+
+```bash
+pip install flashinfer-python
+```
+
+**Package Options:**
+
+- **flashinfer-python**: Core package that compiles/downloads kernels on first use
+- **flashinfer-cubin**: Pre-compiled kernel binaries for all supported GPU architectures
+- **flashinfer-jit-cache**: Pre-built kernel cache for specific CUDA versions
+
+**For faster initialization and offline usage**, install the optional packages to have most kernels pre-compiled:
+
+```bash
+pip install flashinfer-python flashinfer-cubin
+# JIT cache package (replace cu129 with your CUDA version: cu128, cu129, or cu130)
+pip install flashinfer-jit-cache --index-url https://flashinfer.ai/whl/cu129
+```
+
+This eliminates compilation and downloading overhead at runtime.
+
+### Install from Source
+
+Build the core package from source:
+
+```bash
+git clone https://github.com/flashinfer-ai/flashinfer.git --recursive
+cd flashinfer
+python -m pip install -v .
+```
+
+**For development**, install in editable mode:
+
+```bash
+python -m pip install --no-build-isolation -e . -v
+```
+
+**Build optional packages:**
+
+`flashinfer-cubin`:
+
+```bash
+cd flashinfer-cubin
+python -m build --no-isolation --wheel
+python -m pip install dist/*.whl
+```
+
+`flashinfer-jit-cache` (customize `FLASHINFER_CUDA_ARCH_LIST` for your target GPUs):
+
+```bash
+export FLASHINFER_CUDA_ARCH_LIST="7.5 8.0 8.9 9.0a 10.0a 10.3a 11.0a 12.0f"
+cd flashinfer-jit-cache
+python -m build --no-isolation --wheel
+python -m pip install dist/*.whl
+```
+
+For more details, see the [Install from Source documentation](https://docs.flashinfer.ai/installation.html#install-from-source).
+
+### Install Nightly Build
+
+Nightly builds are available for testing the latest features:
+
+```bash
+# Core and cubin packages
+pip install -U --pre flashinfer-python --index-url https://flashinfer.ai/whl/nightly/ --no-deps # Install the nightly package from custom index, without installing dependencies
+pip install flashinfer-python  # Install flashinfer-python's dependencies from PyPI
+pip install -U --pre flashinfer-cubin --index-url https://flashinfer.ai/whl/nightly/
+# JIT cache package (replace cu129 with your CUDA version: cu128, cu129, or cu130)
+pip install -U --pre flashinfer-jit-cache --index-url https://flashinfer.ai/whl/nightly/cu129
+```
+
+### Verify Installation
+
+After installation, verify that FlashInfer is correctly installed and configured:
+
+```bash
+flashinfer show-config
+```
+
+This command displays:
+
+- FlashInfer version and installed packages (flashinfer-python, flashinfer-cubin, flashinfer-jit-cache)
+- PyTorch and CUDA version information
+- Environment variables and artifact paths
+- Downloaded cubin status and module compilation status
+
+### Trying it out
+
+Below is a minimal example of using FlashInfer's single-request decode/append/prefill attention kernels:
+
+```python
+import torch
+import flashinfer
+
+kv_len = 2048
+num_kv_heads = 32
+head_dim = 128
+
+k = torch.randn(kv_len, num_kv_heads, head_dim).half().to(0)
+v = torch.randn(kv_len, num_kv_heads, head_dim).half().to(0)
+
+# decode attention
+
+num_qo_heads = 32
+q = torch.randn(num_qo_heads, head_dim).half().to(0)
+
+o = flashinfer.single_decode_with_kv_cache(q, k, v) # decode attention without RoPE on-the-fly
+o_rope_on_the_fly = flashinfer.single_decode_with_kv_cache(q, k, v, pos_encoding_mode="ROPE_LLAMA") # decode with LLaMA style RoPE on-the-fly
+
+# append attention
+append_qo_len = 128
+q = torch.randn(append_qo_len, num_qo_heads, head_dim).half().to(0) # append attention, the last 128 tokens in the KV-Cache are the new tokens
+o = flashinfer.single_prefill_with_kv_cache(q, k, v, causal=True) # append attention without RoPE on-the-fly, apply causal mask
+o_rope_on_the_fly = flashinfer.single_prefill_with_kv_cache(q, k, v, causal=True, pos_encoding_mode="ROPE_LLAMA") # append attention with LLaMA style RoPE on-the-fly, apply causal mask
+
+# prefill attention
+qo_len = 2048
+q = torch.randn(qo_len, num_qo_heads, head_dim).half().to(0) # prefill attention
+o = flashinfer.single_prefill_with_kv_cache(q, k, v, causal=False) # prefill attention without RoPE on-the-fly, do not apply causal mask
+```
+
+Check out [documentation](https://docs.flashinfer.ai/) for usage of batch decode/append/prefill kernels and shared-prefix cascading kernels.
+
+## Custom Attention Variants
+
+Starting from FlashInfer v0.2, users can customize their own attention variants with additional parameters. For more details, refer to our [JIT examples](https://github.com/flashinfer-ai/flashinfer/blob/main/tests/utils/test_jit_example.py).
+
+## GPU Support
+
+FlashInfer currently provides support for NVIDIA SM architectures 75 and higher and beta support for 103, 110, 120, and 121.
+
+## Adoption
+
+We are thrilled to share that FlashInfer is being adopted by many cutting-edge projects, including but not limited to:
+
+- [MLC-LLM](https://github.com/mlc-ai/mlc-llm)
+- [Punica](https://github.com/punica-ai/punica)
+- [SGLang](https://github.com/sgl-project/sglang)
+- [ScaleLLM](https://github.com/vectorch-ai/ScaleLLM)
+- [vLLM](https://github.com/vllm-project/vllm)
+- [TGI](https://github.com/huggingface/text-generation-inference)
+- [lorax](https://github.com/predibase/lorax)
+- [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM)
+- [LightLLM](https://github.com/ModelTC/lightllm)
+
+## Acknowledgement
+
+FlashInfer is inspired by [FlashAttention 1&2](https://github.com/dao-AILab/flash-attention/), [vLLM](https://github.com/vllm-project/vllm), [stream-K](https://arxiv.org/abs/2301.03598), [cutlass](https://github.com/nvidia/cutlass) and [AITemplate](https://github.com/facebookincubator/AITemplate) projects.
+
+## Citation
+
+If you find FlashInfer helpful in your project or research, please consider citing our [paper](https://arxiv.org/abs/2501.01005):
+
+```bibtex
+@article{ye2025flashinfer,
+    title = {FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving},
+    author = {
+      Ye, Zihao and
+      Chen, Lequn and
+      Lai, Ruihang and
+      Lin, Wuwei and
+      Zhang, Yineng and
+      Wang, Stephanie and
+      Chen, Tianqi and
+      Kasikci, Baris and
+      Grover, Vinod and
+      Krishnamurthy, Arvind and
+      Ceze, Luis
+    },
+    journal = {arXiv preprint arXiv:2501.01005},
+    year = {2025},
+    url = {https://arxiv.org/abs/2501.01005}
+}
+```

flashinfer_python-0.5.3/csrc/batch_mla_binding.cu

@@ -0,0 +1,38 @@
+/*
+ * Copyright (c) 2025 by FlashInfer team.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "batch_mla_config.inc"
+#include "tvm/ffi/container/array.h"
+#include "tvm_ffi_utils.h"
+
+using tvm::ffi::Array;
+using tvm::ffi::Optional;
+
+Array<int64_t> BatchMLAPagedAttentionPlan(TensorView float_workspace_buffer,
+                                          TensorView int_workspace_buffer,
+                                          TensorView page_locked_int_workspace_buffer,
+                                          TensorView qo_indptr, TensorView kv_indptr,
+                                          TensorView kv_len, int64_t num_heads, int64_t head_dim_o,
+                                          bool causal);
+
+void BatchMLAPagedAttentionRun(TensorView float_workspace_buffer, TensorView int_workspace_buffer,
+                               Array<int64_t> plan_info_vec, TensorView q_nope, TensorView q_pe,
+                               TensorView ckv_cache, TensorView kpe_cache, TensorView kv_indices,
+                               TensorView o, Optional<TensorView> maybe_lse, int64_t mask_mode_code,
+                               int64_t num_heads, int64_t page_size, double sm_scale,
+                               bool return_lse_base_on_e);
+
+TVM_FFI_DLL_EXPORT_TYPED_FUNC(plan, BatchMLAPagedAttentionPlan);
+TVM_FFI_DLL_EXPORT_TYPED_FUNC(run, BatchMLAPagedAttentionRun);

flashinfer_python-0.5.3/csrc/batch_mla_run.cu

@@ -0,0 +1,124 @@
+/*
+ * Copyright (c) 2025 by FlashInfer team.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <cstdint>
+#include <flashinfer/attention/mla.cuh>
+#include <flashinfer/attention/scheduler.cuh>
+#include <flashinfer/fastdiv.cuh>
+
+#include "batch_mla_config.inc"
+#include "tvm/ffi/container/array.h"
+#include "tvm_ffi_utils.h"
+
+using namespace flashinfer;
+
+using tvm::ffi::Array;
+using tvm::ffi::Optional;
+
+void BatchMLAPagedAttentionRun(TensorView float_workspace_buffer, TensorView int_workspace_buffer,
+                               Array<int64_t> plan_info_vec, TensorView q_nope, TensorView q_pe,
+                               TensorView ckv_cache, TensorView kpe_cache, TensorView kv_indices,
+                               TensorView o, Optional<TensorView> maybe_lse, int64_t mask_mode_code,
+                               int64_t num_heads, int64_t page_size, double sm_scale,
+                               bool return_lse_base_on_e) {
+  // q_nope: [n, num_heads, head_dim_ckv]
+  // q_pe: [n, num_heads, head_dim_kpe]
+  // ckv_cache: [num_pages, page_size, head_dim_ckv]
+  // kpe_cache: [num_pages, page_size, head_dim_kpe]
+  MLAPlanInfo plan_info;
+  plan_info.FromVector(std::vector<int64_t>(plan_info_vec.begin(), plan_info_vec.end()));
+
+  void* float_buffer_ptr = float_workspace_buffer.data_ptr();
+  void* int_buffer_ptr = int_workspace_buffer.data_ptr();
+
+  const MaskMode mask_mode = static_cast<MaskMode>(mask_mode_code);
+
+  unsigned int q_nope_stride_n = q_nope.stride(0);
+  unsigned int q_nope_stride_h = q_nope.stride(1);
+  unsigned int q_pe_stride_n = q_pe.stride(0);
+  unsigned int q_pe_stride_h = q_pe.stride(1);
+  unsigned int ckv_stride_page = ckv_cache.stride(0);
+  unsigned int ckv_stride_n = ckv_cache.stride(1);
+  unsigned int kpe_stride_page = kpe_cache.stride(0);
+  unsigned int kpe_stride_n = kpe_cache.stride(1);
+  unsigned int o_stride_n = o.stride(0);
+  unsigned int o_stride_h = o.stride(1);
+
+  cudaSetDevice(q_nope.device().device_id);
+  const cudaStream_t stream = get_stream(q_nope.device());
+
+  DISPATCH_context(
+      DTypeQ, DTypeKV, DTypeO, IdType, MASK_MODE, HEAD_DIM_CKV, HEAD_DIM_KPE, Params, [&] {
+        Params params;
+
+        params.q_nope = static_cast<DTypeQ*>(q_nope.data_ptr());
+        params.q_pe = static_cast<DTypeQ*>(q_pe.data_ptr());
+        params.ckv = static_cast<DTypeKV*>(ckv_cache.data_ptr());
+        params.kpe = static_cast<DTypeKV*>(kpe_cache.data_ptr());
+
+        params.q_indptr = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.q_indptr_offset);
+        params.kv_indptr = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.kv_indptr_offset);
+        params.partial_indptr =
+            GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.partial_indptr_offset);
+        params.kv_indices = static_cast<IdType*>(kv_indices.data_ptr());
+        params.q_len = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.q_len_offset);
+        params.kv_len = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.kv_len_offset);
+        params.q_start = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.q_start_offset);
+        params.kv_start = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.kv_start_offset);
+        params.kv_end = GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.kv_end_offset);
+        params.work_indptr =
+            GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.work_indptr_offset);
+        params.merge_packed_offset_start = GetPtrFromBaseOffset<IdType>(
+            int_buffer_ptr, plan_info.merge_packed_offset_start_offset);
+        params.merge_packed_offset_end =
+            GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.merge_packed_offset_end_offset);
+        params.merge_partial_packed_offset_start = GetPtrFromBaseOffset<IdType>(
+            int_buffer_ptr, plan_info.merge_partial_packed_offset_start_offset);
+        params.merge_partial_packed_offset_end = GetPtrFromBaseOffset<IdType>(
+            int_buffer_ptr, plan_info.merge_partial_packed_offset_end_offset);
+        params.merge_partial_stride =
+            GetPtrFromBaseOffset<IdType>(int_buffer_ptr, plan_info.merge_partial_stride_offset);
+        params.final_o = static_cast<DTypeO*>(o.data_ptr());
+        params.final_lse =
+            maybe_lse.has_value() ? static_cast<float*>(maybe_lse.value().data_ptr()) : nullptr;
+        params.partial_o =
+            GetPtrFromBaseOffset<DTypeO>(float_buffer_ptr, plan_info.partial_o_offset);
+        params.partial_lse =
+            GetPtrFromBaseOffset<float>(float_buffer_ptr, plan_info.partial_lse_offset);
+
+        params.num_heads = uint_fastdiv(num_heads);
+        params.block_size = uint_fastdiv(page_size);
+
+        params.q_nope_stride_n = q_nope_stride_n;
+        params.q_nope_stride_h = q_nope_stride_h;
+        params.q_pe_stride_n = q_pe_stride_n;
+        params.q_pe_stride_h = q_pe_stride_h;
+        params.ckv_stride_page = ckv_stride_page;
+        params.ckv_stride_n = ckv_stride_n;
+        params.kpe_stride_page = kpe_stride_page;
+        params.kpe_stride_n = kpe_stride_n;
+        params.o_stride_n = o_stride_n;
+        params.o_stride_h = o_stride_h;
+
+        params.sm_scale = sm_scale;
+        params.return_lse_base_on_e = return_lse_base_on_e;
+
+        cudaError_t status = mla::BatchMLAPagedAttention<MASK_MODE, HEAD_DIM_CKV, HEAD_DIM_KPE>(
+            params, plan_info.num_blks_x, plan_info.num_blks_y, stream);
+
+        TVM_FFI_ICHECK(status == cudaSuccess)
+            << "Failed to run MLA, error: " << cudaGetErrorString(status);
+      });
+}

flashinfer_python-0.5.3/csrc/batch_mla_sm90_binding.cu

@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) 2025 by FlashInfer team.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "batch_mla_sm90_config.inc"
+#include "tvm/ffi/container/array.h"
+#include "tvm_ffi_utils.h"
+
+using tvm::ffi::Array;
+using tvm::ffi::Optional;
+
+Array<int64_t> BatchMLAPagedAttentionSM90Plan(TensorView float_workspace_buffer,
+                                              TensorView int_workspace_buffer,
+                                              TensorView page_locked_int_workspace_buffer,
+                                              TensorView qo_indptr, TensorView kv_indptr,
+                                              TensorView kv_len, int64_t num_heads,
+                                              int64_t head_dim_o, bool causal);
+
+void BatchMLAPagedAttentionSM90Run(TensorView float_workspace_buffer,
+                                   TensorView int_workspace_buffer, Array<int64_t> plan_info_vec,
+                                   TensorView q_nope, TensorView q_pe, TensorView ckv_cache,
+                                   TensorView kpe_cache, TensorView kv_indices, TensorView o,
+                                   Optional<TensorView> maybe_lse, int64_t mask_mode_code,
+                                   int64_t num_heads, int64_t page_size, double sm_scale,
+                                   bool return_lse_base_on_e ADDITIONAL_FUNC_PARAMS);
+
+TVM_FFI_DLL_EXPORT_TYPED_FUNC(plan, BatchMLAPagedAttentionSM90Plan);
+TVM_FFI_DLL_EXPORT_TYPED_FUNC(run, BatchMLAPagedAttentionSM90Run);